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 | ||
0908ad6e AM |
176 | /** |
177 | * is_trap_insn - check if instruction is breakpoint instruction. | |
178 | * @insn: instruction to be checked. | |
179 | * Default implementation of is_trap_insn | |
180 | * Returns true if @insn is a breakpoint instruction. | |
181 | * | |
182 | * This function is needed for the case where an architecture has multiple | |
183 | * trap instructions (like powerpc). | |
184 | */ | |
185 | bool __weak is_trap_insn(uprobe_opcode_t *insn) | |
186 | { | |
187 | return is_swbp_insn(insn); | |
188 | } | |
189 | ||
ab0d805c | 190 | static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len) |
cceb55aa ON |
191 | { |
192 | void *kaddr = kmap_atomic(page); | |
ab0d805c | 193 | memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len); |
cceb55aa ON |
194 | kunmap_atomic(kaddr); |
195 | } | |
196 | ||
5669ccee ON |
197 | static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len) |
198 | { | |
199 | void *kaddr = kmap_atomic(page); | |
200 | memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len); | |
201 | kunmap_atomic(kaddr); | |
202 | } | |
203 | ||
ed6f6a50 ON |
204 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
205 | { | |
206 | uprobe_opcode_t old_opcode; | |
207 | bool is_swbp; | |
208 | ||
0908ad6e AM |
209 | /* |
210 | * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here. | |
211 | * We do not check if it is any other 'trap variant' which could | |
212 | * be conditional trap instruction such as the one powerpc supports. | |
213 | * | |
214 | * The logic is that we do not care if the underlying instruction | |
215 | * is a trap variant; uprobes always wins over any other (gdb) | |
216 | * breakpoint. | |
217 | */ | |
ab0d805c | 218 | copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE); |
ed6f6a50 ON |
219 | is_swbp = is_swbp_insn(&old_opcode); |
220 | ||
221 | if (is_swbp_insn(new_opcode)) { | |
222 | if (is_swbp) /* register: already installed? */ | |
223 | return 0; | |
224 | } else { | |
225 | if (!is_swbp) /* unregister: was it changed by us? */ | |
076a365b | 226 | return 0; |
ed6f6a50 ON |
227 | } |
228 | ||
229 | return 1; | |
230 | } | |
231 | ||
2b144498 SD |
232 | /* |
233 | * NOTE: | |
234 | * Expect the breakpoint instruction to be the smallest size instruction for | |
235 | * the architecture. If an arch has variable length instruction and the | |
236 | * breakpoint instruction is not of the smallest length instruction | |
0908ad6e | 237 | * supported by that architecture then we need to modify is_trap_at_addr and |
2b144498 SD |
238 | * write_opcode accordingly. This would never be a problem for archs that |
239 | * have fixed length instructions. | |
240 | */ | |
241 | ||
242 | /* | |
243 | * write_opcode - write the opcode at a given virtual address. | |
244 | * @mm: the probed process address space. | |
2b144498 SD |
245 | * @vaddr: the virtual address to store the opcode. |
246 | * @opcode: opcode to be written at @vaddr. | |
247 | * | |
248 | * Called with mm->mmap_sem held (for read and with a reference to | |
249 | * mm). | |
250 | * | |
251 | * For mm @mm, write the opcode at @vaddr. | |
252 | * Return 0 (success) or a negative errno. | |
253 | */ | |
cceb55aa ON |
254 | static int write_opcode(struct mm_struct *mm, unsigned long vaddr, |
255 | uprobe_opcode_t opcode) | |
2b144498 SD |
256 | { |
257 | struct page *old_page, *new_page; | |
2b144498 SD |
258 | void *vaddr_old, *vaddr_new; |
259 | struct vm_area_struct *vma; | |
2b144498 | 260 | int ret; |
f403072c | 261 | |
5323ce71 | 262 | retry: |
2b144498 | 263 | /* Read the page with vaddr into memory */ |
75ed82ea | 264 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
265 | if (ret <= 0) |
266 | return ret; | |
7b2d81d4 | 267 | |
ed6f6a50 ON |
268 | ret = verify_opcode(old_page, vaddr, &opcode); |
269 | if (ret <= 0) | |
270 | goto put_old; | |
271 | ||
2b144498 SD |
272 | ret = -ENOMEM; |
273 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
274 | if (!new_page) | |
9f92448c | 275 | goto put_old; |
2b144498 SD |
276 | |
277 | __SetPageUptodate(new_page); | |
278 | ||
2b144498 SD |
279 | /* copy the page now that we've got it stable */ |
280 | vaddr_old = kmap_atomic(old_page); | |
281 | vaddr_new = kmap_atomic(new_page); | |
282 | ||
283 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
d9c4a30e | 284 | memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
285 | |
286 | kunmap_atomic(vaddr_new); | |
287 | kunmap_atomic(vaddr_old); | |
288 | ||
289 | ret = anon_vma_prepare(vma); | |
290 | if (ret) | |
9f92448c | 291 | goto put_new; |
2b144498 | 292 | |
c517ee74 | 293 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 294 | |
9f92448c | 295 | put_new: |
2b144498 | 296 | page_cache_release(new_page); |
9f92448c | 297 | put_old: |
7b2d81d4 IM |
298 | put_page(old_page); |
299 | ||
5323ce71 ON |
300 | if (unlikely(ret == -EAGAIN)) |
301 | goto retry; | |
2b144498 SD |
302 | return ret; |
303 | } | |
304 | ||
2b144498 | 305 | /** |
5cb4ac3a | 306 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 307 | * @auprobe: arch specific probepoint information. |
2b144498 | 308 | * @mm: the probed process address space. |
2b144498 SD |
309 | * @vaddr: the virtual address to insert the opcode. |
310 | * | |
311 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
312 | * Return 0 (success) or a negative errno. | |
313 | */ | |
5cb4ac3a | 314 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 315 | { |
cceb55aa | 316 | return write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
317 | } |
318 | ||
319 | /** | |
320 | * set_orig_insn - Restore the original instruction. | |
321 | * @mm: the probed process address space. | |
e3343e6a | 322 | * @auprobe: arch specific probepoint information. |
2b144498 | 323 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
324 | * |
325 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
326 | * Return 0 (success) or a negative errno. | |
327 | */ | |
7b2d81d4 | 328 | int __weak |
ded86e7c | 329 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 330 | { |
cceb55aa | 331 | return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
332 | } |
333 | ||
334 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
335 | { | |
336 | if (l->inode < r->inode) | |
337 | return -1; | |
7b2d81d4 | 338 | |
2b144498 SD |
339 | if (l->inode > r->inode) |
340 | return 1; | |
2b144498 | 341 | |
7b2d81d4 IM |
342 | if (l->offset < r->offset) |
343 | return -1; | |
344 | ||
345 | if (l->offset > r->offset) | |
346 | return 1; | |
2b144498 SD |
347 | |
348 | return 0; | |
349 | } | |
350 | ||
351 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
352 | { | |
353 | struct uprobe u = { .inode = inode, .offset = offset }; | |
354 | struct rb_node *n = uprobes_tree.rb_node; | |
355 | struct uprobe *uprobe; | |
356 | int match; | |
357 | ||
358 | while (n) { | |
359 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
360 | match = match_uprobe(&u, uprobe); | |
361 | if (!match) { | |
362 | atomic_inc(&uprobe->ref); | |
363 | return uprobe; | |
364 | } | |
7b2d81d4 | 365 | |
2b144498 SD |
366 | if (match < 0) |
367 | n = n->rb_left; | |
368 | else | |
369 | n = n->rb_right; | |
370 | } | |
371 | return NULL; | |
372 | } | |
373 | ||
374 | /* | |
375 | * Find a uprobe corresponding to a given inode:offset | |
376 | * Acquires uprobes_treelock | |
377 | */ | |
378 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
379 | { | |
380 | struct uprobe *uprobe; | |
2b144498 | 381 | |
6f47caa0 | 382 | spin_lock(&uprobes_treelock); |
2b144498 | 383 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 384 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 385 | |
2b144498 SD |
386 | return uprobe; |
387 | } | |
388 | ||
389 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
390 | { | |
391 | struct rb_node **p = &uprobes_tree.rb_node; | |
392 | struct rb_node *parent = NULL; | |
393 | struct uprobe *u; | |
394 | int match; | |
395 | ||
396 | while (*p) { | |
397 | parent = *p; | |
398 | u = rb_entry(parent, struct uprobe, rb_node); | |
399 | match = match_uprobe(uprobe, u); | |
400 | if (!match) { | |
401 | atomic_inc(&u->ref); | |
402 | return u; | |
403 | } | |
404 | ||
405 | if (match < 0) | |
406 | p = &parent->rb_left; | |
407 | else | |
408 | p = &parent->rb_right; | |
409 | ||
410 | } | |
7b2d81d4 | 411 | |
2b144498 SD |
412 | u = NULL; |
413 | rb_link_node(&uprobe->rb_node, parent, p); | |
414 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
415 | /* get access + creation ref */ | |
416 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 417 | |
2b144498 SD |
418 | return u; |
419 | } | |
420 | ||
421 | /* | |
7b2d81d4 | 422 | * Acquire uprobes_treelock. |
2b144498 SD |
423 | * Matching uprobe already exists in rbtree; |
424 | * increment (access refcount) and return the matching uprobe. | |
425 | * | |
426 | * No matching uprobe; insert the uprobe in rb_tree; | |
427 | * get a double refcount (access + creation) and return NULL. | |
428 | */ | |
429 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
430 | { | |
2b144498 SD |
431 | struct uprobe *u; |
432 | ||
6f47caa0 | 433 | spin_lock(&uprobes_treelock); |
2b144498 | 434 | u = __insert_uprobe(uprobe); |
6f47caa0 | 435 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 436 | |
2b144498 SD |
437 | return u; |
438 | } | |
439 | ||
440 | static void put_uprobe(struct uprobe *uprobe) | |
441 | { | |
442 | if (atomic_dec_and_test(&uprobe->ref)) | |
443 | kfree(uprobe); | |
444 | } | |
445 | ||
446 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
447 | { | |
448 | struct uprobe *uprobe, *cur_uprobe; | |
449 | ||
450 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
451 | if (!uprobe) | |
452 | return NULL; | |
453 | ||
454 | uprobe->inode = igrab(inode); | |
455 | uprobe->offset = offset; | |
e591c8d7 | 456 | init_rwsem(&uprobe->register_rwsem); |
2b144498 | 457 | init_rwsem(&uprobe->consumer_rwsem); |
bbc33d05 ON |
458 | /* For now assume that the instruction need not be single-stepped */ |
459 | __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); | |
2b144498 SD |
460 | |
461 | /* add to uprobes_tree, sorted on inode:offset */ | |
462 | cur_uprobe = insert_uprobe(uprobe); | |
463 | ||
464 | /* a uprobe exists for this inode:offset combination */ | |
465 | if (cur_uprobe) { | |
466 | kfree(uprobe); | |
467 | uprobe = cur_uprobe; | |
468 | iput(inode); | |
7b2d81d4 IM |
469 | } |
470 | ||
2b144498 SD |
471 | return uprobe; |
472 | } | |
473 | ||
9a98e03c | 474 | static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
475 | { |
476 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
477 | uc->next = uprobe->consumers; |
478 | uprobe->consumers = uc; | |
2b144498 | 479 | up_write(&uprobe->consumer_rwsem); |
2b144498 SD |
480 | } |
481 | ||
482 | /* | |
e3343e6a SD |
483 | * For uprobe @uprobe, delete the consumer @uc. |
484 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
485 | * or return false. |
486 | */ | |
e3343e6a | 487 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
488 | { |
489 | struct uprobe_consumer **con; | |
490 | bool ret = false; | |
491 | ||
492 | down_write(&uprobe->consumer_rwsem); | |
493 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
494 | if (*con == uc) { |
495 | *con = uc->next; | |
2b144498 SD |
496 | ret = true; |
497 | break; | |
498 | } | |
499 | } | |
500 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 501 | |
2b144498 SD |
502 | return ret; |
503 | } | |
504 | ||
e3343e6a | 505 | static int |
d436615e | 506 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
593609a5 | 507 | unsigned long nbytes, loff_t offset) |
2b144498 | 508 | { |
2b144498 | 509 | struct page *page; |
2b144498 | 510 | |
cc359d18 ON |
511 | if (!mapping->a_ops->readpage) |
512 | return -EIO; | |
2b144498 SD |
513 | /* |
514 | * Ensure that the page that has the original instruction is | |
515 | * populated and in page-cache. | |
516 | */ | |
2edb7b55 | 517 | page = read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT, filp); |
2b144498 SD |
518 | if (IS_ERR(page)) |
519 | return PTR_ERR(page); | |
520 | ||
2edb7b55 | 521 | copy_from_page(page, offset, insn, nbytes); |
2b144498 | 522 | page_cache_release(page); |
7b2d81d4 | 523 | |
2b144498 SD |
524 | return 0; |
525 | } | |
526 | ||
d436615e | 527 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
528 | { |
529 | struct address_space *mapping; | |
2b144498 | 530 | unsigned long nbytes; |
7b2d81d4 | 531 | int bytes; |
2b144498 | 532 | |
d436615e | 533 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
534 | mapping = uprobe->inode->i_mapping; |
535 | ||
536 | /* Instruction at end of binary; copy only available bytes */ | |
537 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
538 | bytes = uprobe->inode->i_size - uprobe->offset; | |
539 | else | |
540 | bytes = MAX_UINSN_BYTES; | |
541 | ||
542 | /* Instruction at the page-boundary; copy bytes in second page */ | |
543 | if (nbytes < bytes) { | |
fc36f595 ON |
544 | int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
545 | bytes - nbytes, uprobe->offset + nbytes); | |
546 | if (err) | |
547 | return err; | |
2b144498 SD |
548 | bytes = nbytes; |
549 | } | |
d436615e | 550 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
551 | } |
552 | ||
cb9a19fe ON |
553 | static int prepare_uprobe(struct uprobe *uprobe, struct file *file, |
554 | struct mm_struct *mm, unsigned long vaddr) | |
555 | { | |
556 | int ret = 0; | |
557 | ||
71434f2f | 558 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
cb9a19fe ON |
559 | return ret; |
560 | ||
d4d3ccc6 ON |
561 | /* TODO: move this into _register, until then we abuse this sem. */ |
562 | down_write(&uprobe->consumer_rwsem); | |
71434f2f | 563 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
564 | goto out; |
565 | ||
cb9a19fe ON |
566 | ret = copy_insn(uprobe, file); |
567 | if (ret) | |
568 | goto out; | |
569 | ||
570 | ret = -ENOTSUPP; | |
0908ad6e | 571 | if (is_trap_insn((uprobe_opcode_t *)uprobe->arch.insn)) |
cb9a19fe ON |
572 | goto out; |
573 | ||
574 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
575 | if (ret) | |
576 | goto out; | |
577 | ||
578 | /* write_opcode() assumes we don't cross page boundary */ | |
579 | BUG_ON((uprobe->offset & ~PAGE_MASK) + | |
580 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
581 | ||
582 | smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ | |
71434f2f | 583 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
584 | |
585 | out: | |
d4d3ccc6 | 586 | up_write(&uprobe->consumer_rwsem); |
4710f05f | 587 | |
cb9a19fe ON |
588 | return ret; |
589 | } | |
590 | ||
8a7f2fa0 ON |
591 | static inline bool consumer_filter(struct uprobe_consumer *uc, |
592 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
806a98bd | 593 | { |
8a7f2fa0 | 594 | return !uc->filter || uc->filter(uc, ctx, mm); |
806a98bd ON |
595 | } |
596 | ||
8a7f2fa0 ON |
597 | static bool filter_chain(struct uprobe *uprobe, |
598 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
63633cbf | 599 | { |
1ff6fee5 ON |
600 | struct uprobe_consumer *uc; |
601 | bool ret = false; | |
602 | ||
603 | down_read(&uprobe->consumer_rwsem); | |
604 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
8a7f2fa0 | 605 | ret = consumer_filter(uc, ctx, mm); |
1ff6fee5 ON |
606 | if (ret) |
607 | break; | |
608 | } | |
609 | up_read(&uprobe->consumer_rwsem); | |
610 | ||
611 | return ret; | |
63633cbf ON |
612 | } |
613 | ||
e3343e6a SD |
614 | static int |
615 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 616 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 617 | { |
f8ac4ec9 | 618 | bool first_uprobe; |
2b144498 SD |
619 | int ret; |
620 | ||
cb9a19fe ON |
621 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
622 | if (ret) | |
623 | return ret; | |
682968e0 | 624 | |
f8ac4ec9 ON |
625 | /* |
626 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
627 | * the task can hit this breakpoint right after __replace_page(). | |
628 | */ | |
629 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
630 | if (first_uprobe) | |
631 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
632 | ||
816c03fb | 633 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
634 | if (!ret) |
635 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
636 | else if (first_uprobe) | |
f8ac4ec9 | 637 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
638 | |
639 | return ret; | |
640 | } | |
641 | ||
076a365b | 642 | static int |
816c03fb | 643 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 644 | { |
9f68f672 | 645 | set_bit(MMF_RECALC_UPROBES, &mm->flags); |
076a365b | 646 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
647 | } |
648 | ||
06b7bcd8 ON |
649 | static inline bool uprobe_is_active(struct uprobe *uprobe) |
650 | { | |
651 | return !RB_EMPTY_NODE(&uprobe->rb_node); | |
652 | } | |
0326f5a9 | 653 | /* |
778b032d ON |
654 | * There could be threads that have already hit the breakpoint. They |
655 | * will recheck the current insn and restart if find_uprobe() fails. | |
656 | * See find_active_uprobe(). | |
0326f5a9 | 657 | */ |
2b144498 SD |
658 | static void delete_uprobe(struct uprobe *uprobe) |
659 | { | |
06b7bcd8 ON |
660 | if (WARN_ON(!uprobe_is_active(uprobe))) |
661 | return; | |
662 | ||
6f47caa0 | 663 | spin_lock(&uprobes_treelock); |
2b144498 | 664 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 665 | spin_unlock(&uprobes_treelock); |
06b7bcd8 | 666 | RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */ |
2b144498 SD |
667 | iput(uprobe->inode); |
668 | put_uprobe(uprobe); | |
2b144498 SD |
669 | } |
670 | ||
26872090 ON |
671 | struct map_info { |
672 | struct map_info *next; | |
673 | struct mm_struct *mm; | |
816c03fb | 674 | unsigned long vaddr; |
26872090 ON |
675 | }; |
676 | ||
677 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 678 | { |
26872090 ON |
679 | struct map_info *next = info->next; |
680 | kfree(info); | |
681 | return next; | |
682 | } | |
683 | ||
684 | static struct map_info * | |
685 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
686 | { | |
687 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 688 | struct vm_area_struct *vma; |
26872090 ON |
689 | struct map_info *curr = NULL; |
690 | struct map_info *prev = NULL; | |
691 | struct map_info *info; | |
692 | int more = 0; | |
2b144498 | 693 | |
26872090 ON |
694 | again: |
695 | mutex_lock(&mapping->i_mmap_mutex); | |
6b2dbba8 | 696 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
697 | if (!valid_vma(vma, is_register)) |
698 | continue; | |
699 | ||
7a5bfb66 ON |
700 | if (!prev && !more) { |
701 | /* | |
702 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
703 | * reclaim. This is optimistic, no harm done if it fails. | |
704 | */ | |
705 | prev = kmalloc(sizeof(struct map_info), | |
706 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
707 | if (prev) | |
708 | prev->next = NULL; | |
709 | } | |
26872090 ON |
710 | if (!prev) { |
711 | more++; | |
712 | continue; | |
2b144498 | 713 | } |
2b144498 | 714 | |
26872090 ON |
715 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
716 | continue; | |
7b2d81d4 | 717 | |
26872090 ON |
718 | info = prev; |
719 | prev = prev->next; | |
720 | info->next = curr; | |
721 | curr = info; | |
2b144498 | 722 | |
26872090 | 723 | info->mm = vma->vm_mm; |
57683f72 | 724 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 725 | } |
2b144498 SD |
726 | mutex_unlock(&mapping->i_mmap_mutex); |
727 | ||
26872090 ON |
728 | if (!more) |
729 | goto out; | |
730 | ||
731 | prev = curr; | |
732 | while (curr) { | |
733 | mmput(curr->mm); | |
734 | curr = curr->next; | |
735 | } | |
7b2d81d4 | 736 | |
26872090 ON |
737 | do { |
738 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
739 | if (!info) { | |
740 | curr = ERR_PTR(-ENOMEM); | |
741 | goto out; | |
742 | } | |
743 | info->next = prev; | |
744 | prev = info; | |
745 | } while (--more); | |
746 | ||
747 | goto again; | |
748 | out: | |
749 | while (prev) | |
750 | prev = free_map_info(prev); | |
751 | return curr; | |
2b144498 SD |
752 | } |
753 | ||
bdf8647c ON |
754 | static int |
755 | register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) | |
2b144498 | 756 | { |
bdf8647c | 757 | bool is_register = !!new; |
26872090 ON |
758 | struct map_info *info; |
759 | int err = 0; | |
2b144498 | 760 | |
32cdba1e | 761 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
762 | info = build_map_info(uprobe->inode->i_mapping, |
763 | uprobe->offset, is_register); | |
32cdba1e ON |
764 | if (IS_ERR(info)) { |
765 | err = PTR_ERR(info); | |
766 | goto out; | |
767 | } | |
7b2d81d4 | 768 | |
26872090 ON |
769 | while (info) { |
770 | struct mm_struct *mm = info->mm; | |
771 | struct vm_area_struct *vma; | |
7b2d81d4 | 772 | |
076a365b | 773 | if (err && is_register) |
26872090 | 774 | goto free; |
7b2d81d4 | 775 | |
77fc4af1 | 776 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
777 | vma = find_vma(mm, info->vaddr); |
778 | if (!vma || !valid_vma(vma, is_register) || | |
f281769e | 779 | file_inode(vma->vm_file) != uprobe->inode) |
26872090 ON |
780 | goto unlock; |
781 | ||
f4d6dfe5 ON |
782 | if (vma->vm_start > info->vaddr || |
783 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 784 | goto unlock; |
2b144498 | 785 | |
806a98bd ON |
786 | if (is_register) { |
787 | /* consult only the "caller", new consumer. */ | |
bdf8647c | 788 | if (consumer_filter(new, |
8a7f2fa0 | 789 | UPROBE_FILTER_REGISTER, mm)) |
806a98bd ON |
790 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
791 | } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) { | |
8a7f2fa0 ON |
792 | if (!filter_chain(uprobe, |
793 | UPROBE_FILTER_UNREGISTER, mm)) | |
806a98bd ON |
794 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
795 | } | |
78f74116 | 796 | |
26872090 ON |
797 | unlock: |
798 | up_write(&mm->mmap_sem); | |
799 | free: | |
800 | mmput(mm); | |
801 | info = free_map_info(info); | |
2b144498 | 802 | } |
32cdba1e ON |
803 | out: |
804 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 805 | return err; |
2b144498 SD |
806 | } |
807 | ||
9a98e03c | 808 | static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 809 | { |
9a98e03c | 810 | consumer_add(uprobe, uc); |
bdf8647c | 811 | return register_for_each_vma(uprobe, uc); |
2b144498 SD |
812 | } |
813 | ||
04aab9b2 | 814 | static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 815 | { |
04aab9b2 ON |
816 | int err; |
817 | ||
818 | if (!consumer_del(uprobe, uc)) /* WARN? */ | |
819 | return; | |
2b144498 | 820 | |
bdf8647c | 821 | err = register_for_each_vma(uprobe, NULL); |
bb929284 ON |
822 | /* TODO : cant unregister? schedule a worker thread */ |
823 | if (!uprobe->consumers && !err) | |
824 | delete_uprobe(uprobe); | |
2b144498 SD |
825 | } |
826 | ||
827 | /* | |
7b2d81d4 | 828 | * uprobe_register - register a probe |
2b144498 SD |
829 | * @inode: the file in which the probe has to be placed. |
830 | * @offset: offset from the start of the file. | |
e3343e6a | 831 | * @uc: information on howto handle the probe.. |
2b144498 | 832 | * |
7b2d81d4 | 833 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
834 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
835 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 836 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 837 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 838 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
839 | * unregisters. |
840 | * | |
841 | * Return errno if it cannot successully install probes | |
842 | * else return 0 (success) | |
843 | */ | |
e3343e6a | 844 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
845 | { |
846 | struct uprobe *uprobe; | |
7b2d81d4 | 847 | int ret; |
2b144498 | 848 | |
f0744af7 | 849 | /* Racy, just to catch the obvious mistakes */ |
2b144498 | 850 | if (offset > i_size_read(inode)) |
7b2d81d4 | 851 | return -EINVAL; |
2b144498 | 852 | |
66d06dff | 853 | retry: |
2b144498 | 854 | uprobe = alloc_uprobe(inode, offset); |
66d06dff ON |
855 | if (!uprobe) |
856 | return -ENOMEM; | |
857 | /* | |
858 | * We can race with uprobe_unregister()->delete_uprobe(). | |
859 | * Check uprobe_is_active() and retry if it is false. | |
860 | */ | |
861 | down_write(&uprobe->register_rwsem); | |
862 | ret = -EAGAIN; | |
863 | if (likely(uprobe_is_active(uprobe))) { | |
9a98e03c ON |
864 | ret = __uprobe_register(uprobe, uc); |
865 | if (ret) | |
04aab9b2 | 866 | __uprobe_unregister(uprobe, uc); |
2b144498 | 867 | } |
66d06dff ON |
868 | up_write(&uprobe->register_rwsem); |
869 | put_uprobe(uprobe); | |
2b144498 | 870 | |
66d06dff ON |
871 | if (unlikely(ret == -EAGAIN)) |
872 | goto retry; | |
2b144498 SD |
873 | return ret; |
874 | } | |
e8440c14 | 875 | EXPORT_SYMBOL_GPL(uprobe_register); |
2b144498 | 876 | |
bdf8647c ON |
877 | /* |
878 | * uprobe_apply - unregister a already registered probe. | |
879 | * @inode: the file in which the probe has to be removed. | |
880 | * @offset: offset from the start of the file. | |
881 | * @uc: consumer which wants to add more or remove some breakpoints | |
882 | * @add: add or remove the breakpoints | |
883 | */ | |
884 | int uprobe_apply(struct inode *inode, loff_t offset, | |
885 | struct uprobe_consumer *uc, bool add) | |
886 | { | |
887 | struct uprobe *uprobe; | |
888 | struct uprobe_consumer *con; | |
889 | int ret = -ENOENT; | |
890 | ||
891 | uprobe = find_uprobe(inode, offset); | |
892 | if (!uprobe) | |
893 | return ret; | |
894 | ||
895 | down_write(&uprobe->register_rwsem); | |
896 | for (con = uprobe->consumers; con && con != uc ; con = con->next) | |
897 | ; | |
898 | if (con) | |
899 | ret = register_for_each_vma(uprobe, add ? uc : NULL); | |
900 | up_write(&uprobe->register_rwsem); | |
901 | put_uprobe(uprobe); | |
902 | ||
903 | return ret; | |
904 | } | |
905 | ||
2b144498 | 906 | /* |
7b2d81d4 | 907 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
908 | * @inode: the file in which the probe has to be removed. |
909 | * @offset: offset from the start of the file. | |
e3343e6a | 910 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 911 | */ |
e3343e6a | 912 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 913 | { |
7b2d81d4 | 914 | struct uprobe *uprobe; |
2b144498 | 915 | |
2b144498 SD |
916 | uprobe = find_uprobe(inode, offset); |
917 | if (!uprobe) | |
918 | return; | |
919 | ||
e591c8d7 | 920 | down_write(&uprobe->register_rwsem); |
04aab9b2 | 921 | __uprobe_unregister(uprobe, uc); |
e591c8d7 | 922 | up_write(&uprobe->register_rwsem); |
c91368c4 | 923 | put_uprobe(uprobe); |
2b144498 | 924 | } |
e8440c14 | 925 | EXPORT_SYMBOL_GPL(uprobe_unregister); |
2b144498 | 926 | |
da1816b1 ON |
927 | static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm) |
928 | { | |
929 | struct vm_area_struct *vma; | |
930 | int err = 0; | |
931 | ||
932 | down_read(&mm->mmap_sem); | |
933 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
934 | unsigned long vaddr; | |
935 | loff_t offset; | |
936 | ||
937 | if (!valid_vma(vma, false) || | |
f281769e | 938 | file_inode(vma->vm_file) != uprobe->inode) |
da1816b1 ON |
939 | continue; |
940 | ||
941 | offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT; | |
942 | if (uprobe->offset < offset || | |
943 | uprobe->offset >= offset + vma->vm_end - vma->vm_start) | |
944 | continue; | |
945 | ||
946 | vaddr = offset_to_vaddr(vma, uprobe->offset); | |
947 | err |= remove_breakpoint(uprobe, mm, vaddr); | |
948 | } | |
949 | up_read(&mm->mmap_sem); | |
950 | ||
951 | return err; | |
952 | } | |
953 | ||
891c3970 ON |
954 | static struct rb_node * |
955 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 956 | { |
2b144498 | 957 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
958 | |
959 | while (n) { | |
891c3970 | 960 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 961 | |
891c3970 | 962 | if (inode < u->inode) { |
2b144498 | 963 | n = n->rb_left; |
891c3970 | 964 | } else if (inode > u->inode) { |
2b144498 | 965 | n = n->rb_right; |
891c3970 ON |
966 | } else { |
967 | if (max < u->offset) | |
968 | n = n->rb_left; | |
969 | else if (min > u->offset) | |
970 | n = n->rb_right; | |
971 | else | |
972 | break; | |
973 | } | |
2b144498 | 974 | } |
7b2d81d4 | 975 | |
891c3970 | 976 | return n; |
2b144498 SD |
977 | } |
978 | ||
979 | /* | |
891c3970 | 980 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 981 | */ |
891c3970 ON |
982 | static void build_probe_list(struct inode *inode, |
983 | struct vm_area_struct *vma, | |
984 | unsigned long start, unsigned long end, | |
985 | struct list_head *head) | |
2b144498 | 986 | { |
891c3970 | 987 | loff_t min, max; |
891c3970 ON |
988 | struct rb_node *n, *t; |
989 | struct uprobe *u; | |
7b2d81d4 | 990 | |
891c3970 | 991 | INIT_LIST_HEAD(head); |
cb113b47 | 992 | min = vaddr_to_offset(vma, start); |
891c3970 | 993 | max = min + (end - start) - 1; |
2b144498 | 994 | |
6f47caa0 | 995 | spin_lock(&uprobes_treelock); |
891c3970 ON |
996 | n = find_node_in_range(inode, min, max); |
997 | if (n) { | |
998 | for (t = n; t; t = rb_prev(t)) { | |
999 | u = rb_entry(t, struct uprobe, rb_node); | |
1000 | if (u->inode != inode || u->offset < min) | |
1001 | break; | |
1002 | list_add(&u->pending_list, head); | |
1003 | atomic_inc(&u->ref); | |
1004 | } | |
1005 | for (t = n; (t = rb_next(t)); ) { | |
1006 | u = rb_entry(t, struct uprobe, rb_node); | |
1007 | if (u->inode != inode || u->offset > max) | |
1008 | break; | |
1009 | list_add(&u->pending_list, head); | |
1010 | atomic_inc(&u->ref); | |
1011 | } | |
2b144498 | 1012 | } |
6f47caa0 | 1013 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
1014 | } |
1015 | ||
1016 | /* | |
5e5be71a | 1017 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 1018 | * |
5e5be71a ON |
1019 | * Currently we ignore all errors and always return 0, the callers |
1020 | * can't handle the failure anyway. | |
2b144498 | 1021 | */ |
7b2d81d4 | 1022 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1023 | { |
1024 | struct list_head tmp_list; | |
665605a2 | 1025 | struct uprobe *uprobe, *u; |
2b144498 | 1026 | struct inode *inode; |
2b144498 | 1027 | |
441f1eb7 | 1028 | if (no_uprobe_events() || !valid_vma(vma, true)) |
7b2d81d4 | 1029 | return 0; |
2b144498 | 1030 | |
f281769e | 1031 | inode = file_inode(vma->vm_file); |
2b144498 | 1032 | if (!inode) |
7b2d81d4 | 1033 | return 0; |
2b144498 | 1034 | |
2b144498 | 1035 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1036 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
806a98bd ON |
1037 | /* |
1038 | * We can race with uprobe_unregister(), this uprobe can be already | |
1039 | * removed. But in this case filter_chain() must return false, all | |
1040 | * consumers have gone away. | |
1041 | */ | |
665605a2 | 1042 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
806a98bd | 1043 | if (!fatal_signal_pending(current) && |
8a7f2fa0 | 1044 | filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) { |
57683f72 | 1045 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 1046 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
1047 | } |
1048 | put_uprobe(uprobe); | |
1049 | } | |
2b144498 SD |
1050 | mutex_unlock(uprobes_mmap_hash(inode)); |
1051 | ||
5e5be71a | 1052 | return 0; |
2b144498 SD |
1053 | } |
1054 | ||
9f68f672 ON |
1055 | static bool |
1056 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1057 | { | |
1058 | loff_t min, max; | |
1059 | struct inode *inode; | |
1060 | struct rb_node *n; | |
1061 | ||
f281769e | 1062 | inode = file_inode(vma->vm_file); |
9f68f672 ON |
1063 | |
1064 | min = vaddr_to_offset(vma, start); | |
1065 | max = min + (end - start) - 1; | |
1066 | ||
1067 | spin_lock(&uprobes_treelock); | |
1068 | n = find_node_in_range(inode, min, max); | |
1069 | spin_unlock(&uprobes_treelock); | |
1070 | ||
1071 | return !!n; | |
1072 | } | |
1073 | ||
682968e0 SD |
1074 | /* |
1075 | * Called in context of a munmap of a vma. | |
1076 | */ | |
cbc91f71 | 1077 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1078 | { |
441f1eb7 | 1079 | if (no_uprobe_events() || !valid_vma(vma, false)) |
682968e0 SD |
1080 | return; |
1081 | ||
2fd611a9 ON |
1082 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1083 | return; | |
1084 | ||
9f68f672 ON |
1085 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1086 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1087 | return; |
1088 | ||
9f68f672 ON |
1089 | if (vma_has_uprobes(vma, start, end)) |
1090 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1091 | } |
1092 | ||
d4b3b638 SD |
1093 | /* Slot allocation for XOL */ |
1094 | static int xol_add_vma(struct xol_area *area) | |
1095 | { | |
c8a82538 ON |
1096 | struct mm_struct *mm = current->mm; |
1097 | int ret = -EALREADY; | |
d4b3b638 SD |
1098 | |
1099 | down_write(&mm->mmap_sem); | |
1100 | if (mm->uprobes_state.xol_area) | |
1101 | goto fail; | |
1102 | ||
1103 | ret = -ENOMEM; | |
d4b3b638 SD |
1104 | /* Try to map as high as possible, this is only a hint. */ |
1105 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1106 | if (area->vaddr & ~PAGE_MASK) { | |
1107 | ret = area->vaddr; | |
1108 | goto fail; | |
1109 | } | |
1110 | ||
1111 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1112 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1113 | if (ret) | |
1114 | goto fail; | |
1115 | ||
1116 | smp_wmb(); /* pairs with get_xol_area() */ | |
1117 | mm->uprobes_state.xol_area = area; | |
1118 | ret = 0; | |
c8a82538 | 1119 | fail: |
d4b3b638 | 1120 | up_write(&mm->mmap_sem); |
d4b3b638 SD |
1121 | |
1122 | return ret; | |
1123 | } | |
1124 | ||
d4b3b638 | 1125 | /* |
9b545df8 ON |
1126 | * get_xol_area - Allocate process's xol_area if necessary. |
1127 | * This area will be used for storing instructions for execution out of line. | |
d4b3b638 SD |
1128 | * |
1129 | * Returns the allocated area or NULL. | |
1130 | */ | |
9b545df8 | 1131 | static struct xol_area *get_xol_area(void) |
d4b3b638 | 1132 | { |
9b545df8 | 1133 | struct mm_struct *mm = current->mm; |
d4b3b638 SD |
1134 | struct xol_area *area; |
1135 | ||
9b545df8 ON |
1136 | area = mm->uprobes_state.xol_area; |
1137 | if (area) | |
1138 | goto ret; | |
1139 | ||
d4b3b638 SD |
1140 | area = kzalloc(sizeof(*area), GFP_KERNEL); |
1141 | if (unlikely(!area)) | |
c8a82538 | 1142 | goto out; |
d4b3b638 SD |
1143 | |
1144 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
d4b3b638 | 1145 | if (!area->bitmap) |
c8a82538 ON |
1146 | goto free_area; |
1147 | ||
1148 | area->page = alloc_page(GFP_HIGHUSER); | |
1149 | if (!area->page) | |
1150 | goto free_bitmap; | |
d4b3b638 SD |
1151 | |
1152 | init_waitqueue_head(&area->wq); | |
1153 | if (!xol_add_vma(area)) | |
1154 | return area; | |
1155 | ||
c8a82538 ON |
1156 | __free_page(area->page); |
1157 | free_bitmap: | |
d4b3b638 | 1158 | kfree(area->bitmap); |
c8a82538 | 1159 | free_area: |
d4b3b638 | 1160 | kfree(area); |
c8a82538 | 1161 | out: |
9b545df8 ON |
1162 | area = mm->uprobes_state.xol_area; |
1163 | ret: | |
1164 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1165 | return area; | |
d4b3b638 SD |
1166 | } |
1167 | ||
1168 | /* | |
1169 | * uprobe_clear_state - Free the area allocated for slots. | |
1170 | */ | |
1171 | void uprobe_clear_state(struct mm_struct *mm) | |
1172 | { | |
1173 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1174 | ||
1175 | if (!area) | |
1176 | return; | |
1177 | ||
1178 | put_page(area->page); | |
1179 | kfree(area->bitmap); | |
1180 | kfree(area); | |
1181 | } | |
1182 | ||
32cdba1e ON |
1183 | void uprobe_start_dup_mmap(void) |
1184 | { | |
1185 | percpu_down_read(&dup_mmap_sem); | |
1186 | } | |
1187 | ||
1188 | void uprobe_end_dup_mmap(void) | |
1189 | { | |
1190 | percpu_up_read(&dup_mmap_sem); | |
1191 | } | |
1192 | ||
f8ac4ec9 ON |
1193 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1194 | { | |
61559a81 ON |
1195 | newmm->uprobes_state.xol_area = NULL; |
1196 | ||
9f68f672 | 1197 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1198 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1199 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1200 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1201 | } | |
f8ac4ec9 ON |
1202 | } |
1203 | ||
d4b3b638 SD |
1204 | /* |
1205 | * - search for a free slot. | |
1206 | */ | |
1207 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1208 | { | |
1209 | unsigned long slot_addr; | |
1210 | int slot_nr; | |
1211 | ||
1212 | do { | |
1213 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1214 | if (slot_nr < UINSNS_PER_PAGE) { | |
1215 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1216 | break; | |
1217 | ||
1218 | slot_nr = UINSNS_PER_PAGE; | |
1219 | continue; | |
1220 | } | |
1221 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1222 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1223 | ||
1224 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1225 | atomic_inc(&area->slot_count); | |
1226 | ||
1227 | return slot_addr; | |
1228 | } | |
1229 | ||
1230 | /* | |
a6cb3f6d | 1231 | * xol_get_insn_slot - allocate a slot for xol. |
d4b3b638 SD |
1232 | * Returns the allocated slot address or 0. |
1233 | */ | |
a6cb3f6d | 1234 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe) |
d4b3b638 SD |
1235 | { |
1236 | struct xol_area *area; | |
a6cb3f6d | 1237 | unsigned long xol_vaddr; |
d4b3b638 | 1238 | |
9b545df8 ON |
1239 | area = get_xol_area(); |
1240 | if (!area) | |
1241 | return 0; | |
d4b3b638 | 1242 | |
a6cb3f6d ON |
1243 | xol_vaddr = xol_take_insn_slot(area); |
1244 | if (unlikely(!xol_vaddr)) | |
d4b3b638 SD |
1245 | return 0; |
1246 | ||
a6cb3f6d | 1247 | /* Initialize the slot */ |
5669ccee | 1248 | copy_to_page(area->page, xol_vaddr, uprobe->arch.insn, MAX_UINSN_BYTES); |
65b6ecc0 RV |
1249 | /* |
1250 | * We probably need flush_icache_user_range() but it needs vma. | |
1251 | * This should work on supported architectures too. | |
1252 | */ | |
1253 | flush_dcache_page(area->page); | |
d4b3b638 | 1254 | |
a6cb3f6d | 1255 | return xol_vaddr; |
d4b3b638 SD |
1256 | } |
1257 | ||
1258 | /* | |
1259 | * xol_free_insn_slot - If slot was earlier allocated by | |
1260 | * @xol_get_insn_slot(), make the slot available for | |
1261 | * subsequent requests. | |
1262 | */ | |
1263 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1264 | { | |
1265 | struct xol_area *area; | |
1266 | unsigned long vma_end; | |
1267 | unsigned long slot_addr; | |
1268 | ||
1269 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1270 | return; | |
1271 | ||
1272 | slot_addr = tsk->utask->xol_vaddr; | |
af4355e9 | 1273 | if (unlikely(!slot_addr)) |
d4b3b638 SD |
1274 | return; |
1275 | ||
1276 | area = tsk->mm->uprobes_state.xol_area; | |
1277 | vma_end = area->vaddr + PAGE_SIZE; | |
1278 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1279 | unsigned long offset; | |
1280 | int slot_nr; | |
1281 | ||
1282 | offset = slot_addr - area->vaddr; | |
1283 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1284 | if (slot_nr >= UINSNS_PER_PAGE) | |
1285 | return; | |
1286 | ||
1287 | clear_bit(slot_nr, area->bitmap); | |
1288 | atomic_dec(&area->slot_count); | |
1289 | if (waitqueue_active(&area->wq)) | |
1290 | wake_up(&area->wq); | |
1291 | ||
1292 | tsk->utask->xol_vaddr = 0; | |
1293 | } | |
1294 | } | |
1295 | ||
0326f5a9 SD |
1296 | /** |
1297 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1298 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1299 | * instruction. | |
1300 | * Return the address of the breakpoint instruction. | |
1301 | */ | |
1302 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1303 | { | |
1304 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * Called with no locks held. | |
1309 | * Called in context of a exiting or a exec-ing thread. | |
1310 | */ | |
1311 | void uprobe_free_utask(struct task_struct *t) | |
1312 | { | |
1313 | struct uprobe_task *utask = t->utask; | |
1314 | ||
0326f5a9 SD |
1315 | if (!utask) |
1316 | return; | |
1317 | ||
1318 | if (utask->active_uprobe) | |
1319 | put_uprobe(utask->active_uprobe); | |
1320 | ||
d4b3b638 | 1321 | xol_free_insn_slot(t); |
0326f5a9 SD |
1322 | kfree(utask); |
1323 | t->utask = NULL; | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Called in context of a new clone/fork from copy_process. | |
1328 | */ | |
1329 | void uprobe_copy_process(struct task_struct *t) | |
1330 | { | |
1331 | t->utask = NULL; | |
0326f5a9 SD |
1332 | } |
1333 | ||
1334 | /* | |
5a2df662 ON |
1335 | * Allocate a uprobe_task object for the task if if necessary. |
1336 | * Called when the thread hits a breakpoint. | |
0326f5a9 SD |
1337 | * |
1338 | * Returns: | |
1339 | * - pointer to new uprobe_task on success | |
1340 | * - NULL otherwise | |
1341 | */ | |
5a2df662 | 1342 | static struct uprobe_task *get_utask(void) |
0326f5a9 | 1343 | { |
5a2df662 ON |
1344 | if (!current->utask) |
1345 | current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1346 | return current->utask; | |
0326f5a9 SD |
1347 | } |
1348 | ||
1349 | /* Prepare to single-step probed instruction out of line. */ | |
1350 | static int | |
a6cb3f6d | 1351 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr) |
0326f5a9 | 1352 | { |
a6cb3f6d ON |
1353 | struct uprobe_task *utask; |
1354 | unsigned long xol_vaddr; | |
aba51024 | 1355 | int err; |
a6cb3f6d | 1356 | |
608e7427 ON |
1357 | utask = get_utask(); |
1358 | if (!utask) | |
1359 | return -ENOMEM; | |
a6cb3f6d ON |
1360 | |
1361 | xol_vaddr = xol_get_insn_slot(uprobe); | |
1362 | if (!xol_vaddr) | |
1363 | return -ENOMEM; | |
1364 | ||
1365 | utask->xol_vaddr = xol_vaddr; | |
1366 | utask->vaddr = bp_vaddr; | |
d4b3b638 | 1367 | |
aba51024 ON |
1368 | err = arch_uprobe_pre_xol(&uprobe->arch, regs); |
1369 | if (unlikely(err)) { | |
1370 | xol_free_insn_slot(current); | |
1371 | return err; | |
1372 | } | |
1373 | ||
608e7427 ON |
1374 | utask->active_uprobe = uprobe; |
1375 | utask->state = UTASK_SSTEP; | |
aba51024 | 1376 | return 0; |
0326f5a9 SD |
1377 | } |
1378 | ||
1379 | /* | |
1380 | * If we are singlestepping, then ensure this thread is not connected to | |
1381 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1382 | * triggers the signal, restart the original insn even if the task is | |
1383 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1384 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1385 | * _same_ instruction should be repeated again after return from the signal | |
1386 | * handler, and SSTEP can never finish in this case. | |
1387 | */ | |
1388 | bool uprobe_deny_signal(void) | |
1389 | { | |
1390 | struct task_struct *t = current; | |
1391 | struct uprobe_task *utask = t->utask; | |
1392 | ||
1393 | if (likely(!utask || !utask->active_uprobe)) | |
1394 | return false; | |
1395 | ||
1396 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1397 | ||
1398 | if (signal_pending(t)) { | |
1399 | spin_lock_irq(&t->sighand->siglock); | |
1400 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1401 | spin_unlock_irq(&t->sighand->siglock); | |
1402 | ||
1403 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1404 | utask->state = UTASK_SSTEP_TRAPPED; | |
1405 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1406 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1407 | } | |
1408 | } | |
1409 | ||
1410 | return true; | |
1411 | } | |
1412 | ||
1413 | /* | |
1414 | * Avoid singlestepping the original instruction if the original instruction | |
1415 | * is a NOP or can be emulated. | |
1416 | */ | |
1417 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1418 | { | |
71434f2f | 1419 | if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) { |
0578a970 ON |
1420 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
1421 | return true; | |
71434f2f | 1422 | clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); |
0578a970 | 1423 | } |
0326f5a9 SD |
1424 | return false; |
1425 | } | |
1426 | ||
499a4f3e ON |
1427 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1428 | { | |
1429 | struct vm_area_struct *vma; | |
1430 | ||
1431 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1432 | if (!valid_vma(vma, false)) | |
1433 | continue; | |
1434 | /* | |
1435 | * This is not strictly accurate, we can race with | |
1436 | * uprobe_unregister() and see the already removed | |
1437 | * uprobe if delete_uprobe() was not yet called. | |
63633cbf | 1438 | * Or this uprobe can be filtered out. |
499a4f3e ON |
1439 | */ |
1440 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1441 | return; | |
1442 | } | |
1443 | ||
1444 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1445 | } | |
1446 | ||
0908ad6e | 1447 | static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr) |
ec75fba9 ON |
1448 | { |
1449 | struct page *page; | |
1450 | uprobe_opcode_t opcode; | |
1451 | int result; | |
1452 | ||
1453 | pagefault_disable(); | |
1454 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
1455 | sizeof(opcode)); | |
1456 | pagefault_enable(); | |
1457 | ||
1458 | if (likely(result == 0)) | |
1459 | goto out; | |
1460 | ||
1461 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); | |
1462 | if (result < 0) | |
1463 | return result; | |
1464 | ||
ab0d805c | 1465 | copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); |
ec75fba9 ON |
1466 | put_page(page); |
1467 | out: | |
0908ad6e AM |
1468 | /* This needs to return true for any variant of the trap insn */ |
1469 | return is_trap_insn(&opcode); | |
ec75fba9 ON |
1470 | } |
1471 | ||
d790d346 | 1472 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1473 | { |
3a9ea052 ON |
1474 | struct mm_struct *mm = current->mm; |
1475 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1476 | struct vm_area_struct *vma; |
0326f5a9 | 1477 | |
0326f5a9 SD |
1478 | down_read(&mm->mmap_sem); |
1479 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1480 | if (vma && vma->vm_start <= bp_vaddr) { |
1481 | if (valid_vma(vma, false)) { | |
f281769e | 1482 | struct inode *inode = file_inode(vma->vm_file); |
cb113b47 | 1483 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); |
0326f5a9 | 1484 | |
3a9ea052 ON |
1485 | uprobe = find_uprobe(inode, offset); |
1486 | } | |
d790d346 ON |
1487 | |
1488 | if (!uprobe) | |
0908ad6e | 1489 | *is_swbp = is_trap_at_addr(mm, bp_vaddr); |
d790d346 ON |
1490 | } else { |
1491 | *is_swbp = -EFAULT; | |
0326f5a9 | 1492 | } |
499a4f3e ON |
1493 | |
1494 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1495 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1496 | up_read(&mm->mmap_sem); |
1497 | ||
3a9ea052 ON |
1498 | return uprobe; |
1499 | } | |
1500 | ||
da1816b1 ON |
1501 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
1502 | { | |
1503 | struct uprobe_consumer *uc; | |
1504 | int remove = UPROBE_HANDLER_REMOVE; | |
1505 | ||
1506 | down_read(&uprobe->register_rwsem); | |
1507 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
1508 | int rc = uc->handler(uc, regs); | |
1509 | ||
1510 | WARN(rc & ~UPROBE_HANDLER_MASK, | |
1511 | "bad rc=0x%x from %pf()\n", rc, uc->handler); | |
1512 | remove &= rc; | |
1513 | } | |
1514 | ||
1515 | if (remove && uprobe->consumers) { | |
1516 | WARN_ON(!uprobe_is_active(uprobe)); | |
1517 | unapply_uprobe(uprobe, current->mm); | |
1518 | } | |
1519 | up_read(&uprobe->register_rwsem); | |
1520 | } | |
1521 | ||
3a9ea052 ON |
1522 | /* |
1523 | * Run handler and ask thread to singlestep. | |
1524 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1525 | */ | |
1526 | static void handle_swbp(struct pt_regs *regs) | |
1527 | { | |
3a9ea052 ON |
1528 | struct uprobe *uprobe; |
1529 | unsigned long bp_vaddr; | |
56bb4cf6 | 1530 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1531 | |
1532 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1533 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1534 | |
0326f5a9 | 1535 | if (!uprobe) { |
56bb4cf6 ON |
1536 | if (is_swbp > 0) { |
1537 | /* No matching uprobe; signal SIGTRAP. */ | |
1538 | send_sig(SIGTRAP, current, 0); | |
1539 | } else { | |
1540 | /* | |
1541 | * Either we raced with uprobe_unregister() or we can't | |
1542 | * access this memory. The latter is only possible if | |
1543 | * another thread plays with our ->mm. In both cases | |
1544 | * we can simply restart. If this vma was unmapped we | |
1545 | * can pretend this insn was not executed yet and get | |
1546 | * the (correct) SIGSEGV after restart. | |
1547 | */ | |
1548 | instruction_pointer_set(regs, bp_vaddr); | |
1549 | } | |
0326f5a9 SD |
1550 | return; |
1551 | } | |
74e59dfc ON |
1552 | |
1553 | /* change it in advance for ->handler() and restart */ | |
1554 | instruction_pointer_set(regs, bp_vaddr); | |
1555 | ||
142b18dd ON |
1556 | /* |
1557 | * TODO: move copy_insn/etc into _register and remove this hack. | |
1558 | * After we hit the bp, _unregister + _register can install the | |
1559 | * new and not-yet-analyzed uprobe at the same address, restart. | |
1560 | */ | |
1561 | smp_rmb(); /* pairs with wmb() in install_breakpoint() */ | |
71434f2f | 1562 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
74e59dfc | 1563 | goto out; |
0326f5a9 | 1564 | |
0326f5a9 | 1565 | handler_chain(uprobe, regs); |
0578a970 ON |
1566 | if (can_skip_sstep(uprobe, regs)) |
1567 | goto out; | |
0326f5a9 | 1568 | |
608e7427 | 1569 | if (!pre_ssout(uprobe, regs, bp_vaddr)) |
0326f5a9 | 1570 | return; |
0326f5a9 | 1571 | |
74e59dfc | 1572 | /* can_skip_sstep() succeeded, or restart if can't singlestep */ |
0578a970 | 1573 | out: |
8bd87445 | 1574 | put_uprobe(uprobe); |
0326f5a9 SD |
1575 | } |
1576 | ||
1577 | /* | |
1578 | * Perform required fix-ups and disable singlestep. | |
1579 | * Allow pending signals to take effect. | |
1580 | */ | |
1581 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1582 | { | |
1583 | struct uprobe *uprobe; | |
1584 | ||
1585 | uprobe = utask->active_uprobe; | |
1586 | if (utask->state == UTASK_SSTEP_ACK) | |
1587 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1588 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1589 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1590 | else | |
1591 | WARN_ON_ONCE(1); | |
1592 | ||
1593 | put_uprobe(uprobe); | |
1594 | utask->active_uprobe = NULL; | |
1595 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1596 | xol_free_insn_slot(current); |
0326f5a9 SD |
1597 | |
1598 | spin_lock_irq(¤t->sighand->siglock); | |
1599 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1600 | spin_unlock_irq(¤t->sighand->siglock); | |
1601 | } | |
1602 | ||
1603 | /* | |
1b08e907 ON |
1604 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1605 | * allows the thread to return from interrupt. After that handle_swbp() | |
1606 | * sets utask->active_uprobe. | |
0326f5a9 | 1607 | * |
1b08e907 ON |
1608 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1609 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1610 | * |
1611 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1612 | * uprobe_notify_resume(). | |
1613 | */ | |
1614 | void uprobe_notify_resume(struct pt_regs *regs) | |
1615 | { | |
1616 | struct uprobe_task *utask; | |
1617 | ||
db023ea5 ON |
1618 | clear_thread_flag(TIF_UPROBE); |
1619 | ||
0326f5a9 | 1620 | utask = current->utask; |
1b08e907 | 1621 | if (utask && utask->active_uprobe) |
0326f5a9 | 1622 | handle_singlestep(utask, regs); |
1b08e907 ON |
1623 | else |
1624 | handle_swbp(regs); | |
0326f5a9 SD |
1625 | } |
1626 | ||
1627 | /* | |
1628 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1629 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1630 | */ | |
1631 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1632 | { | |
f8ac4ec9 | 1633 | if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags)) |
0326f5a9 SD |
1634 | return 0; |
1635 | ||
0326f5a9 | 1636 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1637 | return 1; |
1638 | } | |
1639 | ||
1640 | /* | |
1641 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1642 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1643 | */ | |
1644 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1645 | { | |
1646 | struct uprobe_task *utask = current->utask; | |
1647 | ||
1648 | if (!current->mm || !utask || !utask->active_uprobe) | |
1649 | /* task is currently not uprobed */ | |
1650 | return 0; | |
1651 | ||
1652 | utask->state = UTASK_SSTEP_ACK; | |
1653 | set_thread_flag(TIF_UPROBE); | |
1654 | return 1; | |
1655 | } | |
1656 | ||
1657 | static struct notifier_block uprobe_exception_nb = { | |
1658 | .notifier_call = arch_uprobe_exception_notify, | |
1659 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1660 | }; | |
1661 | ||
2b144498 SD |
1662 | static int __init init_uprobes(void) |
1663 | { | |
1664 | int i; | |
1665 | ||
66d06dff | 1666 | for (i = 0; i < UPROBES_HASH_SZ; i++) |
2b144498 | 1667 | mutex_init(&uprobes_mmap_mutex[i]); |
0326f5a9 | 1668 | |
32cdba1e ON |
1669 | if (percpu_init_rwsem(&dup_mmap_sem)) |
1670 | return -ENOMEM; | |
1671 | ||
0326f5a9 | 1672 | return register_die_notifier(&uprobe_exception_nb); |
2b144498 | 1673 | } |
0326f5a9 | 1674 | module_init(init_uprobes); |
2b144498 SD |
1675 | |
1676 | static void __exit exit_uprobes(void) | |
1677 | { | |
1678 | } | |
2b144498 | 1679 | module_exit(exit_uprobes); |