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; |
7b2d81d4 | 44 | |
2b144498 SD |
45 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
46 | ||
47 | #define UPROBES_HASH_SZ 13 | |
7b2d81d4 | 48 | |
c5784de2 PZ |
49 | /* |
50 | * We need separate register/unregister and mmap/munmap lock hashes because | |
51 | * of mmap_sem nesting. | |
52 | * | |
53 | * uprobe_register() needs to install probes on (potentially) all processes | |
54 | * and thus needs to acquire multiple mmap_sems (consequtively, not | |
55 | * concurrently), whereas uprobe_mmap() is called while holding mmap_sem | |
56 | * for the particular process doing the mmap. | |
57 | * | |
58 | * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem | |
59 | * because of lock order against i_mmap_mutex. This means there's a hole in | |
60 | * the register vma iteration where a mmap() can happen. | |
61 | * | |
62 | * Thus uprobe_register() can race with uprobe_mmap() and we can try and | |
63 | * install a probe where one is already installed. | |
64 | */ | |
65 | ||
2b144498 SD |
66 | /* serialize (un)register */ |
67 | static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 IM |
68 | |
69 | #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) | |
2b144498 SD |
70 | |
71 | /* serialize uprobe->pending_list */ | |
72 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 73 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 | 74 | |
32cdba1e ON |
75 | static struct percpu_rw_semaphore dup_mmap_sem; |
76 | ||
2b144498 | 77 | /* |
7b2d81d4 | 78 | * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe |
2b144498 SD |
79 | * events active at this time. Probably a fine grained per inode count is |
80 | * better? | |
81 | */ | |
82 | static atomic_t uprobe_events = ATOMIC_INIT(0); | |
83 | ||
cb9a19fe | 84 | /* Have a copy of original instruction */ |
71434f2f | 85 | #define UPROBE_COPY_INSN 0 |
cb9a19fe | 86 | /* Dont run handlers when first register/ last unregister in progress*/ |
71434f2f | 87 | #define UPROBE_RUN_HANDLER 1 |
cb9a19fe | 88 | /* Can skip singlestep */ |
71434f2f | 89 | #define UPROBE_SKIP_SSTEP 2 |
cb9a19fe | 90 | |
3ff54efd SD |
91 | struct uprobe { |
92 | struct rb_node rb_node; /* node in the rb tree */ | |
93 | atomic_t ref; | |
94 | struct rw_semaphore consumer_rwsem; | |
4710f05f | 95 | struct mutex copy_mutex; /* TODO: kill me and UPROBE_COPY_INSN */ |
3ff54efd SD |
96 | struct list_head pending_list; |
97 | struct uprobe_consumer *consumers; | |
98 | struct inode *inode; /* Also hold a ref to inode */ | |
99 | loff_t offset; | |
71434f2f | 100 | unsigned long flags; |
3ff54efd SD |
101 | struct arch_uprobe arch; |
102 | }; | |
103 | ||
2b144498 SD |
104 | /* |
105 | * valid_vma: Verify if the specified vma is an executable vma | |
106 | * Relax restrictions while unregistering: vm_flags might have | |
107 | * changed after breakpoint was inserted. | |
108 | * - is_register: indicates if we are in register context. | |
109 | * - Return 1 if the specified virtual address is in an | |
110 | * executable vma. | |
111 | */ | |
112 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
113 | { | |
e40cfce6 | 114 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED; |
2b144498 | 115 | |
e40cfce6 ON |
116 | if (is_register) |
117 | flags |= VM_WRITE; | |
2b144498 | 118 | |
e40cfce6 | 119 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
120 | } |
121 | ||
57683f72 | 122 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 123 | { |
57683f72 | 124 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
125 | } |
126 | ||
cb113b47 ON |
127 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
128 | { | |
129 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
130 | } | |
131 | ||
2b144498 SD |
132 | /** |
133 | * __replace_page - replace page in vma by new page. | |
134 | * based on replace_page in mm/ksm.c | |
135 | * | |
136 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 137 | * @addr: address the old @page is mapped at |
2b144498 SD |
138 | * @page: the cowed page we are replacing by kpage |
139 | * @kpage: the modified page we replace page by | |
140 | * | |
141 | * Returns 0 on success, -EFAULT on failure. | |
142 | */ | |
c517ee74 ON |
143 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
144 | struct page *page, struct page *kpage) | |
2b144498 SD |
145 | { |
146 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
147 | spinlock_t *ptl; |
148 | pte_t *ptep; | |
9f92448c | 149 | int err; |
6bdb913f HE |
150 | /* For mmu_notifiers */ |
151 | const unsigned long mmun_start = addr; | |
152 | const unsigned long mmun_end = addr + PAGE_SIZE; | |
2b144498 | 153 | |
194f8dcb | 154 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
155 | lock_page(page); |
156 | ||
6bdb913f | 157 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
9f92448c | 158 | err = -EAGAIN; |
5323ce71 | 159 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 160 | if (!ptep) |
9f92448c | 161 | goto unlock; |
2b144498 SD |
162 | |
163 | get_page(kpage); | |
164 | page_add_new_anon_rmap(kpage, vma, addr); | |
165 | ||
7396fa81 SD |
166 | if (!PageAnon(page)) { |
167 | dec_mm_counter(mm, MM_FILEPAGES); | |
168 | inc_mm_counter(mm, MM_ANONPAGES); | |
169 | } | |
170 | ||
2b144498 SD |
171 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
172 | ptep_clear_flush(vma, addr, ptep); | |
173 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
174 | ||
175 | page_remove_rmap(page); | |
176 | if (!page_mapped(page)) | |
177 | try_to_free_swap(page); | |
2b144498 | 178 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 179 | |
194f8dcb ON |
180 | if (vma->vm_flags & VM_LOCKED) |
181 | munlock_vma_page(page); | |
182 | put_page(page); | |
183 | ||
9f92448c ON |
184 | err = 0; |
185 | unlock: | |
6bdb913f | 186 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
9f92448c ON |
187 | unlock_page(page); |
188 | return err; | |
2b144498 SD |
189 | } |
190 | ||
191 | /** | |
5cb4ac3a | 192 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 193 | * @insn: instruction to be checked. |
5cb4ac3a | 194 | * Default implementation of is_swbp_insn |
2b144498 SD |
195 | * Returns true if @insn is a breakpoint instruction. |
196 | */ | |
5cb4ac3a | 197 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 198 | { |
5cb4ac3a | 199 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
200 | } |
201 | ||
cceb55aa ON |
202 | static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode) |
203 | { | |
204 | void *kaddr = kmap_atomic(page); | |
205 | memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE); | |
206 | kunmap_atomic(kaddr); | |
207 | } | |
208 | ||
ed6f6a50 ON |
209 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
210 | { | |
211 | uprobe_opcode_t old_opcode; | |
212 | bool is_swbp; | |
213 | ||
214 | copy_opcode(page, vaddr, &old_opcode); | |
215 | is_swbp = is_swbp_insn(&old_opcode); | |
216 | ||
217 | if (is_swbp_insn(new_opcode)) { | |
218 | if (is_swbp) /* register: already installed? */ | |
219 | return 0; | |
220 | } else { | |
221 | if (!is_swbp) /* unregister: was it changed by us? */ | |
076a365b | 222 | return 0; |
ed6f6a50 ON |
223 | } |
224 | ||
225 | return 1; | |
226 | } | |
227 | ||
2b144498 SD |
228 | /* |
229 | * NOTE: | |
230 | * Expect the breakpoint instruction to be the smallest size instruction for | |
231 | * the architecture. If an arch has variable length instruction and the | |
232 | * breakpoint instruction is not of the smallest length instruction | |
cceb55aa | 233 | * supported by that architecture then we need to modify is_swbp_at_addr and |
2b144498 SD |
234 | * write_opcode accordingly. This would never be a problem for archs that |
235 | * have fixed length instructions. | |
236 | */ | |
237 | ||
238 | /* | |
239 | * write_opcode - write the opcode at a given virtual address. | |
240 | * @mm: the probed process address space. | |
2b144498 SD |
241 | * @vaddr: the virtual address to store the opcode. |
242 | * @opcode: opcode to be written at @vaddr. | |
243 | * | |
244 | * Called with mm->mmap_sem held (for read and with a reference to | |
245 | * mm). | |
246 | * | |
247 | * For mm @mm, write the opcode at @vaddr. | |
248 | * Return 0 (success) or a negative errno. | |
249 | */ | |
cceb55aa ON |
250 | static int write_opcode(struct mm_struct *mm, unsigned long vaddr, |
251 | uprobe_opcode_t opcode) | |
2b144498 SD |
252 | { |
253 | struct page *old_page, *new_page; | |
2b144498 SD |
254 | void *vaddr_old, *vaddr_new; |
255 | struct vm_area_struct *vma; | |
2b144498 | 256 | int ret; |
f403072c | 257 | |
5323ce71 | 258 | retry: |
2b144498 | 259 | /* Read the page with vaddr into memory */ |
75ed82ea | 260 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
261 | if (ret <= 0) |
262 | return ret; | |
7b2d81d4 | 263 | |
ed6f6a50 ON |
264 | ret = verify_opcode(old_page, vaddr, &opcode); |
265 | if (ret <= 0) | |
266 | goto put_old; | |
267 | ||
2b144498 SD |
268 | ret = -ENOMEM; |
269 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
270 | if (!new_page) | |
9f92448c | 271 | goto put_old; |
2b144498 SD |
272 | |
273 | __SetPageUptodate(new_page); | |
274 | ||
2b144498 SD |
275 | /* copy the page now that we've got it stable */ |
276 | vaddr_old = kmap_atomic(old_page); | |
277 | vaddr_new = kmap_atomic(new_page); | |
278 | ||
279 | memcpy(vaddr_new, vaddr_old, PAGE_SIZE); | |
d9c4a30e | 280 | memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); |
2b144498 SD |
281 | |
282 | kunmap_atomic(vaddr_new); | |
283 | kunmap_atomic(vaddr_old); | |
284 | ||
285 | ret = anon_vma_prepare(vma); | |
286 | if (ret) | |
9f92448c | 287 | goto put_new; |
2b144498 | 288 | |
c517ee74 | 289 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 290 | |
9f92448c | 291 | put_new: |
2b144498 | 292 | page_cache_release(new_page); |
9f92448c | 293 | put_old: |
7b2d81d4 IM |
294 | put_page(old_page); |
295 | ||
5323ce71 ON |
296 | if (unlikely(ret == -EAGAIN)) |
297 | goto retry; | |
2b144498 SD |
298 | return ret; |
299 | } | |
300 | ||
2b144498 | 301 | /** |
5cb4ac3a | 302 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 303 | * @auprobe: arch specific probepoint information. |
2b144498 | 304 | * @mm: the probed process address space. |
2b144498 SD |
305 | * @vaddr: the virtual address to insert the opcode. |
306 | * | |
307 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
308 | * Return 0 (success) or a negative errno. | |
309 | */ | |
5cb4ac3a | 310 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 311 | { |
cceb55aa | 312 | return write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
313 | } |
314 | ||
315 | /** | |
316 | * set_orig_insn - Restore the original instruction. | |
317 | * @mm: the probed process address space. | |
e3343e6a | 318 | * @auprobe: arch specific probepoint information. |
2b144498 | 319 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
320 | * |
321 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
322 | * Return 0 (success) or a negative errno. | |
323 | */ | |
7b2d81d4 | 324 | int __weak |
ded86e7c | 325 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 326 | { |
cceb55aa | 327 | return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); |
2b144498 SD |
328 | } |
329 | ||
330 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
331 | { | |
332 | if (l->inode < r->inode) | |
333 | return -1; | |
7b2d81d4 | 334 | |
2b144498 SD |
335 | if (l->inode > r->inode) |
336 | return 1; | |
2b144498 | 337 | |
7b2d81d4 IM |
338 | if (l->offset < r->offset) |
339 | return -1; | |
340 | ||
341 | if (l->offset > r->offset) | |
342 | return 1; | |
2b144498 SD |
343 | |
344 | return 0; | |
345 | } | |
346 | ||
347 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
348 | { | |
349 | struct uprobe u = { .inode = inode, .offset = offset }; | |
350 | struct rb_node *n = uprobes_tree.rb_node; | |
351 | struct uprobe *uprobe; | |
352 | int match; | |
353 | ||
354 | while (n) { | |
355 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
356 | match = match_uprobe(&u, uprobe); | |
357 | if (!match) { | |
358 | atomic_inc(&uprobe->ref); | |
359 | return uprobe; | |
360 | } | |
7b2d81d4 | 361 | |
2b144498 SD |
362 | if (match < 0) |
363 | n = n->rb_left; | |
364 | else | |
365 | n = n->rb_right; | |
366 | } | |
367 | return NULL; | |
368 | } | |
369 | ||
370 | /* | |
371 | * Find a uprobe corresponding to a given inode:offset | |
372 | * Acquires uprobes_treelock | |
373 | */ | |
374 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
375 | { | |
376 | struct uprobe *uprobe; | |
2b144498 | 377 | |
6f47caa0 | 378 | spin_lock(&uprobes_treelock); |
2b144498 | 379 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 380 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 381 | |
2b144498 SD |
382 | return uprobe; |
383 | } | |
384 | ||
385 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
386 | { | |
387 | struct rb_node **p = &uprobes_tree.rb_node; | |
388 | struct rb_node *parent = NULL; | |
389 | struct uprobe *u; | |
390 | int match; | |
391 | ||
392 | while (*p) { | |
393 | parent = *p; | |
394 | u = rb_entry(parent, struct uprobe, rb_node); | |
395 | match = match_uprobe(uprobe, u); | |
396 | if (!match) { | |
397 | atomic_inc(&u->ref); | |
398 | return u; | |
399 | } | |
400 | ||
401 | if (match < 0) | |
402 | p = &parent->rb_left; | |
403 | else | |
404 | p = &parent->rb_right; | |
405 | ||
406 | } | |
7b2d81d4 | 407 | |
2b144498 SD |
408 | u = NULL; |
409 | rb_link_node(&uprobe->rb_node, parent, p); | |
410 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
411 | /* get access + creation ref */ | |
412 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 413 | |
2b144498 SD |
414 | return u; |
415 | } | |
416 | ||
417 | /* | |
7b2d81d4 | 418 | * Acquire uprobes_treelock. |
2b144498 SD |
419 | * Matching uprobe already exists in rbtree; |
420 | * increment (access refcount) and return the matching uprobe. | |
421 | * | |
422 | * No matching uprobe; insert the uprobe in rb_tree; | |
423 | * get a double refcount (access + creation) and return NULL. | |
424 | */ | |
425 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
426 | { | |
2b144498 SD |
427 | struct uprobe *u; |
428 | ||
6f47caa0 | 429 | spin_lock(&uprobes_treelock); |
2b144498 | 430 | u = __insert_uprobe(uprobe); |
6f47caa0 | 431 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 432 | |
0326f5a9 | 433 | /* For now assume that the instruction need not be single-stepped */ |
71434f2f | 434 | __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); |
0326f5a9 | 435 | |
2b144498 SD |
436 | return u; |
437 | } | |
438 | ||
439 | static void put_uprobe(struct uprobe *uprobe) | |
440 | { | |
441 | if (atomic_dec_and_test(&uprobe->ref)) | |
442 | kfree(uprobe); | |
443 | } | |
444 | ||
445 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
446 | { | |
447 | struct uprobe *uprobe, *cur_uprobe; | |
448 | ||
449 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
450 | if (!uprobe) | |
451 | return NULL; | |
452 | ||
453 | uprobe->inode = igrab(inode); | |
454 | uprobe->offset = offset; | |
455 | init_rwsem(&uprobe->consumer_rwsem); | |
4710f05f | 456 | mutex_init(&uprobe->copy_mutex); |
2b144498 SD |
457 | |
458 | /* add to uprobes_tree, sorted on inode:offset */ | |
459 | cur_uprobe = insert_uprobe(uprobe); | |
460 | ||
461 | /* a uprobe exists for this inode:offset combination */ | |
462 | if (cur_uprobe) { | |
463 | kfree(uprobe); | |
464 | uprobe = cur_uprobe; | |
465 | iput(inode); | |
7b2d81d4 | 466 | } else { |
2b144498 | 467 | atomic_inc(&uprobe_events); |
7b2d81d4 IM |
468 | } |
469 | ||
2b144498 SD |
470 | return uprobe; |
471 | } | |
472 | ||
0326f5a9 SD |
473 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
474 | { | |
475 | struct uprobe_consumer *uc; | |
476 | ||
71434f2f | 477 | if (!test_bit(UPROBE_RUN_HANDLER, &uprobe->flags)) |
0326f5a9 SD |
478 | return; |
479 | ||
480 | down_read(&uprobe->consumer_rwsem); | |
481 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
482 | if (!uc->filter || uc->filter(uc, current)) | |
483 | uc->handler(uc, regs); | |
484 | } | |
485 | up_read(&uprobe->consumer_rwsem); | |
486 | } | |
487 | ||
2b144498 | 488 | /* Returns the previous consumer */ |
7b2d81d4 | 489 | static struct uprobe_consumer * |
e3343e6a | 490 | consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
491 | { |
492 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
493 | uc->next = uprobe->consumers; |
494 | uprobe->consumers = uc; | |
2b144498 | 495 | up_write(&uprobe->consumer_rwsem); |
7b2d81d4 | 496 | |
e3343e6a | 497 | return uc->next; |
2b144498 SD |
498 | } |
499 | ||
500 | /* | |
e3343e6a SD |
501 | * For uprobe @uprobe, delete the consumer @uc. |
502 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
503 | * or return false. |
504 | */ | |
e3343e6a | 505 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
506 | { |
507 | struct uprobe_consumer **con; | |
508 | bool ret = false; | |
509 | ||
510 | down_write(&uprobe->consumer_rwsem); | |
511 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
512 | if (*con == uc) { |
513 | *con = uc->next; | |
2b144498 SD |
514 | ret = true; |
515 | break; | |
516 | } | |
517 | } | |
518 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 519 | |
2b144498 SD |
520 | return ret; |
521 | } | |
522 | ||
e3343e6a | 523 | static int |
d436615e | 524 | __copy_insn(struct address_space *mapping, struct file *filp, char *insn, |
593609a5 | 525 | unsigned long nbytes, loff_t offset) |
2b144498 | 526 | { |
2b144498 SD |
527 | struct page *page; |
528 | void *vaddr; | |
593609a5 ON |
529 | unsigned long off; |
530 | pgoff_t idx; | |
2b144498 SD |
531 | |
532 | if (!filp) | |
533 | return -EINVAL; | |
534 | ||
cc359d18 ON |
535 | if (!mapping->a_ops->readpage) |
536 | return -EIO; | |
537 | ||
593609a5 ON |
538 | idx = offset >> PAGE_CACHE_SHIFT; |
539 | off = offset & ~PAGE_MASK; | |
2b144498 SD |
540 | |
541 | /* | |
542 | * Ensure that the page that has the original instruction is | |
543 | * populated and in page-cache. | |
544 | */ | |
545 | page = read_mapping_page(mapping, idx, filp); | |
546 | if (IS_ERR(page)) | |
547 | return PTR_ERR(page); | |
548 | ||
549 | vaddr = kmap_atomic(page); | |
593609a5 | 550 | memcpy(insn, vaddr + off, nbytes); |
2b144498 SD |
551 | kunmap_atomic(vaddr); |
552 | page_cache_release(page); | |
7b2d81d4 | 553 | |
2b144498 SD |
554 | return 0; |
555 | } | |
556 | ||
d436615e | 557 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 SD |
558 | { |
559 | struct address_space *mapping; | |
2b144498 | 560 | unsigned long nbytes; |
7b2d81d4 | 561 | int bytes; |
2b144498 | 562 | |
d436615e | 563 | nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); |
2b144498 SD |
564 | mapping = uprobe->inode->i_mapping; |
565 | ||
566 | /* Instruction at end of binary; copy only available bytes */ | |
567 | if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) | |
568 | bytes = uprobe->inode->i_size - uprobe->offset; | |
569 | else | |
570 | bytes = MAX_UINSN_BYTES; | |
571 | ||
572 | /* Instruction at the page-boundary; copy bytes in second page */ | |
573 | if (nbytes < bytes) { | |
fc36f595 ON |
574 | int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, |
575 | bytes - nbytes, uprobe->offset + nbytes); | |
576 | if (err) | |
577 | return err; | |
2b144498 SD |
578 | bytes = nbytes; |
579 | } | |
d436615e | 580 | return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); |
2b144498 SD |
581 | } |
582 | ||
cb9a19fe ON |
583 | static int prepare_uprobe(struct uprobe *uprobe, struct file *file, |
584 | struct mm_struct *mm, unsigned long vaddr) | |
585 | { | |
586 | int ret = 0; | |
587 | ||
71434f2f | 588 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
cb9a19fe ON |
589 | return ret; |
590 | ||
4710f05f | 591 | mutex_lock(&uprobe->copy_mutex); |
71434f2f | 592 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
593 | goto out; |
594 | ||
cb9a19fe ON |
595 | ret = copy_insn(uprobe, file); |
596 | if (ret) | |
597 | goto out; | |
598 | ||
599 | ret = -ENOTSUPP; | |
600 | if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) | |
601 | goto out; | |
602 | ||
603 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
604 | if (ret) | |
605 | goto out; | |
606 | ||
607 | /* write_opcode() assumes we don't cross page boundary */ | |
608 | BUG_ON((uprobe->offset & ~PAGE_MASK) + | |
609 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
610 | ||
611 | smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ | |
71434f2f | 612 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
613 | |
614 | out: | |
4710f05f ON |
615 | mutex_unlock(&uprobe->copy_mutex); |
616 | ||
cb9a19fe ON |
617 | return ret; |
618 | } | |
619 | ||
e3343e6a SD |
620 | static int |
621 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 622 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 623 | { |
f8ac4ec9 | 624 | bool first_uprobe; |
2b144498 SD |
625 | int ret; |
626 | ||
627 | /* | |
628 | * If probe is being deleted, unregister thread could be done with | |
629 | * the vma-rmap-walk through. Adding a probe now can be fatal since | |
630 | * nobody will be able to cleanup. Also we could be from fork or | |
631 | * mremap path, where the probe might have already been inserted. | |
632 | * Hence behave as if probe already existed. | |
633 | */ | |
634 | if (!uprobe->consumers) | |
78f74116 | 635 | return 0; |
2b144498 | 636 | |
cb9a19fe ON |
637 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
638 | if (ret) | |
639 | return ret; | |
682968e0 | 640 | |
f8ac4ec9 ON |
641 | /* |
642 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
643 | * the task can hit this breakpoint right after __replace_page(). | |
644 | */ | |
645 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
646 | if (first_uprobe) | |
647 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
648 | ||
816c03fb | 649 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
650 | if (!ret) |
651 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
652 | else if (first_uprobe) | |
f8ac4ec9 | 653 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
654 | |
655 | return ret; | |
656 | } | |
657 | ||
076a365b | 658 | static int |
816c03fb | 659 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 660 | { |
9f68f672 ON |
661 | /* can happen if uprobe_register() fails */ |
662 | if (!test_bit(MMF_HAS_UPROBES, &mm->flags)) | |
076a365b | 663 | return 0; |
9f68f672 ON |
664 | |
665 | set_bit(MMF_RECALC_UPROBES, &mm->flags); | |
076a365b | 666 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
667 | } |
668 | ||
0326f5a9 | 669 | /* |
778b032d ON |
670 | * There could be threads that have already hit the breakpoint. They |
671 | * will recheck the current insn and restart if find_uprobe() fails. | |
672 | * See find_active_uprobe(). | |
0326f5a9 | 673 | */ |
2b144498 SD |
674 | static void delete_uprobe(struct uprobe *uprobe) |
675 | { | |
6f47caa0 | 676 | spin_lock(&uprobes_treelock); |
2b144498 | 677 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 678 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
679 | iput(uprobe->inode); |
680 | put_uprobe(uprobe); | |
681 | atomic_dec(&uprobe_events); | |
682 | } | |
683 | ||
26872090 ON |
684 | struct map_info { |
685 | struct map_info *next; | |
686 | struct mm_struct *mm; | |
816c03fb | 687 | unsigned long vaddr; |
26872090 ON |
688 | }; |
689 | ||
690 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 691 | { |
26872090 ON |
692 | struct map_info *next = info->next; |
693 | kfree(info); | |
694 | return next; | |
695 | } | |
696 | ||
697 | static struct map_info * | |
698 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
699 | { | |
700 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 701 | struct vm_area_struct *vma; |
26872090 ON |
702 | struct map_info *curr = NULL; |
703 | struct map_info *prev = NULL; | |
704 | struct map_info *info; | |
705 | int more = 0; | |
2b144498 | 706 | |
26872090 ON |
707 | again: |
708 | mutex_lock(&mapping->i_mmap_mutex); | |
6b2dbba8 | 709 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
710 | if (!valid_vma(vma, is_register)) |
711 | continue; | |
712 | ||
7a5bfb66 ON |
713 | if (!prev && !more) { |
714 | /* | |
715 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
716 | * reclaim. This is optimistic, no harm done if it fails. | |
717 | */ | |
718 | prev = kmalloc(sizeof(struct map_info), | |
719 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
720 | if (prev) | |
721 | prev->next = NULL; | |
722 | } | |
26872090 ON |
723 | if (!prev) { |
724 | more++; | |
725 | continue; | |
2b144498 | 726 | } |
2b144498 | 727 | |
26872090 ON |
728 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
729 | continue; | |
7b2d81d4 | 730 | |
26872090 ON |
731 | info = prev; |
732 | prev = prev->next; | |
733 | info->next = curr; | |
734 | curr = info; | |
2b144498 | 735 | |
26872090 | 736 | info->mm = vma->vm_mm; |
57683f72 | 737 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 738 | } |
2b144498 SD |
739 | mutex_unlock(&mapping->i_mmap_mutex); |
740 | ||
26872090 ON |
741 | if (!more) |
742 | goto out; | |
743 | ||
744 | prev = curr; | |
745 | while (curr) { | |
746 | mmput(curr->mm); | |
747 | curr = curr->next; | |
748 | } | |
7b2d81d4 | 749 | |
26872090 ON |
750 | do { |
751 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
752 | if (!info) { | |
753 | curr = ERR_PTR(-ENOMEM); | |
754 | goto out; | |
755 | } | |
756 | info->next = prev; | |
757 | prev = info; | |
758 | } while (--more); | |
759 | ||
760 | goto again; | |
761 | out: | |
762 | while (prev) | |
763 | prev = free_map_info(prev); | |
764 | return curr; | |
2b144498 SD |
765 | } |
766 | ||
767 | static int register_for_each_vma(struct uprobe *uprobe, bool is_register) | |
768 | { | |
26872090 ON |
769 | struct map_info *info; |
770 | int err = 0; | |
2b144498 | 771 | |
32cdba1e | 772 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
773 | info = build_map_info(uprobe->inode->i_mapping, |
774 | uprobe->offset, is_register); | |
32cdba1e ON |
775 | if (IS_ERR(info)) { |
776 | err = PTR_ERR(info); | |
777 | goto out; | |
778 | } | |
7b2d81d4 | 779 | |
26872090 ON |
780 | while (info) { |
781 | struct mm_struct *mm = info->mm; | |
782 | struct vm_area_struct *vma; | |
7b2d81d4 | 783 | |
076a365b | 784 | if (err && is_register) |
26872090 | 785 | goto free; |
7b2d81d4 | 786 | |
77fc4af1 | 787 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
788 | vma = find_vma(mm, info->vaddr); |
789 | if (!vma || !valid_vma(vma, is_register) || | |
790 | vma->vm_file->f_mapping->host != uprobe->inode) | |
26872090 ON |
791 | goto unlock; |
792 | ||
f4d6dfe5 ON |
793 | if (vma->vm_start > info->vaddr || |
794 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 795 | goto unlock; |
2b144498 | 796 | |
78f74116 | 797 | if (is_register) |
26872090 | 798 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
78f74116 | 799 | else |
076a365b | 800 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
78f74116 | 801 | |
26872090 ON |
802 | unlock: |
803 | up_write(&mm->mmap_sem); | |
804 | free: | |
805 | mmput(mm); | |
806 | info = free_map_info(info); | |
2b144498 | 807 | } |
32cdba1e ON |
808 | out: |
809 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 810 | return err; |
2b144498 SD |
811 | } |
812 | ||
7b2d81d4 | 813 | static int __uprobe_register(struct uprobe *uprobe) |
2b144498 SD |
814 | { |
815 | return register_for_each_vma(uprobe, true); | |
816 | } | |
817 | ||
7b2d81d4 | 818 | static void __uprobe_unregister(struct uprobe *uprobe) |
2b144498 SD |
819 | { |
820 | if (!register_for_each_vma(uprobe, false)) | |
821 | delete_uprobe(uprobe); | |
822 | ||
823 | /* TODO : cant unregister? schedule a worker thread */ | |
824 | } | |
825 | ||
826 | /* | |
7b2d81d4 | 827 | * uprobe_register - register a probe |
2b144498 SD |
828 | * @inode: the file in which the probe has to be placed. |
829 | * @offset: offset from the start of the file. | |
e3343e6a | 830 | * @uc: information on howto handle the probe.. |
2b144498 | 831 | * |
7b2d81d4 | 832 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
833 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
834 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 835 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 836 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 837 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
838 | * unregisters. |
839 | * | |
840 | * Return errno if it cannot successully install probes | |
841 | * else return 0 (success) | |
842 | */ | |
e3343e6a | 843 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
844 | { |
845 | struct uprobe *uprobe; | |
7b2d81d4 | 846 | int ret; |
2b144498 | 847 | |
e3343e6a | 848 | if (!inode || !uc || uc->next) |
7b2d81d4 | 849 | return -EINVAL; |
2b144498 SD |
850 | |
851 | if (offset > i_size_read(inode)) | |
7b2d81d4 | 852 | return -EINVAL; |
2b144498 SD |
853 | |
854 | ret = 0; | |
855 | mutex_lock(uprobes_hash(inode)); | |
856 | uprobe = alloc_uprobe(inode, offset); | |
7b2d81d4 | 857 | |
a5f658b7 ON |
858 | if (!uprobe) { |
859 | ret = -ENOMEM; | |
860 | } else if (!consumer_add(uprobe, uc)) { | |
7b2d81d4 | 861 | ret = __uprobe_register(uprobe); |
2b144498 SD |
862 | if (ret) { |
863 | uprobe->consumers = NULL; | |
7b2d81d4 IM |
864 | __uprobe_unregister(uprobe); |
865 | } else { | |
71434f2f | 866 | set_bit(UPROBE_RUN_HANDLER, &uprobe->flags); |
7b2d81d4 | 867 | } |
2b144498 SD |
868 | } |
869 | ||
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 | |
e3343e6a | 887 | if (!inode || !uc) |
2b144498 SD |
888 | return; |
889 | ||
890 | uprobe = find_uprobe(inode, offset); | |
891 | if (!uprobe) | |
892 | return; | |
893 | ||
894 | mutex_lock(uprobes_hash(inode)); | |
2b144498 | 895 | |
e3343e6a | 896 | if (consumer_del(uprobe, uc)) { |
7b2d81d4 IM |
897 | if (!uprobe->consumers) { |
898 | __uprobe_unregister(uprobe); | |
71434f2f | 899 | clear_bit(UPROBE_RUN_HANDLER, &uprobe->flags); |
7b2d81d4 | 900 | } |
2b144498 SD |
901 | } |
902 | ||
2b144498 | 903 | mutex_unlock(uprobes_hash(inode)); |
c91368c4 | 904 | put_uprobe(uprobe); |
2b144498 SD |
905 | } |
906 | ||
891c3970 ON |
907 | static struct rb_node * |
908 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 909 | { |
2b144498 | 910 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
911 | |
912 | while (n) { | |
891c3970 | 913 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 914 | |
891c3970 | 915 | if (inode < u->inode) { |
2b144498 | 916 | n = n->rb_left; |
891c3970 | 917 | } else if (inode > u->inode) { |
2b144498 | 918 | n = n->rb_right; |
891c3970 ON |
919 | } else { |
920 | if (max < u->offset) | |
921 | n = n->rb_left; | |
922 | else if (min > u->offset) | |
923 | n = n->rb_right; | |
924 | else | |
925 | break; | |
926 | } | |
2b144498 | 927 | } |
7b2d81d4 | 928 | |
891c3970 | 929 | return n; |
2b144498 SD |
930 | } |
931 | ||
932 | /* | |
891c3970 | 933 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 934 | */ |
891c3970 ON |
935 | static void build_probe_list(struct inode *inode, |
936 | struct vm_area_struct *vma, | |
937 | unsigned long start, unsigned long end, | |
938 | struct list_head *head) | |
2b144498 | 939 | { |
891c3970 | 940 | loff_t min, max; |
891c3970 ON |
941 | struct rb_node *n, *t; |
942 | struct uprobe *u; | |
7b2d81d4 | 943 | |
891c3970 | 944 | INIT_LIST_HEAD(head); |
cb113b47 | 945 | min = vaddr_to_offset(vma, start); |
891c3970 | 946 | max = min + (end - start) - 1; |
2b144498 | 947 | |
6f47caa0 | 948 | spin_lock(&uprobes_treelock); |
891c3970 ON |
949 | n = find_node_in_range(inode, min, max); |
950 | if (n) { | |
951 | for (t = n; t; t = rb_prev(t)) { | |
952 | u = rb_entry(t, struct uprobe, rb_node); | |
953 | if (u->inode != inode || u->offset < min) | |
954 | break; | |
955 | list_add(&u->pending_list, head); | |
956 | atomic_inc(&u->ref); | |
957 | } | |
958 | for (t = n; (t = rb_next(t)); ) { | |
959 | u = rb_entry(t, struct uprobe, rb_node); | |
960 | if (u->inode != inode || u->offset > max) | |
961 | break; | |
962 | list_add(&u->pending_list, head); | |
963 | atomic_inc(&u->ref); | |
964 | } | |
2b144498 | 965 | } |
6f47caa0 | 966 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
967 | } |
968 | ||
969 | /* | |
5e5be71a | 970 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 971 | * |
5e5be71a ON |
972 | * Currently we ignore all errors and always return 0, the callers |
973 | * can't handle the failure anyway. | |
2b144498 | 974 | */ |
7b2d81d4 | 975 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
976 | { |
977 | struct list_head tmp_list; | |
665605a2 | 978 | struct uprobe *uprobe, *u; |
2b144498 | 979 | struct inode *inode; |
2b144498 SD |
980 | |
981 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) | |
7b2d81d4 | 982 | return 0; |
2b144498 SD |
983 | |
984 | inode = vma->vm_file->f_mapping->host; | |
985 | if (!inode) | |
7b2d81d4 | 986 | return 0; |
2b144498 | 987 | |
2b144498 | 988 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 989 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
7b2d81d4 | 990 | |
665605a2 | 991 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
5e5be71a | 992 | if (!fatal_signal_pending(current)) { |
57683f72 | 993 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 994 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
995 | } |
996 | put_uprobe(uprobe); | |
997 | } | |
2b144498 SD |
998 | mutex_unlock(uprobes_mmap_hash(inode)); |
999 | ||
5e5be71a | 1000 | return 0; |
2b144498 SD |
1001 | } |
1002 | ||
9f68f672 ON |
1003 | static bool |
1004 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1005 | { | |
1006 | loff_t min, max; | |
1007 | struct inode *inode; | |
1008 | struct rb_node *n; | |
1009 | ||
1010 | inode = vma->vm_file->f_mapping->host; | |
1011 | ||
1012 | min = vaddr_to_offset(vma, start); | |
1013 | max = min + (end - start) - 1; | |
1014 | ||
1015 | spin_lock(&uprobes_treelock); | |
1016 | n = find_node_in_range(inode, min, max); | |
1017 | spin_unlock(&uprobes_treelock); | |
1018 | ||
1019 | return !!n; | |
1020 | } | |
1021 | ||
682968e0 SD |
1022 | /* |
1023 | * Called in context of a munmap of a vma. | |
1024 | */ | |
cbc91f71 | 1025 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1026 | { |
682968e0 SD |
1027 | if (!atomic_read(&uprobe_events) || !valid_vma(vma, false)) |
1028 | return; | |
1029 | ||
2fd611a9 ON |
1030 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1031 | return; | |
1032 | ||
9f68f672 ON |
1033 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1034 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1035 | return; |
1036 | ||
9f68f672 ON |
1037 | if (vma_has_uprobes(vma, start, end)) |
1038 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1039 | } |
1040 | ||
d4b3b638 SD |
1041 | /* Slot allocation for XOL */ |
1042 | static int xol_add_vma(struct xol_area *area) | |
1043 | { | |
1044 | struct mm_struct *mm; | |
1045 | int ret; | |
1046 | ||
1047 | area->page = alloc_page(GFP_HIGHUSER); | |
1048 | if (!area->page) | |
1049 | return -ENOMEM; | |
1050 | ||
1051 | ret = -EALREADY; | |
1052 | mm = current->mm; | |
1053 | ||
1054 | down_write(&mm->mmap_sem); | |
1055 | if (mm->uprobes_state.xol_area) | |
1056 | goto fail; | |
1057 | ||
1058 | ret = -ENOMEM; | |
1059 | ||
1060 | /* Try to map as high as possible, this is only a hint. */ | |
1061 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); | |
1062 | if (area->vaddr & ~PAGE_MASK) { | |
1063 | ret = area->vaddr; | |
1064 | goto fail; | |
1065 | } | |
1066 | ||
1067 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1068 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1069 | if (ret) | |
1070 | goto fail; | |
1071 | ||
1072 | smp_wmb(); /* pairs with get_xol_area() */ | |
1073 | mm->uprobes_state.xol_area = area; | |
1074 | ret = 0; | |
1075 | ||
1076 | fail: | |
1077 | up_write(&mm->mmap_sem); | |
1078 | if (ret) | |
1079 | __free_page(area->page); | |
1080 | ||
1081 | return ret; | |
1082 | } | |
1083 | ||
1084 | static struct xol_area *get_xol_area(struct mm_struct *mm) | |
1085 | { | |
1086 | struct xol_area *area; | |
1087 | ||
1088 | area = mm->uprobes_state.xol_area; | |
1089 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ | |
1090 | ||
1091 | return area; | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * xol_alloc_area - Allocate process's xol_area. | |
1096 | * This area will be used for storing instructions for execution out of | |
1097 | * line. | |
1098 | * | |
1099 | * Returns the allocated area or NULL. | |
1100 | */ | |
1101 | static struct xol_area *xol_alloc_area(void) | |
1102 | { | |
1103 | struct xol_area *area; | |
1104 | ||
1105 | area = kzalloc(sizeof(*area), GFP_KERNEL); | |
1106 | if (unlikely(!area)) | |
1107 | return NULL; | |
1108 | ||
1109 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
1110 | ||
1111 | if (!area->bitmap) | |
1112 | goto fail; | |
1113 | ||
1114 | init_waitqueue_head(&area->wq); | |
1115 | if (!xol_add_vma(area)) | |
1116 | return area; | |
1117 | ||
1118 | fail: | |
1119 | kfree(area->bitmap); | |
1120 | kfree(area); | |
1121 | ||
1122 | return get_xol_area(current->mm); | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * uprobe_clear_state - Free the area allocated for slots. | |
1127 | */ | |
1128 | void uprobe_clear_state(struct mm_struct *mm) | |
1129 | { | |
1130 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1131 | ||
1132 | if (!area) | |
1133 | return; | |
1134 | ||
1135 | put_page(area->page); | |
1136 | kfree(area->bitmap); | |
1137 | kfree(area); | |
1138 | } | |
1139 | ||
32cdba1e ON |
1140 | void uprobe_start_dup_mmap(void) |
1141 | { | |
1142 | percpu_down_read(&dup_mmap_sem); | |
1143 | } | |
1144 | ||
1145 | void uprobe_end_dup_mmap(void) | |
1146 | { | |
1147 | percpu_up_read(&dup_mmap_sem); | |
1148 | } | |
1149 | ||
f8ac4ec9 ON |
1150 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1151 | { | |
61559a81 ON |
1152 | newmm->uprobes_state.xol_area = NULL; |
1153 | ||
9f68f672 | 1154 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1155 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1156 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1157 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1158 | } | |
f8ac4ec9 ON |
1159 | } |
1160 | ||
d4b3b638 SD |
1161 | /* |
1162 | * - search for a free slot. | |
1163 | */ | |
1164 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1165 | { | |
1166 | unsigned long slot_addr; | |
1167 | int slot_nr; | |
1168 | ||
1169 | do { | |
1170 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1171 | if (slot_nr < UINSNS_PER_PAGE) { | |
1172 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1173 | break; | |
1174 | ||
1175 | slot_nr = UINSNS_PER_PAGE; | |
1176 | continue; | |
1177 | } | |
1178 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1179 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1180 | ||
1181 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1182 | atomic_inc(&area->slot_count); | |
1183 | ||
1184 | return slot_addr; | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * xol_get_insn_slot - If was not allocated a slot, then | |
1189 | * allocate a slot. | |
1190 | * Returns the allocated slot address or 0. | |
1191 | */ | |
1192 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) | |
1193 | { | |
1194 | struct xol_area *area; | |
1195 | unsigned long offset; | |
1196 | void *vaddr; | |
1197 | ||
1198 | area = get_xol_area(current->mm); | |
1199 | if (!area) { | |
1200 | area = xol_alloc_area(); | |
1201 | if (!area) | |
1202 | return 0; | |
1203 | } | |
1204 | current->utask->xol_vaddr = xol_take_insn_slot(area); | |
1205 | ||
1206 | /* | |
1207 | * Initialize the slot if xol_vaddr points to valid | |
1208 | * instruction slot. | |
1209 | */ | |
1210 | if (unlikely(!current->utask->xol_vaddr)) | |
1211 | return 0; | |
1212 | ||
1213 | current->utask->vaddr = slot_addr; | |
1214 | offset = current->utask->xol_vaddr & ~PAGE_MASK; | |
1215 | vaddr = kmap_atomic(area->page); | |
1216 | memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); | |
1217 | kunmap_atomic(vaddr); | |
65b6ecc0 RV |
1218 | /* |
1219 | * We probably need flush_icache_user_range() but it needs vma. | |
1220 | * This should work on supported architectures too. | |
1221 | */ | |
1222 | flush_dcache_page(area->page); | |
d4b3b638 SD |
1223 | |
1224 | return current->utask->xol_vaddr; | |
1225 | } | |
1226 | ||
1227 | /* | |
1228 | * xol_free_insn_slot - If slot was earlier allocated by | |
1229 | * @xol_get_insn_slot(), make the slot available for | |
1230 | * subsequent requests. | |
1231 | */ | |
1232 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1233 | { | |
1234 | struct xol_area *area; | |
1235 | unsigned long vma_end; | |
1236 | unsigned long slot_addr; | |
1237 | ||
1238 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1239 | return; | |
1240 | ||
1241 | slot_addr = tsk->utask->xol_vaddr; | |
1242 | ||
1243 | if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) | |
1244 | return; | |
1245 | ||
1246 | area = tsk->mm->uprobes_state.xol_area; | |
1247 | vma_end = area->vaddr + PAGE_SIZE; | |
1248 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1249 | unsigned long offset; | |
1250 | int slot_nr; | |
1251 | ||
1252 | offset = slot_addr - area->vaddr; | |
1253 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1254 | if (slot_nr >= UINSNS_PER_PAGE) | |
1255 | return; | |
1256 | ||
1257 | clear_bit(slot_nr, area->bitmap); | |
1258 | atomic_dec(&area->slot_count); | |
1259 | if (waitqueue_active(&area->wq)) | |
1260 | wake_up(&area->wq); | |
1261 | ||
1262 | tsk->utask->xol_vaddr = 0; | |
1263 | } | |
1264 | } | |
1265 | ||
0326f5a9 SD |
1266 | /** |
1267 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1268 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1269 | * instruction. | |
1270 | * Return the address of the breakpoint instruction. | |
1271 | */ | |
1272 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1273 | { | |
1274 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Called with no locks held. | |
1279 | * Called in context of a exiting or a exec-ing thread. | |
1280 | */ | |
1281 | void uprobe_free_utask(struct task_struct *t) | |
1282 | { | |
1283 | struct uprobe_task *utask = t->utask; | |
1284 | ||
0326f5a9 SD |
1285 | if (!utask) |
1286 | return; | |
1287 | ||
1288 | if (utask->active_uprobe) | |
1289 | put_uprobe(utask->active_uprobe); | |
1290 | ||
d4b3b638 | 1291 | xol_free_insn_slot(t); |
0326f5a9 SD |
1292 | kfree(utask); |
1293 | t->utask = NULL; | |
1294 | } | |
1295 | ||
1296 | /* | |
1297 | * Called in context of a new clone/fork from copy_process. | |
1298 | */ | |
1299 | void uprobe_copy_process(struct task_struct *t) | |
1300 | { | |
1301 | t->utask = NULL; | |
0326f5a9 SD |
1302 | } |
1303 | ||
1304 | /* | |
1305 | * Allocate a uprobe_task object for the task. | |
1306 | * Called when the thread hits a breakpoint for the first time. | |
1307 | * | |
1308 | * Returns: | |
1309 | * - pointer to new uprobe_task on success | |
1310 | * - NULL otherwise | |
1311 | */ | |
1312 | static struct uprobe_task *add_utask(void) | |
1313 | { | |
1314 | struct uprobe_task *utask; | |
1315 | ||
1316 | utask = kzalloc(sizeof *utask, GFP_KERNEL); | |
1317 | if (unlikely(!utask)) | |
1318 | return NULL; | |
1319 | ||
0326f5a9 SD |
1320 | current->utask = utask; |
1321 | return utask; | |
1322 | } | |
1323 | ||
1324 | /* Prepare to single-step probed instruction out of line. */ | |
1325 | static int | |
1326 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) | |
1327 | { | |
d4b3b638 SD |
1328 | if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) |
1329 | return 0; | |
1330 | ||
0326f5a9 SD |
1331 | return -EFAULT; |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * If we are singlestepping, then ensure this thread is not connected to | |
1336 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1337 | * triggers the signal, restart the original insn even if the task is | |
1338 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1339 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1340 | * _same_ instruction should be repeated again after return from the signal | |
1341 | * handler, and SSTEP can never finish in this case. | |
1342 | */ | |
1343 | bool uprobe_deny_signal(void) | |
1344 | { | |
1345 | struct task_struct *t = current; | |
1346 | struct uprobe_task *utask = t->utask; | |
1347 | ||
1348 | if (likely(!utask || !utask->active_uprobe)) | |
1349 | return false; | |
1350 | ||
1351 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1352 | ||
1353 | if (signal_pending(t)) { | |
1354 | spin_lock_irq(&t->sighand->siglock); | |
1355 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1356 | spin_unlock_irq(&t->sighand->siglock); | |
1357 | ||
1358 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1359 | utask->state = UTASK_SSTEP_TRAPPED; | |
1360 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1361 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1362 | } | |
1363 | } | |
1364 | ||
1365 | return true; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * Avoid singlestepping the original instruction if the original instruction | |
1370 | * is a NOP or can be emulated. | |
1371 | */ | |
1372 | static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) | |
1373 | { | |
71434f2f | 1374 | if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) { |
0578a970 ON |
1375 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
1376 | return true; | |
71434f2f | 1377 | clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); |
0578a970 | 1378 | } |
0326f5a9 SD |
1379 | return false; |
1380 | } | |
1381 | ||
499a4f3e ON |
1382 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1383 | { | |
1384 | struct vm_area_struct *vma; | |
1385 | ||
1386 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1387 | if (!valid_vma(vma, false)) | |
1388 | continue; | |
1389 | /* | |
1390 | * This is not strictly accurate, we can race with | |
1391 | * uprobe_unregister() and see the already removed | |
1392 | * uprobe if delete_uprobe() was not yet called. | |
1393 | */ | |
1394 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1395 | return; | |
1396 | } | |
1397 | ||
1398 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1399 | } | |
1400 | ||
ec75fba9 ON |
1401 | static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) |
1402 | { | |
1403 | struct page *page; | |
1404 | uprobe_opcode_t opcode; | |
1405 | int result; | |
1406 | ||
1407 | pagefault_disable(); | |
1408 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
1409 | sizeof(opcode)); | |
1410 | pagefault_enable(); | |
1411 | ||
1412 | if (likely(result == 0)) | |
1413 | goto out; | |
1414 | ||
1415 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); | |
1416 | if (result < 0) | |
1417 | return result; | |
1418 | ||
1419 | copy_opcode(page, vaddr, &opcode); | |
1420 | put_page(page); | |
1421 | out: | |
1422 | return is_swbp_insn(&opcode); | |
1423 | } | |
1424 | ||
d790d346 | 1425 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1426 | { |
3a9ea052 ON |
1427 | struct mm_struct *mm = current->mm; |
1428 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1429 | struct vm_area_struct *vma; |
0326f5a9 | 1430 | |
0326f5a9 SD |
1431 | down_read(&mm->mmap_sem); |
1432 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1433 | if (vma && vma->vm_start <= bp_vaddr) { |
1434 | if (valid_vma(vma, false)) { | |
cb113b47 ON |
1435 | struct inode *inode = vma->vm_file->f_mapping->host; |
1436 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); | |
0326f5a9 | 1437 | |
3a9ea052 ON |
1438 | uprobe = find_uprobe(inode, offset); |
1439 | } | |
d790d346 ON |
1440 | |
1441 | if (!uprobe) | |
1442 | *is_swbp = is_swbp_at_addr(mm, bp_vaddr); | |
1443 | } else { | |
1444 | *is_swbp = -EFAULT; | |
0326f5a9 | 1445 | } |
499a4f3e ON |
1446 | |
1447 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1448 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1449 | up_read(&mm->mmap_sem); |
1450 | ||
3a9ea052 ON |
1451 | return uprobe; |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * Run handler and ask thread to singlestep. | |
1456 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1457 | */ | |
1458 | static void handle_swbp(struct pt_regs *regs) | |
1459 | { | |
1460 | struct uprobe_task *utask; | |
1461 | struct uprobe *uprobe; | |
1462 | unsigned long bp_vaddr; | |
56bb4cf6 | 1463 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1464 | |
1465 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
d790d346 | 1466 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); |
3a9ea052 | 1467 | |
0326f5a9 | 1468 | if (!uprobe) { |
56bb4cf6 ON |
1469 | if (is_swbp > 0) { |
1470 | /* No matching uprobe; signal SIGTRAP. */ | |
1471 | send_sig(SIGTRAP, current, 0); | |
1472 | } else { | |
1473 | /* | |
1474 | * Either we raced with uprobe_unregister() or we can't | |
1475 | * access this memory. The latter is only possible if | |
1476 | * another thread plays with our ->mm. In both cases | |
1477 | * we can simply restart. If this vma was unmapped we | |
1478 | * can pretend this insn was not executed yet and get | |
1479 | * the (correct) SIGSEGV after restart. | |
1480 | */ | |
1481 | instruction_pointer_set(regs, bp_vaddr); | |
1482 | } | |
0326f5a9 SD |
1483 | return; |
1484 | } | |
142b18dd ON |
1485 | /* |
1486 | * TODO: move copy_insn/etc into _register and remove this hack. | |
1487 | * After we hit the bp, _unregister + _register can install the | |
1488 | * new and not-yet-analyzed uprobe at the same address, restart. | |
1489 | */ | |
1490 | smp_rmb(); /* pairs with wmb() in install_breakpoint() */ | |
71434f2f | 1491 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
142b18dd | 1492 | goto restart; |
0326f5a9 SD |
1493 | |
1494 | utask = current->utask; | |
1495 | if (!utask) { | |
1496 | utask = add_utask(); | |
1497 | /* Cannot allocate; re-execute the instruction. */ | |
1498 | if (!utask) | |
0578a970 | 1499 | goto restart; |
0326f5a9 | 1500 | } |
746a9e6b | 1501 | |
0326f5a9 | 1502 | handler_chain(uprobe, regs); |
0578a970 ON |
1503 | if (can_skip_sstep(uprobe, regs)) |
1504 | goto out; | |
0326f5a9 | 1505 | |
0326f5a9 | 1506 | if (!pre_ssout(uprobe, regs, bp_vaddr)) { |
746a9e6b ON |
1507 | utask->active_uprobe = uprobe; |
1508 | utask->state = UTASK_SSTEP; | |
0326f5a9 SD |
1509 | return; |
1510 | } | |
1511 | ||
0578a970 ON |
1512 | restart: |
1513 | /* | |
1514 | * cannot singlestep; cannot skip instruction; | |
1515 | * re-execute the instruction. | |
1516 | */ | |
1517 | instruction_pointer_set(regs, bp_vaddr); | |
1518 | out: | |
8bd87445 | 1519 | put_uprobe(uprobe); |
0326f5a9 SD |
1520 | } |
1521 | ||
1522 | /* | |
1523 | * Perform required fix-ups and disable singlestep. | |
1524 | * Allow pending signals to take effect. | |
1525 | */ | |
1526 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1527 | { | |
1528 | struct uprobe *uprobe; | |
1529 | ||
1530 | uprobe = utask->active_uprobe; | |
1531 | if (utask->state == UTASK_SSTEP_ACK) | |
1532 | arch_uprobe_post_xol(&uprobe->arch, regs); | |
1533 | else if (utask->state == UTASK_SSTEP_TRAPPED) | |
1534 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1535 | else | |
1536 | WARN_ON_ONCE(1); | |
1537 | ||
1538 | put_uprobe(uprobe); | |
1539 | utask->active_uprobe = NULL; | |
1540 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1541 | xol_free_insn_slot(current); |
0326f5a9 SD |
1542 | |
1543 | spin_lock_irq(¤t->sighand->siglock); | |
1544 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1545 | spin_unlock_irq(¤t->sighand->siglock); | |
1546 | } | |
1547 | ||
1548 | /* | |
1b08e907 ON |
1549 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1550 | * allows the thread to return from interrupt. After that handle_swbp() | |
1551 | * sets utask->active_uprobe. | |
0326f5a9 | 1552 | * |
1b08e907 ON |
1553 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1554 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1555 | * |
1556 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1557 | * uprobe_notify_resume(). | |
1558 | */ | |
1559 | void uprobe_notify_resume(struct pt_regs *regs) | |
1560 | { | |
1561 | struct uprobe_task *utask; | |
1562 | ||
db023ea5 ON |
1563 | clear_thread_flag(TIF_UPROBE); |
1564 | ||
0326f5a9 | 1565 | utask = current->utask; |
1b08e907 | 1566 | if (utask && utask->active_uprobe) |
0326f5a9 | 1567 | handle_singlestep(utask, regs); |
1b08e907 ON |
1568 | else |
1569 | handle_swbp(regs); | |
0326f5a9 SD |
1570 | } |
1571 | ||
1572 | /* | |
1573 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1574 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1575 | */ | |
1576 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1577 | { | |
f8ac4ec9 | 1578 | if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags)) |
0326f5a9 SD |
1579 | return 0; |
1580 | ||
0326f5a9 | 1581 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1582 | return 1; |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1587 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1588 | */ | |
1589 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1590 | { | |
1591 | struct uprobe_task *utask = current->utask; | |
1592 | ||
1593 | if (!current->mm || !utask || !utask->active_uprobe) | |
1594 | /* task is currently not uprobed */ | |
1595 | return 0; | |
1596 | ||
1597 | utask->state = UTASK_SSTEP_ACK; | |
1598 | set_thread_flag(TIF_UPROBE); | |
1599 | return 1; | |
1600 | } | |
1601 | ||
1602 | static struct notifier_block uprobe_exception_nb = { | |
1603 | .notifier_call = arch_uprobe_exception_notify, | |
1604 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1605 | }; | |
1606 | ||
2b144498 SD |
1607 | static int __init init_uprobes(void) |
1608 | { | |
1609 | int i; | |
1610 | ||
1611 | for (i = 0; i < UPROBES_HASH_SZ; i++) { | |
1612 | mutex_init(&uprobes_mutex[i]); | |
1613 | mutex_init(&uprobes_mmap_mutex[i]); | |
1614 | } | |
0326f5a9 | 1615 | |
32cdba1e ON |
1616 | if (percpu_init_rwsem(&dup_mmap_sem)) |
1617 | return -ENOMEM; | |
1618 | ||
0326f5a9 | 1619 | return register_die_notifier(&uprobe_exception_nb); |
2b144498 | 1620 | } |
0326f5a9 | 1621 | module_init(init_uprobes); |
2b144498 SD |
1622 | |
1623 | static void __exit exit_uprobes(void) | |
1624 | { | |
1625 | } | |
2b144498 | 1626 | module_exit(exit_uprobes); |