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> |
aa59c53f | 38 | #include <linux/task_work.h> |
40814f68 | 39 | #include <linux/shmem_fs.h> |
7b2d81d4 | 40 | |
2b144498 SD |
41 | #include <linux/uprobes.h> |
42 | ||
d4b3b638 SD |
43 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
44 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
45 | ||
2b144498 | 46 | static struct rb_root uprobes_tree = RB_ROOT; |
441f1eb7 ON |
47 | /* |
48 | * allows us to skip the uprobe_mmap if there are no uprobe events active | |
49 | * at this time. Probably a fine grained per inode count is better? | |
50 | */ | |
51 | #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree) | |
7b2d81d4 | 52 | |
2b144498 SD |
53 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
54 | ||
55 | #define UPROBES_HASH_SZ 13 | |
2b144498 SD |
56 | /* serialize uprobe->pending_list */ |
57 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 58 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 | 59 | |
32cdba1e ON |
60 | static struct percpu_rw_semaphore dup_mmap_sem; |
61 | ||
cb9a19fe | 62 | /* Have a copy of original instruction */ |
71434f2f | 63 | #define UPROBE_COPY_INSN 0 |
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; |
ad439356 ON |
75 | |
76 | /* | |
77 | * The generic code assumes that it has two members of unknown type | |
78 | * owned by the arch-specific code: | |
79 | * | |
80 | * insn - copy_insn() saves the original instruction here for | |
81 | * arch_uprobe_analyze_insn(). | |
82 | * | |
83 | * ixol - potentially modified instruction to execute out of | |
84 | * line, copied to xol_area by xol_get_insn_slot(). | |
85 | */ | |
3ff54efd SD |
86 | struct arch_uprobe arch; |
87 | }; | |
88 | ||
c912dae6 | 89 | /* |
ad439356 ON |
90 | * Execute out of line area: anonymous executable mapping installed |
91 | * by the probed task to execute the copy of the original instruction | |
92 | * mangled by set_swbp(). | |
93 | * | |
c912dae6 ON |
94 | * On a breakpoint hit, thread contests for a slot. It frees the |
95 | * slot after singlestep. Currently a fixed number of slots are | |
96 | * allocated. | |
97 | */ | |
98 | struct xol_area { | |
99 | wait_queue_head_t wq; /* if all slots are busy */ | |
100 | atomic_t slot_count; /* number of in-use slots */ | |
101 | unsigned long *bitmap; /* 0 = free slot */ | |
f58bea2f | 102 | struct page *pages[2]; |
c912dae6 ON |
103 | |
104 | /* | |
105 | * We keep the vma's vm_start rather than a pointer to the vma | |
106 | * itself. The probed process or a naughty kernel module could make | |
107 | * the vma go away, and we must handle that reasonably gracefully. | |
108 | */ | |
109 | unsigned long vaddr; /* Page(s) of instruction slots */ | |
110 | }; | |
111 | ||
2b144498 SD |
112 | /* |
113 | * valid_vma: Verify if the specified vma is an executable vma | |
114 | * Relax restrictions while unregistering: vm_flags might have | |
115 | * changed after breakpoint was inserted. | |
116 | * - is_register: indicates if we are in register context. | |
117 | * - Return 1 if the specified virtual address is in an | |
118 | * executable vma. | |
119 | */ | |
120 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
121 | { | |
13f59c5e | 122 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE; |
2b144498 | 123 | |
e40cfce6 ON |
124 | if (is_register) |
125 | flags |= VM_WRITE; | |
2b144498 | 126 | |
e40cfce6 | 127 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
128 | } |
129 | ||
57683f72 | 130 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 131 | { |
57683f72 | 132 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
133 | } |
134 | ||
cb113b47 ON |
135 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
136 | { | |
137 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
138 | } | |
139 | ||
2b144498 SD |
140 | /** |
141 | * __replace_page - replace page in vma by new page. | |
142 | * based on replace_page in mm/ksm.c | |
143 | * | |
144 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 145 | * @addr: address the old @page is mapped at |
2b144498 SD |
146 | * @page: the cowed page we are replacing by kpage |
147 | * @kpage: the modified page we replace page by | |
148 | * | |
149 | * Returns 0 on success, -EFAULT on failure. | |
150 | */ | |
c517ee74 ON |
151 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
152 | struct page *page, struct page *kpage) | |
2b144498 SD |
153 | { |
154 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
155 | spinlock_t *ptl; |
156 | pte_t *ptep; | |
9f92448c | 157 | int err; |
6bdb913f HE |
158 | /* For mmu_notifiers */ |
159 | const unsigned long mmun_start = addr; | |
160 | const unsigned long mmun_end = addr + PAGE_SIZE; | |
00501b53 JW |
161 | struct mem_cgroup *memcg; |
162 | ||
163 | err = mem_cgroup_try_charge(kpage, vma->vm_mm, GFP_KERNEL, &memcg); | |
164 | if (err) | |
165 | return err; | |
2b144498 | 166 | |
194f8dcb | 167 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
168 | lock_page(page); |
169 | ||
6bdb913f | 170 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
9f92448c | 171 | err = -EAGAIN; |
5323ce71 | 172 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 173 | if (!ptep) |
9f92448c | 174 | goto unlock; |
2b144498 SD |
175 | |
176 | get_page(kpage); | |
177 | page_add_new_anon_rmap(kpage, vma, addr); | |
00501b53 JW |
178 | mem_cgroup_commit_charge(kpage, memcg, false); |
179 | lru_cache_add_active_or_unevictable(kpage, vma); | |
2b144498 | 180 | |
7396fa81 SD |
181 | if (!PageAnon(page)) { |
182 | dec_mm_counter(mm, MM_FILEPAGES); | |
183 | inc_mm_counter(mm, MM_ANONPAGES); | |
184 | } | |
185 | ||
2b144498 | 186 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
34ee645e | 187 | ptep_clear_flush_notify(vma, addr, ptep); |
2b144498 SD |
188 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); |
189 | ||
190 | page_remove_rmap(page); | |
191 | if (!page_mapped(page)) | |
192 | try_to_free_swap(page); | |
2b144498 | 193 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 194 | |
194f8dcb ON |
195 | if (vma->vm_flags & VM_LOCKED) |
196 | munlock_vma_page(page); | |
197 | put_page(page); | |
198 | ||
9f92448c ON |
199 | err = 0; |
200 | unlock: | |
00501b53 | 201 | mem_cgroup_cancel_charge(kpage, memcg); |
6bdb913f | 202 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
9f92448c ON |
203 | unlock_page(page); |
204 | return err; | |
2b144498 SD |
205 | } |
206 | ||
207 | /** | |
5cb4ac3a | 208 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 209 | * @insn: instruction to be checked. |
5cb4ac3a | 210 | * Default implementation of is_swbp_insn |
2b144498 SD |
211 | * Returns true if @insn is a breakpoint instruction. |
212 | */ | |
5cb4ac3a | 213 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 214 | { |
5cb4ac3a | 215 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
216 | } |
217 | ||
0908ad6e AM |
218 | /** |
219 | * is_trap_insn - check if instruction is breakpoint instruction. | |
220 | * @insn: instruction to be checked. | |
221 | * Default implementation of is_trap_insn | |
222 | * Returns true if @insn is a breakpoint instruction. | |
223 | * | |
224 | * This function is needed for the case where an architecture has multiple | |
225 | * trap instructions (like powerpc). | |
226 | */ | |
227 | bool __weak is_trap_insn(uprobe_opcode_t *insn) | |
228 | { | |
229 | return is_swbp_insn(insn); | |
230 | } | |
231 | ||
ab0d805c | 232 | static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len) |
cceb55aa ON |
233 | { |
234 | void *kaddr = kmap_atomic(page); | |
ab0d805c | 235 | memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len); |
cceb55aa ON |
236 | kunmap_atomic(kaddr); |
237 | } | |
238 | ||
5669ccee ON |
239 | static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len) |
240 | { | |
241 | void *kaddr = kmap_atomic(page); | |
242 | memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len); | |
243 | kunmap_atomic(kaddr); | |
244 | } | |
245 | ||
ed6f6a50 ON |
246 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
247 | { | |
248 | uprobe_opcode_t old_opcode; | |
249 | bool is_swbp; | |
250 | ||
0908ad6e AM |
251 | /* |
252 | * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here. | |
253 | * We do not check if it is any other 'trap variant' which could | |
254 | * be conditional trap instruction such as the one powerpc supports. | |
255 | * | |
256 | * The logic is that we do not care if the underlying instruction | |
257 | * is a trap variant; uprobes always wins over any other (gdb) | |
258 | * breakpoint. | |
259 | */ | |
ab0d805c | 260 | copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE); |
ed6f6a50 ON |
261 | is_swbp = is_swbp_insn(&old_opcode); |
262 | ||
263 | if (is_swbp_insn(new_opcode)) { | |
264 | if (is_swbp) /* register: already installed? */ | |
265 | return 0; | |
266 | } else { | |
267 | if (!is_swbp) /* unregister: was it changed by us? */ | |
076a365b | 268 | return 0; |
ed6f6a50 ON |
269 | } |
270 | ||
271 | return 1; | |
272 | } | |
273 | ||
2b144498 SD |
274 | /* |
275 | * NOTE: | |
276 | * Expect the breakpoint instruction to be the smallest size instruction for | |
277 | * the architecture. If an arch has variable length instruction and the | |
278 | * breakpoint instruction is not of the smallest length instruction | |
0908ad6e | 279 | * supported by that architecture then we need to modify is_trap_at_addr and |
f72d41fa ON |
280 | * uprobe_write_opcode accordingly. This would never be a problem for archs |
281 | * that have fixed length instructions. | |
29dedee0 | 282 | * |
f72d41fa | 283 | * uprobe_write_opcode - write the opcode at a given virtual address. |
2b144498 | 284 | * @mm: the probed process address space. |
2b144498 SD |
285 | * @vaddr: the virtual address to store the opcode. |
286 | * @opcode: opcode to be written at @vaddr. | |
287 | * | |
29dedee0 | 288 | * Called with mm->mmap_sem held for write. |
2b144498 SD |
289 | * Return 0 (success) or a negative errno. |
290 | */ | |
f72d41fa | 291 | int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr, |
cceb55aa | 292 | uprobe_opcode_t opcode) |
2b144498 SD |
293 | { |
294 | struct page *old_page, *new_page; | |
2b144498 | 295 | struct vm_area_struct *vma; |
2b144498 | 296 | int ret; |
f403072c | 297 | |
5323ce71 | 298 | retry: |
2b144498 | 299 | /* Read the page with vaddr into memory */ |
75ed82ea | 300 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
301 | if (ret <= 0) |
302 | return ret; | |
7b2d81d4 | 303 | |
ed6f6a50 ON |
304 | ret = verify_opcode(old_page, vaddr, &opcode); |
305 | if (ret <= 0) | |
306 | goto put_old; | |
307 | ||
29dedee0 ON |
308 | ret = anon_vma_prepare(vma); |
309 | if (ret) | |
310 | goto put_old; | |
311 | ||
2b144498 SD |
312 | ret = -ENOMEM; |
313 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
314 | if (!new_page) | |
9f92448c | 315 | goto put_old; |
2b144498 | 316 | |
29dedee0 | 317 | __SetPageUptodate(new_page); |
3f47107c ON |
318 | copy_highpage(new_page, old_page); |
319 | copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); | |
2b144498 | 320 | |
c517ee74 | 321 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 322 | page_cache_release(new_page); |
9f92448c | 323 | put_old: |
7b2d81d4 IM |
324 | put_page(old_page); |
325 | ||
5323ce71 ON |
326 | if (unlikely(ret == -EAGAIN)) |
327 | goto retry; | |
2b144498 SD |
328 | return ret; |
329 | } | |
330 | ||
2b144498 | 331 | /** |
5cb4ac3a | 332 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 333 | * @auprobe: arch specific probepoint information. |
2b144498 | 334 | * @mm: the probed process address space. |
2b144498 SD |
335 | * @vaddr: the virtual address to insert the opcode. |
336 | * | |
337 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
338 | * Return 0 (success) or a negative errno. | |
339 | */ | |
5cb4ac3a | 340 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 341 | { |
f72d41fa | 342 | return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
343 | } |
344 | ||
345 | /** | |
346 | * set_orig_insn - Restore the original instruction. | |
347 | * @mm: the probed process address space. | |
e3343e6a | 348 | * @auprobe: arch specific probepoint information. |
2b144498 | 349 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
350 | * |
351 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
352 | * Return 0 (success) or a negative errno. | |
353 | */ | |
7b2d81d4 | 354 | int __weak |
ded86e7c | 355 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 356 | { |
803200e2 | 357 | return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn); |
2b144498 SD |
358 | } |
359 | ||
f231722a ON |
360 | static struct uprobe *get_uprobe(struct uprobe *uprobe) |
361 | { | |
362 | atomic_inc(&uprobe->ref); | |
363 | return uprobe; | |
364 | } | |
365 | ||
366 | static void put_uprobe(struct uprobe *uprobe) | |
367 | { | |
368 | if (atomic_dec_and_test(&uprobe->ref)) | |
369 | kfree(uprobe); | |
370 | } | |
371 | ||
2b144498 SD |
372 | static int match_uprobe(struct uprobe *l, struct uprobe *r) |
373 | { | |
374 | if (l->inode < r->inode) | |
375 | return -1; | |
7b2d81d4 | 376 | |
2b144498 SD |
377 | if (l->inode > r->inode) |
378 | return 1; | |
2b144498 | 379 | |
7b2d81d4 IM |
380 | if (l->offset < r->offset) |
381 | return -1; | |
382 | ||
383 | if (l->offset > r->offset) | |
384 | return 1; | |
2b144498 SD |
385 | |
386 | return 0; | |
387 | } | |
388 | ||
389 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
390 | { | |
391 | struct uprobe u = { .inode = inode, .offset = offset }; | |
392 | struct rb_node *n = uprobes_tree.rb_node; | |
393 | struct uprobe *uprobe; | |
394 | int match; | |
395 | ||
396 | while (n) { | |
397 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
398 | match = match_uprobe(&u, uprobe); | |
f231722a ON |
399 | if (!match) |
400 | return get_uprobe(uprobe); | |
7b2d81d4 | 401 | |
2b144498 SD |
402 | if (match < 0) |
403 | n = n->rb_left; | |
404 | else | |
405 | n = n->rb_right; | |
406 | } | |
407 | return NULL; | |
408 | } | |
409 | ||
410 | /* | |
411 | * Find a uprobe corresponding to a given inode:offset | |
412 | * Acquires uprobes_treelock | |
413 | */ | |
414 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
415 | { | |
416 | struct uprobe *uprobe; | |
2b144498 | 417 | |
6f47caa0 | 418 | spin_lock(&uprobes_treelock); |
2b144498 | 419 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 420 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 421 | |
2b144498 SD |
422 | return uprobe; |
423 | } | |
424 | ||
425 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
426 | { | |
427 | struct rb_node **p = &uprobes_tree.rb_node; | |
428 | struct rb_node *parent = NULL; | |
429 | struct uprobe *u; | |
430 | int match; | |
431 | ||
432 | while (*p) { | |
433 | parent = *p; | |
434 | u = rb_entry(parent, struct uprobe, rb_node); | |
435 | match = match_uprobe(uprobe, u); | |
f231722a ON |
436 | if (!match) |
437 | return get_uprobe(u); | |
2b144498 SD |
438 | |
439 | if (match < 0) | |
440 | p = &parent->rb_left; | |
441 | else | |
442 | p = &parent->rb_right; | |
443 | ||
444 | } | |
7b2d81d4 | 445 | |
2b144498 SD |
446 | u = NULL; |
447 | rb_link_node(&uprobe->rb_node, parent, p); | |
448 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
449 | /* get access + creation ref */ | |
450 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 451 | |
2b144498 SD |
452 | return u; |
453 | } | |
454 | ||
455 | /* | |
7b2d81d4 | 456 | * Acquire uprobes_treelock. |
2b144498 SD |
457 | * Matching uprobe already exists in rbtree; |
458 | * increment (access refcount) and return the matching uprobe. | |
459 | * | |
460 | * No matching uprobe; insert the uprobe in rb_tree; | |
461 | * get a double refcount (access + creation) and return NULL. | |
462 | */ | |
463 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
464 | { | |
2b144498 SD |
465 | struct uprobe *u; |
466 | ||
6f47caa0 | 467 | spin_lock(&uprobes_treelock); |
2b144498 | 468 | u = __insert_uprobe(uprobe); |
6f47caa0 | 469 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 470 | |
2b144498 SD |
471 | return u; |
472 | } | |
473 | ||
2b144498 SD |
474 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) |
475 | { | |
476 | struct uprobe *uprobe, *cur_uprobe; | |
477 | ||
478 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
479 | if (!uprobe) | |
480 | return NULL; | |
481 | ||
482 | uprobe->inode = igrab(inode); | |
483 | uprobe->offset = offset; | |
e591c8d7 | 484 | init_rwsem(&uprobe->register_rwsem); |
2b144498 | 485 | init_rwsem(&uprobe->consumer_rwsem); |
2b144498 SD |
486 | |
487 | /* add to uprobes_tree, sorted on inode:offset */ | |
488 | cur_uprobe = insert_uprobe(uprobe); | |
2b144498 SD |
489 | /* a uprobe exists for this inode:offset combination */ |
490 | if (cur_uprobe) { | |
491 | kfree(uprobe); | |
492 | uprobe = cur_uprobe; | |
493 | iput(inode); | |
7b2d81d4 IM |
494 | } |
495 | ||
2b144498 SD |
496 | return uprobe; |
497 | } | |
498 | ||
9a98e03c | 499 | static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
500 | { |
501 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
502 | uc->next = uprobe->consumers; |
503 | uprobe->consumers = uc; | |
2b144498 | 504 | up_write(&uprobe->consumer_rwsem); |
2b144498 SD |
505 | } |
506 | ||
507 | /* | |
e3343e6a SD |
508 | * For uprobe @uprobe, delete the consumer @uc. |
509 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
510 | * or return false. |
511 | */ | |
e3343e6a | 512 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
513 | { |
514 | struct uprobe_consumer **con; | |
515 | bool ret = false; | |
516 | ||
517 | down_write(&uprobe->consumer_rwsem); | |
518 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
519 | if (*con == uc) { |
520 | *con = uc->next; | |
2b144498 SD |
521 | ret = true; |
522 | break; | |
523 | } | |
524 | } | |
525 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 526 | |
2b144498 SD |
527 | return ret; |
528 | } | |
529 | ||
2ded0980 ON |
530 | static int __copy_insn(struct address_space *mapping, struct file *filp, |
531 | void *insn, int nbytes, loff_t offset) | |
2b144498 | 532 | { |
2b144498 | 533 | struct page *page; |
2b144498 | 534 | /* |
40814f68 ON |
535 | * Ensure that the page that has the original instruction is populated |
536 | * and in page-cache. If ->readpage == NULL it must be shmem_mapping(), | |
537 | * see uprobe_register(). | |
2b144498 | 538 | */ |
40814f68 ON |
539 | if (mapping->a_ops->readpage) |
540 | page = read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT, filp); | |
541 | else | |
542 | page = shmem_read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT); | |
2b144498 SD |
543 | if (IS_ERR(page)) |
544 | return PTR_ERR(page); | |
545 | ||
2edb7b55 | 546 | copy_from_page(page, offset, insn, nbytes); |
2b144498 | 547 | page_cache_release(page); |
7b2d81d4 | 548 | |
2b144498 SD |
549 | return 0; |
550 | } | |
551 | ||
d436615e | 552 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 | 553 | { |
2ded0980 ON |
554 | struct address_space *mapping = uprobe->inode->i_mapping; |
555 | loff_t offs = uprobe->offset; | |
803200e2 ON |
556 | void *insn = &uprobe->arch.insn; |
557 | int size = sizeof(uprobe->arch.insn); | |
2ded0980 ON |
558 | int len, err = -EIO; |
559 | ||
560 | /* Copy only available bytes, -EIO if nothing was read */ | |
561 | do { | |
562 | if (offs >= i_size_read(uprobe->inode)) | |
563 | break; | |
564 | ||
565 | len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK)); | |
566 | err = __copy_insn(mapping, filp, insn, len, offs); | |
fc36f595 | 567 | if (err) |
2ded0980 ON |
568 | break; |
569 | ||
570 | insn += len; | |
571 | offs += len; | |
572 | size -= len; | |
573 | } while (size); | |
574 | ||
575 | return err; | |
2b144498 SD |
576 | } |
577 | ||
cb9a19fe ON |
578 | static int prepare_uprobe(struct uprobe *uprobe, struct file *file, |
579 | struct mm_struct *mm, unsigned long vaddr) | |
580 | { | |
581 | int ret = 0; | |
582 | ||
71434f2f | 583 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
cb9a19fe ON |
584 | return ret; |
585 | ||
d4d3ccc6 ON |
586 | /* TODO: move this into _register, until then we abuse this sem. */ |
587 | down_write(&uprobe->consumer_rwsem); | |
71434f2f | 588 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
589 | goto out; |
590 | ||
cb9a19fe ON |
591 | ret = copy_insn(uprobe, file); |
592 | if (ret) | |
593 | goto out; | |
594 | ||
595 | ret = -ENOTSUPP; | |
803200e2 | 596 | if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn)) |
cb9a19fe ON |
597 | goto out; |
598 | ||
599 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
600 | if (ret) | |
601 | goto out; | |
602 | ||
f72d41fa | 603 | /* uprobe_write_opcode() assumes we don't cross page boundary */ |
cb9a19fe ON |
604 | BUG_ON((uprobe->offset & ~PAGE_MASK) + |
605 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
606 | ||
607 | smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ | |
71434f2f | 608 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
609 | |
610 | out: | |
d4d3ccc6 | 611 | up_write(&uprobe->consumer_rwsem); |
4710f05f | 612 | |
cb9a19fe ON |
613 | return ret; |
614 | } | |
615 | ||
8a7f2fa0 ON |
616 | static inline bool consumer_filter(struct uprobe_consumer *uc, |
617 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
806a98bd | 618 | { |
8a7f2fa0 | 619 | return !uc->filter || uc->filter(uc, ctx, mm); |
806a98bd ON |
620 | } |
621 | ||
8a7f2fa0 ON |
622 | static bool filter_chain(struct uprobe *uprobe, |
623 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
63633cbf | 624 | { |
1ff6fee5 ON |
625 | struct uprobe_consumer *uc; |
626 | bool ret = false; | |
627 | ||
628 | down_read(&uprobe->consumer_rwsem); | |
629 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
8a7f2fa0 | 630 | ret = consumer_filter(uc, ctx, mm); |
1ff6fee5 ON |
631 | if (ret) |
632 | break; | |
633 | } | |
634 | up_read(&uprobe->consumer_rwsem); | |
635 | ||
636 | return ret; | |
63633cbf ON |
637 | } |
638 | ||
e3343e6a SD |
639 | static int |
640 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 641 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 642 | { |
f8ac4ec9 | 643 | bool first_uprobe; |
2b144498 SD |
644 | int ret; |
645 | ||
cb9a19fe ON |
646 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
647 | if (ret) | |
648 | return ret; | |
682968e0 | 649 | |
f8ac4ec9 ON |
650 | /* |
651 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
652 | * the task can hit this breakpoint right after __replace_page(). | |
653 | */ | |
654 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
655 | if (first_uprobe) | |
656 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
657 | ||
816c03fb | 658 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
659 | if (!ret) |
660 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
661 | else if (first_uprobe) | |
f8ac4ec9 | 662 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
663 | |
664 | return ret; | |
665 | } | |
666 | ||
076a365b | 667 | static int |
816c03fb | 668 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 669 | { |
9f68f672 | 670 | set_bit(MMF_RECALC_UPROBES, &mm->flags); |
076a365b | 671 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
672 | } |
673 | ||
06b7bcd8 ON |
674 | static inline bool uprobe_is_active(struct uprobe *uprobe) |
675 | { | |
676 | return !RB_EMPTY_NODE(&uprobe->rb_node); | |
677 | } | |
0326f5a9 | 678 | /* |
778b032d ON |
679 | * There could be threads that have already hit the breakpoint. They |
680 | * will recheck the current insn and restart if find_uprobe() fails. | |
681 | * See find_active_uprobe(). | |
0326f5a9 | 682 | */ |
2b144498 SD |
683 | static void delete_uprobe(struct uprobe *uprobe) |
684 | { | |
06b7bcd8 ON |
685 | if (WARN_ON(!uprobe_is_active(uprobe))) |
686 | return; | |
687 | ||
6f47caa0 | 688 | spin_lock(&uprobes_treelock); |
2b144498 | 689 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 690 | spin_unlock(&uprobes_treelock); |
06b7bcd8 | 691 | RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */ |
2b144498 SD |
692 | iput(uprobe->inode); |
693 | put_uprobe(uprobe); | |
2b144498 SD |
694 | } |
695 | ||
26872090 ON |
696 | struct map_info { |
697 | struct map_info *next; | |
698 | struct mm_struct *mm; | |
816c03fb | 699 | unsigned long vaddr; |
26872090 ON |
700 | }; |
701 | ||
702 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 703 | { |
26872090 ON |
704 | struct map_info *next = info->next; |
705 | kfree(info); | |
706 | return next; | |
707 | } | |
708 | ||
709 | static struct map_info * | |
710 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
711 | { | |
712 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 713 | struct vm_area_struct *vma; |
26872090 ON |
714 | struct map_info *curr = NULL; |
715 | struct map_info *prev = NULL; | |
716 | struct map_info *info; | |
717 | int more = 0; | |
2b144498 | 718 | |
26872090 | 719 | again: |
4a23717a | 720 | i_mmap_lock_read(mapping); |
6b2dbba8 | 721 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
722 | if (!valid_vma(vma, is_register)) |
723 | continue; | |
724 | ||
7a5bfb66 ON |
725 | if (!prev && !more) { |
726 | /* | |
c8c06efa | 727 | * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through |
7a5bfb66 ON |
728 | * reclaim. This is optimistic, no harm done if it fails. |
729 | */ | |
730 | prev = kmalloc(sizeof(struct map_info), | |
731 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
732 | if (prev) | |
733 | prev->next = NULL; | |
734 | } | |
26872090 ON |
735 | if (!prev) { |
736 | more++; | |
737 | continue; | |
2b144498 | 738 | } |
2b144498 | 739 | |
26872090 ON |
740 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
741 | continue; | |
7b2d81d4 | 742 | |
26872090 ON |
743 | info = prev; |
744 | prev = prev->next; | |
745 | info->next = curr; | |
746 | curr = info; | |
2b144498 | 747 | |
26872090 | 748 | info->mm = vma->vm_mm; |
57683f72 | 749 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 750 | } |
4a23717a | 751 | i_mmap_unlock_read(mapping); |
2b144498 | 752 | |
26872090 ON |
753 | if (!more) |
754 | goto out; | |
755 | ||
756 | prev = curr; | |
757 | while (curr) { | |
758 | mmput(curr->mm); | |
759 | curr = curr->next; | |
760 | } | |
7b2d81d4 | 761 | |
26872090 ON |
762 | do { |
763 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
764 | if (!info) { | |
765 | curr = ERR_PTR(-ENOMEM); | |
766 | goto out; | |
767 | } | |
768 | info->next = prev; | |
769 | prev = info; | |
770 | } while (--more); | |
771 | ||
772 | goto again; | |
773 | out: | |
774 | while (prev) | |
775 | prev = free_map_info(prev); | |
776 | return curr; | |
2b144498 SD |
777 | } |
778 | ||
bdf8647c ON |
779 | static int |
780 | register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) | |
2b144498 | 781 | { |
bdf8647c | 782 | bool is_register = !!new; |
26872090 ON |
783 | struct map_info *info; |
784 | int err = 0; | |
2b144498 | 785 | |
32cdba1e | 786 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
787 | info = build_map_info(uprobe->inode->i_mapping, |
788 | uprobe->offset, is_register); | |
32cdba1e ON |
789 | if (IS_ERR(info)) { |
790 | err = PTR_ERR(info); | |
791 | goto out; | |
792 | } | |
7b2d81d4 | 793 | |
26872090 ON |
794 | while (info) { |
795 | struct mm_struct *mm = info->mm; | |
796 | struct vm_area_struct *vma; | |
7b2d81d4 | 797 | |
076a365b | 798 | if (err && is_register) |
26872090 | 799 | goto free; |
7b2d81d4 | 800 | |
77fc4af1 | 801 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
802 | vma = find_vma(mm, info->vaddr); |
803 | if (!vma || !valid_vma(vma, is_register) || | |
f281769e | 804 | file_inode(vma->vm_file) != uprobe->inode) |
26872090 ON |
805 | goto unlock; |
806 | ||
f4d6dfe5 ON |
807 | if (vma->vm_start > info->vaddr || |
808 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 809 | goto unlock; |
2b144498 | 810 | |
806a98bd ON |
811 | if (is_register) { |
812 | /* consult only the "caller", new consumer. */ | |
bdf8647c | 813 | if (consumer_filter(new, |
8a7f2fa0 | 814 | UPROBE_FILTER_REGISTER, mm)) |
806a98bd ON |
815 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
816 | } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) { | |
8a7f2fa0 ON |
817 | if (!filter_chain(uprobe, |
818 | UPROBE_FILTER_UNREGISTER, mm)) | |
806a98bd ON |
819 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
820 | } | |
78f74116 | 821 | |
26872090 ON |
822 | unlock: |
823 | up_write(&mm->mmap_sem); | |
824 | free: | |
825 | mmput(mm); | |
826 | info = free_map_info(info); | |
2b144498 | 827 | } |
32cdba1e ON |
828 | out: |
829 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 830 | return err; |
2b144498 SD |
831 | } |
832 | ||
9a98e03c | 833 | static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 834 | { |
9a98e03c | 835 | consumer_add(uprobe, uc); |
bdf8647c | 836 | return register_for_each_vma(uprobe, uc); |
2b144498 SD |
837 | } |
838 | ||
04aab9b2 | 839 | static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 840 | { |
04aab9b2 ON |
841 | int err; |
842 | ||
06d07139 | 843 | if (WARN_ON(!consumer_del(uprobe, uc))) |
04aab9b2 | 844 | return; |
2b144498 | 845 | |
bdf8647c | 846 | err = register_for_each_vma(uprobe, NULL); |
bb929284 ON |
847 | /* TODO : cant unregister? schedule a worker thread */ |
848 | if (!uprobe->consumers && !err) | |
849 | delete_uprobe(uprobe); | |
2b144498 SD |
850 | } |
851 | ||
852 | /* | |
7b2d81d4 | 853 | * uprobe_register - register a probe |
2b144498 SD |
854 | * @inode: the file in which the probe has to be placed. |
855 | * @offset: offset from the start of the file. | |
e3343e6a | 856 | * @uc: information on howto handle the probe.. |
2b144498 | 857 | * |
7b2d81d4 | 858 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
859 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
860 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 861 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 862 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 863 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
864 | * unregisters. |
865 | * | |
866 | * Return errno if it cannot successully install probes | |
867 | * else return 0 (success) | |
868 | */ | |
e3343e6a | 869 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
870 | { |
871 | struct uprobe *uprobe; | |
7b2d81d4 | 872 | int ret; |
2b144498 | 873 | |
ea024870 AA |
874 | /* Uprobe must have at least one set consumer */ |
875 | if (!uc->handler && !uc->ret_handler) | |
876 | return -EINVAL; | |
877 | ||
40814f68 ON |
878 | /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */ |
879 | if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping)) | |
41ccba02 | 880 | return -EIO; |
f0744af7 | 881 | /* Racy, just to catch the obvious mistakes */ |
2b144498 | 882 | if (offset > i_size_read(inode)) |
7b2d81d4 | 883 | return -EINVAL; |
2b144498 | 884 | |
66d06dff | 885 | retry: |
2b144498 | 886 | uprobe = alloc_uprobe(inode, offset); |
66d06dff ON |
887 | if (!uprobe) |
888 | return -ENOMEM; | |
889 | /* | |
890 | * We can race with uprobe_unregister()->delete_uprobe(). | |
891 | * Check uprobe_is_active() and retry if it is false. | |
892 | */ | |
893 | down_write(&uprobe->register_rwsem); | |
894 | ret = -EAGAIN; | |
895 | if (likely(uprobe_is_active(uprobe))) { | |
9a98e03c ON |
896 | ret = __uprobe_register(uprobe, uc); |
897 | if (ret) | |
04aab9b2 | 898 | __uprobe_unregister(uprobe, uc); |
2b144498 | 899 | } |
66d06dff ON |
900 | up_write(&uprobe->register_rwsem); |
901 | put_uprobe(uprobe); | |
2b144498 | 902 | |
66d06dff ON |
903 | if (unlikely(ret == -EAGAIN)) |
904 | goto retry; | |
2b144498 SD |
905 | return ret; |
906 | } | |
e8440c14 | 907 | EXPORT_SYMBOL_GPL(uprobe_register); |
2b144498 | 908 | |
bdf8647c ON |
909 | /* |
910 | * uprobe_apply - unregister a already registered probe. | |
911 | * @inode: the file in which the probe has to be removed. | |
912 | * @offset: offset from the start of the file. | |
913 | * @uc: consumer which wants to add more or remove some breakpoints | |
914 | * @add: add or remove the breakpoints | |
915 | */ | |
916 | int uprobe_apply(struct inode *inode, loff_t offset, | |
917 | struct uprobe_consumer *uc, bool add) | |
918 | { | |
919 | struct uprobe *uprobe; | |
920 | struct uprobe_consumer *con; | |
921 | int ret = -ENOENT; | |
922 | ||
923 | uprobe = find_uprobe(inode, offset); | |
06d07139 | 924 | if (WARN_ON(!uprobe)) |
bdf8647c ON |
925 | return ret; |
926 | ||
927 | down_write(&uprobe->register_rwsem); | |
928 | for (con = uprobe->consumers; con && con != uc ; con = con->next) | |
929 | ; | |
930 | if (con) | |
931 | ret = register_for_each_vma(uprobe, add ? uc : NULL); | |
932 | up_write(&uprobe->register_rwsem); | |
933 | put_uprobe(uprobe); | |
934 | ||
935 | return ret; | |
936 | } | |
937 | ||
2b144498 | 938 | /* |
7b2d81d4 | 939 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
940 | * @inode: the file in which the probe has to be removed. |
941 | * @offset: offset from the start of the file. | |
e3343e6a | 942 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 943 | */ |
e3343e6a | 944 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 945 | { |
7b2d81d4 | 946 | struct uprobe *uprobe; |
2b144498 | 947 | |
2b144498 | 948 | uprobe = find_uprobe(inode, offset); |
06d07139 | 949 | if (WARN_ON(!uprobe)) |
2b144498 SD |
950 | return; |
951 | ||
e591c8d7 | 952 | down_write(&uprobe->register_rwsem); |
04aab9b2 | 953 | __uprobe_unregister(uprobe, uc); |
e591c8d7 | 954 | up_write(&uprobe->register_rwsem); |
c91368c4 | 955 | put_uprobe(uprobe); |
2b144498 | 956 | } |
e8440c14 | 957 | EXPORT_SYMBOL_GPL(uprobe_unregister); |
2b144498 | 958 | |
da1816b1 ON |
959 | static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm) |
960 | { | |
961 | struct vm_area_struct *vma; | |
962 | int err = 0; | |
963 | ||
964 | down_read(&mm->mmap_sem); | |
965 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
966 | unsigned long vaddr; | |
967 | loff_t offset; | |
968 | ||
969 | if (!valid_vma(vma, false) || | |
f281769e | 970 | file_inode(vma->vm_file) != uprobe->inode) |
da1816b1 ON |
971 | continue; |
972 | ||
973 | offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT; | |
974 | if (uprobe->offset < offset || | |
975 | uprobe->offset >= offset + vma->vm_end - vma->vm_start) | |
976 | continue; | |
977 | ||
978 | vaddr = offset_to_vaddr(vma, uprobe->offset); | |
979 | err |= remove_breakpoint(uprobe, mm, vaddr); | |
980 | } | |
981 | up_read(&mm->mmap_sem); | |
982 | ||
983 | return err; | |
984 | } | |
985 | ||
891c3970 ON |
986 | static struct rb_node * |
987 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 988 | { |
2b144498 | 989 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
990 | |
991 | while (n) { | |
891c3970 | 992 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 993 | |
891c3970 | 994 | if (inode < u->inode) { |
2b144498 | 995 | n = n->rb_left; |
891c3970 | 996 | } else if (inode > u->inode) { |
2b144498 | 997 | n = n->rb_right; |
891c3970 ON |
998 | } else { |
999 | if (max < u->offset) | |
1000 | n = n->rb_left; | |
1001 | else if (min > u->offset) | |
1002 | n = n->rb_right; | |
1003 | else | |
1004 | break; | |
1005 | } | |
2b144498 | 1006 | } |
7b2d81d4 | 1007 | |
891c3970 | 1008 | return n; |
2b144498 SD |
1009 | } |
1010 | ||
1011 | /* | |
891c3970 | 1012 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 1013 | */ |
891c3970 ON |
1014 | static void build_probe_list(struct inode *inode, |
1015 | struct vm_area_struct *vma, | |
1016 | unsigned long start, unsigned long end, | |
1017 | struct list_head *head) | |
2b144498 | 1018 | { |
891c3970 | 1019 | loff_t min, max; |
891c3970 ON |
1020 | struct rb_node *n, *t; |
1021 | struct uprobe *u; | |
7b2d81d4 | 1022 | |
891c3970 | 1023 | INIT_LIST_HEAD(head); |
cb113b47 | 1024 | min = vaddr_to_offset(vma, start); |
891c3970 | 1025 | max = min + (end - start) - 1; |
2b144498 | 1026 | |
6f47caa0 | 1027 | spin_lock(&uprobes_treelock); |
891c3970 ON |
1028 | n = find_node_in_range(inode, min, max); |
1029 | if (n) { | |
1030 | for (t = n; t; t = rb_prev(t)) { | |
1031 | u = rb_entry(t, struct uprobe, rb_node); | |
1032 | if (u->inode != inode || u->offset < min) | |
1033 | break; | |
1034 | list_add(&u->pending_list, head); | |
f231722a | 1035 | get_uprobe(u); |
891c3970 ON |
1036 | } |
1037 | for (t = n; (t = rb_next(t)); ) { | |
1038 | u = rb_entry(t, struct uprobe, rb_node); | |
1039 | if (u->inode != inode || u->offset > max) | |
1040 | break; | |
1041 | list_add(&u->pending_list, head); | |
f231722a | 1042 | get_uprobe(u); |
891c3970 | 1043 | } |
2b144498 | 1044 | } |
6f47caa0 | 1045 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
1046 | } |
1047 | ||
1048 | /* | |
5e5be71a | 1049 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 1050 | * |
5e5be71a ON |
1051 | * Currently we ignore all errors and always return 0, the callers |
1052 | * can't handle the failure anyway. | |
2b144498 | 1053 | */ |
7b2d81d4 | 1054 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1055 | { |
1056 | struct list_head tmp_list; | |
665605a2 | 1057 | struct uprobe *uprobe, *u; |
2b144498 | 1058 | struct inode *inode; |
2b144498 | 1059 | |
441f1eb7 | 1060 | if (no_uprobe_events() || !valid_vma(vma, true)) |
7b2d81d4 | 1061 | return 0; |
2b144498 | 1062 | |
f281769e | 1063 | inode = file_inode(vma->vm_file); |
2b144498 | 1064 | if (!inode) |
7b2d81d4 | 1065 | return 0; |
2b144498 | 1066 | |
2b144498 | 1067 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1068 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
806a98bd ON |
1069 | /* |
1070 | * We can race with uprobe_unregister(), this uprobe can be already | |
1071 | * removed. But in this case filter_chain() must return false, all | |
1072 | * consumers have gone away. | |
1073 | */ | |
665605a2 | 1074 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
806a98bd | 1075 | if (!fatal_signal_pending(current) && |
8a7f2fa0 | 1076 | filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) { |
57683f72 | 1077 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 1078 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
1079 | } |
1080 | put_uprobe(uprobe); | |
1081 | } | |
2b144498 SD |
1082 | mutex_unlock(uprobes_mmap_hash(inode)); |
1083 | ||
5e5be71a | 1084 | return 0; |
2b144498 SD |
1085 | } |
1086 | ||
9f68f672 ON |
1087 | static bool |
1088 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1089 | { | |
1090 | loff_t min, max; | |
1091 | struct inode *inode; | |
1092 | struct rb_node *n; | |
1093 | ||
f281769e | 1094 | inode = file_inode(vma->vm_file); |
9f68f672 ON |
1095 | |
1096 | min = vaddr_to_offset(vma, start); | |
1097 | max = min + (end - start) - 1; | |
1098 | ||
1099 | spin_lock(&uprobes_treelock); | |
1100 | n = find_node_in_range(inode, min, max); | |
1101 | spin_unlock(&uprobes_treelock); | |
1102 | ||
1103 | return !!n; | |
1104 | } | |
1105 | ||
682968e0 SD |
1106 | /* |
1107 | * Called in context of a munmap of a vma. | |
1108 | */ | |
cbc91f71 | 1109 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1110 | { |
441f1eb7 | 1111 | if (no_uprobe_events() || !valid_vma(vma, false)) |
682968e0 SD |
1112 | return; |
1113 | ||
2fd611a9 ON |
1114 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1115 | return; | |
1116 | ||
9f68f672 ON |
1117 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1118 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1119 | return; |
1120 | ||
9f68f672 ON |
1121 | if (vma_has_uprobes(vma, start, end)) |
1122 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1123 | } |
1124 | ||
d4b3b638 | 1125 | /* Slot allocation for XOL */ |
6441ec8b | 1126 | static int xol_add_vma(struct mm_struct *mm, struct xol_area *area) |
d4b3b638 | 1127 | { |
c8a82538 | 1128 | int ret = -EALREADY; |
d4b3b638 SD |
1129 | |
1130 | down_write(&mm->mmap_sem); | |
1131 | if (mm->uprobes_state.xol_area) | |
1132 | goto fail; | |
1133 | ||
af0d95af ON |
1134 | if (!area->vaddr) { |
1135 | /* Try to map as high as possible, this is only a hint. */ | |
1136 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, | |
1137 | PAGE_SIZE, 0, 0); | |
1138 | if (area->vaddr & ~PAGE_MASK) { | |
1139 | ret = area->vaddr; | |
1140 | goto fail; | |
1141 | } | |
d4b3b638 SD |
1142 | } |
1143 | ||
1144 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
f58bea2f | 1145 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, area->pages); |
d4b3b638 SD |
1146 | if (ret) |
1147 | goto fail; | |
1148 | ||
1149 | smp_wmb(); /* pairs with get_xol_area() */ | |
1150 | mm->uprobes_state.xol_area = area; | |
c8a82538 | 1151 | fail: |
d4b3b638 | 1152 | up_write(&mm->mmap_sem); |
d4b3b638 SD |
1153 | |
1154 | return ret; | |
1155 | } | |
1156 | ||
af0d95af | 1157 | static struct xol_area *__create_xol_area(unsigned long vaddr) |
d4b3b638 | 1158 | { |
9b545df8 | 1159 | struct mm_struct *mm = current->mm; |
e78aebfd | 1160 | uprobe_opcode_t insn = UPROBE_SWBP_INSN; |
6441ec8b | 1161 | struct xol_area *area; |
9b545df8 | 1162 | |
af0d95af | 1163 | area = kmalloc(sizeof(*area), GFP_KERNEL); |
d4b3b638 | 1164 | if (unlikely(!area)) |
c8a82538 | 1165 | goto out; |
d4b3b638 SD |
1166 | |
1167 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
d4b3b638 | 1168 | if (!area->bitmap) |
c8a82538 ON |
1169 | goto free_area; |
1170 | ||
f58bea2f ON |
1171 | area->pages[0] = alloc_page(GFP_HIGHUSER); |
1172 | if (!area->pages[0]) | |
c8a82538 | 1173 | goto free_bitmap; |
f58bea2f | 1174 | area->pages[1] = NULL; |
d4b3b638 | 1175 | |
af0d95af | 1176 | area->vaddr = vaddr; |
6441ec8b ON |
1177 | init_waitqueue_head(&area->wq); |
1178 | /* Reserve the 1st slot for get_trampoline_vaddr() */ | |
e78aebfd | 1179 | set_bit(0, area->bitmap); |
e78aebfd | 1180 | atomic_set(&area->slot_count, 1); |
f58bea2f | 1181 | copy_to_page(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE); |
e78aebfd | 1182 | |
6441ec8b | 1183 | if (!xol_add_vma(mm, area)) |
d4b3b638 SD |
1184 | return area; |
1185 | ||
f58bea2f | 1186 | __free_page(area->pages[0]); |
c8a82538 | 1187 | free_bitmap: |
d4b3b638 | 1188 | kfree(area->bitmap); |
c8a82538 | 1189 | free_area: |
d4b3b638 | 1190 | kfree(area); |
c8a82538 | 1191 | out: |
6441ec8b ON |
1192 | return NULL; |
1193 | } | |
1194 | ||
1195 | /* | |
1196 | * get_xol_area - Allocate process's xol_area if necessary. | |
1197 | * This area will be used for storing instructions for execution out of line. | |
1198 | * | |
1199 | * Returns the allocated area or NULL. | |
1200 | */ | |
1201 | static struct xol_area *get_xol_area(void) | |
1202 | { | |
1203 | struct mm_struct *mm = current->mm; | |
1204 | struct xol_area *area; | |
1205 | ||
1206 | if (!mm->uprobes_state.xol_area) | |
af0d95af | 1207 | __create_xol_area(0); |
6441ec8b | 1208 | |
9b545df8 | 1209 | area = mm->uprobes_state.xol_area; |
6441ec8b | 1210 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ |
9b545df8 | 1211 | return area; |
d4b3b638 SD |
1212 | } |
1213 | ||
1214 | /* | |
1215 | * uprobe_clear_state - Free the area allocated for slots. | |
1216 | */ | |
1217 | void uprobe_clear_state(struct mm_struct *mm) | |
1218 | { | |
1219 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1220 | ||
1221 | if (!area) | |
1222 | return; | |
1223 | ||
f58bea2f | 1224 | put_page(area->pages[0]); |
d4b3b638 SD |
1225 | kfree(area->bitmap); |
1226 | kfree(area); | |
1227 | } | |
1228 | ||
32cdba1e ON |
1229 | void uprobe_start_dup_mmap(void) |
1230 | { | |
1231 | percpu_down_read(&dup_mmap_sem); | |
1232 | } | |
1233 | ||
1234 | void uprobe_end_dup_mmap(void) | |
1235 | { | |
1236 | percpu_up_read(&dup_mmap_sem); | |
1237 | } | |
1238 | ||
f8ac4ec9 ON |
1239 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1240 | { | |
61559a81 ON |
1241 | newmm->uprobes_state.xol_area = NULL; |
1242 | ||
9f68f672 | 1243 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1244 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1245 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1246 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1247 | } | |
f8ac4ec9 ON |
1248 | } |
1249 | ||
d4b3b638 SD |
1250 | /* |
1251 | * - search for a free slot. | |
1252 | */ | |
1253 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1254 | { | |
1255 | unsigned long slot_addr; | |
1256 | int slot_nr; | |
1257 | ||
1258 | do { | |
1259 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1260 | if (slot_nr < UINSNS_PER_PAGE) { | |
1261 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1262 | break; | |
1263 | ||
1264 | slot_nr = UINSNS_PER_PAGE; | |
1265 | continue; | |
1266 | } | |
1267 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1268 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1269 | ||
1270 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1271 | atomic_inc(&area->slot_count); | |
1272 | ||
1273 | return slot_addr; | |
1274 | } | |
1275 | ||
1276 | /* | |
a6cb3f6d | 1277 | * xol_get_insn_slot - allocate a slot for xol. |
d4b3b638 SD |
1278 | * Returns the allocated slot address or 0. |
1279 | */ | |
a6cb3f6d | 1280 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe) |
d4b3b638 SD |
1281 | { |
1282 | struct xol_area *area; | |
a6cb3f6d | 1283 | unsigned long xol_vaddr; |
d4b3b638 | 1284 | |
9b545df8 ON |
1285 | area = get_xol_area(); |
1286 | if (!area) | |
1287 | return 0; | |
d4b3b638 | 1288 | |
a6cb3f6d ON |
1289 | xol_vaddr = xol_take_insn_slot(area); |
1290 | if (unlikely(!xol_vaddr)) | |
d4b3b638 SD |
1291 | return 0; |
1292 | ||
f58bea2f | 1293 | arch_uprobe_copy_ixol(area->pages[0], xol_vaddr, |
72e6ae28 | 1294 | &uprobe->arch.ixol, sizeof(uprobe->arch.ixol)); |
d4b3b638 | 1295 | |
a6cb3f6d | 1296 | return xol_vaddr; |
d4b3b638 SD |
1297 | } |
1298 | ||
1299 | /* | |
1300 | * xol_free_insn_slot - If slot was earlier allocated by | |
1301 | * @xol_get_insn_slot(), make the slot available for | |
1302 | * subsequent requests. | |
1303 | */ | |
1304 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1305 | { | |
1306 | struct xol_area *area; | |
1307 | unsigned long vma_end; | |
1308 | unsigned long slot_addr; | |
1309 | ||
1310 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1311 | return; | |
1312 | ||
1313 | slot_addr = tsk->utask->xol_vaddr; | |
af4355e9 | 1314 | if (unlikely(!slot_addr)) |
d4b3b638 SD |
1315 | return; |
1316 | ||
1317 | area = tsk->mm->uprobes_state.xol_area; | |
1318 | vma_end = area->vaddr + PAGE_SIZE; | |
1319 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1320 | unsigned long offset; | |
1321 | int slot_nr; | |
1322 | ||
1323 | offset = slot_addr - area->vaddr; | |
1324 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1325 | if (slot_nr >= UINSNS_PER_PAGE) | |
1326 | return; | |
1327 | ||
1328 | clear_bit(slot_nr, area->bitmap); | |
1329 | atomic_dec(&area->slot_count); | |
1330 | if (waitqueue_active(&area->wq)) | |
1331 | wake_up(&area->wq); | |
1332 | ||
1333 | tsk->utask->xol_vaddr = 0; | |
1334 | } | |
1335 | } | |
1336 | ||
72e6ae28 VK |
1337 | void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr, |
1338 | void *src, unsigned long len) | |
1339 | { | |
1340 | /* Initialize the slot */ | |
1341 | copy_to_page(page, vaddr, src, len); | |
1342 | ||
1343 | /* | |
1344 | * We probably need flush_icache_user_range() but it needs vma. | |
1345 | * This should work on most of architectures by default. If | |
1346 | * architecture needs to do something different it can define | |
1347 | * its own version of the function. | |
1348 | */ | |
1349 | flush_dcache_page(page); | |
1350 | } | |
1351 | ||
0326f5a9 SD |
1352 | /** |
1353 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1354 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1355 | * instruction. | |
1356 | * Return the address of the breakpoint instruction. | |
1357 | */ | |
1358 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1359 | { | |
1360 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1361 | } | |
1362 | ||
b02ef20a ON |
1363 | unsigned long uprobe_get_trap_addr(struct pt_regs *regs) |
1364 | { | |
1365 | struct uprobe_task *utask = current->utask; | |
1366 | ||
1367 | if (unlikely(utask && utask->active_uprobe)) | |
1368 | return utask->vaddr; | |
1369 | ||
1370 | return instruction_pointer(regs); | |
1371 | } | |
1372 | ||
2bb5e840 ON |
1373 | static struct return_instance *free_ret_instance(struct return_instance *ri) |
1374 | { | |
1375 | struct return_instance *next = ri->next; | |
1376 | put_uprobe(ri->uprobe); | |
1377 | kfree(ri); | |
1378 | return next; | |
1379 | } | |
1380 | ||
0326f5a9 SD |
1381 | /* |
1382 | * Called with no locks held. | |
1383 | * Called in context of a exiting or a exec-ing thread. | |
1384 | */ | |
1385 | void uprobe_free_utask(struct task_struct *t) | |
1386 | { | |
1387 | struct uprobe_task *utask = t->utask; | |
2bb5e840 | 1388 | struct return_instance *ri; |
0326f5a9 | 1389 | |
0326f5a9 SD |
1390 | if (!utask) |
1391 | return; | |
1392 | ||
1393 | if (utask->active_uprobe) | |
1394 | put_uprobe(utask->active_uprobe); | |
1395 | ||
0dfd0eb8 | 1396 | ri = utask->return_instances; |
2bb5e840 ON |
1397 | while (ri) |
1398 | ri = free_ret_instance(ri); | |
0dfd0eb8 | 1399 | |
d4b3b638 | 1400 | xol_free_insn_slot(t); |
0326f5a9 SD |
1401 | kfree(utask); |
1402 | t->utask = NULL; | |
1403 | } | |
1404 | ||
0326f5a9 | 1405 | /* |
5a2df662 ON |
1406 | * Allocate a uprobe_task object for the task if if necessary. |
1407 | * Called when the thread hits a breakpoint. | |
0326f5a9 SD |
1408 | * |
1409 | * Returns: | |
1410 | * - pointer to new uprobe_task on success | |
1411 | * - NULL otherwise | |
1412 | */ | |
5a2df662 | 1413 | static struct uprobe_task *get_utask(void) |
0326f5a9 | 1414 | { |
5a2df662 ON |
1415 | if (!current->utask) |
1416 | current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1417 | return current->utask; | |
0326f5a9 SD |
1418 | } |
1419 | ||
248d3a7b ON |
1420 | static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask) |
1421 | { | |
1422 | struct uprobe_task *n_utask; | |
1423 | struct return_instance **p, *o, *n; | |
1424 | ||
1425 | n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1426 | if (!n_utask) | |
1427 | return -ENOMEM; | |
1428 | t->utask = n_utask; | |
1429 | ||
1430 | p = &n_utask->return_instances; | |
1431 | for (o = o_utask->return_instances; o; o = o->next) { | |
1432 | n = kmalloc(sizeof(struct return_instance), GFP_KERNEL); | |
1433 | if (!n) | |
1434 | return -ENOMEM; | |
1435 | ||
1436 | *n = *o; | |
f231722a | 1437 | get_uprobe(n->uprobe); |
248d3a7b ON |
1438 | n->next = NULL; |
1439 | ||
1440 | *p = n; | |
1441 | p = &n->next; | |
1442 | n_utask->depth++; | |
1443 | } | |
1444 | ||
1445 | return 0; | |
1446 | } | |
1447 | ||
1448 | static void uprobe_warn(struct task_struct *t, const char *msg) | |
1449 | { | |
1450 | pr_warn("uprobe: %s:%d failed to %s\n", | |
1451 | current->comm, current->pid, msg); | |
1452 | } | |
1453 | ||
aa59c53f ON |
1454 | static void dup_xol_work(struct callback_head *work) |
1455 | { | |
aa59c53f ON |
1456 | if (current->flags & PF_EXITING) |
1457 | return; | |
1458 | ||
32473431 | 1459 | if (!__create_xol_area(current->utask->dup_xol_addr)) |
aa59c53f ON |
1460 | uprobe_warn(current, "dup xol area"); |
1461 | } | |
1462 | ||
b68e0749 ON |
1463 | /* |
1464 | * Called in context of a new clone/fork from copy_process. | |
1465 | */ | |
3ab67966 | 1466 | void uprobe_copy_process(struct task_struct *t, unsigned long flags) |
b68e0749 | 1467 | { |
248d3a7b ON |
1468 | struct uprobe_task *utask = current->utask; |
1469 | struct mm_struct *mm = current->mm; | |
aa59c53f | 1470 | struct xol_area *area; |
248d3a7b | 1471 | |
b68e0749 | 1472 | t->utask = NULL; |
248d3a7b | 1473 | |
3ab67966 ON |
1474 | if (!utask || !utask->return_instances) |
1475 | return; | |
1476 | ||
1477 | if (mm == t->mm && !(flags & CLONE_VFORK)) | |
248d3a7b ON |
1478 | return; |
1479 | ||
1480 | if (dup_utask(t, utask)) | |
1481 | return uprobe_warn(t, "dup ret instances"); | |
aa59c53f ON |
1482 | |
1483 | /* The task can fork() after dup_xol_work() fails */ | |
1484 | area = mm->uprobes_state.xol_area; | |
1485 | if (!area) | |
1486 | return uprobe_warn(t, "dup xol area"); | |
1487 | ||
3ab67966 ON |
1488 | if (mm == t->mm) |
1489 | return; | |
1490 | ||
32473431 ON |
1491 | t->utask->dup_xol_addr = area->vaddr; |
1492 | init_task_work(&t->utask->dup_xol_work, dup_xol_work); | |
1493 | task_work_add(t, &t->utask->dup_xol_work, true); | |
b68e0749 ON |
1494 | } |
1495 | ||
e78aebfd AA |
1496 | /* |
1497 | * Current area->vaddr notion assume the trampoline address is always | |
1498 | * equal area->vaddr. | |
1499 | * | |
1500 | * Returns -1 in case the xol_area is not allocated. | |
1501 | */ | |
1502 | static unsigned long get_trampoline_vaddr(void) | |
1503 | { | |
1504 | struct xol_area *area; | |
1505 | unsigned long trampoline_vaddr = -1; | |
1506 | ||
1507 | area = current->mm->uprobes_state.xol_area; | |
1508 | smp_read_barrier_depends(); | |
1509 | if (area) | |
1510 | trampoline_vaddr = area->vaddr; | |
1511 | ||
1512 | return trampoline_vaddr; | |
1513 | } | |
1514 | ||
db087ef6 ON |
1515 | static void cleanup_return_instances(struct uprobe_task *utask, bool chained, |
1516 | struct pt_regs *regs) | |
a5b7e1a8 ON |
1517 | { |
1518 | struct return_instance *ri = utask->return_instances; | |
db087ef6 | 1519 | enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL; |
86dcb702 ON |
1520 | |
1521 | while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) { | |
a5b7e1a8 ON |
1522 | ri = free_ret_instance(ri); |
1523 | utask->depth--; | |
1524 | } | |
1525 | utask->return_instances = ri; | |
1526 | } | |
1527 | ||
0dfd0eb8 AA |
1528 | static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs) |
1529 | { | |
1530 | struct return_instance *ri; | |
1531 | struct uprobe_task *utask; | |
1532 | unsigned long orig_ret_vaddr, trampoline_vaddr; | |
db087ef6 | 1533 | bool chained; |
0dfd0eb8 AA |
1534 | |
1535 | if (!get_xol_area()) | |
1536 | return; | |
1537 | ||
1538 | utask = get_utask(); | |
1539 | if (!utask) | |
1540 | return; | |
1541 | ||
ded49c55 AA |
1542 | if (utask->depth >= MAX_URETPROBE_DEPTH) { |
1543 | printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to" | |
1544 | " nestedness limit pid/tgid=%d/%d\n", | |
1545 | current->pid, current->tgid); | |
1546 | return; | |
1547 | } | |
1548 | ||
6c58d0e4 | 1549 | ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL); |
0dfd0eb8 | 1550 | if (!ri) |
6c58d0e4 | 1551 | return; |
0dfd0eb8 AA |
1552 | |
1553 | trampoline_vaddr = get_trampoline_vaddr(); | |
1554 | orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs); | |
1555 | if (orig_ret_vaddr == -1) | |
1556 | goto fail; | |
1557 | ||
a5b7e1a8 | 1558 | /* drop the entries invalidated by longjmp() */ |
db087ef6 ON |
1559 | chained = (orig_ret_vaddr == trampoline_vaddr); |
1560 | cleanup_return_instances(utask, chained, regs); | |
a5b7e1a8 | 1561 | |
0dfd0eb8 AA |
1562 | /* |
1563 | * We don't want to keep trampoline address in stack, rather keep the | |
1564 | * original return address of first caller thru all the consequent | |
1565 | * instances. This also makes breakpoint unwrapping easier. | |
1566 | */ | |
db087ef6 | 1567 | if (chained) { |
0dfd0eb8 AA |
1568 | if (!utask->return_instances) { |
1569 | /* | |
1570 | * This situation is not possible. Likely we have an | |
1571 | * attack from user-space. | |
1572 | */ | |
6c58d0e4 | 1573 | uprobe_warn(current, "handle tail call"); |
0dfd0eb8 AA |
1574 | goto fail; |
1575 | } | |
0dfd0eb8 AA |
1576 | orig_ret_vaddr = utask->return_instances->orig_ret_vaddr; |
1577 | } | |
1578 | ||
f231722a | 1579 | ri->uprobe = get_uprobe(uprobe); |
0dfd0eb8 | 1580 | ri->func = instruction_pointer(regs); |
7b868e48 | 1581 | ri->stack = user_stack_pointer(regs); |
0dfd0eb8 AA |
1582 | ri->orig_ret_vaddr = orig_ret_vaddr; |
1583 | ri->chained = chained; | |
1584 | ||
ded49c55 | 1585 | utask->depth++; |
0dfd0eb8 AA |
1586 | ri->next = utask->return_instances; |
1587 | utask->return_instances = ri; | |
1588 | ||
1589 | return; | |
0dfd0eb8 AA |
1590 | fail: |
1591 | kfree(ri); | |
1592 | } | |
1593 | ||
0326f5a9 SD |
1594 | /* Prepare to single-step probed instruction out of line. */ |
1595 | static int | |
a6cb3f6d | 1596 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr) |
0326f5a9 | 1597 | { |
a6cb3f6d ON |
1598 | struct uprobe_task *utask; |
1599 | unsigned long xol_vaddr; | |
aba51024 | 1600 | int err; |
a6cb3f6d | 1601 | |
608e7427 ON |
1602 | utask = get_utask(); |
1603 | if (!utask) | |
1604 | return -ENOMEM; | |
a6cb3f6d ON |
1605 | |
1606 | xol_vaddr = xol_get_insn_slot(uprobe); | |
1607 | if (!xol_vaddr) | |
1608 | return -ENOMEM; | |
1609 | ||
1610 | utask->xol_vaddr = xol_vaddr; | |
1611 | utask->vaddr = bp_vaddr; | |
d4b3b638 | 1612 | |
aba51024 ON |
1613 | err = arch_uprobe_pre_xol(&uprobe->arch, regs); |
1614 | if (unlikely(err)) { | |
1615 | xol_free_insn_slot(current); | |
1616 | return err; | |
1617 | } | |
1618 | ||
608e7427 ON |
1619 | utask->active_uprobe = uprobe; |
1620 | utask->state = UTASK_SSTEP; | |
aba51024 | 1621 | return 0; |
0326f5a9 SD |
1622 | } |
1623 | ||
1624 | /* | |
1625 | * If we are singlestepping, then ensure this thread is not connected to | |
1626 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1627 | * triggers the signal, restart the original insn even if the task is | |
1628 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1629 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1630 | * _same_ instruction should be repeated again after return from the signal | |
1631 | * handler, and SSTEP can never finish in this case. | |
1632 | */ | |
1633 | bool uprobe_deny_signal(void) | |
1634 | { | |
1635 | struct task_struct *t = current; | |
1636 | struct uprobe_task *utask = t->utask; | |
1637 | ||
1638 | if (likely(!utask || !utask->active_uprobe)) | |
1639 | return false; | |
1640 | ||
1641 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1642 | ||
1643 | if (signal_pending(t)) { | |
1644 | spin_lock_irq(&t->sighand->siglock); | |
1645 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1646 | spin_unlock_irq(&t->sighand->siglock); | |
1647 | ||
1648 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1649 | utask->state = UTASK_SSTEP_TRAPPED; | |
1650 | set_tsk_thread_flag(t, TIF_UPROBE); | |
0326f5a9 SD |
1651 | } |
1652 | } | |
1653 | ||
1654 | return true; | |
1655 | } | |
1656 | ||
499a4f3e ON |
1657 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1658 | { | |
1659 | struct vm_area_struct *vma; | |
1660 | ||
1661 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1662 | if (!valid_vma(vma, false)) | |
1663 | continue; | |
1664 | /* | |
1665 | * This is not strictly accurate, we can race with | |
1666 | * uprobe_unregister() and see the already removed | |
1667 | * uprobe if delete_uprobe() was not yet called. | |
63633cbf | 1668 | * Or this uprobe can be filtered out. |
499a4f3e ON |
1669 | */ |
1670 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1671 | return; | |
1672 | } | |
1673 | ||
1674 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1675 | } | |
1676 | ||
0908ad6e | 1677 | static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr) |
ec75fba9 ON |
1678 | { |
1679 | struct page *page; | |
1680 | uprobe_opcode_t opcode; | |
1681 | int result; | |
1682 | ||
1683 | pagefault_disable(); | |
1684 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
1685 | sizeof(opcode)); | |
1686 | pagefault_enable(); | |
1687 | ||
1688 | if (likely(result == 0)) | |
1689 | goto out; | |
1690 | ||
1691 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); | |
1692 | if (result < 0) | |
1693 | return result; | |
1694 | ||
ab0d805c | 1695 | copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); |
ec75fba9 ON |
1696 | put_page(page); |
1697 | out: | |
0908ad6e AM |
1698 | /* This needs to return true for any variant of the trap insn */ |
1699 | return is_trap_insn(&opcode); | |
ec75fba9 ON |
1700 | } |
1701 | ||
d790d346 | 1702 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1703 | { |
3a9ea052 ON |
1704 | struct mm_struct *mm = current->mm; |
1705 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1706 | struct vm_area_struct *vma; |
0326f5a9 | 1707 | |
0326f5a9 SD |
1708 | down_read(&mm->mmap_sem); |
1709 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1710 | if (vma && vma->vm_start <= bp_vaddr) { |
1711 | if (valid_vma(vma, false)) { | |
f281769e | 1712 | struct inode *inode = file_inode(vma->vm_file); |
cb113b47 | 1713 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); |
0326f5a9 | 1714 | |
3a9ea052 ON |
1715 | uprobe = find_uprobe(inode, offset); |
1716 | } | |
d790d346 ON |
1717 | |
1718 | if (!uprobe) | |
0908ad6e | 1719 | *is_swbp = is_trap_at_addr(mm, bp_vaddr); |
d790d346 ON |
1720 | } else { |
1721 | *is_swbp = -EFAULT; | |
0326f5a9 | 1722 | } |
499a4f3e ON |
1723 | |
1724 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1725 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1726 | up_read(&mm->mmap_sem); |
1727 | ||
3a9ea052 ON |
1728 | return uprobe; |
1729 | } | |
1730 | ||
da1816b1 ON |
1731 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
1732 | { | |
1733 | struct uprobe_consumer *uc; | |
1734 | int remove = UPROBE_HANDLER_REMOVE; | |
0dfd0eb8 | 1735 | bool need_prep = false; /* prepare return uprobe, when needed */ |
da1816b1 ON |
1736 | |
1737 | down_read(&uprobe->register_rwsem); | |
1738 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
ea024870 | 1739 | int rc = 0; |
da1816b1 | 1740 | |
ea024870 AA |
1741 | if (uc->handler) { |
1742 | rc = uc->handler(uc, regs); | |
1743 | WARN(rc & ~UPROBE_HANDLER_MASK, | |
1744 | "bad rc=0x%x from %pf()\n", rc, uc->handler); | |
1745 | } | |
0dfd0eb8 AA |
1746 | |
1747 | if (uc->ret_handler) | |
1748 | need_prep = true; | |
1749 | ||
da1816b1 ON |
1750 | remove &= rc; |
1751 | } | |
1752 | ||
0dfd0eb8 AA |
1753 | if (need_prep && !remove) |
1754 | prepare_uretprobe(uprobe, regs); /* put bp at return */ | |
1755 | ||
da1816b1 ON |
1756 | if (remove && uprobe->consumers) { |
1757 | WARN_ON(!uprobe_is_active(uprobe)); | |
1758 | unapply_uprobe(uprobe, current->mm); | |
1759 | } | |
1760 | up_read(&uprobe->register_rwsem); | |
1761 | } | |
1762 | ||
fec8898d AA |
1763 | static void |
1764 | handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs) | |
1765 | { | |
1766 | struct uprobe *uprobe = ri->uprobe; | |
1767 | struct uprobe_consumer *uc; | |
1768 | ||
1769 | down_read(&uprobe->register_rwsem); | |
1770 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
1771 | if (uc->ret_handler) | |
1772 | uc->ret_handler(uc, ri->func, regs); | |
1773 | } | |
1774 | up_read(&uprobe->register_rwsem); | |
1775 | } | |
1776 | ||
a83cfeb9 ON |
1777 | static struct return_instance *find_next_ret_chain(struct return_instance *ri) |
1778 | { | |
1779 | bool chained; | |
1780 | ||
1781 | do { | |
1782 | chained = ri->chained; | |
1783 | ri = ri->next; /* can't be NULL if chained */ | |
1784 | } while (chained); | |
1785 | ||
1786 | return ri; | |
1787 | } | |
1788 | ||
0b5256c7 | 1789 | static void handle_trampoline(struct pt_regs *regs) |
fec8898d AA |
1790 | { |
1791 | struct uprobe_task *utask; | |
a83cfeb9 | 1792 | struct return_instance *ri, *next; |
5eeb50de | 1793 | bool valid; |
fec8898d AA |
1794 | |
1795 | utask = current->utask; | |
1796 | if (!utask) | |
0b5256c7 | 1797 | goto sigill; |
fec8898d AA |
1798 | |
1799 | ri = utask->return_instances; | |
1800 | if (!ri) | |
0b5256c7 | 1801 | goto sigill; |
fec8898d | 1802 | |
a83cfeb9 | 1803 | do { |
5eeb50de ON |
1804 | /* |
1805 | * We should throw out the frames invalidated by longjmp(). | |
1806 | * If this chain is valid, then the next one should be alive | |
1807 | * or NULL; the latter case means that nobody but ri->func | |
1808 | * could hit this trampoline on return. TODO: sigaltstack(). | |
1809 | */ | |
1810 | next = find_next_ret_chain(ri); | |
86dcb702 | 1811 | valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs); |
5eeb50de ON |
1812 | |
1813 | instruction_pointer_set(regs, ri->orig_ret_vaddr); | |
1814 | do { | |
1815 | if (valid) | |
1816 | handle_uretprobe_chain(ri, regs); | |
1817 | ri = free_ret_instance(ri); | |
1818 | utask->depth--; | |
1819 | } while (ri != next); | |
1820 | } while (!valid); | |
fec8898d AA |
1821 | |
1822 | utask->return_instances = ri; | |
0b5256c7 ON |
1823 | return; |
1824 | ||
1825 | sigill: | |
1826 | uprobe_warn(current, "handle uretprobe, sending SIGILL."); | |
1827 | force_sig_info(SIGILL, SEND_SIG_FORCED, current); | |
fec8898d | 1828 | |
fec8898d AA |
1829 | } |
1830 | ||
6fe50a28 DL |
1831 | bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs) |
1832 | { | |
1833 | return false; | |
1834 | } | |
1835 | ||
86dcb702 ON |
1836 | bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx, |
1837 | struct pt_regs *regs) | |
97da8976 ON |
1838 | { |
1839 | return true; | |
1840 | } | |
1841 | ||
3a9ea052 ON |
1842 | /* |
1843 | * Run handler and ask thread to singlestep. | |
1844 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1845 | */ | |
1846 | static void handle_swbp(struct pt_regs *regs) | |
1847 | { | |
3a9ea052 ON |
1848 | struct uprobe *uprobe; |
1849 | unsigned long bp_vaddr; | |
56bb4cf6 | 1850 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1851 | |
1852 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
0b5256c7 ON |
1853 | if (bp_vaddr == get_trampoline_vaddr()) |
1854 | return handle_trampoline(regs); | |
fec8898d AA |
1855 | |
1856 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); | |
0326f5a9 | 1857 | if (!uprobe) { |
56bb4cf6 ON |
1858 | if (is_swbp > 0) { |
1859 | /* No matching uprobe; signal SIGTRAP. */ | |
1860 | send_sig(SIGTRAP, current, 0); | |
1861 | } else { | |
1862 | /* | |
1863 | * Either we raced with uprobe_unregister() or we can't | |
1864 | * access this memory. The latter is only possible if | |
1865 | * another thread plays with our ->mm. In both cases | |
1866 | * we can simply restart. If this vma was unmapped we | |
1867 | * can pretend this insn was not executed yet and get | |
1868 | * the (correct) SIGSEGV after restart. | |
1869 | */ | |
1870 | instruction_pointer_set(regs, bp_vaddr); | |
1871 | } | |
0326f5a9 SD |
1872 | return; |
1873 | } | |
74e59dfc ON |
1874 | |
1875 | /* change it in advance for ->handler() and restart */ | |
1876 | instruction_pointer_set(regs, bp_vaddr); | |
1877 | ||
142b18dd ON |
1878 | /* |
1879 | * TODO: move copy_insn/etc into _register and remove this hack. | |
1880 | * After we hit the bp, _unregister + _register can install the | |
1881 | * new and not-yet-analyzed uprobe at the same address, restart. | |
1882 | */ | |
1883 | smp_rmb(); /* pairs with wmb() in install_breakpoint() */ | |
71434f2f | 1884 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
74e59dfc | 1885 | goto out; |
0326f5a9 | 1886 | |
72fd293a ON |
1887 | /* Tracing handlers use ->utask to communicate with fetch methods */ |
1888 | if (!get_utask()) | |
1889 | goto out; | |
1890 | ||
6fe50a28 DL |
1891 | if (arch_uprobe_ignore(&uprobe->arch, regs)) |
1892 | goto out; | |
1893 | ||
0326f5a9 | 1894 | handler_chain(uprobe, regs); |
6fe50a28 | 1895 | |
8a6b1732 | 1896 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
0578a970 | 1897 | goto out; |
0326f5a9 | 1898 | |
608e7427 | 1899 | if (!pre_ssout(uprobe, regs, bp_vaddr)) |
0326f5a9 | 1900 | return; |
0326f5a9 | 1901 | |
8a6b1732 | 1902 | /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */ |
0578a970 | 1903 | out: |
8bd87445 | 1904 | put_uprobe(uprobe); |
0326f5a9 SD |
1905 | } |
1906 | ||
1907 | /* | |
1908 | * Perform required fix-ups and disable singlestep. | |
1909 | * Allow pending signals to take effect. | |
1910 | */ | |
1911 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1912 | { | |
1913 | struct uprobe *uprobe; | |
014940ba | 1914 | int err = 0; |
0326f5a9 SD |
1915 | |
1916 | uprobe = utask->active_uprobe; | |
1917 | if (utask->state == UTASK_SSTEP_ACK) | |
014940ba | 1918 | err = arch_uprobe_post_xol(&uprobe->arch, regs); |
0326f5a9 SD |
1919 | else if (utask->state == UTASK_SSTEP_TRAPPED) |
1920 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1921 | else | |
1922 | WARN_ON_ONCE(1); | |
1923 | ||
1924 | put_uprobe(uprobe); | |
1925 | utask->active_uprobe = NULL; | |
1926 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1927 | xol_free_insn_slot(current); |
0326f5a9 SD |
1928 | |
1929 | spin_lock_irq(¤t->sighand->siglock); | |
1930 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1931 | spin_unlock_irq(¤t->sighand->siglock); | |
014940ba ON |
1932 | |
1933 | if (unlikely(err)) { | |
1934 | uprobe_warn(current, "execute the probed insn, sending SIGILL."); | |
1935 | force_sig_info(SIGILL, SEND_SIG_FORCED, current); | |
1936 | } | |
0326f5a9 SD |
1937 | } |
1938 | ||
1939 | /* | |
1b08e907 ON |
1940 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1941 | * allows the thread to return from interrupt. After that handle_swbp() | |
1942 | * sets utask->active_uprobe. | |
0326f5a9 | 1943 | * |
1b08e907 ON |
1944 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1945 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1946 | * |
1947 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1948 | * uprobe_notify_resume(). | |
1949 | */ | |
1950 | void uprobe_notify_resume(struct pt_regs *regs) | |
1951 | { | |
1952 | struct uprobe_task *utask; | |
1953 | ||
db023ea5 ON |
1954 | clear_thread_flag(TIF_UPROBE); |
1955 | ||
0326f5a9 | 1956 | utask = current->utask; |
1b08e907 | 1957 | if (utask && utask->active_uprobe) |
0326f5a9 | 1958 | handle_singlestep(utask, regs); |
1b08e907 ON |
1959 | else |
1960 | handle_swbp(regs); | |
0326f5a9 SD |
1961 | } |
1962 | ||
1963 | /* | |
1964 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1965 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1966 | */ | |
1967 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1968 | { | |
0dfd0eb8 AA |
1969 | if (!current->mm) |
1970 | return 0; | |
1971 | ||
1972 | if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) && | |
1973 | (!current->utask || !current->utask->return_instances)) | |
0326f5a9 SD |
1974 | return 0; |
1975 | ||
0326f5a9 | 1976 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1977 | return 1; |
1978 | } | |
1979 | ||
1980 | /* | |
1981 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1982 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1983 | */ | |
1984 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1985 | { | |
1986 | struct uprobe_task *utask = current->utask; | |
1987 | ||
1988 | if (!current->mm || !utask || !utask->active_uprobe) | |
1989 | /* task is currently not uprobed */ | |
1990 | return 0; | |
1991 | ||
1992 | utask->state = UTASK_SSTEP_ACK; | |
1993 | set_thread_flag(TIF_UPROBE); | |
1994 | return 1; | |
1995 | } | |
1996 | ||
1997 | static struct notifier_block uprobe_exception_nb = { | |
1998 | .notifier_call = arch_uprobe_exception_notify, | |
1999 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
2000 | }; | |
2001 | ||
2b144498 SD |
2002 | static int __init init_uprobes(void) |
2003 | { | |
2004 | int i; | |
2005 | ||
66d06dff | 2006 | for (i = 0; i < UPROBES_HASH_SZ; i++) |
2b144498 | 2007 | mutex_init(&uprobes_mmap_mutex[i]); |
0326f5a9 | 2008 | |
32cdba1e ON |
2009 | if (percpu_init_rwsem(&dup_mmap_sem)) |
2010 | return -ENOMEM; | |
2011 | ||
0326f5a9 | 2012 | return register_die_notifier(&uprobe_exception_nb); |
2b144498 | 2013 | } |
736e89d9 | 2014 | __initcall(init_uprobes); |