]>
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> |
7b2d81d4 | 39 | |
2b144498 SD |
40 | #include <linux/uprobes.h> |
41 | ||
d4b3b638 SD |
42 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
43 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
44 | ||
2b144498 | 45 | static struct rb_root uprobes_tree = RB_ROOT; |
441f1eb7 ON |
46 | /* |
47 | * allows us to skip the uprobe_mmap if there are no uprobe events active | |
48 | * at this time. Probably a fine grained per inode count is better? | |
49 | */ | |
50 | #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree) | |
7b2d81d4 | 51 | |
2b144498 SD |
52 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
53 | ||
54 | #define UPROBES_HASH_SZ 13 | |
2b144498 SD |
55 | /* serialize uprobe->pending_list */ |
56 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 57 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 | 58 | |
32cdba1e ON |
59 | static struct percpu_rw_semaphore dup_mmap_sem; |
60 | ||
cb9a19fe | 61 | /* Have a copy of original instruction */ |
71434f2f | 62 | #define UPROBE_COPY_INSN 0 |
cb9a19fe | 63 | |
3ff54efd SD |
64 | struct uprobe { |
65 | struct rb_node rb_node; /* node in the rb tree */ | |
66 | atomic_t ref; | |
e591c8d7 | 67 | struct rw_semaphore register_rwsem; |
3ff54efd SD |
68 | struct rw_semaphore consumer_rwsem; |
69 | struct list_head pending_list; | |
70 | struct uprobe_consumer *consumers; | |
71 | struct inode *inode; /* Also hold a ref to inode */ | |
72 | loff_t offset; | |
71434f2f | 73 | unsigned long flags; |
ad439356 ON |
74 | |
75 | /* | |
76 | * The generic code assumes that it has two members of unknown type | |
77 | * owned by the arch-specific code: | |
78 | * | |
79 | * insn - copy_insn() saves the original instruction here for | |
80 | * arch_uprobe_analyze_insn(). | |
81 | * | |
82 | * ixol - potentially modified instruction to execute out of | |
83 | * line, copied to xol_area by xol_get_insn_slot(). | |
84 | */ | |
3ff54efd SD |
85 | struct arch_uprobe arch; |
86 | }; | |
87 | ||
0dfd0eb8 AA |
88 | struct return_instance { |
89 | struct uprobe *uprobe; | |
90 | unsigned long func; | |
91 | unsigned long orig_ret_vaddr; /* original return address */ | |
92 | bool chained; /* true, if instance is nested */ | |
93 | ||
94 | struct return_instance *next; /* keep as stack */ | |
95 | }; | |
96 | ||
c912dae6 | 97 | /* |
ad439356 ON |
98 | * Execute out of line area: anonymous executable mapping installed |
99 | * by the probed task to execute the copy of the original instruction | |
100 | * mangled by set_swbp(). | |
101 | * | |
c912dae6 ON |
102 | * On a breakpoint hit, thread contests for a slot. It frees the |
103 | * slot after singlestep. Currently a fixed number of slots are | |
104 | * allocated. | |
105 | */ | |
106 | struct xol_area { | |
107 | wait_queue_head_t wq; /* if all slots are busy */ | |
108 | atomic_t slot_count; /* number of in-use slots */ | |
109 | unsigned long *bitmap; /* 0 = free slot */ | |
110 | struct page *page; | |
111 | ||
112 | /* | |
113 | * We keep the vma's vm_start rather than a pointer to the vma | |
114 | * itself. The probed process or a naughty kernel module could make | |
115 | * the vma go away, and we must handle that reasonably gracefully. | |
116 | */ | |
117 | unsigned long vaddr; /* Page(s) of instruction slots */ | |
118 | }; | |
119 | ||
2b144498 SD |
120 | /* |
121 | * valid_vma: Verify if the specified vma is an executable vma | |
122 | * Relax restrictions while unregistering: vm_flags might have | |
123 | * changed after breakpoint was inserted. | |
124 | * - is_register: indicates if we are in register context. | |
125 | * - Return 1 if the specified virtual address is in an | |
126 | * executable vma. | |
127 | */ | |
128 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
129 | { | |
e40cfce6 | 130 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED; |
2b144498 | 131 | |
e40cfce6 ON |
132 | if (is_register) |
133 | flags |= VM_WRITE; | |
2b144498 | 134 | |
e40cfce6 | 135 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
136 | } |
137 | ||
57683f72 | 138 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 139 | { |
57683f72 | 140 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
141 | } |
142 | ||
cb113b47 ON |
143 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
144 | { | |
145 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
146 | } | |
147 | ||
2b144498 SD |
148 | /** |
149 | * __replace_page - replace page in vma by new page. | |
150 | * based on replace_page in mm/ksm.c | |
151 | * | |
152 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 153 | * @addr: address the old @page is mapped at |
2b144498 SD |
154 | * @page: the cowed page we are replacing by kpage |
155 | * @kpage: the modified page we replace page by | |
156 | * | |
157 | * Returns 0 on success, -EFAULT on failure. | |
158 | */ | |
c517ee74 ON |
159 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
160 | struct page *page, struct page *kpage) | |
2b144498 SD |
161 | { |
162 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
163 | spinlock_t *ptl; |
164 | pte_t *ptep; | |
9f92448c | 165 | int err; |
6bdb913f HE |
166 | /* For mmu_notifiers */ |
167 | const unsigned long mmun_start = addr; | |
168 | const unsigned long mmun_end = addr + PAGE_SIZE; | |
2b144498 | 169 | |
194f8dcb | 170 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
171 | lock_page(page); |
172 | ||
6bdb913f | 173 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
9f92448c | 174 | err = -EAGAIN; |
5323ce71 | 175 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 176 | if (!ptep) |
9f92448c | 177 | goto unlock; |
2b144498 SD |
178 | |
179 | get_page(kpage); | |
180 | page_add_new_anon_rmap(kpage, vma, addr); | |
181 | ||
7396fa81 SD |
182 | if (!PageAnon(page)) { |
183 | dec_mm_counter(mm, MM_FILEPAGES); | |
184 | inc_mm_counter(mm, MM_ANONPAGES); | |
185 | } | |
186 | ||
2b144498 SD |
187 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
188 | ptep_clear_flush(vma, addr, ptep); | |
189 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); | |
190 | ||
191 | page_remove_rmap(page); | |
192 | if (!page_mapped(page)) | |
193 | try_to_free_swap(page); | |
2b144498 | 194 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 195 | |
194f8dcb ON |
196 | if (vma->vm_flags & VM_LOCKED) |
197 | munlock_vma_page(page); | |
198 | put_page(page); | |
199 | ||
9f92448c ON |
200 | err = 0; |
201 | unlock: | |
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. | |
2b144498 SD |
282 | */ |
283 | ||
284 | /* | |
f72d41fa | 285 | * uprobe_write_opcode - write the opcode at a given virtual address. |
2b144498 | 286 | * @mm: the probed process address space. |
2b144498 SD |
287 | * @vaddr: the virtual address to store the opcode. |
288 | * @opcode: opcode to be written at @vaddr. | |
289 | * | |
290 | * Called with mm->mmap_sem held (for read and with a reference to | |
291 | * mm). | |
292 | * | |
293 | * For mm @mm, write the opcode at @vaddr. | |
294 | * Return 0 (success) or a negative errno. | |
295 | */ | |
f72d41fa | 296 | int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr, |
cceb55aa | 297 | uprobe_opcode_t opcode) |
2b144498 SD |
298 | { |
299 | struct page *old_page, *new_page; | |
2b144498 | 300 | struct vm_area_struct *vma; |
2b144498 | 301 | int ret; |
f403072c | 302 | |
5323ce71 | 303 | retry: |
2b144498 | 304 | /* Read the page with vaddr into memory */ |
75ed82ea | 305 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
306 | if (ret <= 0) |
307 | return ret; | |
7b2d81d4 | 308 | |
ed6f6a50 ON |
309 | ret = verify_opcode(old_page, vaddr, &opcode); |
310 | if (ret <= 0) | |
311 | goto put_old; | |
312 | ||
2b144498 SD |
313 | ret = -ENOMEM; |
314 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
315 | if (!new_page) | |
9f92448c | 316 | goto put_old; |
2b144498 SD |
317 | |
318 | __SetPageUptodate(new_page); | |
319 | ||
3f47107c ON |
320 | copy_highpage(new_page, old_page); |
321 | copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); | |
2b144498 SD |
322 | |
323 | ret = anon_vma_prepare(vma); | |
324 | if (ret) | |
9f92448c | 325 | goto put_new; |
2b144498 | 326 | |
c517ee74 | 327 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 328 | |
9f92448c | 329 | put_new: |
2b144498 | 330 | page_cache_release(new_page); |
9f92448c | 331 | put_old: |
7b2d81d4 IM |
332 | put_page(old_page); |
333 | ||
5323ce71 ON |
334 | if (unlikely(ret == -EAGAIN)) |
335 | goto retry; | |
2b144498 SD |
336 | return ret; |
337 | } | |
338 | ||
2b144498 | 339 | /** |
5cb4ac3a | 340 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 341 | * @auprobe: arch specific probepoint information. |
2b144498 | 342 | * @mm: the probed process address space. |
2b144498 SD |
343 | * @vaddr: the virtual address to insert the opcode. |
344 | * | |
345 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
346 | * Return 0 (success) or a negative errno. | |
347 | */ | |
5cb4ac3a | 348 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 349 | { |
f72d41fa | 350 | return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
351 | } |
352 | ||
353 | /** | |
354 | * set_orig_insn - Restore the original instruction. | |
355 | * @mm: the probed process address space. | |
e3343e6a | 356 | * @auprobe: arch specific probepoint information. |
2b144498 | 357 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
358 | * |
359 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
360 | * Return 0 (success) or a negative errno. | |
361 | */ | |
7b2d81d4 | 362 | int __weak |
ded86e7c | 363 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 364 | { |
803200e2 | 365 | return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn); |
2b144498 SD |
366 | } |
367 | ||
368 | static int match_uprobe(struct uprobe *l, struct uprobe *r) | |
369 | { | |
370 | if (l->inode < r->inode) | |
371 | return -1; | |
7b2d81d4 | 372 | |
2b144498 SD |
373 | if (l->inode > r->inode) |
374 | return 1; | |
2b144498 | 375 | |
7b2d81d4 IM |
376 | if (l->offset < r->offset) |
377 | return -1; | |
378 | ||
379 | if (l->offset > r->offset) | |
380 | return 1; | |
2b144498 SD |
381 | |
382 | return 0; | |
383 | } | |
384 | ||
385 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
386 | { | |
387 | struct uprobe u = { .inode = inode, .offset = offset }; | |
388 | struct rb_node *n = uprobes_tree.rb_node; | |
389 | struct uprobe *uprobe; | |
390 | int match; | |
391 | ||
392 | while (n) { | |
393 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
394 | match = match_uprobe(&u, uprobe); | |
395 | if (!match) { | |
396 | atomic_inc(&uprobe->ref); | |
397 | return uprobe; | |
398 | } | |
7b2d81d4 | 399 | |
2b144498 SD |
400 | if (match < 0) |
401 | n = n->rb_left; | |
402 | else | |
403 | n = n->rb_right; | |
404 | } | |
405 | return NULL; | |
406 | } | |
407 | ||
408 | /* | |
409 | * Find a uprobe corresponding to a given inode:offset | |
410 | * Acquires uprobes_treelock | |
411 | */ | |
412 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
413 | { | |
414 | struct uprobe *uprobe; | |
2b144498 | 415 | |
6f47caa0 | 416 | spin_lock(&uprobes_treelock); |
2b144498 | 417 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 418 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 419 | |
2b144498 SD |
420 | return uprobe; |
421 | } | |
422 | ||
423 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
424 | { | |
425 | struct rb_node **p = &uprobes_tree.rb_node; | |
426 | struct rb_node *parent = NULL; | |
427 | struct uprobe *u; | |
428 | int match; | |
429 | ||
430 | while (*p) { | |
431 | parent = *p; | |
432 | u = rb_entry(parent, struct uprobe, rb_node); | |
433 | match = match_uprobe(uprobe, u); | |
434 | if (!match) { | |
435 | atomic_inc(&u->ref); | |
436 | return u; | |
437 | } | |
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 | ||
474 | static void put_uprobe(struct uprobe *uprobe) | |
475 | { | |
476 | if (atomic_dec_and_test(&uprobe->ref)) | |
477 | kfree(uprobe); | |
478 | } | |
479 | ||
480 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) | |
481 | { | |
482 | struct uprobe *uprobe, *cur_uprobe; | |
483 | ||
484 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
485 | if (!uprobe) | |
486 | return NULL; | |
487 | ||
488 | uprobe->inode = igrab(inode); | |
489 | uprobe->offset = offset; | |
e591c8d7 | 490 | init_rwsem(&uprobe->register_rwsem); |
2b144498 | 491 | init_rwsem(&uprobe->consumer_rwsem); |
2b144498 SD |
492 | |
493 | /* add to uprobes_tree, sorted on inode:offset */ | |
494 | cur_uprobe = insert_uprobe(uprobe); | |
2b144498 SD |
495 | /* a uprobe exists for this inode:offset combination */ |
496 | if (cur_uprobe) { | |
497 | kfree(uprobe); | |
498 | uprobe = cur_uprobe; | |
499 | iput(inode); | |
7b2d81d4 IM |
500 | } |
501 | ||
2b144498 SD |
502 | return uprobe; |
503 | } | |
504 | ||
9a98e03c | 505 | static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
506 | { |
507 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
508 | uc->next = uprobe->consumers; |
509 | uprobe->consumers = uc; | |
2b144498 | 510 | up_write(&uprobe->consumer_rwsem); |
2b144498 SD |
511 | } |
512 | ||
513 | /* | |
e3343e6a SD |
514 | * For uprobe @uprobe, delete the consumer @uc. |
515 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
516 | * or return false. |
517 | */ | |
e3343e6a | 518 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
519 | { |
520 | struct uprobe_consumer **con; | |
521 | bool ret = false; | |
522 | ||
523 | down_write(&uprobe->consumer_rwsem); | |
524 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
525 | if (*con == uc) { |
526 | *con = uc->next; | |
2b144498 SD |
527 | ret = true; |
528 | break; | |
529 | } | |
530 | } | |
531 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 532 | |
2b144498 SD |
533 | return ret; |
534 | } | |
535 | ||
2ded0980 ON |
536 | static int __copy_insn(struct address_space *mapping, struct file *filp, |
537 | void *insn, int nbytes, loff_t offset) | |
2b144498 | 538 | { |
2b144498 | 539 | struct page *page; |
2b144498 | 540 | |
cc359d18 ON |
541 | if (!mapping->a_ops->readpage) |
542 | return -EIO; | |
2b144498 SD |
543 | /* |
544 | * Ensure that the page that has the original instruction is | |
545 | * populated and in page-cache. | |
546 | */ | |
2edb7b55 | 547 | page = read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT, filp); |
2b144498 SD |
548 | if (IS_ERR(page)) |
549 | return PTR_ERR(page); | |
550 | ||
2edb7b55 | 551 | copy_from_page(page, offset, insn, nbytes); |
2b144498 | 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 | 558 | { |
2ded0980 ON |
559 | struct address_space *mapping = uprobe->inode->i_mapping; |
560 | loff_t offs = uprobe->offset; | |
803200e2 ON |
561 | void *insn = &uprobe->arch.insn; |
562 | int size = sizeof(uprobe->arch.insn); | |
2ded0980 ON |
563 | int len, err = -EIO; |
564 | ||
565 | /* Copy only available bytes, -EIO if nothing was read */ | |
566 | do { | |
567 | if (offs >= i_size_read(uprobe->inode)) | |
568 | break; | |
569 | ||
570 | len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK)); | |
571 | err = __copy_insn(mapping, filp, insn, len, offs); | |
fc36f595 | 572 | if (err) |
2ded0980 ON |
573 | break; |
574 | ||
575 | insn += len; | |
576 | offs += len; | |
577 | size -= len; | |
578 | } while (size); | |
579 | ||
580 | return err; | |
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 | ||
d4d3ccc6 ON |
591 | /* TODO: move this into _register, until then we abuse this sem. */ |
592 | down_write(&uprobe->consumer_rwsem); | |
71434f2f | 593 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
594 | goto out; |
595 | ||
cb9a19fe ON |
596 | ret = copy_insn(uprobe, file); |
597 | if (ret) | |
598 | goto out; | |
599 | ||
600 | ret = -ENOTSUPP; | |
803200e2 | 601 | if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn)) |
cb9a19fe ON |
602 | goto out; |
603 | ||
604 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
605 | if (ret) | |
606 | goto out; | |
607 | ||
f72d41fa | 608 | /* uprobe_write_opcode() assumes we don't cross page boundary */ |
cb9a19fe ON |
609 | BUG_ON((uprobe->offset & ~PAGE_MASK) + |
610 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
611 | ||
612 | smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ | |
71434f2f | 613 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
614 | |
615 | out: | |
d4d3ccc6 | 616 | up_write(&uprobe->consumer_rwsem); |
4710f05f | 617 | |
cb9a19fe ON |
618 | return ret; |
619 | } | |
620 | ||
8a7f2fa0 ON |
621 | static inline bool consumer_filter(struct uprobe_consumer *uc, |
622 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
806a98bd | 623 | { |
8a7f2fa0 | 624 | return !uc->filter || uc->filter(uc, ctx, mm); |
806a98bd ON |
625 | } |
626 | ||
8a7f2fa0 ON |
627 | static bool filter_chain(struct uprobe *uprobe, |
628 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
63633cbf | 629 | { |
1ff6fee5 ON |
630 | struct uprobe_consumer *uc; |
631 | bool ret = false; | |
632 | ||
633 | down_read(&uprobe->consumer_rwsem); | |
634 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
8a7f2fa0 | 635 | ret = consumer_filter(uc, ctx, mm); |
1ff6fee5 ON |
636 | if (ret) |
637 | break; | |
638 | } | |
639 | up_read(&uprobe->consumer_rwsem); | |
640 | ||
641 | return ret; | |
63633cbf ON |
642 | } |
643 | ||
e3343e6a SD |
644 | static int |
645 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 646 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 647 | { |
f8ac4ec9 | 648 | bool first_uprobe; |
2b144498 SD |
649 | int ret; |
650 | ||
cb9a19fe ON |
651 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
652 | if (ret) | |
653 | return ret; | |
682968e0 | 654 | |
f8ac4ec9 ON |
655 | /* |
656 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
657 | * the task can hit this breakpoint right after __replace_page(). | |
658 | */ | |
659 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
660 | if (first_uprobe) | |
661 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
662 | ||
816c03fb | 663 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
664 | if (!ret) |
665 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
666 | else if (first_uprobe) | |
f8ac4ec9 | 667 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
668 | |
669 | return ret; | |
670 | } | |
671 | ||
076a365b | 672 | static int |
816c03fb | 673 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 674 | { |
9f68f672 | 675 | set_bit(MMF_RECALC_UPROBES, &mm->flags); |
076a365b | 676 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
677 | } |
678 | ||
06b7bcd8 ON |
679 | static inline bool uprobe_is_active(struct uprobe *uprobe) |
680 | { | |
681 | return !RB_EMPTY_NODE(&uprobe->rb_node); | |
682 | } | |
0326f5a9 | 683 | /* |
778b032d ON |
684 | * There could be threads that have already hit the breakpoint. They |
685 | * will recheck the current insn and restart if find_uprobe() fails. | |
686 | * See find_active_uprobe(). | |
0326f5a9 | 687 | */ |
2b144498 SD |
688 | static void delete_uprobe(struct uprobe *uprobe) |
689 | { | |
06b7bcd8 ON |
690 | if (WARN_ON(!uprobe_is_active(uprobe))) |
691 | return; | |
692 | ||
6f47caa0 | 693 | spin_lock(&uprobes_treelock); |
2b144498 | 694 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 695 | spin_unlock(&uprobes_treelock); |
06b7bcd8 | 696 | RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */ |
2b144498 SD |
697 | iput(uprobe->inode); |
698 | put_uprobe(uprobe); | |
2b144498 SD |
699 | } |
700 | ||
26872090 ON |
701 | struct map_info { |
702 | struct map_info *next; | |
703 | struct mm_struct *mm; | |
816c03fb | 704 | unsigned long vaddr; |
26872090 ON |
705 | }; |
706 | ||
707 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 708 | { |
26872090 ON |
709 | struct map_info *next = info->next; |
710 | kfree(info); | |
711 | return next; | |
712 | } | |
713 | ||
714 | static struct map_info * | |
715 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
716 | { | |
717 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 718 | struct vm_area_struct *vma; |
26872090 ON |
719 | struct map_info *curr = NULL; |
720 | struct map_info *prev = NULL; | |
721 | struct map_info *info; | |
722 | int more = 0; | |
2b144498 | 723 | |
26872090 ON |
724 | again: |
725 | mutex_lock(&mapping->i_mmap_mutex); | |
6b2dbba8 | 726 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
727 | if (!valid_vma(vma, is_register)) |
728 | continue; | |
729 | ||
7a5bfb66 ON |
730 | if (!prev && !more) { |
731 | /* | |
732 | * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through | |
733 | * reclaim. This is optimistic, no harm done if it fails. | |
734 | */ | |
735 | prev = kmalloc(sizeof(struct map_info), | |
736 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
737 | if (prev) | |
738 | prev->next = NULL; | |
739 | } | |
26872090 ON |
740 | if (!prev) { |
741 | more++; | |
742 | continue; | |
2b144498 | 743 | } |
2b144498 | 744 | |
26872090 ON |
745 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
746 | continue; | |
7b2d81d4 | 747 | |
26872090 ON |
748 | info = prev; |
749 | prev = prev->next; | |
750 | info->next = curr; | |
751 | curr = info; | |
2b144498 | 752 | |
26872090 | 753 | info->mm = vma->vm_mm; |
57683f72 | 754 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 755 | } |
2b144498 SD |
756 | mutex_unlock(&mapping->i_mmap_mutex); |
757 | ||
26872090 ON |
758 | if (!more) |
759 | goto out; | |
760 | ||
761 | prev = curr; | |
762 | while (curr) { | |
763 | mmput(curr->mm); | |
764 | curr = curr->next; | |
765 | } | |
7b2d81d4 | 766 | |
26872090 ON |
767 | do { |
768 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
769 | if (!info) { | |
770 | curr = ERR_PTR(-ENOMEM); | |
771 | goto out; | |
772 | } | |
773 | info->next = prev; | |
774 | prev = info; | |
775 | } while (--more); | |
776 | ||
777 | goto again; | |
778 | out: | |
779 | while (prev) | |
780 | prev = free_map_info(prev); | |
781 | return curr; | |
2b144498 SD |
782 | } |
783 | ||
bdf8647c ON |
784 | static int |
785 | register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) | |
2b144498 | 786 | { |
bdf8647c | 787 | bool is_register = !!new; |
26872090 ON |
788 | struct map_info *info; |
789 | int err = 0; | |
2b144498 | 790 | |
32cdba1e | 791 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
792 | info = build_map_info(uprobe->inode->i_mapping, |
793 | uprobe->offset, is_register); | |
32cdba1e ON |
794 | if (IS_ERR(info)) { |
795 | err = PTR_ERR(info); | |
796 | goto out; | |
797 | } | |
7b2d81d4 | 798 | |
26872090 ON |
799 | while (info) { |
800 | struct mm_struct *mm = info->mm; | |
801 | struct vm_area_struct *vma; | |
7b2d81d4 | 802 | |
076a365b | 803 | if (err && is_register) |
26872090 | 804 | goto free; |
7b2d81d4 | 805 | |
77fc4af1 | 806 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
807 | vma = find_vma(mm, info->vaddr); |
808 | if (!vma || !valid_vma(vma, is_register) || | |
f281769e | 809 | file_inode(vma->vm_file) != uprobe->inode) |
26872090 ON |
810 | goto unlock; |
811 | ||
f4d6dfe5 ON |
812 | if (vma->vm_start > info->vaddr || |
813 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 814 | goto unlock; |
2b144498 | 815 | |
806a98bd ON |
816 | if (is_register) { |
817 | /* consult only the "caller", new consumer. */ | |
bdf8647c | 818 | if (consumer_filter(new, |
8a7f2fa0 | 819 | UPROBE_FILTER_REGISTER, mm)) |
806a98bd ON |
820 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
821 | } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) { | |
8a7f2fa0 ON |
822 | if (!filter_chain(uprobe, |
823 | UPROBE_FILTER_UNREGISTER, mm)) | |
806a98bd ON |
824 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
825 | } | |
78f74116 | 826 | |
26872090 ON |
827 | unlock: |
828 | up_write(&mm->mmap_sem); | |
829 | free: | |
830 | mmput(mm); | |
831 | info = free_map_info(info); | |
2b144498 | 832 | } |
32cdba1e ON |
833 | out: |
834 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 835 | return err; |
2b144498 SD |
836 | } |
837 | ||
9a98e03c | 838 | static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 839 | { |
9a98e03c | 840 | consumer_add(uprobe, uc); |
bdf8647c | 841 | return register_for_each_vma(uprobe, uc); |
2b144498 SD |
842 | } |
843 | ||
04aab9b2 | 844 | static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 845 | { |
04aab9b2 ON |
846 | int err; |
847 | ||
848 | if (!consumer_del(uprobe, uc)) /* WARN? */ | |
849 | return; | |
2b144498 | 850 | |
bdf8647c | 851 | err = register_for_each_vma(uprobe, NULL); |
bb929284 ON |
852 | /* TODO : cant unregister? schedule a worker thread */ |
853 | if (!uprobe->consumers && !err) | |
854 | delete_uprobe(uprobe); | |
2b144498 SD |
855 | } |
856 | ||
857 | /* | |
7b2d81d4 | 858 | * uprobe_register - register a probe |
2b144498 SD |
859 | * @inode: the file in which the probe has to be placed. |
860 | * @offset: offset from the start of the file. | |
e3343e6a | 861 | * @uc: information on howto handle the probe.. |
2b144498 | 862 | * |
7b2d81d4 | 863 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
864 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
865 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 866 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 867 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 868 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
869 | * unregisters. |
870 | * | |
871 | * Return errno if it cannot successully install probes | |
872 | * else return 0 (success) | |
873 | */ | |
e3343e6a | 874 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
875 | { |
876 | struct uprobe *uprobe; | |
7b2d81d4 | 877 | int ret; |
2b144498 | 878 | |
ea024870 AA |
879 | /* Uprobe must have at least one set consumer */ |
880 | if (!uc->handler && !uc->ret_handler) | |
881 | return -EINVAL; | |
882 | ||
f0744af7 | 883 | /* Racy, just to catch the obvious mistakes */ |
2b144498 | 884 | if (offset > i_size_read(inode)) |
7b2d81d4 | 885 | return -EINVAL; |
2b144498 | 886 | |
66d06dff | 887 | retry: |
2b144498 | 888 | uprobe = alloc_uprobe(inode, offset); |
66d06dff ON |
889 | if (!uprobe) |
890 | return -ENOMEM; | |
891 | /* | |
892 | * We can race with uprobe_unregister()->delete_uprobe(). | |
893 | * Check uprobe_is_active() and retry if it is false. | |
894 | */ | |
895 | down_write(&uprobe->register_rwsem); | |
896 | ret = -EAGAIN; | |
897 | if (likely(uprobe_is_active(uprobe))) { | |
9a98e03c ON |
898 | ret = __uprobe_register(uprobe, uc); |
899 | if (ret) | |
04aab9b2 | 900 | __uprobe_unregister(uprobe, uc); |
2b144498 | 901 | } |
66d06dff ON |
902 | up_write(&uprobe->register_rwsem); |
903 | put_uprobe(uprobe); | |
2b144498 | 904 | |
66d06dff ON |
905 | if (unlikely(ret == -EAGAIN)) |
906 | goto retry; | |
2b144498 SD |
907 | return ret; |
908 | } | |
e8440c14 | 909 | EXPORT_SYMBOL_GPL(uprobe_register); |
2b144498 | 910 | |
bdf8647c ON |
911 | /* |
912 | * uprobe_apply - unregister a already registered probe. | |
913 | * @inode: the file in which the probe has to be removed. | |
914 | * @offset: offset from the start of the file. | |
915 | * @uc: consumer which wants to add more or remove some breakpoints | |
916 | * @add: add or remove the breakpoints | |
917 | */ | |
918 | int uprobe_apply(struct inode *inode, loff_t offset, | |
919 | struct uprobe_consumer *uc, bool add) | |
920 | { | |
921 | struct uprobe *uprobe; | |
922 | struct uprobe_consumer *con; | |
923 | int ret = -ENOENT; | |
924 | ||
925 | uprobe = find_uprobe(inode, offset); | |
926 | if (!uprobe) | |
927 | return ret; | |
928 | ||
929 | down_write(&uprobe->register_rwsem); | |
930 | for (con = uprobe->consumers; con && con != uc ; con = con->next) | |
931 | ; | |
932 | if (con) | |
933 | ret = register_for_each_vma(uprobe, add ? uc : NULL); | |
934 | up_write(&uprobe->register_rwsem); | |
935 | put_uprobe(uprobe); | |
936 | ||
937 | return ret; | |
938 | } | |
939 | ||
2b144498 | 940 | /* |
7b2d81d4 | 941 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
942 | * @inode: the file in which the probe has to be removed. |
943 | * @offset: offset from the start of the file. | |
e3343e6a | 944 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 945 | */ |
e3343e6a | 946 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 947 | { |
7b2d81d4 | 948 | struct uprobe *uprobe; |
2b144498 | 949 | |
2b144498 SD |
950 | uprobe = find_uprobe(inode, offset); |
951 | if (!uprobe) | |
952 | return; | |
953 | ||
e591c8d7 | 954 | down_write(&uprobe->register_rwsem); |
04aab9b2 | 955 | __uprobe_unregister(uprobe, uc); |
e591c8d7 | 956 | up_write(&uprobe->register_rwsem); |
c91368c4 | 957 | put_uprobe(uprobe); |
2b144498 | 958 | } |
e8440c14 | 959 | EXPORT_SYMBOL_GPL(uprobe_unregister); |
2b144498 | 960 | |
da1816b1 ON |
961 | static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm) |
962 | { | |
963 | struct vm_area_struct *vma; | |
964 | int err = 0; | |
965 | ||
966 | down_read(&mm->mmap_sem); | |
967 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
968 | unsigned long vaddr; | |
969 | loff_t offset; | |
970 | ||
971 | if (!valid_vma(vma, false) || | |
f281769e | 972 | file_inode(vma->vm_file) != uprobe->inode) |
da1816b1 ON |
973 | continue; |
974 | ||
975 | offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT; | |
976 | if (uprobe->offset < offset || | |
977 | uprobe->offset >= offset + vma->vm_end - vma->vm_start) | |
978 | continue; | |
979 | ||
980 | vaddr = offset_to_vaddr(vma, uprobe->offset); | |
981 | err |= remove_breakpoint(uprobe, mm, vaddr); | |
982 | } | |
983 | up_read(&mm->mmap_sem); | |
984 | ||
985 | return err; | |
986 | } | |
987 | ||
891c3970 ON |
988 | static struct rb_node * |
989 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 990 | { |
2b144498 | 991 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
992 | |
993 | while (n) { | |
891c3970 | 994 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 995 | |
891c3970 | 996 | if (inode < u->inode) { |
2b144498 | 997 | n = n->rb_left; |
891c3970 | 998 | } else if (inode > u->inode) { |
2b144498 | 999 | n = n->rb_right; |
891c3970 ON |
1000 | } else { |
1001 | if (max < u->offset) | |
1002 | n = n->rb_left; | |
1003 | else if (min > u->offset) | |
1004 | n = n->rb_right; | |
1005 | else | |
1006 | break; | |
1007 | } | |
2b144498 | 1008 | } |
7b2d81d4 | 1009 | |
891c3970 | 1010 | return n; |
2b144498 SD |
1011 | } |
1012 | ||
1013 | /* | |
891c3970 | 1014 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 1015 | */ |
891c3970 ON |
1016 | static void build_probe_list(struct inode *inode, |
1017 | struct vm_area_struct *vma, | |
1018 | unsigned long start, unsigned long end, | |
1019 | struct list_head *head) | |
2b144498 | 1020 | { |
891c3970 | 1021 | loff_t min, max; |
891c3970 ON |
1022 | struct rb_node *n, *t; |
1023 | struct uprobe *u; | |
7b2d81d4 | 1024 | |
891c3970 | 1025 | INIT_LIST_HEAD(head); |
cb113b47 | 1026 | min = vaddr_to_offset(vma, start); |
891c3970 | 1027 | max = min + (end - start) - 1; |
2b144498 | 1028 | |
6f47caa0 | 1029 | spin_lock(&uprobes_treelock); |
891c3970 ON |
1030 | n = find_node_in_range(inode, min, max); |
1031 | if (n) { | |
1032 | for (t = n; t; t = rb_prev(t)) { | |
1033 | u = rb_entry(t, struct uprobe, rb_node); | |
1034 | if (u->inode != inode || u->offset < min) | |
1035 | break; | |
1036 | list_add(&u->pending_list, head); | |
1037 | atomic_inc(&u->ref); | |
1038 | } | |
1039 | for (t = n; (t = rb_next(t)); ) { | |
1040 | u = rb_entry(t, struct uprobe, rb_node); | |
1041 | if (u->inode != inode || u->offset > max) | |
1042 | break; | |
1043 | list_add(&u->pending_list, head); | |
1044 | atomic_inc(&u->ref); | |
1045 | } | |
2b144498 | 1046 | } |
6f47caa0 | 1047 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
1048 | } |
1049 | ||
1050 | /* | |
5e5be71a | 1051 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 1052 | * |
5e5be71a ON |
1053 | * Currently we ignore all errors and always return 0, the callers |
1054 | * can't handle the failure anyway. | |
2b144498 | 1055 | */ |
7b2d81d4 | 1056 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1057 | { |
1058 | struct list_head tmp_list; | |
665605a2 | 1059 | struct uprobe *uprobe, *u; |
2b144498 | 1060 | struct inode *inode; |
2b144498 | 1061 | |
441f1eb7 | 1062 | if (no_uprobe_events() || !valid_vma(vma, true)) |
7b2d81d4 | 1063 | return 0; |
2b144498 | 1064 | |
f281769e | 1065 | inode = file_inode(vma->vm_file); |
2b144498 | 1066 | if (!inode) |
7b2d81d4 | 1067 | return 0; |
2b144498 | 1068 | |
2b144498 | 1069 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1070 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
806a98bd ON |
1071 | /* |
1072 | * We can race with uprobe_unregister(), this uprobe can be already | |
1073 | * removed. But in this case filter_chain() must return false, all | |
1074 | * consumers have gone away. | |
1075 | */ | |
665605a2 | 1076 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
806a98bd | 1077 | if (!fatal_signal_pending(current) && |
8a7f2fa0 | 1078 | filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) { |
57683f72 | 1079 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 1080 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
1081 | } |
1082 | put_uprobe(uprobe); | |
1083 | } | |
2b144498 SD |
1084 | mutex_unlock(uprobes_mmap_hash(inode)); |
1085 | ||
5e5be71a | 1086 | return 0; |
2b144498 SD |
1087 | } |
1088 | ||
9f68f672 ON |
1089 | static bool |
1090 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1091 | { | |
1092 | loff_t min, max; | |
1093 | struct inode *inode; | |
1094 | struct rb_node *n; | |
1095 | ||
f281769e | 1096 | inode = file_inode(vma->vm_file); |
9f68f672 ON |
1097 | |
1098 | min = vaddr_to_offset(vma, start); | |
1099 | max = min + (end - start) - 1; | |
1100 | ||
1101 | spin_lock(&uprobes_treelock); | |
1102 | n = find_node_in_range(inode, min, max); | |
1103 | spin_unlock(&uprobes_treelock); | |
1104 | ||
1105 | return !!n; | |
1106 | } | |
1107 | ||
682968e0 SD |
1108 | /* |
1109 | * Called in context of a munmap of a vma. | |
1110 | */ | |
cbc91f71 | 1111 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1112 | { |
441f1eb7 | 1113 | if (no_uprobe_events() || !valid_vma(vma, false)) |
682968e0 SD |
1114 | return; |
1115 | ||
2fd611a9 ON |
1116 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1117 | return; | |
1118 | ||
9f68f672 ON |
1119 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1120 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1121 | return; |
1122 | ||
9f68f672 ON |
1123 | if (vma_has_uprobes(vma, start, end)) |
1124 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1125 | } |
1126 | ||
d4b3b638 | 1127 | /* Slot allocation for XOL */ |
6441ec8b | 1128 | static int xol_add_vma(struct mm_struct *mm, struct xol_area *area) |
d4b3b638 | 1129 | { |
c8a82538 | 1130 | int ret = -EALREADY; |
d4b3b638 SD |
1131 | |
1132 | down_write(&mm->mmap_sem); | |
1133 | if (mm->uprobes_state.xol_area) | |
1134 | goto fail; | |
1135 | ||
af0d95af ON |
1136 | if (!area->vaddr) { |
1137 | /* Try to map as high as possible, this is only a hint. */ | |
1138 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, | |
1139 | PAGE_SIZE, 0, 0); | |
1140 | if (area->vaddr & ~PAGE_MASK) { | |
1141 | ret = area->vaddr; | |
1142 | goto fail; | |
1143 | } | |
d4b3b638 SD |
1144 | } |
1145 | ||
1146 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1147 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1148 | if (ret) | |
1149 | goto fail; | |
1150 | ||
1151 | smp_wmb(); /* pairs with get_xol_area() */ | |
1152 | mm->uprobes_state.xol_area = area; | |
c8a82538 | 1153 | fail: |
d4b3b638 | 1154 | up_write(&mm->mmap_sem); |
d4b3b638 SD |
1155 | |
1156 | return ret; | |
1157 | } | |
1158 | ||
af0d95af | 1159 | static struct xol_area *__create_xol_area(unsigned long vaddr) |
d4b3b638 | 1160 | { |
9b545df8 | 1161 | struct mm_struct *mm = current->mm; |
e78aebfd | 1162 | uprobe_opcode_t insn = UPROBE_SWBP_INSN; |
6441ec8b | 1163 | struct xol_area *area; |
9b545df8 | 1164 | |
af0d95af | 1165 | area = kmalloc(sizeof(*area), GFP_KERNEL); |
d4b3b638 | 1166 | if (unlikely(!area)) |
c8a82538 | 1167 | goto out; |
d4b3b638 SD |
1168 | |
1169 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
d4b3b638 | 1170 | if (!area->bitmap) |
c8a82538 ON |
1171 | goto free_area; |
1172 | ||
1173 | area->page = alloc_page(GFP_HIGHUSER); | |
1174 | if (!area->page) | |
1175 | goto free_bitmap; | |
d4b3b638 | 1176 | |
af0d95af | 1177 | area->vaddr = vaddr; |
6441ec8b ON |
1178 | init_waitqueue_head(&area->wq); |
1179 | /* Reserve the 1st slot for get_trampoline_vaddr() */ | |
e78aebfd | 1180 | set_bit(0, area->bitmap); |
e78aebfd | 1181 | atomic_set(&area->slot_count, 1); |
6441ec8b | 1182 | copy_to_page(area->page, 0, &insn, UPROBE_SWBP_INSN_SIZE); |
e78aebfd | 1183 | |
6441ec8b | 1184 | if (!xol_add_vma(mm, area)) |
d4b3b638 SD |
1185 | return area; |
1186 | ||
c8a82538 ON |
1187 | __free_page(area->page); |
1188 | free_bitmap: | |
d4b3b638 | 1189 | kfree(area->bitmap); |
c8a82538 | 1190 | free_area: |
d4b3b638 | 1191 | kfree(area); |
c8a82538 | 1192 | out: |
6441ec8b ON |
1193 | return NULL; |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * get_xol_area - Allocate process's xol_area if necessary. | |
1198 | * This area will be used for storing instructions for execution out of line. | |
1199 | * | |
1200 | * Returns the allocated area or NULL. | |
1201 | */ | |
1202 | static struct xol_area *get_xol_area(void) | |
1203 | { | |
1204 | struct mm_struct *mm = current->mm; | |
1205 | struct xol_area *area; | |
1206 | ||
1207 | if (!mm->uprobes_state.xol_area) | |
af0d95af | 1208 | __create_xol_area(0); |
6441ec8b | 1209 | |
9b545df8 | 1210 | area = mm->uprobes_state.xol_area; |
6441ec8b | 1211 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ |
9b545df8 | 1212 | return area; |
d4b3b638 SD |
1213 | } |
1214 | ||
1215 | /* | |
1216 | * uprobe_clear_state - Free the area allocated for slots. | |
1217 | */ | |
1218 | void uprobe_clear_state(struct mm_struct *mm) | |
1219 | { | |
1220 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1221 | ||
1222 | if (!area) | |
1223 | return; | |
1224 | ||
1225 | put_page(area->page); | |
1226 | kfree(area->bitmap); | |
1227 | kfree(area); | |
1228 | } | |
1229 | ||
32cdba1e ON |
1230 | void uprobe_start_dup_mmap(void) |
1231 | { | |
1232 | percpu_down_read(&dup_mmap_sem); | |
1233 | } | |
1234 | ||
1235 | void uprobe_end_dup_mmap(void) | |
1236 | { | |
1237 | percpu_up_read(&dup_mmap_sem); | |
1238 | } | |
1239 | ||
f8ac4ec9 ON |
1240 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1241 | { | |
61559a81 ON |
1242 | newmm->uprobes_state.xol_area = NULL; |
1243 | ||
9f68f672 | 1244 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1245 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1246 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1247 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1248 | } | |
f8ac4ec9 ON |
1249 | } |
1250 | ||
d4b3b638 SD |
1251 | /* |
1252 | * - search for a free slot. | |
1253 | */ | |
1254 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1255 | { | |
1256 | unsigned long slot_addr; | |
1257 | int slot_nr; | |
1258 | ||
1259 | do { | |
1260 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1261 | if (slot_nr < UINSNS_PER_PAGE) { | |
1262 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1263 | break; | |
1264 | ||
1265 | slot_nr = UINSNS_PER_PAGE; | |
1266 | continue; | |
1267 | } | |
1268 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1269 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1270 | ||
1271 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1272 | atomic_inc(&area->slot_count); | |
1273 | ||
1274 | return slot_addr; | |
1275 | } | |
1276 | ||
1277 | /* | |
a6cb3f6d | 1278 | * xol_get_insn_slot - allocate a slot for xol. |
d4b3b638 SD |
1279 | * Returns the allocated slot address or 0. |
1280 | */ | |
a6cb3f6d | 1281 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe) |
d4b3b638 SD |
1282 | { |
1283 | struct xol_area *area; | |
a6cb3f6d | 1284 | unsigned long xol_vaddr; |
d4b3b638 | 1285 | |
9b545df8 ON |
1286 | area = get_xol_area(); |
1287 | if (!area) | |
1288 | return 0; | |
d4b3b638 | 1289 | |
a6cb3f6d ON |
1290 | xol_vaddr = xol_take_insn_slot(area); |
1291 | if (unlikely(!xol_vaddr)) | |
d4b3b638 SD |
1292 | return 0; |
1293 | ||
a6cb3f6d | 1294 | /* Initialize the slot */ |
8a8de66c | 1295 | copy_to_page(area->page, xol_vaddr, |
803200e2 | 1296 | &uprobe->arch.ixol, sizeof(uprobe->arch.ixol)); |
65b6ecc0 RV |
1297 | /* |
1298 | * We probably need flush_icache_user_range() but it needs vma. | |
1299 | * This should work on supported architectures too. | |
1300 | */ | |
1301 | flush_dcache_page(area->page); | |
d4b3b638 | 1302 | |
a6cb3f6d | 1303 | return xol_vaddr; |
d4b3b638 SD |
1304 | } |
1305 | ||
1306 | /* | |
1307 | * xol_free_insn_slot - If slot was earlier allocated by | |
1308 | * @xol_get_insn_slot(), make the slot available for | |
1309 | * subsequent requests. | |
1310 | */ | |
1311 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1312 | { | |
1313 | struct xol_area *area; | |
1314 | unsigned long vma_end; | |
1315 | unsigned long slot_addr; | |
1316 | ||
1317 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1318 | return; | |
1319 | ||
1320 | slot_addr = tsk->utask->xol_vaddr; | |
af4355e9 | 1321 | if (unlikely(!slot_addr)) |
d4b3b638 SD |
1322 | return; |
1323 | ||
1324 | area = tsk->mm->uprobes_state.xol_area; | |
1325 | vma_end = area->vaddr + PAGE_SIZE; | |
1326 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1327 | unsigned long offset; | |
1328 | int slot_nr; | |
1329 | ||
1330 | offset = slot_addr - area->vaddr; | |
1331 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1332 | if (slot_nr >= UINSNS_PER_PAGE) | |
1333 | return; | |
1334 | ||
1335 | clear_bit(slot_nr, area->bitmap); | |
1336 | atomic_dec(&area->slot_count); | |
1337 | if (waitqueue_active(&area->wq)) | |
1338 | wake_up(&area->wq); | |
1339 | ||
1340 | tsk->utask->xol_vaddr = 0; | |
1341 | } | |
1342 | } | |
1343 | ||
0326f5a9 SD |
1344 | /** |
1345 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1346 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1347 | * instruction. | |
1348 | * Return the address of the breakpoint instruction. | |
1349 | */ | |
1350 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1351 | { | |
1352 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * Called with no locks held. | |
1357 | * Called in context of a exiting or a exec-ing thread. | |
1358 | */ | |
1359 | void uprobe_free_utask(struct task_struct *t) | |
1360 | { | |
1361 | struct uprobe_task *utask = t->utask; | |
0dfd0eb8 | 1362 | struct return_instance *ri, *tmp; |
0326f5a9 | 1363 | |
0326f5a9 SD |
1364 | if (!utask) |
1365 | return; | |
1366 | ||
1367 | if (utask->active_uprobe) | |
1368 | put_uprobe(utask->active_uprobe); | |
1369 | ||
0dfd0eb8 AA |
1370 | ri = utask->return_instances; |
1371 | while (ri) { | |
1372 | tmp = ri; | |
1373 | ri = ri->next; | |
1374 | ||
1375 | put_uprobe(tmp->uprobe); | |
1376 | kfree(tmp); | |
1377 | } | |
1378 | ||
d4b3b638 | 1379 | xol_free_insn_slot(t); |
0326f5a9 SD |
1380 | kfree(utask); |
1381 | t->utask = NULL; | |
1382 | } | |
1383 | ||
0326f5a9 | 1384 | /* |
5a2df662 ON |
1385 | * Allocate a uprobe_task object for the task if if necessary. |
1386 | * Called when the thread hits a breakpoint. | |
0326f5a9 SD |
1387 | * |
1388 | * Returns: | |
1389 | * - pointer to new uprobe_task on success | |
1390 | * - NULL otherwise | |
1391 | */ | |
5a2df662 | 1392 | static struct uprobe_task *get_utask(void) |
0326f5a9 | 1393 | { |
5a2df662 ON |
1394 | if (!current->utask) |
1395 | current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1396 | return current->utask; | |
0326f5a9 SD |
1397 | } |
1398 | ||
248d3a7b ON |
1399 | static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask) |
1400 | { | |
1401 | struct uprobe_task *n_utask; | |
1402 | struct return_instance **p, *o, *n; | |
1403 | ||
1404 | n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1405 | if (!n_utask) | |
1406 | return -ENOMEM; | |
1407 | t->utask = n_utask; | |
1408 | ||
1409 | p = &n_utask->return_instances; | |
1410 | for (o = o_utask->return_instances; o; o = o->next) { | |
1411 | n = kmalloc(sizeof(struct return_instance), GFP_KERNEL); | |
1412 | if (!n) | |
1413 | return -ENOMEM; | |
1414 | ||
1415 | *n = *o; | |
1416 | atomic_inc(&n->uprobe->ref); | |
1417 | n->next = NULL; | |
1418 | ||
1419 | *p = n; | |
1420 | p = &n->next; | |
1421 | n_utask->depth++; | |
1422 | } | |
1423 | ||
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | static void uprobe_warn(struct task_struct *t, const char *msg) | |
1428 | { | |
1429 | pr_warn("uprobe: %s:%d failed to %s\n", | |
1430 | current->comm, current->pid, msg); | |
1431 | } | |
1432 | ||
aa59c53f ON |
1433 | static void dup_xol_work(struct callback_head *work) |
1434 | { | |
aa59c53f ON |
1435 | if (current->flags & PF_EXITING) |
1436 | return; | |
1437 | ||
32473431 | 1438 | if (!__create_xol_area(current->utask->dup_xol_addr)) |
aa59c53f ON |
1439 | uprobe_warn(current, "dup xol area"); |
1440 | } | |
1441 | ||
b68e0749 ON |
1442 | /* |
1443 | * Called in context of a new clone/fork from copy_process. | |
1444 | */ | |
3ab67966 | 1445 | void uprobe_copy_process(struct task_struct *t, unsigned long flags) |
b68e0749 | 1446 | { |
248d3a7b ON |
1447 | struct uprobe_task *utask = current->utask; |
1448 | struct mm_struct *mm = current->mm; | |
aa59c53f | 1449 | struct xol_area *area; |
248d3a7b | 1450 | |
b68e0749 | 1451 | t->utask = NULL; |
248d3a7b | 1452 | |
3ab67966 ON |
1453 | if (!utask || !utask->return_instances) |
1454 | return; | |
1455 | ||
1456 | if (mm == t->mm && !(flags & CLONE_VFORK)) | |
248d3a7b ON |
1457 | return; |
1458 | ||
1459 | if (dup_utask(t, utask)) | |
1460 | return uprobe_warn(t, "dup ret instances"); | |
aa59c53f ON |
1461 | |
1462 | /* The task can fork() after dup_xol_work() fails */ | |
1463 | area = mm->uprobes_state.xol_area; | |
1464 | if (!area) | |
1465 | return uprobe_warn(t, "dup xol area"); | |
1466 | ||
3ab67966 ON |
1467 | if (mm == t->mm) |
1468 | return; | |
1469 | ||
32473431 ON |
1470 | t->utask->dup_xol_addr = area->vaddr; |
1471 | init_task_work(&t->utask->dup_xol_work, dup_xol_work); | |
1472 | task_work_add(t, &t->utask->dup_xol_work, true); | |
b68e0749 ON |
1473 | } |
1474 | ||
e78aebfd AA |
1475 | /* |
1476 | * Current area->vaddr notion assume the trampoline address is always | |
1477 | * equal area->vaddr. | |
1478 | * | |
1479 | * Returns -1 in case the xol_area is not allocated. | |
1480 | */ | |
1481 | static unsigned long get_trampoline_vaddr(void) | |
1482 | { | |
1483 | struct xol_area *area; | |
1484 | unsigned long trampoline_vaddr = -1; | |
1485 | ||
1486 | area = current->mm->uprobes_state.xol_area; | |
1487 | smp_read_barrier_depends(); | |
1488 | if (area) | |
1489 | trampoline_vaddr = area->vaddr; | |
1490 | ||
1491 | return trampoline_vaddr; | |
1492 | } | |
1493 | ||
0dfd0eb8 AA |
1494 | static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs) |
1495 | { | |
1496 | struct return_instance *ri; | |
1497 | struct uprobe_task *utask; | |
1498 | unsigned long orig_ret_vaddr, trampoline_vaddr; | |
1499 | bool chained = false; | |
1500 | ||
1501 | if (!get_xol_area()) | |
1502 | return; | |
1503 | ||
1504 | utask = get_utask(); | |
1505 | if (!utask) | |
1506 | return; | |
1507 | ||
ded49c55 AA |
1508 | if (utask->depth >= MAX_URETPROBE_DEPTH) { |
1509 | printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to" | |
1510 | " nestedness limit pid/tgid=%d/%d\n", | |
1511 | current->pid, current->tgid); | |
1512 | return; | |
1513 | } | |
1514 | ||
0dfd0eb8 AA |
1515 | ri = kzalloc(sizeof(struct return_instance), GFP_KERNEL); |
1516 | if (!ri) | |
1517 | goto fail; | |
1518 | ||
1519 | trampoline_vaddr = get_trampoline_vaddr(); | |
1520 | orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs); | |
1521 | if (orig_ret_vaddr == -1) | |
1522 | goto fail; | |
1523 | ||
1524 | /* | |
1525 | * We don't want to keep trampoline address in stack, rather keep the | |
1526 | * original return address of first caller thru all the consequent | |
1527 | * instances. This also makes breakpoint unwrapping easier. | |
1528 | */ | |
1529 | if (orig_ret_vaddr == trampoline_vaddr) { | |
1530 | if (!utask->return_instances) { | |
1531 | /* | |
1532 | * This situation is not possible. Likely we have an | |
1533 | * attack from user-space. | |
1534 | */ | |
1535 | pr_warn("uprobe: unable to set uretprobe pid/tgid=%d/%d\n", | |
1536 | current->pid, current->tgid); | |
1537 | goto fail; | |
1538 | } | |
1539 | ||
1540 | chained = true; | |
1541 | orig_ret_vaddr = utask->return_instances->orig_ret_vaddr; | |
1542 | } | |
1543 | ||
1544 | atomic_inc(&uprobe->ref); | |
1545 | ri->uprobe = uprobe; | |
1546 | ri->func = instruction_pointer(regs); | |
1547 | ri->orig_ret_vaddr = orig_ret_vaddr; | |
1548 | ri->chained = chained; | |
1549 | ||
ded49c55 AA |
1550 | utask->depth++; |
1551 | ||
0dfd0eb8 AA |
1552 | /* add instance to the stack */ |
1553 | ri->next = utask->return_instances; | |
1554 | utask->return_instances = ri; | |
1555 | ||
1556 | return; | |
1557 | ||
1558 | fail: | |
1559 | kfree(ri); | |
1560 | } | |
1561 | ||
0326f5a9 SD |
1562 | /* Prepare to single-step probed instruction out of line. */ |
1563 | static int | |
a6cb3f6d | 1564 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr) |
0326f5a9 | 1565 | { |
a6cb3f6d ON |
1566 | struct uprobe_task *utask; |
1567 | unsigned long xol_vaddr; | |
aba51024 | 1568 | int err; |
a6cb3f6d | 1569 | |
608e7427 ON |
1570 | utask = get_utask(); |
1571 | if (!utask) | |
1572 | return -ENOMEM; | |
a6cb3f6d ON |
1573 | |
1574 | xol_vaddr = xol_get_insn_slot(uprobe); | |
1575 | if (!xol_vaddr) | |
1576 | return -ENOMEM; | |
1577 | ||
1578 | utask->xol_vaddr = xol_vaddr; | |
1579 | utask->vaddr = bp_vaddr; | |
d4b3b638 | 1580 | |
aba51024 ON |
1581 | err = arch_uprobe_pre_xol(&uprobe->arch, regs); |
1582 | if (unlikely(err)) { | |
1583 | xol_free_insn_slot(current); | |
1584 | return err; | |
1585 | } | |
1586 | ||
608e7427 ON |
1587 | utask->active_uprobe = uprobe; |
1588 | utask->state = UTASK_SSTEP; | |
aba51024 | 1589 | return 0; |
0326f5a9 SD |
1590 | } |
1591 | ||
1592 | /* | |
1593 | * If we are singlestepping, then ensure this thread is not connected to | |
1594 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1595 | * triggers the signal, restart the original insn even if the task is | |
1596 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1597 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1598 | * _same_ instruction should be repeated again after return from the signal | |
1599 | * handler, and SSTEP can never finish in this case. | |
1600 | */ | |
1601 | bool uprobe_deny_signal(void) | |
1602 | { | |
1603 | struct task_struct *t = current; | |
1604 | struct uprobe_task *utask = t->utask; | |
1605 | ||
1606 | if (likely(!utask || !utask->active_uprobe)) | |
1607 | return false; | |
1608 | ||
1609 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1610 | ||
1611 | if (signal_pending(t)) { | |
1612 | spin_lock_irq(&t->sighand->siglock); | |
1613 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1614 | spin_unlock_irq(&t->sighand->siglock); | |
1615 | ||
1616 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1617 | utask->state = UTASK_SSTEP_TRAPPED; | |
1618 | set_tsk_thread_flag(t, TIF_UPROBE); | |
1619 | set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); | |
1620 | } | |
1621 | } | |
1622 | ||
1623 | return true; | |
1624 | } | |
1625 | ||
499a4f3e ON |
1626 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1627 | { | |
1628 | struct vm_area_struct *vma; | |
1629 | ||
1630 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1631 | if (!valid_vma(vma, false)) | |
1632 | continue; | |
1633 | /* | |
1634 | * This is not strictly accurate, we can race with | |
1635 | * uprobe_unregister() and see the already removed | |
1636 | * uprobe if delete_uprobe() was not yet called. | |
63633cbf | 1637 | * Or this uprobe can be filtered out. |
499a4f3e ON |
1638 | */ |
1639 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1640 | return; | |
1641 | } | |
1642 | ||
1643 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1644 | } | |
1645 | ||
0908ad6e | 1646 | static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr) |
ec75fba9 ON |
1647 | { |
1648 | struct page *page; | |
1649 | uprobe_opcode_t opcode; | |
1650 | int result; | |
1651 | ||
1652 | pagefault_disable(); | |
1653 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
1654 | sizeof(opcode)); | |
1655 | pagefault_enable(); | |
1656 | ||
1657 | if (likely(result == 0)) | |
1658 | goto out; | |
1659 | ||
1660 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); | |
1661 | if (result < 0) | |
1662 | return result; | |
1663 | ||
ab0d805c | 1664 | copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); |
ec75fba9 ON |
1665 | put_page(page); |
1666 | out: | |
0908ad6e AM |
1667 | /* This needs to return true for any variant of the trap insn */ |
1668 | return is_trap_insn(&opcode); | |
ec75fba9 ON |
1669 | } |
1670 | ||
d790d346 | 1671 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1672 | { |
3a9ea052 ON |
1673 | struct mm_struct *mm = current->mm; |
1674 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1675 | struct vm_area_struct *vma; |
0326f5a9 | 1676 | |
0326f5a9 SD |
1677 | down_read(&mm->mmap_sem); |
1678 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1679 | if (vma && vma->vm_start <= bp_vaddr) { |
1680 | if (valid_vma(vma, false)) { | |
f281769e | 1681 | struct inode *inode = file_inode(vma->vm_file); |
cb113b47 | 1682 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); |
0326f5a9 | 1683 | |
3a9ea052 ON |
1684 | uprobe = find_uprobe(inode, offset); |
1685 | } | |
d790d346 ON |
1686 | |
1687 | if (!uprobe) | |
0908ad6e | 1688 | *is_swbp = is_trap_at_addr(mm, bp_vaddr); |
d790d346 ON |
1689 | } else { |
1690 | *is_swbp = -EFAULT; | |
0326f5a9 | 1691 | } |
499a4f3e ON |
1692 | |
1693 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1694 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1695 | up_read(&mm->mmap_sem); |
1696 | ||
3a9ea052 ON |
1697 | return uprobe; |
1698 | } | |
1699 | ||
da1816b1 ON |
1700 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
1701 | { | |
1702 | struct uprobe_consumer *uc; | |
1703 | int remove = UPROBE_HANDLER_REMOVE; | |
0dfd0eb8 | 1704 | bool need_prep = false; /* prepare return uprobe, when needed */ |
da1816b1 ON |
1705 | |
1706 | down_read(&uprobe->register_rwsem); | |
1707 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
ea024870 | 1708 | int rc = 0; |
da1816b1 | 1709 | |
ea024870 AA |
1710 | if (uc->handler) { |
1711 | rc = uc->handler(uc, regs); | |
1712 | WARN(rc & ~UPROBE_HANDLER_MASK, | |
1713 | "bad rc=0x%x from %pf()\n", rc, uc->handler); | |
1714 | } | |
0dfd0eb8 AA |
1715 | |
1716 | if (uc->ret_handler) | |
1717 | need_prep = true; | |
1718 | ||
da1816b1 ON |
1719 | remove &= rc; |
1720 | } | |
1721 | ||
0dfd0eb8 AA |
1722 | if (need_prep && !remove) |
1723 | prepare_uretprobe(uprobe, regs); /* put bp at return */ | |
1724 | ||
da1816b1 ON |
1725 | if (remove && uprobe->consumers) { |
1726 | WARN_ON(!uprobe_is_active(uprobe)); | |
1727 | unapply_uprobe(uprobe, current->mm); | |
1728 | } | |
1729 | up_read(&uprobe->register_rwsem); | |
1730 | } | |
1731 | ||
fec8898d AA |
1732 | static void |
1733 | handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs) | |
1734 | { | |
1735 | struct uprobe *uprobe = ri->uprobe; | |
1736 | struct uprobe_consumer *uc; | |
1737 | ||
1738 | down_read(&uprobe->register_rwsem); | |
1739 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
1740 | if (uc->ret_handler) | |
1741 | uc->ret_handler(uc, ri->func, regs); | |
1742 | } | |
1743 | up_read(&uprobe->register_rwsem); | |
1744 | } | |
1745 | ||
1746 | static bool handle_trampoline(struct pt_regs *regs) | |
1747 | { | |
1748 | struct uprobe_task *utask; | |
1749 | struct return_instance *ri, *tmp; | |
1750 | bool chained; | |
1751 | ||
1752 | utask = current->utask; | |
1753 | if (!utask) | |
1754 | return false; | |
1755 | ||
1756 | ri = utask->return_instances; | |
1757 | if (!ri) | |
1758 | return false; | |
1759 | ||
1760 | /* | |
1761 | * TODO: we should throw out return_instance's invalidated by | |
1762 | * longjmp(), currently we assume that the probed function always | |
1763 | * returns. | |
1764 | */ | |
1765 | instruction_pointer_set(regs, ri->orig_ret_vaddr); | |
1766 | ||
1767 | for (;;) { | |
1768 | handle_uretprobe_chain(ri, regs); | |
1769 | ||
1770 | chained = ri->chained; | |
1771 | put_uprobe(ri->uprobe); | |
1772 | ||
1773 | tmp = ri; | |
1774 | ri = ri->next; | |
1775 | kfree(tmp); | |
878b5a6e | 1776 | utask->depth--; |
fec8898d AA |
1777 | |
1778 | if (!chained) | |
1779 | break; | |
fec8898d AA |
1780 | BUG_ON(!ri); |
1781 | } | |
1782 | ||
1783 | utask->return_instances = ri; | |
1784 | ||
1785 | return true; | |
1786 | } | |
1787 | ||
6fe50a28 DL |
1788 | bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs) |
1789 | { | |
1790 | return false; | |
1791 | } | |
1792 | ||
3a9ea052 ON |
1793 | /* |
1794 | * Run handler and ask thread to singlestep. | |
1795 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1796 | */ | |
1797 | static void handle_swbp(struct pt_regs *regs) | |
1798 | { | |
3a9ea052 ON |
1799 | struct uprobe *uprobe; |
1800 | unsigned long bp_vaddr; | |
56bb4cf6 | 1801 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1802 | |
1803 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
fec8898d AA |
1804 | if (bp_vaddr == get_trampoline_vaddr()) { |
1805 | if (handle_trampoline(regs)) | |
1806 | return; | |
3a9ea052 | 1807 | |
fec8898d AA |
1808 | pr_warn("uprobe: unable to handle uretprobe pid/tgid=%d/%d\n", |
1809 | current->pid, current->tgid); | |
1810 | } | |
1811 | ||
1812 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); | |
0326f5a9 | 1813 | if (!uprobe) { |
56bb4cf6 ON |
1814 | if (is_swbp > 0) { |
1815 | /* No matching uprobe; signal SIGTRAP. */ | |
1816 | send_sig(SIGTRAP, current, 0); | |
1817 | } else { | |
1818 | /* | |
1819 | * Either we raced with uprobe_unregister() or we can't | |
1820 | * access this memory. The latter is only possible if | |
1821 | * another thread plays with our ->mm. In both cases | |
1822 | * we can simply restart. If this vma was unmapped we | |
1823 | * can pretend this insn was not executed yet and get | |
1824 | * the (correct) SIGSEGV after restart. | |
1825 | */ | |
1826 | instruction_pointer_set(regs, bp_vaddr); | |
1827 | } | |
0326f5a9 SD |
1828 | return; |
1829 | } | |
74e59dfc ON |
1830 | |
1831 | /* change it in advance for ->handler() and restart */ | |
1832 | instruction_pointer_set(regs, bp_vaddr); | |
1833 | ||
142b18dd ON |
1834 | /* |
1835 | * TODO: move copy_insn/etc into _register and remove this hack. | |
1836 | * After we hit the bp, _unregister + _register can install the | |
1837 | * new and not-yet-analyzed uprobe at the same address, restart. | |
1838 | */ | |
1839 | smp_rmb(); /* pairs with wmb() in install_breakpoint() */ | |
71434f2f | 1840 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
74e59dfc | 1841 | goto out; |
0326f5a9 | 1842 | |
72fd293a ON |
1843 | /* Tracing handlers use ->utask to communicate with fetch methods */ |
1844 | if (!get_utask()) | |
1845 | goto out; | |
1846 | ||
6fe50a28 DL |
1847 | if (arch_uprobe_ignore(&uprobe->arch, regs)) |
1848 | goto out; | |
1849 | ||
0326f5a9 | 1850 | handler_chain(uprobe, regs); |
6fe50a28 | 1851 | |
8a6b1732 | 1852 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
0578a970 | 1853 | goto out; |
0326f5a9 | 1854 | |
608e7427 | 1855 | if (!pre_ssout(uprobe, regs, bp_vaddr)) |
0326f5a9 | 1856 | return; |
0326f5a9 | 1857 | |
8a6b1732 | 1858 | /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */ |
0578a970 | 1859 | out: |
8bd87445 | 1860 | put_uprobe(uprobe); |
0326f5a9 SD |
1861 | } |
1862 | ||
1863 | /* | |
1864 | * Perform required fix-ups and disable singlestep. | |
1865 | * Allow pending signals to take effect. | |
1866 | */ | |
1867 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1868 | { | |
1869 | struct uprobe *uprobe; | |
014940ba | 1870 | int err = 0; |
0326f5a9 SD |
1871 | |
1872 | uprobe = utask->active_uprobe; | |
1873 | if (utask->state == UTASK_SSTEP_ACK) | |
014940ba | 1874 | err = arch_uprobe_post_xol(&uprobe->arch, regs); |
0326f5a9 SD |
1875 | else if (utask->state == UTASK_SSTEP_TRAPPED) |
1876 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1877 | else | |
1878 | WARN_ON_ONCE(1); | |
1879 | ||
1880 | put_uprobe(uprobe); | |
1881 | utask->active_uprobe = NULL; | |
1882 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1883 | xol_free_insn_slot(current); |
0326f5a9 SD |
1884 | |
1885 | spin_lock_irq(¤t->sighand->siglock); | |
1886 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1887 | spin_unlock_irq(¤t->sighand->siglock); | |
014940ba ON |
1888 | |
1889 | if (unlikely(err)) { | |
1890 | uprobe_warn(current, "execute the probed insn, sending SIGILL."); | |
1891 | force_sig_info(SIGILL, SEND_SIG_FORCED, current); | |
1892 | } | |
0326f5a9 SD |
1893 | } |
1894 | ||
1895 | /* | |
1b08e907 ON |
1896 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1897 | * allows the thread to return from interrupt. After that handle_swbp() | |
1898 | * sets utask->active_uprobe. | |
0326f5a9 | 1899 | * |
1b08e907 ON |
1900 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1901 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1902 | * |
1903 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1904 | * uprobe_notify_resume(). | |
1905 | */ | |
1906 | void uprobe_notify_resume(struct pt_regs *regs) | |
1907 | { | |
1908 | struct uprobe_task *utask; | |
1909 | ||
db023ea5 ON |
1910 | clear_thread_flag(TIF_UPROBE); |
1911 | ||
0326f5a9 | 1912 | utask = current->utask; |
1b08e907 | 1913 | if (utask && utask->active_uprobe) |
0326f5a9 | 1914 | handle_singlestep(utask, regs); |
1b08e907 ON |
1915 | else |
1916 | handle_swbp(regs); | |
0326f5a9 SD |
1917 | } |
1918 | ||
1919 | /* | |
1920 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1921 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1922 | */ | |
1923 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1924 | { | |
0dfd0eb8 AA |
1925 | if (!current->mm) |
1926 | return 0; | |
1927 | ||
1928 | if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) && | |
1929 | (!current->utask || !current->utask->return_instances)) | |
0326f5a9 SD |
1930 | return 0; |
1931 | ||
0326f5a9 | 1932 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1933 | return 1; |
1934 | } | |
1935 | ||
1936 | /* | |
1937 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1938 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1939 | */ | |
1940 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1941 | { | |
1942 | struct uprobe_task *utask = current->utask; | |
1943 | ||
1944 | if (!current->mm || !utask || !utask->active_uprobe) | |
1945 | /* task is currently not uprobed */ | |
1946 | return 0; | |
1947 | ||
1948 | utask->state = UTASK_SSTEP_ACK; | |
1949 | set_thread_flag(TIF_UPROBE); | |
1950 | return 1; | |
1951 | } | |
1952 | ||
1953 | static struct notifier_block uprobe_exception_nb = { | |
1954 | .notifier_call = arch_uprobe_exception_notify, | |
1955 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1956 | }; | |
1957 | ||
2b144498 SD |
1958 | static int __init init_uprobes(void) |
1959 | { | |
1960 | int i; | |
1961 | ||
66d06dff | 1962 | for (i = 0; i < UPROBES_HASH_SZ; i++) |
2b144498 | 1963 | mutex_init(&uprobes_mmap_mutex[i]); |
0326f5a9 | 1964 | |
32cdba1e ON |
1965 | if (percpu_init_rwsem(&dup_mmap_sem)) |
1966 | return -ENOMEM; | |
1967 | ||
0326f5a9 | 1968 | return register_die_notifier(&uprobe_exception_nb); |
2b144498 | 1969 | } |
736e89d9 | 1970 | __initcall(init_uprobes); |