]>
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 | ||
0dfd0eb8 AA |
89 | struct return_instance { |
90 | struct uprobe *uprobe; | |
91 | unsigned long func; | |
92 | unsigned long orig_ret_vaddr; /* original return address */ | |
93 | bool chained; /* true, if instance is nested */ | |
94 | ||
95 | struct return_instance *next; /* keep as stack */ | |
96 | }; | |
97 | ||
c912dae6 | 98 | /* |
ad439356 ON |
99 | * Execute out of line area: anonymous executable mapping installed |
100 | * by the probed task to execute the copy of the original instruction | |
101 | * mangled by set_swbp(). | |
102 | * | |
c912dae6 ON |
103 | * On a breakpoint hit, thread contests for a slot. It frees the |
104 | * slot after singlestep. Currently a fixed number of slots are | |
105 | * allocated. | |
106 | */ | |
107 | struct xol_area { | |
108 | wait_queue_head_t wq; /* if all slots are busy */ | |
109 | atomic_t slot_count; /* number of in-use slots */ | |
110 | unsigned long *bitmap; /* 0 = free slot */ | |
111 | struct page *page; | |
112 | ||
113 | /* | |
114 | * We keep the vma's vm_start rather than a pointer to the vma | |
115 | * itself. The probed process or a naughty kernel module could make | |
116 | * the vma go away, and we must handle that reasonably gracefully. | |
117 | */ | |
118 | unsigned long vaddr; /* Page(s) of instruction slots */ | |
119 | }; | |
120 | ||
2b144498 SD |
121 | /* |
122 | * valid_vma: Verify if the specified vma is an executable vma | |
123 | * Relax restrictions while unregistering: vm_flags might have | |
124 | * changed after breakpoint was inserted. | |
125 | * - is_register: indicates if we are in register context. | |
126 | * - Return 1 if the specified virtual address is in an | |
127 | * executable vma. | |
128 | */ | |
129 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
130 | { | |
13f59c5e | 131 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE; |
2b144498 | 132 | |
e40cfce6 ON |
133 | if (is_register) |
134 | flags |= VM_WRITE; | |
2b144498 | 135 | |
e40cfce6 | 136 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
137 | } |
138 | ||
57683f72 | 139 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 140 | { |
57683f72 | 141 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
142 | } |
143 | ||
cb113b47 ON |
144 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
145 | { | |
146 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
147 | } | |
148 | ||
2b144498 SD |
149 | /** |
150 | * __replace_page - replace page in vma by new page. | |
151 | * based on replace_page in mm/ksm.c | |
152 | * | |
153 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 154 | * @addr: address the old @page is mapped at |
2b144498 SD |
155 | * @page: the cowed page we are replacing by kpage |
156 | * @kpage: the modified page we replace page by | |
157 | * | |
158 | * Returns 0 on success, -EFAULT on failure. | |
159 | */ | |
c517ee74 ON |
160 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
161 | struct page *page, struct page *kpage) | |
2b144498 SD |
162 | { |
163 | struct mm_struct *mm = vma->vm_mm; | |
5323ce71 ON |
164 | spinlock_t *ptl; |
165 | pte_t *ptep; | |
9f92448c | 166 | int err; |
6bdb913f HE |
167 | /* For mmu_notifiers */ |
168 | const unsigned long mmun_start = addr; | |
169 | const unsigned long mmun_end = addr + PAGE_SIZE; | |
00501b53 JW |
170 | struct mem_cgroup *memcg; |
171 | ||
172 | err = mem_cgroup_try_charge(kpage, vma->vm_mm, GFP_KERNEL, &memcg); | |
173 | if (err) | |
174 | return err; | |
2b144498 | 175 | |
194f8dcb | 176 | /* For try_to_free_swap() and munlock_vma_page() below */ |
9f92448c ON |
177 | lock_page(page); |
178 | ||
6bdb913f | 179 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
9f92448c | 180 | err = -EAGAIN; |
5323ce71 | 181 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
2b144498 | 182 | if (!ptep) |
9f92448c | 183 | goto unlock; |
2b144498 SD |
184 | |
185 | get_page(kpage); | |
186 | page_add_new_anon_rmap(kpage, vma, addr); | |
00501b53 JW |
187 | mem_cgroup_commit_charge(kpage, memcg, false); |
188 | lru_cache_add_active_or_unevictable(kpage, vma); | |
2b144498 | 189 | |
7396fa81 SD |
190 | if (!PageAnon(page)) { |
191 | dec_mm_counter(mm, MM_FILEPAGES); | |
192 | inc_mm_counter(mm, MM_ANONPAGES); | |
193 | } | |
194 | ||
2b144498 | 195 | flush_cache_page(vma, addr, pte_pfn(*ptep)); |
34ee645e | 196 | ptep_clear_flush_notify(vma, addr, ptep); |
2b144498 SD |
197 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); |
198 | ||
199 | page_remove_rmap(page); | |
200 | if (!page_mapped(page)) | |
201 | try_to_free_swap(page); | |
2b144498 | 202 | pte_unmap_unlock(ptep, ptl); |
2b144498 | 203 | |
194f8dcb ON |
204 | if (vma->vm_flags & VM_LOCKED) |
205 | munlock_vma_page(page); | |
206 | put_page(page); | |
207 | ||
9f92448c ON |
208 | err = 0; |
209 | unlock: | |
00501b53 | 210 | mem_cgroup_cancel_charge(kpage, memcg); |
6bdb913f | 211 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
9f92448c ON |
212 | unlock_page(page); |
213 | return err; | |
2b144498 SD |
214 | } |
215 | ||
216 | /** | |
5cb4ac3a | 217 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 218 | * @insn: instruction to be checked. |
5cb4ac3a | 219 | * Default implementation of is_swbp_insn |
2b144498 SD |
220 | * Returns true if @insn is a breakpoint instruction. |
221 | */ | |
5cb4ac3a | 222 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 223 | { |
5cb4ac3a | 224 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
225 | } |
226 | ||
0908ad6e AM |
227 | /** |
228 | * is_trap_insn - check if instruction is breakpoint instruction. | |
229 | * @insn: instruction to be checked. | |
230 | * Default implementation of is_trap_insn | |
231 | * Returns true if @insn is a breakpoint instruction. | |
232 | * | |
233 | * This function is needed for the case where an architecture has multiple | |
234 | * trap instructions (like powerpc). | |
235 | */ | |
236 | bool __weak is_trap_insn(uprobe_opcode_t *insn) | |
237 | { | |
238 | return is_swbp_insn(insn); | |
239 | } | |
240 | ||
ab0d805c | 241 | static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len) |
cceb55aa ON |
242 | { |
243 | void *kaddr = kmap_atomic(page); | |
ab0d805c | 244 | memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len); |
cceb55aa ON |
245 | kunmap_atomic(kaddr); |
246 | } | |
247 | ||
5669ccee ON |
248 | static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len) |
249 | { | |
250 | void *kaddr = kmap_atomic(page); | |
251 | memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len); | |
252 | kunmap_atomic(kaddr); | |
253 | } | |
254 | ||
ed6f6a50 ON |
255 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
256 | { | |
257 | uprobe_opcode_t old_opcode; | |
258 | bool is_swbp; | |
259 | ||
0908ad6e AM |
260 | /* |
261 | * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here. | |
262 | * We do not check if it is any other 'trap variant' which could | |
263 | * be conditional trap instruction such as the one powerpc supports. | |
264 | * | |
265 | * The logic is that we do not care if the underlying instruction | |
266 | * is a trap variant; uprobes always wins over any other (gdb) | |
267 | * breakpoint. | |
268 | */ | |
ab0d805c | 269 | copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE); |
ed6f6a50 ON |
270 | is_swbp = is_swbp_insn(&old_opcode); |
271 | ||
272 | if (is_swbp_insn(new_opcode)) { | |
273 | if (is_swbp) /* register: already installed? */ | |
274 | return 0; | |
275 | } else { | |
276 | if (!is_swbp) /* unregister: was it changed by us? */ | |
076a365b | 277 | return 0; |
ed6f6a50 ON |
278 | } |
279 | ||
280 | return 1; | |
281 | } | |
282 | ||
2b144498 SD |
283 | /* |
284 | * NOTE: | |
285 | * Expect the breakpoint instruction to be the smallest size instruction for | |
286 | * the architecture. If an arch has variable length instruction and the | |
287 | * breakpoint instruction is not of the smallest length instruction | |
0908ad6e | 288 | * supported by that architecture then we need to modify is_trap_at_addr and |
f72d41fa ON |
289 | * uprobe_write_opcode accordingly. This would never be a problem for archs |
290 | * that have fixed length instructions. | |
29dedee0 | 291 | * |
f72d41fa | 292 | * uprobe_write_opcode - write the opcode at a given virtual address. |
2b144498 | 293 | * @mm: the probed process address space. |
2b144498 SD |
294 | * @vaddr: the virtual address to store the opcode. |
295 | * @opcode: opcode to be written at @vaddr. | |
296 | * | |
29dedee0 | 297 | * Called with mm->mmap_sem held for write. |
2b144498 SD |
298 | * Return 0 (success) or a negative errno. |
299 | */ | |
f72d41fa | 300 | int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr, |
cceb55aa | 301 | uprobe_opcode_t opcode) |
2b144498 SD |
302 | { |
303 | struct page *old_page, *new_page; | |
2b144498 | 304 | struct vm_area_struct *vma; |
2b144498 | 305 | int ret; |
f403072c | 306 | |
5323ce71 | 307 | retry: |
2b144498 | 308 | /* Read the page with vaddr into memory */ |
75ed82ea | 309 | ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); |
2b144498 SD |
310 | if (ret <= 0) |
311 | return ret; | |
7b2d81d4 | 312 | |
ed6f6a50 ON |
313 | ret = verify_opcode(old_page, vaddr, &opcode); |
314 | if (ret <= 0) | |
315 | goto put_old; | |
316 | ||
29dedee0 ON |
317 | ret = anon_vma_prepare(vma); |
318 | if (ret) | |
319 | goto put_old; | |
320 | ||
2b144498 SD |
321 | ret = -ENOMEM; |
322 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
323 | if (!new_page) | |
9f92448c | 324 | goto put_old; |
2b144498 | 325 | |
29dedee0 | 326 | __SetPageUptodate(new_page); |
3f47107c ON |
327 | copy_highpage(new_page, old_page); |
328 | copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); | |
2b144498 | 329 | |
c517ee74 | 330 | ret = __replace_page(vma, vaddr, old_page, new_page); |
2b144498 | 331 | page_cache_release(new_page); |
9f92448c | 332 | put_old: |
7b2d81d4 IM |
333 | put_page(old_page); |
334 | ||
5323ce71 ON |
335 | if (unlikely(ret == -EAGAIN)) |
336 | goto retry; | |
2b144498 SD |
337 | return ret; |
338 | } | |
339 | ||
2b144498 | 340 | /** |
5cb4ac3a | 341 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 342 | * @auprobe: arch specific probepoint information. |
2b144498 | 343 | * @mm: the probed process address space. |
2b144498 SD |
344 | * @vaddr: the virtual address to insert the opcode. |
345 | * | |
346 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
347 | * Return 0 (success) or a negative errno. | |
348 | */ | |
5cb4ac3a | 349 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 350 | { |
f72d41fa | 351 | return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
352 | } |
353 | ||
354 | /** | |
355 | * set_orig_insn - Restore the original instruction. | |
356 | * @mm: the probed process address space. | |
e3343e6a | 357 | * @auprobe: arch specific probepoint information. |
2b144498 | 358 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
359 | * |
360 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
361 | * Return 0 (success) or a negative errno. | |
362 | */ | |
7b2d81d4 | 363 | int __weak |
ded86e7c | 364 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 365 | { |
803200e2 | 366 | return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn); |
2b144498 SD |
367 | } |
368 | ||
f231722a ON |
369 | static struct uprobe *get_uprobe(struct uprobe *uprobe) |
370 | { | |
371 | atomic_inc(&uprobe->ref); | |
372 | return uprobe; | |
373 | } | |
374 | ||
375 | static void put_uprobe(struct uprobe *uprobe) | |
376 | { | |
377 | if (atomic_dec_and_test(&uprobe->ref)) | |
378 | kfree(uprobe); | |
379 | } | |
380 | ||
2b144498 SD |
381 | static int match_uprobe(struct uprobe *l, struct uprobe *r) |
382 | { | |
383 | if (l->inode < r->inode) | |
384 | return -1; | |
7b2d81d4 | 385 | |
2b144498 SD |
386 | if (l->inode > r->inode) |
387 | return 1; | |
2b144498 | 388 | |
7b2d81d4 IM |
389 | if (l->offset < r->offset) |
390 | return -1; | |
391 | ||
392 | if (l->offset > r->offset) | |
393 | return 1; | |
2b144498 SD |
394 | |
395 | return 0; | |
396 | } | |
397 | ||
398 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
399 | { | |
400 | struct uprobe u = { .inode = inode, .offset = offset }; | |
401 | struct rb_node *n = uprobes_tree.rb_node; | |
402 | struct uprobe *uprobe; | |
403 | int match; | |
404 | ||
405 | while (n) { | |
406 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
407 | match = match_uprobe(&u, uprobe); | |
f231722a ON |
408 | if (!match) |
409 | return get_uprobe(uprobe); | |
7b2d81d4 | 410 | |
2b144498 SD |
411 | if (match < 0) |
412 | n = n->rb_left; | |
413 | else | |
414 | n = n->rb_right; | |
415 | } | |
416 | return NULL; | |
417 | } | |
418 | ||
419 | /* | |
420 | * Find a uprobe corresponding to a given inode:offset | |
421 | * Acquires uprobes_treelock | |
422 | */ | |
423 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
424 | { | |
425 | struct uprobe *uprobe; | |
2b144498 | 426 | |
6f47caa0 | 427 | spin_lock(&uprobes_treelock); |
2b144498 | 428 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 429 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 430 | |
2b144498 SD |
431 | return uprobe; |
432 | } | |
433 | ||
434 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
435 | { | |
436 | struct rb_node **p = &uprobes_tree.rb_node; | |
437 | struct rb_node *parent = NULL; | |
438 | struct uprobe *u; | |
439 | int match; | |
440 | ||
441 | while (*p) { | |
442 | parent = *p; | |
443 | u = rb_entry(parent, struct uprobe, rb_node); | |
444 | match = match_uprobe(uprobe, u); | |
f231722a ON |
445 | if (!match) |
446 | return get_uprobe(u); | |
2b144498 SD |
447 | |
448 | if (match < 0) | |
449 | p = &parent->rb_left; | |
450 | else | |
451 | p = &parent->rb_right; | |
452 | ||
453 | } | |
7b2d81d4 | 454 | |
2b144498 SD |
455 | u = NULL; |
456 | rb_link_node(&uprobe->rb_node, parent, p); | |
457 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
458 | /* get access + creation ref */ | |
459 | atomic_set(&uprobe->ref, 2); | |
7b2d81d4 | 460 | |
2b144498 SD |
461 | return u; |
462 | } | |
463 | ||
464 | /* | |
7b2d81d4 | 465 | * Acquire uprobes_treelock. |
2b144498 SD |
466 | * Matching uprobe already exists in rbtree; |
467 | * increment (access refcount) and return the matching uprobe. | |
468 | * | |
469 | * No matching uprobe; insert the uprobe in rb_tree; | |
470 | * get a double refcount (access + creation) and return NULL. | |
471 | */ | |
472 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
473 | { | |
2b144498 SD |
474 | struct uprobe *u; |
475 | ||
6f47caa0 | 476 | spin_lock(&uprobes_treelock); |
2b144498 | 477 | u = __insert_uprobe(uprobe); |
6f47caa0 | 478 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 479 | |
2b144498 SD |
480 | return u; |
481 | } | |
482 | ||
2b144498 SD |
483 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) |
484 | { | |
485 | struct uprobe *uprobe, *cur_uprobe; | |
486 | ||
487 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
488 | if (!uprobe) | |
489 | return NULL; | |
490 | ||
491 | uprobe->inode = igrab(inode); | |
492 | uprobe->offset = offset; | |
e591c8d7 | 493 | init_rwsem(&uprobe->register_rwsem); |
2b144498 | 494 | init_rwsem(&uprobe->consumer_rwsem); |
2b144498 SD |
495 | |
496 | /* add to uprobes_tree, sorted on inode:offset */ | |
497 | cur_uprobe = insert_uprobe(uprobe); | |
2b144498 SD |
498 | /* a uprobe exists for this inode:offset combination */ |
499 | if (cur_uprobe) { | |
500 | kfree(uprobe); | |
501 | uprobe = cur_uprobe; | |
502 | iput(inode); | |
7b2d81d4 IM |
503 | } |
504 | ||
2b144498 SD |
505 | return uprobe; |
506 | } | |
507 | ||
9a98e03c | 508 | static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
509 | { |
510 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
511 | uc->next = uprobe->consumers; |
512 | uprobe->consumers = uc; | |
2b144498 | 513 | up_write(&uprobe->consumer_rwsem); |
2b144498 SD |
514 | } |
515 | ||
516 | /* | |
e3343e6a SD |
517 | * For uprobe @uprobe, delete the consumer @uc. |
518 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
519 | * or return false. |
520 | */ | |
e3343e6a | 521 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
522 | { |
523 | struct uprobe_consumer **con; | |
524 | bool ret = false; | |
525 | ||
526 | down_write(&uprobe->consumer_rwsem); | |
527 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
528 | if (*con == uc) { |
529 | *con = uc->next; | |
2b144498 SD |
530 | ret = true; |
531 | break; | |
532 | } | |
533 | } | |
534 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 535 | |
2b144498 SD |
536 | return ret; |
537 | } | |
538 | ||
2ded0980 ON |
539 | static int __copy_insn(struct address_space *mapping, struct file *filp, |
540 | void *insn, int nbytes, loff_t offset) | |
2b144498 | 541 | { |
2b144498 | 542 | struct page *page; |
2b144498 | 543 | /* |
40814f68 ON |
544 | * Ensure that the page that has the original instruction is populated |
545 | * and in page-cache. If ->readpage == NULL it must be shmem_mapping(), | |
546 | * see uprobe_register(). | |
2b144498 | 547 | */ |
40814f68 ON |
548 | if (mapping->a_ops->readpage) |
549 | page = read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT, filp); | |
550 | else | |
551 | page = shmem_read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT); | |
2b144498 SD |
552 | if (IS_ERR(page)) |
553 | return PTR_ERR(page); | |
554 | ||
2edb7b55 | 555 | copy_from_page(page, offset, insn, nbytes); |
2b144498 | 556 | page_cache_release(page); |
7b2d81d4 | 557 | |
2b144498 SD |
558 | return 0; |
559 | } | |
560 | ||
d436615e | 561 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 | 562 | { |
2ded0980 ON |
563 | struct address_space *mapping = uprobe->inode->i_mapping; |
564 | loff_t offs = uprobe->offset; | |
803200e2 ON |
565 | void *insn = &uprobe->arch.insn; |
566 | int size = sizeof(uprobe->arch.insn); | |
2ded0980 ON |
567 | int len, err = -EIO; |
568 | ||
569 | /* Copy only available bytes, -EIO if nothing was read */ | |
570 | do { | |
571 | if (offs >= i_size_read(uprobe->inode)) | |
572 | break; | |
573 | ||
574 | len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK)); | |
575 | err = __copy_insn(mapping, filp, insn, len, offs); | |
fc36f595 | 576 | if (err) |
2ded0980 ON |
577 | break; |
578 | ||
579 | insn += len; | |
580 | offs += len; | |
581 | size -= len; | |
582 | } while (size); | |
583 | ||
584 | return err; | |
2b144498 SD |
585 | } |
586 | ||
cb9a19fe ON |
587 | static int prepare_uprobe(struct uprobe *uprobe, struct file *file, |
588 | struct mm_struct *mm, unsigned long vaddr) | |
589 | { | |
590 | int ret = 0; | |
591 | ||
71434f2f | 592 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
cb9a19fe ON |
593 | return ret; |
594 | ||
d4d3ccc6 ON |
595 | /* TODO: move this into _register, until then we abuse this sem. */ |
596 | down_write(&uprobe->consumer_rwsem); | |
71434f2f | 597 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
598 | goto out; |
599 | ||
cb9a19fe ON |
600 | ret = copy_insn(uprobe, file); |
601 | if (ret) | |
602 | goto out; | |
603 | ||
604 | ret = -ENOTSUPP; | |
803200e2 | 605 | if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn)) |
cb9a19fe ON |
606 | goto out; |
607 | ||
608 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
609 | if (ret) | |
610 | goto out; | |
611 | ||
f72d41fa | 612 | /* uprobe_write_opcode() assumes we don't cross page boundary */ |
cb9a19fe ON |
613 | BUG_ON((uprobe->offset & ~PAGE_MASK) + |
614 | UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); | |
615 | ||
616 | smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ | |
71434f2f | 617 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
618 | |
619 | out: | |
d4d3ccc6 | 620 | up_write(&uprobe->consumer_rwsem); |
4710f05f | 621 | |
cb9a19fe ON |
622 | return ret; |
623 | } | |
624 | ||
8a7f2fa0 ON |
625 | static inline bool consumer_filter(struct uprobe_consumer *uc, |
626 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
806a98bd | 627 | { |
8a7f2fa0 | 628 | return !uc->filter || uc->filter(uc, ctx, mm); |
806a98bd ON |
629 | } |
630 | ||
8a7f2fa0 ON |
631 | static bool filter_chain(struct uprobe *uprobe, |
632 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
63633cbf | 633 | { |
1ff6fee5 ON |
634 | struct uprobe_consumer *uc; |
635 | bool ret = false; | |
636 | ||
637 | down_read(&uprobe->consumer_rwsem); | |
638 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
8a7f2fa0 | 639 | ret = consumer_filter(uc, ctx, mm); |
1ff6fee5 ON |
640 | if (ret) |
641 | break; | |
642 | } | |
643 | up_read(&uprobe->consumer_rwsem); | |
644 | ||
645 | return ret; | |
63633cbf ON |
646 | } |
647 | ||
e3343e6a SD |
648 | static int |
649 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 650 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 651 | { |
f8ac4ec9 | 652 | bool first_uprobe; |
2b144498 SD |
653 | int ret; |
654 | ||
cb9a19fe ON |
655 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
656 | if (ret) | |
657 | return ret; | |
682968e0 | 658 | |
f8ac4ec9 ON |
659 | /* |
660 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
661 | * the task can hit this breakpoint right after __replace_page(). | |
662 | */ | |
663 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
664 | if (first_uprobe) | |
665 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
666 | ||
816c03fb | 667 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
668 | if (!ret) |
669 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
670 | else if (first_uprobe) | |
f8ac4ec9 | 671 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
672 | |
673 | return ret; | |
674 | } | |
675 | ||
076a365b | 676 | static int |
816c03fb | 677 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 678 | { |
9f68f672 | 679 | set_bit(MMF_RECALC_UPROBES, &mm->flags); |
076a365b | 680 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
681 | } |
682 | ||
06b7bcd8 ON |
683 | static inline bool uprobe_is_active(struct uprobe *uprobe) |
684 | { | |
685 | return !RB_EMPTY_NODE(&uprobe->rb_node); | |
686 | } | |
0326f5a9 | 687 | /* |
778b032d ON |
688 | * There could be threads that have already hit the breakpoint. They |
689 | * will recheck the current insn and restart if find_uprobe() fails. | |
690 | * See find_active_uprobe(). | |
0326f5a9 | 691 | */ |
2b144498 SD |
692 | static void delete_uprobe(struct uprobe *uprobe) |
693 | { | |
06b7bcd8 ON |
694 | if (WARN_ON(!uprobe_is_active(uprobe))) |
695 | return; | |
696 | ||
6f47caa0 | 697 | spin_lock(&uprobes_treelock); |
2b144498 | 698 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 699 | spin_unlock(&uprobes_treelock); |
06b7bcd8 | 700 | RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */ |
2b144498 SD |
701 | iput(uprobe->inode); |
702 | put_uprobe(uprobe); | |
2b144498 SD |
703 | } |
704 | ||
26872090 ON |
705 | struct map_info { |
706 | struct map_info *next; | |
707 | struct mm_struct *mm; | |
816c03fb | 708 | unsigned long vaddr; |
26872090 ON |
709 | }; |
710 | ||
711 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 712 | { |
26872090 ON |
713 | struct map_info *next = info->next; |
714 | kfree(info); | |
715 | return next; | |
716 | } | |
717 | ||
718 | static struct map_info * | |
719 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
720 | { | |
721 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 722 | struct vm_area_struct *vma; |
26872090 ON |
723 | struct map_info *curr = NULL; |
724 | struct map_info *prev = NULL; | |
725 | struct map_info *info; | |
726 | int more = 0; | |
2b144498 | 727 | |
26872090 | 728 | again: |
4a23717a | 729 | i_mmap_lock_read(mapping); |
6b2dbba8 | 730 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
731 | if (!valid_vma(vma, is_register)) |
732 | continue; | |
733 | ||
7a5bfb66 ON |
734 | if (!prev && !more) { |
735 | /* | |
c8c06efa | 736 | * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through |
7a5bfb66 ON |
737 | * reclaim. This is optimistic, no harm done if it fails. |
738 | */ | |
739 | prev = kmalloc(sizeof(struct map_info), | |
740 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
741 | if (prev) | |
742 | prev->next = NULL; | |
743 | } | |
26872090 ON |
744 | if (!prev) { |
745 | more++; | |
746 | continue; | |
2b144498 | 747 | } |
2b144498 | 748 | |
26872090 ON |
749 | if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) |
750 | continue; | |
7b2d81d4 | 751 | |
26872090 ON |
752 | info = prev; |
753 | prev = prev->next; | |
754 | info->next = curr; | |
755 | curr = info; | |
2b144498 | 756 | |
26872090 | 757 | info->mm = vma->vm_mm; |
57683f72 | 758 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 759 | } |
4a23717a | 760 | i_mmap_unlock_read(mapping); |
2b144498 | 761 | |
26872090 ON |
762 | if (!more) |
763 | goto out; | |
764 | ||
765 | prev = curr; | |
766 | while (curr) { | |
767 | mmput(curr->mm); | |
768 | curr = curr->next; | |
769 | } | |
7b2d81d4 | 770 | |
26872090 ON |
771 | do { |
772 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
773 | if (!info) { | |
774 | curr = ERR_PTR(-ENOMEM); | |
775 | goto out; | |
776 | } | |
777 | info->next = prev; | |
778 | prev = info; | |
779 | } while (--more); | |
780 | ||
781 | goto again; | |
782 | out: | |
783 | while (prev) | |
784 | prev = free_map_info(prev); | |
785 | return curr; | |
2b144498 SD |
786 | } |
787 | ||
bdf8647c ON |
788 | static int |
789 | register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) | |
2b144498 | 790 | { |
bdf8647c | 791 | bool is_register = !!new; |
26872090 ON |
792 | struct map_info *info; |
793 | int err = 0; | |
2b144498 | 794 | |
32cdba1e | 795 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
796 | info = build_map_info(uprobe->inode->i_mapping, |
797 | uprobe->offset, is_register); | |
32cdba1e ON |
798 | if (IS_ERR(info)) { |
799 | err = PTR_ERR(info); | |
800 | goto out; | |
801 | } | |
7b2d81d4 | 802 | |
26872090 ON |
803 | while (info) { |
804 | struct mm_struct *mm = info->mm; | |
805 | struct vm_area_struct *vma; | |
7b2d81d4 | 806 | |
076a365b | 807 | if (err && is_register) |
26872090 | 808 | goto free; |
7b2d81d4 | 809 | |
77fc4af1 | 810 | down_write(&mm->mmap_sem); |
f4d6dfe5 ON |
811 | vma = find_vma(mm, info->vaddr); |
812 | if (!vma || !valid_vma(vma, is_register) || | |
f281769e | 813 | file_inode(vma->vm_file) != uprobe->inode) |
26872090 ON |
814 | goto unlock; |
815 | ||
f4d6dfe5 ON |
816 | if (vma->vm_start > info->vaddr || |
817 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 818 | goto unlock; |
2b144498 | 819 | |
806a98bd ON |
820 | if (is_register) { |
821 | /* consult only the "caller", new consumer. */ | |
bdf8647c | 822 | if (consumer_filter(new, |
8a7f2fa0 | 823 | UPROBE_FILTER_REGISTER, mm)) |
806a98bd ON |
824 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
825 | } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) { | |
8a7f2fa0 ON |
826 | if (!filter_chain(uprobe, |
827 | UPROBE_FILTER_UNREGISTER, mm)) | |
806a98bd ON |
828 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
829 | } | |
78f74116 | 830 | |
26872090 ON |
831 | unlock: |
832 | up_write(&mm->mmap_sem); | |
833 | free: | |
834 | mmput(mm); | |
835 | info = free_map_info(info); | |
2b144498 | 836 | } |
32cdba1e ON |
837 | out: |
838 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 839 | return err; |
2b144498 SD |
840 | } |
841 | ||
9a98e03c | 842 | static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 843 | { |
9a98e03c | 844 | consumer_add(uprobe, uc); |
bdf8647c | 845 | return register_for_each_vma(uprobe, uc); |
2b144498 SD |
846 | } |
847 | ||
04aab9b2 | 848 | static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 | 849 | { |
04aab9b2 ON |
850 | int err; |
851 | ||
06d07139 | 852 | if (WARN_ON(!consumer_del(uprobe, uc))) |
04aab9b2 | 853 | return; |
2b144498 | 854 | |
bdf8647c | 855 | err = register_for_each_vma(uprobe, NULL); |
bb929284 ON |
856 | /* TODO : cant unregister? schedule a worker thread */ |
857 | if (!uprobe->consumers && !err) | |
858 | delete_uprobe(uprobe); | |
2b144498 SD |
859 | } |
860 | ||
861 | /* | |
7b2d81d4 | 862 | * uprobe_register - register a probe |
2b144498 SD |
863 | * @inode: the file in which the probe has to be placed. |
864 | * @offset: offset from the start of the file. | |
e3343e6a | 865 | * @uc: information on howto handle the probe.. |
2b144498 | 866 | * |
7b2d81d4 | 867 | * Apart from the access refcount, uprobe_register() takes a creation |
2b144498 SD |
868 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
869 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 870 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 871 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 872 | * refcount is released when the last @uc for the @uprobe |
2b144498 SD |
873 | * unregisters. |
874 | * | |
875 | * Return errno if it cannot successully install probes | |
876 | * else return 0 (success) | |
877 | */ | |
e3343e6a | 878 | int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 SD |
879 | { |
880 | struct uprobe *uprobe; | |
7b2d81d4 | 881 | int ret; |
2b144498 | 882 | |
ea024870 AA |
883 | /* Uprobe must have at least one set consumer */ |
884 | if (!uc->handler && !uc->ret_handler) | |
885 | return -EINVAL; | |
886 | ||
40814f68 ON |
887 | /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */ |
888 | if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping)) | |
41ccba02 | 889 | return -EIO; |
f0744af7 | 890 | /* Racy, just to catch the obvious mistakes */ |
2b144498 | 891 | if (offset > i_size_read(inode)) |
7b2d81d4 | 892 | return -EINVAL; |
2b144498 | 893 | |
66d06dff | 894 | retry: |
2b144498 | 895 | uprobe = alloc_uprobe(inode, offset); |
66d06dff ON |
896 | if (!uprobe) |
897 | return -ENOMEM; | |
898 | /* | |
899 | * We can race with uprobe_unregister()->delete_uprobe(). | |
900 | * Check uprobe_is_active() and retry if it is false. | |
901 | */ | |
902 | down_write(&uprobe->register_rwsem); | |
903 | ret = -EAGAIN; | |
904 | if (likely(uprobe_is_active(uprobe))) { | |
9a98e03c ON |
905 | ret = __uprobe_register(uprobe, uc); |
906 | if (ret) | |
04aab9b2 | 907 | __uprobe_unregister(uprobe, uc); |
2b144498 | 908 | } |
66d06dff ON |
909 | up_write(&uprobe->register_rwsem); |
910 | put_uprobe(uprobe); | |
2b144498 | 911 | |
66d06dff ON |
912 | if (unlikely(ret == -EAGAIN)) |
913 | goto retry; | |
2b144498 SD |
914 | return ret; |
915 | } | |
e8440c14 | 916 | EXPORT_SYMBOL_GPL(uprobe_register); |
2b144498 | 917 | |
bdf8647c ON |
918 | /* |
919 | * uprobe_apply - unregister a already registered probe. | |
920 | * @inode: the file in which the probe has to be removed. | |
921 | * @offset: offset from the start of the file. | |
922 | * @uc: consumer which wants to add more or remove some breakpoints | |
923 | * @add: add or remove the breakpoints | |
924 | */ | |
925 | int uprobe_apply(struct inode *inode, loff_t offset, | |
926 | struct uprobe_consumer *uc, bool add) | |
927 | { | |
928 | struct uprobe *uprobe; | |
929 | struct uprobe_consumer *con; | |
930 | int ret = -ENOENT; | |
931 | ||
932 | uprobe = find_uprobe(inode, offset); | |
06d07139 | 933 | if (WARN_ON(!uprobe)) |
bdf8647c ON |
934 | return ret; |
935 | ||
936 | down_write(&uprobe->register_rwsem); | |
937 | for (con = uprobe->consumers; con && con != uc ; con = con->next) | |
938 | ; | |
939 | if (con) | |
940 | ret = register_for_each_vma(uprobe, add ? uc : NULL); | |
941 | up_write(&uprobe->register_rwsem); | |
942 | put_uprobe(uprobe); | |
943 | ||
944 | return ret; | |
945 | } | |
946 | ||
2b144498 | 947 | /* |
7b2d81d4 | 948 | * uprobe_unregister - unregister a already registered probe. |
2b144498 SD |
949 | * @inode: the file in which the probe has to be removed. |
950 | * @offset: offset from the start of the file. | |
e3343e6a | 951 | * @uc: identify which probe if multiple probes are colocated. |
2b144498 | 952 | */ |
e3343e6a | 953 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) |
2b144498 | 954 | { |
7b2d81d4 | 955 | struct uprobe *uprobe; |
2b144498 | 956 | |
2b144498 | 957 | uprobe = find_uprobe(inode, offset); |
06d07139 | 958 | if (WARN_ON(!uprobe)) |
2b144498 SD |
959 | return; |
960 | ||
e591c8d7 | 961 | down_write(&uprobe->register_rwsem); |
04aab9b2 | 962 | __uprobe_unregister(uprobe, uc); |
e591c8d7 | 963 | up_write(&uprobe->register_rwsem); |
c91368c4 | 964 | put_uprobe(uprobe); |
2b144498 | 965 | } |
e8440c14 | 966 | EXPORT_SYMBOL_GPL(uprobe_unregister); |
2b144498 | 967 | |
da1816b1 ON |
968 | static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm) |
969 | { | |
970 | struct vm_area_struct *vma; | |
971 | int err = 0; | |
972 | ||
973 | down_read(&mm->mmap_sem); | |
974 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
975 | unsigned long vaddr; | |
976 | loff_t offset; | |
977 | ||
978 | if (!valid_vma(vma, false) || | |
f281769e | 979 | file_inode(vma->vm_file) != uprobe->inode) |
da1816b1 ON |
980 | continue; |
981 | ||
982 | offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT; | |
983 | if (uprobe->offset < offset || | |
984 | uprobe->offset >= offset + vma->vm_end - vma->vm_start) | |
985 | continue; | |
986 | ||
987 | vaddr = offset_to_vaddr(vma, uprobe->offset); | |
988 | err |= remove_breakpoint(uprobe, mm, vaddr); | |
989 | } | |
990 | up_read(&mm->mmap_sem); | |
991 | ||
992 | return err; | |
993 | } | |
994 | ||
891c3970 ON |
995 | static struct rb_node * |
996 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 997 | { |
2b144498 | 998 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
999 | |
1000 | while (n) { | |
891c3970 | 1001 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 1002 | |
891c3970 | 1003 | if (inode < u->inode) { |
2b144498 | 1004 | n = n->rb_left; |
891c3970 | 1005 | } else if (inode > u->inode) { |
2b144498 | 1006 | n = n->rb_right; |
891c3970 ON |
1007 | } else { |
1008 | if (max < u->offset) | |
1009 | n = n->rb_left; | |
1010 | else if (min > u->offset) | |
1011 | n = n->rb_right; | |
1012 | else | |
1013 | break; | |
1014 | } | |
2b144498 | 1015 | } |
7b2d81d4 | 1016 | |
891c3970 | 1017 | return n; |
2b144498 SD |
1018 | } |
1019 | ||
1020 | /* | |
891c3970 | 1021 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 1022 | */ |
891c3970 ON |
1023 | static void build_probe_list(struct inode *inode, |
1024 | struct vm_area_struct *vma, | |
1025 | unsigned long start, unsigned long end, | |
1026 | struct list_head *head) | |
2b144498 | 1027 | { |
891c3970 | 1028 | loff_t min, max; |
891c3970 ON |
1029 | struct rb_node *n, *t; |
1030 | struct uprobe *u; | |
7b2d81d4 | 1031 | |
891c3970 | 1032 | INIT_LIST_HEAD(head); |
cb113b47 | 1033 | min = vaddr_to_offset(vma, start); |
891c3970 | 1034 | max = min + (end - start) - 1; |
2b144498 | 1035 | |
6f47caa0 | 1036 | spin_lock(&uprobes_treelock); |
891c3970 ON |
1037 | n = find_node_in_range(inode, min, max); |
1038 | if (n) { | |
1039 | for (t = n; t; t = rb_prev(t)) { | |
1040 | u = rb_entry(t, struct uprobe, rb_node); | |
1041 | if (u->inode != inode || u->offset < min) | |
1042 | break; | |
1043 | list_add(&u->pending_list, head); | |
f231722a | 1044 | get_uprobe(u); |
891c3970 ON |
1045 | } |
1046 | for (t = n; (t = rb_next(t)); ) { | |
1047 | u = rb_entry(t, struct uprobe, rb_node); | |
1048 | if (u->inode != inode || u->offset > max) | |
1049 | break; | |
1050 | list_add(&u->pending_list, head); | |
f231722a | 1051 | get_uprobe(u); |
891c3970 | 1052 | } |
2b144498 | 1053 | } |
6f47caa0 | 1054 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
1055 | } |
1056 | ||
1057 | /* | |
5e5be71a | 1058 | * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. |
2b144498 | 1059 | * |
5e5be71a ON |
1060 | * Currently we ignore all errors and always return 0, the callers |
1061 | * can't handle the failure anyway. | |
2b144498 | 1062 | */ |
7b2d81d4 | 1063 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1064 | { |
1065 | struct list_head tmp_list; | |
665605a2 | 1066 | struct uprobe *uprobe, *u; |
2b144498 | 1067 | struct inode *inode; |
2b144498 | 1068 | |
441f1eb7 | 1069 | if (no_uprobe_events() || !valid_vma(vma, true)) |
7b2d81d4 | 1070 | return 0; |
2b144498 | 1071 | |
f281769e | 1072 | inode = file_inode(vma->vm_file); |
2b144498 | 1073 | if (!inode) |
7b2d81d4 | 1074 | return 0; |
2b144498 | 1075 | |
2b144498 | 1076 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1077 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
806a98bd ON |
1078 | /* |
1079 | * We can race with uprobe_unregister(), this uprobe can be already | |
1080 | * removed. But in this case filter_chain() must return false, all | |
1081 | * consumers have gone away. | |
1082 | */ | |
665605a2 | 1083 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
806a98bd | 1084 | if (!fatal_signal_pending(current) && |
8a7f2fa0 | 1085 | filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) { |
57683f72 | 1086 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 1087 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
1088 | } |
1089 | put_uprobe(uprobe); | |
1090 | } | |
2b144498 SD |
1091 | mutex_unlock(uprobes_mmap_hash(inode)); |
1092 | ||
5e5be71a | 1093 | return 0; |
2b144498 SD |
1094 | } |
1095 | ||
9f68f672 ON |
1096 | static bool |
1097 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1098 | { | |
1099 | loff_t min, max; | |
1100 | struct inode *inode; | |
1101 | struct rb_node *n; | |
1102 | ||
f281769e | 1103 | inode = file_inode(vma->vm_file); |
9f68f672 ON |
1104 | |
1105 | min = vaddr_to_offset(vma, start); | |
1106 | max = min + (end - start) - 1; | |
1107 | ||
1108 | spin_lock(&uprobes_treelock); | |
1109 | n = find_node_in_range(inode, min, max); | |
1110 | spin_unlock(&uprobes_treelock); | |
1111 | ||
1112 | return !!n; | |
1113 | } | |
1114 | ||
682968e0 SD |
1115 | /* |
1116 | * Called in context of a munmap of a vma. | |
1117 | */ | |
cbc91f71 | 1118 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1119 | { |
441f1eb7 | 1120 | if (no_uprobe_events() || !valid_vma(vma, false)) |
682968e0 SD |
1121 | return; |
1122 | ||
2fd611a9 ON |
1123 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1124 | return; | |
1125 | ||
9f68f672 ON |
1126 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1127 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1128 | return; |
1129 | ||
9f68f672 ON |
1130 | if (vma_has_uprobes(vma, start, end)) |
1131 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1132 | } |
1133 | ||
d4b3b638 | 1134 | /* Slot allocation for XOL */ |
6441ec8b | 1135 | static int xol_add_vma(struct mm_struct *mm, struct xol_area *area) |
d4b3b638 | 1136 | { |
c8a82538 | 1137 | int ret = -EALREADY; |
d4b3b638 SD |
1138 | |
1139 | down_write(&mm->mmap_sem); | |
1140 | if (mm->uprobes_state.xol_area) | |
1141 | goto fail; | |
1142 | ||
af0d95af ON |
1143 | if (!area->vaddr) { |
1144 | /* Try to map as high as possible, this is only a hint. */ | |
1145 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, | |
1146 | PAGE_SIZE, 0, 0); | |
1147 | if (area->vaddr & ~PAGE_MASK) { | |
1148 | ret = area->vaddr; | |
1149 | goto fail; | |
1150 | } | |
d4b3b638 SD |
1151 | } |
1152 | ||
1153 | ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, | |
1154 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); | |
1155 | if (ret) | |
1156 | goto fail; | |
1157 | ||
1158 | smp_wmb(); /* pairs with get_xol_area() */ | |
1159 | mm->uprobes_state.xol_area = area; | |
c8a82538 | 1160 | fail: |
d4b3b638 | 1161 | up_write(&mm->mmap_sem); |
d4b3b638 SD |
1162 | |
1163 | return ret; | |
1164 | } | |
1165 | ||
af0d95af | 1166 | static struct xol_area *__create_xol_area(unsigned long vaddr) |
d4b3b638 | 1167 | { |
9b545df8 | 1168 | struct mm_struct *mm = current->mm; |
e78aebfd | 1169 | uprobe_opcode_t insn = UPROBE_SWBP_INSN; |
6441ec8b | 1170 | struct xol_area *area; |
9b545df8 | 1171 | |
af0d95af | 1172 | area = kmalloc(sizeof(*area), GFP_KERNEL); |
d4b3b638 | 1173 | if (unlikely(!area)) |
c8a82538 | 1174 | goto out; |
d4b3b638 SD |
1175 | |
1176 | area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); | |
d4b3b638 | 1177 | if (!area->bitmap) |
c8a82538 ON |
1178 | goto free_area; |
1179 | ||
1180 | area->page = alloc_page(GFP_HIGHUSER); | |
1181 | if (!area->page) | |
1182 | goto free_bitmap; | |
d4b3b638 | 1183 | |
af0d95af | 1184 | area->vaddr = vaddr; |
6441ec8b ON |
1185 | init_waitqueue_head(&area->wq); |
1186 | /* Reserve the 1st slot for get_trampoline_vaddr() */ | |
e78aebfd | 1187 | set_bit(0, area->bitmap); |
e78aebfd | 1188 | atomic_set(&area->slot_count, 1); |
6441ec8b | 1189 | copy_to_page(area->page, 0, &insn, UPROBE_SWBP_INSN_SIZE); |
e78aebfd | 1190 | |
6441ec8b | 1191 | if (!xol_add_vma(mm, area)) |
d4b3b638 SD |
1192 | return area; |
1193 | ||
c8a82538 ON |
1194 | __free_page(area->page); |
1195 | free_bitmap: | |
d4b3b638 | 1196 | kfree(area->bitmap); |
c8a82538 | 1197 | free_area: |
d4b3b638 | 1198 | kfree(area); |
c8a82538 | 1199 | out: |
6441ec8b ON |
1200 | return NULL; |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * get_xol_area - Allocate process's xol_area if necessary. | |
1205 | * This area will be used for storing instructions for execution out of line. | |
1206 | * | |
1207 | * Returns the allocated area or NULL. | |
1208 | */ | |
1209 | static struct xol_area *get_xol_area(void) | |
1210 | { | |
1211 | struct mm_struct *mm = current->mm; | |
1212 | struct xol_area *area; | |
1213 | ||
1214 | if (!mm->uprobes_state.xol_area) | |
af0d95af | 1215 | __create_xol_area(0); |
6441ec8b | 1216 | |
9b545df8 | 1217 | area = mm->uprobes_state.xol_area; |
6441ec8b | 1218 | smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ |
9b545df8 | 1219 | return area; |
d4b3b638 SD |
1220 | } |
1221 | ||
1222 | /* | |
1223 | * uprobe_clear_state - Free the area allocated for slots. | |
1224 | */ | |
1225 | void uprobe_clear_state(struct mm_struct *mm) | |
1226 | { | |
1227 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1228 | ||
1229 | if (!area) | |
1230 | return; | |
1231 | ||
1232 | put_page(area->page); | |
1233 | kfree(area->bitmap); | |
1234 | kfree(area); | |
1235 | } | |
1236 | ||
32cdba1e ON |
1237 | void uprobe_start_dup_mmap(void) |
1238 | { | |
1239 | percpu_down_read(&dup_mmap_sem); | |
1240 | } | |
1241 | ||
1242 | void uprobe_end_dup_mmap(void) | |
1243 | { | |
1244 | percpu_up_read(&dup_mmap_sem); | |
1245 | } | |
1246 | ||
f8ac4ec9 ON |
1247 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1248 | { | |
61559a81 ON |
1249 | newmm->uprobes_state.xol_area = NULL; |
1250 | ||
9f68f672 | 1251 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1252 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1253 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1254 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1255 | } | |
f8ac4ec9 ON |
1256 | } |
1257 | ||
d4b3b638 SD |
1258 | /* |
1259 | * - search for a free slot. | |
1260 | */ | |
1261 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1262 | { | |
1263 | unsigned long slot_addr; | |
1264 | int slot_nr; | |
1265 | ||
1266 | do { | |
1267 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1268 | if (slot_nr < UINSNS_PER_PAGE) { | |
1269 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1270 | break; | |
1271 | ||
1272 | slot_nr = UINSNS_PER_PAGE; | |
1273 | continue; | |
1274 | } | |
1275 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1276 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1277 | ||
1278 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1279 | atomic_inc(&area->slot_count); | |
1280 | ||
1281 | return slot_addr; | |
1282 | } | |
1283 | ||
1284 | /* | |
a6cb3f6d | 1285 | * xol_get_insn_slot - allocate a slot for xol. |
d4b3b638 SD |
1286 | * Returns the allocated slot address or 0. |
1287 | */ | |
a6cb3f6d | 1288 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe) |
d4b3b638 SD |
1289 | { |
1290 | struct xol_area *area; | |
a6cb3f6d | 1291 | unsigned long xol_vaddr; |
d4b3b638 | 1292 | |
9b545df8 ON |
1293 | area = get_xol_area(); |
1294 | if (!area) | |
1295 | return 0; | |
d4b3b638 | 1296 | |
a6cb3f6d ON |
1297 | xol_vaddr = xol_take_insn_slot(area); |
1298 | if (unlikely(!xol_vaddr)) | |
d4b3b638 SD |
1299 | return 0; |
1300 | ||
72e6ae28 VK |
1301 | arch_uprobe_copy_ixol(area->page, xol_vaddr, |
1302 | &uprobe->arch.ixol, sizeof(uprobe->arch.ixol)); | |
d4b3b638 | 1303 | |
a6cb3f6d | 1304 | return xol_vaddr; |
d4b3b638 SD |
1305 | } |
1306 | ||
1307 | /* | |
1308 | * xol_free_insn_slot - If slot was earlier allocated by | |
1309 | * @xol_get_insn_slot(), make the slot available for | |
1310 | * subsequent requests. | |
1311 | */ | |
1312 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1313 | { | |
1314 | struct xol_area *area; | |
1315 | unsigned long vma_end; | |
1316 | unsigned long slot_addr; | |
1317 | ||
1318 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1319 | return; | |
1320 | ||
1321 | slot_addr = tsk->utask->xol_vaddr; | |
af4355e9 | 1322 | if (unlikely(!slot_addr)) |
d4b3b638 SD |
1323 | return; |
1324 | ||
1325 | area = tsk->mm->uprobes_state.xol_area; | |
1326 | vma_end = area->vaddr + PAGE_SIZE; | |
1327 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1328 | unsigned long offset; | |
1329 | int slot_nr; | |
1330 | ||
1331 | offset = slot_addr - area->vaddr; | |
1332 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1333 | if (slot_nr >= UINSNS_PER_PAGE) | |
1334 | return; | |
1335 | ||
1336 | clear_bit(slot_nr, area->bitmap); | |
1337 | atomic_dec(&area->slot_count); | |
1338 | if (waitqueue_active(&area->wq)) | |
1339 | wake_up(&area->wq); | |
1340 | ||
1341 | tsk->utask->xol_vaddr = 0; | |
1342 | } | |
1343 | } | |
1344 | ||
72e6ae28 VK |
1345 | void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr, |
1346 | void *src, unsigned long len) | |
1347 | { | |
1348 | /* Initialize the slot */ | |
1349 | copy_to_page(page, vaddr, src, len); | |
1350 | ||
1351 | /* | |
1352 | * We probably need flush_icache_user_range() but it needs vma. | |
1353 | * This should work on most of architectures by default. If | |
1354 | * architecture needs to do something different it can define | |
1355 | * its own version of the function. | |
1356 | */ | |
1357 | flush_dcache_page(page); | |
1358 | } | |
1359 | ||
0326f5a9 SD |
1360 | /** |
1361 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1362 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1363 | * instruction. | |
1364 | * Return the address of the breakpoint instruction. | |
1365 | */ | |
1366 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1367 | { | |
1368 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1369 | } | |
1370 | ||
b02ef20a ON |
1371 | unsigned long uprobe_get_trap_addr(struct pt_regs *regs) |
1372 | { | |
1373 | struct uprobe_task *utask = current->utask; | |
1374 | ||
1375 | if (unlikely(utask && utask->active_uprobe)) | |
1376 | return utask->vaddr; | |
1377 | ||
1378 | return instruction_pointer(regs); | |
1379 | } | |
1380 | ||
2bb5e840 ON |
1381 | static struct return_instance *free_ret_instance(struct return_instance *ri) |
1382 | { | |
1383 | struct return_instance *next = ri->next; | |
1384 | put_uprobe(ri->uprobe); | |
1385 | kfree(ri); | |
1386 | return next; | |
1387 | } | |
1388 | ||
0326f5a9 SD |
1389 | /* |
1390 | * Called with no locks held. | |
1391 | * Called in context of a exiting or a exec-ing thread. | |
1392 | */ | |
1393 | void uprobe_free_utask(struct task_struct *t) | |
1394 | { | |
1395 | struct uprobe_task *utask = t->utask; | |
2bb5e840 | 1396 | struct return_instance *ri; |
0326f5a9 | 1397 | |
0326f5a9 SD |
1398 | if (!utask) |
1399 | return; | |
1400 | ||
1401 | if (utask->active_uprobe) | |
1402 | put_uprobe(utask->active_uprobe); | |
1403 | ||
0dfd0eb8 | 1404 | ri = utask->return_instances; |
2bb5e840 ON |
1405 | while (ri) |
1406 | ri = free_ret_instance(ri); | |
0dfd0eb8 | 1407 | |
d4b3b638 | 1408 | xol_free_insn_slot(t); |
0326f5a9 SD |
1409 | kfree(utask); |
1410 | t->utask = NULL; | |
1411 | } | |
1412 | ||
0326f5a9 | 1413 | /* |
5a2df662 ON |
1414 | * Allocate a uprobe_task object for the task if if necessary. |
1415 | * Called when the thread hits a breakpoint. | |
0326f5a9 SD |
1416 | * |
1417 | * Returns: | |
1418 | * - pointer to new uprobe_task on success | |
1419 | * - NULL otherwise | |
1420 | */ | |
5a2df662 | 1421 | static struct uprobe_task *get_utask(void) |
0326f5a9 | 1422 | { |
5a2df662 ON |
1423 | if (!current->utask) |
1424 | current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1425 | return current->utask; | |
0326f5a9 SD |
1426 | } |
1427 | ||
248d3a7b ON |
1428 | static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask) |
1429 | { | |
1430 | struct uprobe_task *n_utask; | |
1431 | struct return_instance **p, *o, *n; | |
1432 | ||
1433 | n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1434 | if (!n_utask) | |
1435 | return -ENOMEM; | |
1436 | t->utask = n_utask; | |
1437 | ||
1438 | p = &n_utask->return_instances; | |
1439 | for (o = o_utask->return_instances; o; o = o->next) { | |
1440 | n = kmalloc(sizeof(struct return_instance), GFP_KERNEL); | |
1441 | if (!n) | |
1442 | return -ENOMEM; | |
1443 | ||
1444 | *n = *o; | |
f231722a | 1445 | get_uprobe(n->uprobe); |
248d3a7b ON |
1446 | n->next = NULL; |
1447 | ||
1448 | *p = n; | |
1449 | p = &n->next; | |
1450 | n_utask->depth++; | |
1451 | } | |
1452 | ||
1453 | return 0; | |
1454 | } | |
1455 | ||
1456 | static void uprobe_warn(struct task_struct *t, const char *msg) | |
1457 | { | |
1458 | pr_warn("uprobe: %s:%d failed to %s\n", | |
1459 | current->comm, current->pid, msg); | |
1460 | } | |
1461 | ||
aa59c53f ON |
1462 | static void dup_xol_work(struct callback_head *work) |
1463 | { | |
aa59c53f ON |
1464 | if (current->flags & PF_EXITING) |
1465 | return; | |
1466 | ||
32473431 | 1467 | if (!__create_xol_area(current->utask->dup_xol_addr)) |
aa59c53f ON |
1468 | uprobe_warn(current, "dup xol area"); |
1469 | } | |
1470 | ||
b68e0749 ON |
1471 | /* |
1472 | * Called in context of a new clone/fork from copy_process. | |
1473 | */ | |
3ab67966 | 1474 | void uprobe_copy_process(struct task_struct *t, unsigned long flags) |
b68e0749 | 1475 | { |
248d3a7b ON |
1476 | struct uprobe_task *utask = current->utask; |
1477 | struct mm_struct *mm = current->mm; | |
aa59c53f | 1478 | struct xol_area *area; |
248d3a7b | 1479 | |
b68e0749 | 1480 | t->utask = NULL; |
248d3a7b | 1481 | |
3ab67966 ON |
1482 | if (!utask || !utask->return_instances) |
1483 | return; | |
1484 | ||
1485 | if (mm == t->mm && !(flags & CLONE_VFORK)) | |
248d3a7b ON |
1486 | return; |
1487 | ||
1488 | if (dup_utask(t, utask)) | |
1489 | return uprobe_warn(t, "dup ret instances"); | |
aa59c53f ON |
1490 | |
1491 | /* The task can fork() after dup_xol_work() fails */ | |
1492 | area = mm->uprobes_state.xol_area; | |
1493 | if (!area) | |
1494 | return uprobe_warn(t, "dup xol area"); | |
1495 | ||
3ab67966 ON |
1496 | if (mm == t->mm) |
1497 | return; | |
1498 | ||
32473431 ON |
1499 | t->utask->dup_xol_addr = area->vaddr; |
1500 | init_task_work(&t->utask->dup_xol_work, dup_xol_work); | |
1501 | task_work_add(t, &t->utask->dup_xol_work, true); | |
b68e0749 ON |
1502 | } |
1503 | ||
e78aebfd AA |
1504 | /* |
1505 | * Current area->vaddr notion assume the trampoline address is always | |
1506 | * equal area->vaddr. | |
1507 | * | |
1508 | * Returns -1 in case the xol_area is not allocated. | |
1509 | */ | |
1510 | static unsigned long get_trampoline_vaddr(void) | |
1511 | { | |
1512 | struct xol_area *area; | |
1513 | unsigned long trampoline_vaddr = -1; | |
1514 | ||
1515 | area = current->mm->uprobes_state.xol_area; | |
1516 | smp_read_barrier_depends(); | |
1517 | if (area) | |
1518 | trampoline_vaddr = area->vaddr; | |
1519 | ||
1520 | return trampoline_vaddr; | |
1521 | } | |
1522 | ||
0dfd0eb8 AA |
1523 | static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs) |
1524 | { | |
1525 | struct return_instance *ri; | |
1526 | struct uprobe_task *utask; | |
1527 | unsigned long orig_ret_vaddr, trampoline_vaddr; | |
1528 | bool chained = false; | |
1529 | ||
1530 | if (!get_xol_area()) | |
1531 | return; | |
1532 | ||
1533 | utask = get_utask(); | |
1534 | if (!utask) | |
1535 | return; | |
1536 | ||
ded49c55 AA |
1537 | if (utask->depth >= MAX_URETPROBE_DEPTH) { |
1538 | printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to" | |
1539 | " nestedness limit pid/tgid=%d/%d\n", | |
1540 | current->pid, current->tgid); | |
1541 | return; | |
1542 | } | |
1543 | ||
0dfd0eb8 AA |
1544 | ri = kzalloc(sizeof(struct return_instance), GFP_KERNEL); |
1545 | if (!ri) | |
1546 | goto fail; | |
1547 | ||
1548 | trampoline_vaddr = get_trampoline_vaddr(); | |
1549 | orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs); | |
1550 | if (orig_ret_vaddr == -1) | |
1551 | goto fail; | |
1552 | ||
1553 | /* | |
1554 | * We don't want to keep trampoline address in stack, rather keep the | |
1555 | * original return address of first caller thru all the consequent | |
1556 | * instances. This also makes breakpoint unwrapping easier. | |
1557 | */ | |
1558 | if (orig_ret_vaddr == trampoline_vaddr) { | |
1559 | if (!utask->return_instances) { | |
1560 | /* | |
1561 | * This situation is not possible. Likely we have an | |
1562 | * attack from user-space. | |
1563 | */ | |
1564 | pr_warn("uprobe: unable to set uretprobe pid/tgid=%d/%d\n", | |
1565 | current->pid, current->tgid); | |
1566 | goto fail; | |
1567 | } | |
1568 | ||
1569 | chained = true; | |
1570 | orig_ret_vaddr = utask->return_instances->orig_ret_vaddr; | |
1571 | } | |
1572 | ||
f231722a | 1573 | ri->uprobe = get_uprobe(uprobe); |
0dfd0eb8 AA |
1574 | ri->func = instruction_pointer(regs); |
1575 | ri->orig_ret_vaddr = orig_ret_vaddr; | |
1576 | ri->chained = chained; | |
1577 | ||
ded49c55 AA |
1578 | utask->depth++; |
1579 | ||
0dfd0eb8 AA |
1580 | /* add instance to the stack */ |
1581 | ri->next = utask->return_instances; | |
1582 | utask->return_instances = ri; | |
1583 | ||
1584 | return; | |
1585 | ||
1586 | fail: | |
1587 | kfree(ri); | |
1588 | } | |
1589 | ||
0326f5a9 SD |
1590 | /* Prepare to single-step probed instruction out of line. */ |
1591 | static int | |
a6cb3f6d | 1592 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr) |
0326f5a9 | 1593 | { |
a6cb3f6d ON |
1594 | struct uprobe_task *utask; |
1595 | unsigned long xol_vaddr; | |
aba51024 | 1596 | int err; |
a6cb3f6d | 1597 | |
608e7427 ON |
1598 | utask = get_utask(); |
1599 | if (!utask) | |
1600 | return -ENOMEM; | |
a6cb3f6d ON |
1601 | |
1602 | xol_vaddr = xol_get_insn_slot(uprobe); | |
1603 | if (!xol_vaddr) | |
1604 | return -ENOMEM; | |
1605 | ||
1606 | utask->xol_vaddr = xol_vaddr; | |
1607 | utask->vaddr = bp_vaddr; | |
d4b3b638 | 1608 | |
aba51024 ON |
1609 | err = arch_uprobe_pre_xol(&uprobe->arch, regs); |
1610 | if (unlikely(err)) { | |
1611 | xol_free_insn_slot(current); | |
1612 | return err; | |
1613 | } | |
1614 | ||
608e7427 ON |
1615 | utask->active_uprobe = uprobe; |
1616 | utask->state = UTASK_SSTEP; | |
aba51024 | 1617 | return 0; |
0326f5a9 SD |
1618 | } |
1619 | ||
1620 | /* | |
1621 | * If we are singlestepping, then ensure this thread is not connected to | |
1622 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1623 | * triggers the signal, restart the original insn even if the task is | |
1624 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1625 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1626 | * _same_ instruction should be repeated again after return from the signal | |
1627 | * handler, and SSTEP can never finish in this case. | |
1628 | */ | |
1629 | bool uprobe_deny_signal(void) | |
1630 | { | |
1631 | struct task_struct *t = current; | |
1632 | struct uprobe_task *utask = t->utask; | |
1633 | ||
1634 | if (likely(!utask || !utask->active_uprobe)) | |
1635 | return false; | |
1636 | ||
1637 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1638 | ||
1639 | if (signal_pending(t)) { | |
1640 | spin_lock_irq(&t->sighand->siglock); | |
1641 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1642 | spin_unlock_irq(&t->sighand->siglock); | |
1643 | ||
1644 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1645 | utask->state = UTASK_SSTEP_TRAPPED; | |
1646 | set_tsk_thread_flag(t, TIF_UPROBE); | |
0326f5a9 SD |
1647 | } |
1648 | } | |
1649 | ||
1650 | return true; | |
1651 | } | |
1652 | ||
499a4f3e ON |
1653 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1654 | { | |
1655 | struct vm_area_struct *vma; | |
1656 | ||
1657 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1658 | if (!valid_vma(vma, false)) | |
1659 | continue; | |
1660 | /* | |
1661 | * This is not strictly accurate, we can race with | |
1662 | * uprobe_unregister() and see the already removed | |
1663 | * uprobe if delete_uprobe() was not yet called. | |
63633cbf | 1664 | * Or this uprobe can be filtered out. |
499a4f3e ON |
1665 | */ |
1666 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
1667 | return; | |
1668 | } | |
1669 | ||
1670 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
1671 | } | |
1672 | ||
0908ad6e | 1673 | static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr) |
ec75fba9 ON |
1674 | { |
1675 | struct page *page; | |
1676 | uprobe_opcode_t opcode; | |
1677 | int result; | |
1678 | ||
1679 | pagefault_disable(); | |
1680 | result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, | |
1681 | sizeof(opcode)); | |
1682 | pagefault_enable(); | |
1683 | ||
1684 | if (likely(result == 0)) | |
1685 | goto out; | |
1686 | ||
1687 | result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); | |
1688 | if (result < 0) | |
1689 | return result; | |
1690 | ||
ab0d805c | 1691 | copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); |
ec75fba9 ON |
1692 | put_page(page); |
1693 | out: | |
0908ad6e AM |
1694 | /* This needs to return true for any variant of the trap insn */ |
1695 | return is_trap_insn(&opcode); | |
ec75fba9 ON |
1696 | } |
1697 | ||
d790d346 | 1698 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 1699 | { |
3a9ea052 ON |
1700 | struct mm_struct *mm = current->mm; |
1701 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 1702 | struct vm_area_struct *vma; |
0326f5a9 | 1703 | |
0326f5a9 SD |
1704 | down_read(&mm->mmap_sem); |
1705 | vma = find_vma(mm, bp_vaddr); | |
3a9ea052 ON |
1706 | if (vma && vma->vm_start <= bp_vaddr) { |
1707 | if (valid_vma(vma, false)) { | |
f281769e | 1708 | struct inode *inode = file_inode(vma->vm_file); |
cb113b47 | 1709 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); |
0326f5a9 | 1710 | |
3a9ea052 ON |
1711 | uprobe = find_uprobe(inode, offset); |
1712 | } | |
d790d346 ON |
1713 | |
1714 | if (!uprobe) | |
0908ad6e | 1715 | *is_swbp = is_trap_at_addr(mm, bp_vaddr); |
d790d346 ON |
1716 | } else { |
1717 | *is_swbp = -EFAULT; | |
0326f5a9 | 1718 | } |
499a4f3e ON |
1719 | |
1720 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
1721 | mmf_recalc_uprobes(mm); | |
0326f5a9 SD |
1722 | up_read(&mm->mmap_sem); |
1723 | ||
3a9ea052 ON |
1724 | return uprobe; |
1725 | } | |
1726 | ||
da1816b1 ON |
1727 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
1728 | { | |
1729 | struct uprobe_consumer *uc; | |
1730 | int remove = UPROBE_HANDLER_REMOVE; | |
0dfd0eb8 | 1731 | bool need_prep = false; /* prepare return uprobe, when needed */ |
da1816b1 ON |
1732 | |
1733 | down_read(&uprobe->register_rwsem); | |
1734 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
ea024870 | 1735 | int rc = 0; |
da1816b1 | 1736 | |
ea024870 AA |
1737 | if (uc->handler) { |
1738 | rc = uc->handler(uc, regs); | |
1739 | WARN(rc & ~UPROBE_HANDLER_MASK, | |
1740 | "bad rc=0x%x from %pf()\n", rc, uc->handler); | |
1741 | } | |
0dfd0eb8 AA |
1742 | |
1743 | if (uc->ret_handler) | |
1744 | need_prep = true; | |
1745 | ||
da1816b1 ON |
1746 | remove &= rc; |
1747 | } | |
1748 | ||
0dfd0eb8 AA |
1749 | if (need_prep && !remove) |
1750 | prepare_uretprobe(uprobe, regs); /* put bp at return */ | |
1751 | ||
da1816b1 ON |
1752 | if (remove && uprobe->consumers) { |
1753 | WARN_ON(!uprobe_is_active(uprobe)); | |
1754 | unapply_uprobe(uprobe, current->mm); | |
1755 | } | |
1756 | up_read(&uprobe->register_rwsem); | |
1757 | } | |
1758 | ||
fec8898d AA |
1759 | static void |
1760 | handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs) | |
1761 | { | |
1762 | struct uprobe *uprobe = ri->uprobe; | |
1763 | struct uprobe_consumer *uc; | |
1764 | ||
1765 | down_read(&uprobe->register_rwsem); | |
1766 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
1767 | if (uc->ret_handler) | |
1768 | uc->ret_handler(uc, ri->func, regs); | |
1769 | } | |
1770 | up_read(&uprobe->register_rwsem); | |
1771 | } | |
1772 | ||
1773 | static bool handle_trampoline(struct pt_regs *regs) | |
1774 | { | |
1775 | struct uprobe_task *utask; | |
2bb5e840 | 1776 | struct return_instance *ri; |
fec8898d AA |
1777 | bool chained; |
1778 | ||
1779 | utask = current->utask; | |
1780 | if (!utask) | |
1781 | return false; | |
1782 | ||
1783 | ri = utask->return_instances; | |
1784 | if (!ri) | |
1785 | return false; | |
1786 | ||
1787 | /* | |
1788 | * TODO: we should throw out return_instance's invalidated by | |
1789 | * longjmp(), currently we assume that the probed function always | |
1790 | * returns. | |
1791 | */ | |
1792 | instruction_pointer_set(regs, ri->orig_ret_vaddr); | |
1793 | ||
1794 | for (;;) { | |
1795 | handle_uretprobe_chain(ri, regs); | |
1796 | ||
1797 | chained = ri->chained; | |
2bb5e840 | 1798 | ri = free_ret_instance(ri); |
878b5a6e | 1799 | utask->depth--; |
fec8898d AA |
1800 | |
1801 | if (!chained) | |
1802 | break; | |
fec8898d AA |
1803 | BUG_ON(!ri); |
1804 | } | |
1805 | ||
1806 | utask->return_instances = ri; | |
1807 | ||
1808 | return true; | |
1809 | } | |
1810 | ||
6fe50a28 DL |
1811 | bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs) |
1812 | { | |
1813 | return false; | |
1814 | } | |
1815 | ||
3a9ea052 ON |
1816 | /* |
1817 | * Run handler and ask thread to singlestep. | |
1818 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
1819 | */ | |
1820 | static void handle_swbp(struct pt_regs *regs) | |
1821 | { | |
3a9ea052 ON |
1822 | struct uprobe *uprobe; |
1823 | unsigned long bp_vaddr; | |
56bb4cf6 | 1824 | int uninitialized_var(is_swbp); |
3a9ea052 ON |
1825 | |
1826 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
fec8898d AA |
1827 | if (bp_vaddr == get_trampoline_vaddr()) { |
1828 | if (handle_trampoline(regs)) | |
1829 | return; | |
3a9ea052 | 1830 | |
fec8898d AA |
1831 | pr_warn("uprobe: unable to handle uretprobe pid/tgid=%d/%d\n", |
1832 | current->pid, current->tgid); | |
1833 | } | |
1834 | ||
1835 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); | |
0326f5a9 | 1836 | if (!uprobe) { |
56bb4cf6 ON |
1837 | if (is_swbp > 0) { |
1838 | /* No matching uprobe; signal SIGTRAP. */ | |
1839 | send_sig(SIGTRAP, current, 0); | |
1840 | } else { | |
1841 | /* | |
1842 | * Either we raced with uprobe_unregister() or we can't | |
1843 | * access this memory. The latter is only possible if | |
1844 | * another thread plays with our ->mm. In both cases | |
1845 | * we can simply restart. If this vma was unmapped we | |
1846 | * can pretend this insn was not executed yet and get | |
1847 | * the (correct) SIGSEGV after restart. | |
1848 | */ | |
1849 | instruction_pointer_set(regs, bp_vaddr); | |
1850 | } | |
0326f5a9 SD |
1851 | return; |
1852 | } | |
74e59dfc ON |
1853 | |
1854 | /* change it in advance for ->handler() and restart */ | |
1855 | instruction_pointer_set(regs, bp_vaddr); | |
1856 | ||
142b18dd ON |
1857 | /* |
1858 | * TODO: move copy_insn/etc into _register and remove this hack. | |
1859 | * After we hit the bp, _unregister + _register can install the | |
1860 | * new and not-yet-analyzed uprobe at the same address, restart. | |
1861 | */ | |
1862 | smp_rmb(); /* pairs with wmb() in install_breakpoint() */ | |
71434f2f | 1863 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
74e59dfc | 1864 | goto out; |
0326f5a9 | 1865 | |
72fd293a ON |
1866 | /* Tracing handlers use ->utask to communicate with fetch methods */ |
1867 | if (!get_utask()) | |
1868 | goto out; | |
1869 | ||
6fe50a28 DL |
1870 | if (arch_uprobe_ignore(&uprobe->arch, regs)) |
1871 | goto out; | |
1872 | ||
0326f5a9 | 1873 | handler_chain(uprobe, regs); |
6fe50a28 | 1874 | |
8a6b1732 | 1875 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
0578a970 | 1876 | goto out; |
0326f5a9 | 1877 | |
608e7427 | 1878 | if (!pre_ssout(uprobe, regs, bp_vaddr)) |
0326f5a9 | 1879 | return; |
0326f5a9 | 1880 | |
8a6b1732 | 1881 | /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */ |
0578a970 | 1882 | out: |
8bd87445 | 1883 | put_uprobe(uprobe); |
0326f5a9 SD |
1884 | } |
1885 | ||
1886 | /* | |
1887 | * Perform required fix-ups and disable singlestep. | |
1888 | * Allow pending signals to take effect. | |
1889 | */ | |
1890 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
1891 | { | |
1892 | struct uprobe *uprobe; | |
014940ba | 1893 | int err = 0; |
0326f5a9 SD |
1894 | |
1895 | uprobe = utask->active_uprobe; | |
1896 | if (utask->state == UTASK_SSTEP_ACK) | |
014940ba | 1897 | err = arch_uprobe_post_xol(&uprobe->arch, regs); |
0326f5a9 SD |
1898 | else if (utask->state == UTASK_SSTEP_TRAPPED) |
1899 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
1900 | else | |
1901 | WARN_ON_ONCE(1); | |
1902 | ||
1903 | put_uprobe(uprobe); | |
1904 | utask->active_uprobe = NULL; | |
1905 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 1906 | xol_free_insn_slot(current); |
0326f5a9 SD |
1907 | |
1908 | spin_lock_irq(¤t->sighand->siglock); | |
1909 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
1910 | spin_unlock_irq(¤t->sighand->siglock); | |
014940ba ON |
1911 | |
1912 | if (unlikely(err)) { | |
1913 | uprobe_warn(current, "execute the probed insn, sending SIGILL."); | |
1914 | force_sig_info(SIGILL, SEND_SIG_FORCED, current); | |
1915 | } | |
0326f5a9 SD |
1916 | } |
1917 | ||
1918 | /* | |
1b08e907 ON |
1919 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
1920 | * allows the thread to return from interrupt. After that handle_swbp() | |
1921 | * sets utask->active_uprobe. | |
0326f5a9 | 1922 | * |
1b08e907 ON |
1923 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
1924 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
1925 | * |
1926 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
1927 | * uprobe_notify_resume(). | |
1928 | */ | |
1929 | void uprobe_notify_resume(struct pt_regs *regs) | |
1930 | { | |
1931 | struct uprobe_task *utask; | |
1932 | ||
db023ea5 ON |
1933 | clear_thread_flag(TIF_UPROBE); |
1934 | ||
0326f5a9 | 1935 | utask = current->utask; |
1b08e907 | 1936 | if (utask && utask->active_uprobe) |
0326f5a9 | 1937 | handle_singlestep(utask, regs); |
1b08e907 ON |
1938 | else |
1939 | handle_swbp(regs); | |
0326f5a9 SD |
1940 | } |
1941 | ||
1942 | /* | |
1943 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
1944 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
1945 | */ | |
1946 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
1947 | { | |
0dfd0eb8 AA |
1948 | if (!current->mm) |
1949 | return 0; | |
1950 | ||
1951 | if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) && | |
1952 | (!current->utask || !current->utask->return_instances)) | |
0326f5a9 SD |
1953 | return 0; |
1954 | ||
0326f5a9 | 1955 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
1956 | return 1; |
1957 | } | |
1958 | ||
1959 | /* | |
1960 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
1961 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
1962 | */ | |
1963 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
1964 | { | |
1965 | struct uprobe_task *utask = current->utask; | |
1966 | ||
1967 | if (!current->mm || !utask || !utask->active_uprobe) | |
1968 | /* task is currently not uprobed */ | |
1969 | return 0; | |
1970 | ||
1971 | utask->state = UTASK_SSTEP_ACK; | |
1972 | set_thread_flag(TIF_UPROBE); | |
1973 | return 1; | |
1974 | } | |
1975 | ||
1976 | static struct notifier_block uprobe_exception_nb = { | |
1977 | .notifier_call = arch_uprobe_exception_notify, | |
1978 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
1979 | }; | |
1980 | ||
2b144498 SD |
1981 | static int __init init_uprobes(void) |
1982 | { | |
1983 | int i; | |
1984 | ||
66d06dff | 1985 | for (i = 0; i < UPROBES_HASH_SZ; i++) |
2b144498 | 1986 | mutex_init(&uprobes_mmap_mutex[i]); |
0326f5a9 | 1987 | |
32cdba1e ON |
1988 | if (percpu_init_rwsem(&dup_mmap_sem)) |
1989 | return -ENOMEM; | |
1990 | ||
0326f5a9 | 1991 | return register_die_notifier(&uprobe_exception_nb); |
2b144498 | 1992 | } |
736e89d9 | 1993 | __initcall(init_uprobes); |