1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * User-space Probes (UProbes)
4 *
5 * Copyright (C) IBM Corporation, 2008-2012
6 * Authors:
7 * Srikar Dronamraju
8 * Jim Keniston
9 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra
10 */
11
12 #include <linux/kernel.h>
13 #include <linux/highmem.h>
14 #include <linux/pagemap.h> /* read_mapping_page */
15 #include <linux/slab.h>
16 #include <linux/sched.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/coredump.h>
19 #include <linux/export.h>
20 #include <linux/rmap.h> /* anon_vma_prepare */
21 #include <linux/mmu_notifier.h>
22 #include <linux/swap.h> /* folio_free_swap */
23 #include <linux/ptrace.h> /* user_enable_single_step */
24 #include <linux/kdebug.h> /* notifier mechanism */
25 #include <linux/percpu-rwsem.h>
26 #include <linux/task_work.h>
27 #include <linux/shmem_fs.h>
28 #include <linux/khugepaged.h>
29
30 #include <linux/uprobes.h>
31
32 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
33 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
34
35 static struct rb_root uprobes_tree = RB_ROOT;
36 /*
37 * allows us to skip the uprobe_mmap if there are no uprobe events active
38 * at this time. Probably a fine grained per inode count is better?
39 */
40 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
41
42 static DEFINE_RWLOCK(uprobes_treelock); /* serialize rbtree access */
43
44 #define UPROBES_HASH_SZ 13
45 /* serialize uprobe->pending_list */
46 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
47 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
48
49 DEFINE_STATIC_PERCPU_RWSEM(dup_mmap_sem);
50
51 /* Have a copy of original instruction */
52 #define UPROBE_COPY_INSN 0
53
54 struct uprobe {
55 struct rb_node rb_node; /* node in the rb tree */
56 refcount_t ref;
57 struct rw_semaphore register_rwsem;
58 struct rw_semaphore consumer_rwsem;
59 struct list_head pending_list;
60 struct uprobe_consumer *consumers;
61 struct inode *inode; /* Also hold a ref to inode */
62 loff_t offset;
63 loff_t ref_ctr_offset;
64 unsigned long flags;
65
66 /*
67 * The generic code assumes that it has two members of unknown type
68 * owned by the arch-specific code:
69 *
70 * insn - copy_insn() saves the original instruction here for
71 * arch_uprobe_analyze_insn().
72 *
73 * ixol - potentially modified instruction to execute out of
74 * line, copied to xol_area by xol_get_insn_slot().
75 */
76 struct arch_uprobe arch;
77 };
78
79 struct delayed_uprobe {
80 struct list_head list;
81 struct uprobe *uprobe;
82 struct mm_struct *mm;
83 };
84
85 static DEFINE_MUTEX(delayed_uprobe_lock);
86 static LIST_HEAD(delayed_uprobe_list);
87
88 /*
89 * Execute out of line area: anonymous executable mapping installed
90 * by the probed task to execute the copy of the original instruction
91 * mangled by set_swbp().
92 *
93 * On a breakpoint hit, thread contests for a slot. It frees the
94 * slot after singlestep. Currently a fixed number of slots are
95 * allocated.
96 */
97 struct xol_area {
98 wait_queue_head_t wq; /* if all slots are busy */
99 atomic_t slot_count; /* number of in-use slots */
100 unsigned long *bitmap; /* 0 = free slot */
101
102 struct vm_special_mapping xol_mapping;
103 struct page *pages[2];
104 /*
105 * We keep the vma's vm_start rather than a pointer to the vma
106 * itself. The probed process or a naughty kernel module could make
107 * the vma go away, and we must handle that reasonably gracefully.
108 */
109 unsigned long vaddr; /* Page(s) of instruction slots */
110 };
111
112 /*
113 * valid_vma: Verify if the specified vma is an executable vma
114 * Relax restrictions while unregistering: vm_flags might have
115 * changed after breakpoint was inserted.
116 * - is_register: indicates if we are in register context.
117 * - Return 1 if the specified virtual address is in an
118 * executable vma.
119 */
120 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
121 {
122 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
123
124 if (is_register)
125 flags |= VM_WRITE;
126
127 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
128 }
129
130 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
131 {
132 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
133 }
134
135 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
136 {
137 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
138 }
139
140 /**
141 * __replace_page - replace page in vma by new page.
142 * based on replace_page in mm/ksm.c
143 *
144 * @vma: vma that holds the pte pointing to page
145 * @addr: address the old @page is mapped at
146 * @old_page: the page we are replacing by new_page
147 * @new_page: the modified page we replace page by
148 *
149 * If @new_page is NULL, only unmap @old_page.
150 *
151 * Returns 0 on success, negative error code otherwise.
152 */
153 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
154 struct page *old_page, struct page *new_page)
155 {
156 struct folio *old_folio = page_folio(old_page);
157 struct folio *new_folio;
158 struct mm_struct *mm = vma->vm_mm;
159 DEFINE_FOLIO_VMA_WALK(pvmw, old_folio, vma, addr, 0);
160 int err;
161 struct mmu_notifier_range range;
162
163 mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, addr,
164 addr + PAGE_SIZE);
165
166 if (new_page) {
167 new_folio = page_folio(new_page);
168 err = mem_cgroup_charge(new_folio, vma->vm_mm, GFP_KERNEL);
169 if (err)
170 return err;
171 }
172
173 /* For folio_free_swap() below */
174 folio_lock(old_folio);
175
176 mmu_notifier_invalidate_range_start(&range);
177 err = -EAGAIN;
178 if (!page_vma_mapped_walk(&pvmw))
179 goto unlock;
180 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
181
182 if (new_page) {
183 folio_get(new_folio);
184 folio_add_new_anon_rmap(new_folio, vma, addr, RMAP_EXCLUSIVE);
185 folio_add_lru_vma(new_folio, vma);
186 } else
187 /* no new page, just dec_mm_counter for old_page */
188 dec_mm_counter(mm, MM_ANONPAGES);
189
190 if (!folio_test_anon(old_folio)) {
191 dec_mm_counter(mm, mm_counter_file(old_folio));
192 inc_mm_counter(mm, MM_ANONPAGES);
193 }
194
195 flush_cache_page(vma, addr, pte_pfn(ptep_get(pvmw.pte)));
196 ptep_clear_flush(vma, addr, pvmw.pte);
197 if (new_page)
198 set_pte_at(mm, addr, pvmw.pte,
199 mk_pte(new_page, vma->vm_page_prot));
200
201 folio_remove_rmap_pte(old_folio, old_page, vma);
202 if (!folio_mapped(old_folio))
203 folio_free_swap(old_folio);
204 page_vma_mapped_walk_done(&pvmw);
205 folio_put(old_folio);
206
207 err = 0;
208 unlock:
209 mmu_notifier_invalidate_range_end(&range);
210 folio_unlock(old_folio);
211 return err;
212 }
213
214 /**
215 * is_swbp_insn - check if instruction is breakpoint instruction.
216 * @insn: instruction to be checked.
217 * Default implementation of is_swbp_insn
218 * Returns true if @insn is a breakpoint instruction.
219 */
220 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
221 {
222 return *insn == UPROBE_SWBP_INSN;
223 }
224
225 /**
226 * is_trap_insn - check if instruction is breakpoint instruction.
227 * @insn: instruction to be checked.
228 * Default implementation of is_trap_insn
229 * Returns true if @insn is a breakpoint instruction.
230 *
231 * This function is needed for the case where an architecture has multiple
232 * trap instructions (like powerpc).
233 */
234 bool __weak is_trap_insn(uprobe_opcode_t *insn)
235 {
236 return is_swbp_insn(insn);
237 }
238
239 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
240 {
241 void *kaddr = kmap_atomic(page);
242 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
243 kunmap_atomic(kaddr);
244 }
245
246 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
247 {
248 void *kaddr = kmap_atomic(page);
249 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
250 kunmap_atomic(kaddr);
251 }
252
253 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
254 {
255 uprobe_opcode_t old_opcode;
256 bool is_swbp;
257
258 /*
259 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
260 * We do not check if it is any other 'trap variant' which could
261 * be conditional trap instruction such as the one powerpc supports.
262 *
263 * The logic is that we do not care if the underlying instruction
264 * is a trap variant; uprobes always wins over any other (gdb)
265 * breakpoint.
266 */
267 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
268 is_swbp = is_swbp_insn(&old_opcode);
269
270 if (is_swbp_insn(new_opcode)) {
271 if (is_swbp) /* register: already installed? */
272 return 0;
273 } else {
274 if (!is_swbp) /* unregister: was it changed by us? */
275 return 0;
276 }
277
278 return 1;
279 }
280
281 static struct delayed_uprobe *
282 delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm)
283 {
284 struct delayed_uprobe *du;
285
286 list_for_each_entry(du, &delayed_uprobe_list, list)
287 if (du->uprobe == uprobe && du->mm == mm)
288 return du;
289 return NULL;
290 }
291
292 static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
293 {
294 struct delayed_uprobe *du;
295
296 if (delayed_uprobe_check(uprobe, mm))
297 return 0;
298
299 du = kzalloc(sizeof(*du), GFP_KERNEL);
300 if (!du)
301 return -ENOMEM;
302
303 du->uprobe = uprobe;
304 du->mm = mm;
305 list_add(&du->list, &delayed_uprobe_list);
306 return 0;
307 }
308
309 static void delayed_uprobe_delete(struct delayed_uprobe *du)
310 {
311 if (WARN_ON(!du))
312 return;
313 list_del(&du->list);
314 kfree(du);
315 }
316
317 static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm)
318 {
319 struct list_head *pos, *q;
320 struct delayed_uprobe *du;
321
322 if (!uprobe && !mm)
323 return;
324
325 list_for_each_safe(pos, q, &delayed_uprobe_list) {
326 du = list_entry(pos, struct delayed_uprobe, list);
327
328 if (uprobe && du->uprobe != uprobe)
329 continue;
330 if (mm && du->mm != mm)
331 continue;
332
333 delayed_uprobe_delete(du);
334 }
335 }
336
337 static bool valid_ref_ctr_vma(struct uprobe *uprobe,
338 struct vm_area_struct *vma)
339 {
340 unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
341
342 return uprobe->ref_ctr_offset &&
343 vma->vm_file &&
344 file_inode(vma->vm_file) == uprobe->inode &&
345 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
346 vma->vm_start <= vaddr &&
347 vma->vm_end > vaddr;
348 }
349
350 static struct vm_area_struct *
351 find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm)
352 {
353 VMA_ITERATOR(vmi, mm, 0);
354 struct vm_area_struct *tmp;
355
356 for_each_vma(vmi, tmp)
357 if (valid_ref_ctr_vma(uprobe, tmp))
358 return tmp;
359
360 return NULL;
361 }
362
363 static int
364 __update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d)
365 {
366 void *kaddr;
367 struct page *page;
368 int ret;
369 short *ptr;
370
371 if (!vaddr || !d)
372 return -EINVAL;
373
374 ret = get_user_pages_remote(mm, vaddr, 1,
375 FOLL_WRITE, &page, NULL);
376 if (unlikely(ret <= 0)) {
377 /*
378 * We are asking for 1 page. If get_user_pages_remote() fails,
379 * it may return 0, in that case we have to return error.
380 */
381 return ret == 0 ? -EBUSY : ret;
382 }
383
384 kaddr = kmap_atomic(page);
385 ptr = kaddr + (vaddr & ~PAGE_MASK);
386
387 if (unlikely(*ptr + d < 0)) {
388 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
389 "curr val: %d, delta: %d\n", vaddr, *ptr, d);
390 ret = -EINVAL;
391 goto out;
392 }
393
394 *ptr += d;
395 ret = 0;
396 out:
397 kunmap_atomic(kaddr);
398 put_page(page);
399 return ret;
400 }
401
402 static void update_ref_ctr_warn(struct uprobe *uprobe,
403 struct mm_struct *mm, short d)
404 {
405 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
406 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
407 d > 0 ? "increment" : "decrement", uprobe->inode->i_ino,
408 (unsigned long long) uprobe->offset,
409 (unsigned long long) uprobe->ref_ctr_offset, mm);
410 }
411
412 static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
413 short d)
414 {
415 struct vm_area_struct *rc_vma;
416 unsigned long rc_vaddr;
417 int ret = 0;
418
419 rc_vma = find_ref_ctr_vma(uprobe, mm);
420
421 if (rc_vma) {
422 rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset);
423 ret = __update_ref_ctr(mm, rc_vaddr, d);
424 if (ret)
425 update_ref_ctr_warn(uprobe, mm, d);
426
427 if (d > 0)
428 return ret;
429 }
430
431 mutex_lock(&delayed_uprobe_lock);
432 if (d > 0)
433 ret = delayed_uprobe_add(uprobe, mm);
434 else
435 delayed_uprobe_remove(uprobe, mm);
436 mutex_unlock(&delayed_uprobe_lock);
437
438 return ret;
439 }
440
441 /*
442 * NOTE:
443 * Expect the breakpoint instruction to be the smallest size instruction for
444 * the architecture. If an arch has variable length instruction and the
445 * breakpoint instruction is not of the smallest length instruction
446 * supported by that architecture then we need to modify is_trap_at_addr and
447 * uprobe_write_opcode accordingly. This would never be a problem for archs
448 * that have fixed length instructions.
449 *
450 * uprobe_write_opcode - write the opcode at a given virtual address.
451 * @auprobe: arch specific probepoint information.
452 * @mm: the probed process address space.
453 * @vaddr: the virtual address to store the opcode.
454 * @opcode: opcode to be written at @vaddr.
455 *
456 * Called with mm->mmap_lock held for write.
457 * Return 0 (success) or a negative errno.
458 */
459 int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
460 unsigned long vaddr, uprobe_opcode_t opcode)
461 {
462 struct uprobe *uprobe;
463 struct page *old_page, *new_page;
464 struct vm_area_struct *vma;
465 int ret, is_register, ref_ctr_updated = 0;
466 bool orig_page_huge = false;
467 unsigned int gup_flags = FOLL_FORCE;
468
469 is_register = is_swbp_insn(&opcode);
470 uprobe = container_of(auprobe, struct uprobe, arch);
471
472 retry:
473 if (is_register)
474 gup_flags |= FOLL_SPLIT_PMD;
475 /* Read the page with vaddr into memory */
476 old_page = get_user_page_vma_remote(mm, vaddr, gup_flags, &vma);
477 if (IS_ERR(old_page))
478 return PTR_ERR(old_page);
479
480 ret = verify_opcode(old_page, vaddr, &opcode);
481 if (ret <= 0)
482 goto put_old;
483
484 if (WARN(!is_register && PageCompound(old_page),
485 "uprobe unregister should never work on compound page\n")) {
486 ret = -EINVAL;
487 goto put_old;
488 }
489
490 /* We are going to replace instruction, update ref_ctr. */
491 if (!ref_ctr_updated && uprobe->ref_ctr_offset) {
492 ret = update_ref_ctr(uprobe, mm, is_register ? 1 : -1);
493 if (ret)
494 goto put_old;
495
496 ref_ctr_updated = 1;
497 }
498
499 ret = 0;
500 if (!is_register && !PageAnon(old_page))
501 goto put_old;
502
503 ret = anon_vma_prepare(vma);
504 if (ret)
505 goto put_old;
506
507 ret = -ENOMEM;
508 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
509 if (!new_page)
510 goto put_old;
511
512 __SetPageUptodate(new_page);
513 copy_highpage(new_page, old_page);
514 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
515
516 if (!is_register) {
517 struct page *orig_page;
518 pgoff_t index;
519
520 VM_BUG_ON_PAGE(!PageAnon(old_page), old_page);
521
522 index = vaddr_to_offset(vma, vaddr & PAGE_MASK) >> PAGE_SHIFT;
523 orig_page = find_get_page(vma->vm_file->f_inode->i_mapping,
524 index);
525
526 if (orig_page) {
527 if (PageUptodate(orig_page) &&
528 pages_identical(new_page, orig_page)) {
529 /* let go new_page */
530 put_page(new_page);
531 new_page = NULL;
532
533 if (PageCompound(orig_page))
534 orig_page_huge = true;
535 }
536 put_page(orig_page);
537 }
538 }
539
540 ret = __replace_page(vma, vaddr & PAGE_MASK, old_page, new_page);
541 if (new_page)
542 put_page(new_page);
543 put_old:
544 put_page(old_page);
545
546 if (unlikely(ret == -EAGAIN))
547 goto retry;
548
549 /* Revert back reference counter if instruction update failed. */
550 if (ret && is_register && ref_ctr_updated)
551 update_ref_ctr(uprobe, mm, -1);
552
553 /* try collapse pmd for compound page */
554 if (!ret && orig_page_huge)
555 collapse_pte_mapped_thp(mm, vaddr, false);
556
557 return ret;
558 }
559
560 /**
561 * set_swbp - store breakpoint at a given address.
562 * @auprobe: arch specific probepoint information.
563 * @mm: the probed process address space.
564 * @vaddr: the virtual address to insert the opcode.
565 *
566 * For mm @mm, store the breakpoint instruction at @vaddr.
567 * Return 0 (success) or a negative errno.
568 */
569 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
570 {
571 return uprobe_write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
572 }
573
574 /**
575 * set_orig_insn - Restore the original instruction.
576 * @mm: the probed process address space.
577 * @auprobe: arch specific probepoint information.
578 * @vaddr: the virtual address to insert the opcode.
579 *
580 * For mm @mm, restore the original opcode (opcode) at @vaddr.
581 * Return 0 (success) or a negative errno.
582 */
583 int __weak
584 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
585 {
586 return uprobe_write_opcode(auprobe, mm, vaddr,
587 *(uprobe_opcode_t *)&auprobe->insn);
588 }
589
590 static struct uprobe *get_uprobe(struct uprobe *uprobe)
591 {
592 refcount_inc(&uprobe->ref);
593 return uprobe;
594 }
595
596 static void put_uprobe(struct uprobe *uprobe)
597 {
598 if (refcount_dec_and_test(&uprobe->ref)) {
599 /*
600 * If application munmap(exec_vma) before uprobe_unregister()
601 * gets called, we don't get a chance to remove uprobe from
602 * delayed_uprobe_list from remove_breakpoint(). Do it here.
603 */
604 mutex_lock(&delayed_uprobe_lock);
605 delayed_uprobe_remove(uprobe, NULL);
606 mutex_unlock(&delayed_uprobe_lock);
607 kfree(uprobe);
608 }
609 }
610
611 static __always_inline
612 int uprobe_cmp(const struct inode *l_inode, const loff_t l_offset,
613 const struct uprobe *r)
614 {
615 if (l_inode < r->inode)
616 return -1;
617
618 if (l_inode > r->inode)
619 return 1;
620
621 if (l_offset < r->offset)
622 return -1;
623
624 if (l_offset > r->offset)
625 return 1;
626
627 return 0;
628 }
629
630 #define __node_2_uprobe(node) \
631 rb_entry((node), struct uprobe, rb_node)
632
633 struct __uprobe_key {
634 struct inode *inode;
635 loff_t offset;
636 };
637
638 static inline int __uprobe_cmp_key(const void *key, const struct rb_node *b)
639 {
640 const struct __uprobe_key *a = key;
641 return uprobe_cmp(a->inode, a->offset, __node_2_uprobe(b));
642 }
643
644 static inline int __uprobe_cmp(struct rb_node *a, const struct rb_node *b)
645 {
646 struct uprobe *u = __node_2_uprobe(a);
647 return uprobe_cmp(u->inode, u->offset, __node_2_uprobe(b));
648 }
649
650 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
651 {
652 struct __uprobe_key key = {
653 .inode = inode,
654 .offset = offset,
655 };
656 struct rb_node *node = rb_find(&key, &uprobes_tree, __uprobe_cmp_key);
657
658 if (node)
659 return get_uprobe(__node_2_uprobe(node));
660
661 return NULL;
662 }
663
664 /*
665 * Find a uprobe corresponding to a given inode:offset
666 * Acquires uprobes_treelock
667 */
668 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
669 {
670 struct uprobe *uprobe;
671
672 read_lock(&uprobes_treelock);
673 uprobe = __find_uprobe(inode, offset);
674 read_unlock(&uprobes_treelock);
675
676 return uprobe;
677 }
678
679 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
680 {
681 struct rb_node *node;
682
683 node = rb_find_add(&uprobe->rb_node, &uprobes_tree, __uprobe_cmp);
684 if (node)
685 return get_uprobe(__node_2_uprobe(node));
686
687 /* get access + creation ref */
688 refcount_set(&uprobe->ref, 2);
689 return NULL;
690 }
691
692 /*
693 * Acquire uprobes_treelock.
694 * Matching uprobe already exists in rbtree;
695 * increment (access refcount) and return the matching uprobe.
696 *
697 * No matching uprobe; insert the uprobe in rb_tree;
698 * get a double refcount (access + creation) and return NULL.
699 */
700 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
701 {
702 struct uprobe *u;
703
704 write_lock(&uprobes_treelock);
705 u = __insert_uprobe(uprobe);
706 write_unlock(&uprobes_treelock);
707
708 return u;
709 }
710
711 static void
712 ref_ctr_mismatch_warn(struct uprobe *cur_uprobe, struct uprobe *uprobe)
713 {
714 pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx "
715 "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n",
716 uprobe->inode->i_ino, (unsigned long long) uprobe->offset,
717 (unsigned long long) cur_uprobe->ref_ctr_offset,
718 (unsigned long long) uprobe->ref_ctr_offset);
719 }
720
721 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
722 loff_t ref_ctr_offset)
723 {
724 struct uprobe *uprobe, *cur_uprobe;
725
726 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
727 if (!uprobe)
728 return NULL;
729
730 uprobe->inode = inode;
731 uprobe->offset = offset;
732 uprobe->ref_ctr_offset = ref_ctr_offset;
733 init_rwsem(&uprobe->register_rwsem);
734 init_rwsem(&uprobe->consumer_rwsem);
735
736 /* add to uprobes_tree, sorted on inode:offset */
737 cur_uprobe = insert_uprobe(uprobe);
738 /* a uprobe exists for this inode:offset combination */
739 if (cur_uprobe) {
740 if (cur_uprobe->ref_ctr_offset != uprobe->ref_ctr_offset) {
741 ref_ctr_mismatch_warn(cur_uprobe, uprobe);
742 put_uprobe(cur_uprobe);
743 kfree(uprobe);
744 return ERR_PTR(-EINVAL);
745 }
746 kfree(uprobe);
747 uprobe = cur_uprobe;
748 }
749
750 return uprobe;
751 }
752
753 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
754 {
755 down_write(&uprobe->consumer_rwsem);
756 uc->next = uprobe->consumers;
757 uprobe->consumers = uc;
758 up_write(&uprobe->consumer_rwsem);
759 }
760
761 /*
762 * For uprobe @uprobe, delete the consumer @uc.
763 * Return true if the @uc is deleted successfully
764 * or return false.
765 */
766 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
767 {
768 struct uprobe_consumer **con;
769 bool ret = false;
770
771 down_write(&uprobe->consumer_rwsem);
772 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
773 if (*con == uc) {
774 *con = uc->next;
775 ret = true;
776 break;
777 }
778 }
779 up_write(&uprobe->consumer_rwsem);
780
781 return ret;
782 }
783
784 static int __copy_insn(struct address_space *mapping, struct file *filp,
785 void *insn, int nbytes, loff_t offset)
786 {
787 struct page *page;
788 /*
789 * Ensure that the page that has the original instruction is populated
790 * and in page-cache. If ->read_folio == NULL it must be shmem_mapping(),
791 * see uprobe_register().
792 */
793 if (mapping->a_ops->read_folio)
794 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
795 else
796 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
797 if (IS_ERR(page))
798 return PTR_ERR(page);
799
800 copy_from_page(page, offset, insn, nbytes);
801 put_page(page);
802
803 return 0;
804 }
805
806 static int copy_insn(struct uprobe *uprobe, struct file *filp)
807 {
808 struct address_space *mapping = uprobe->inode->i_mapping;
809 loff_t offs = uprobe->offset;
810 void *insn = &uprobe->arch.insn;
811 int size = sizeof(uprobe->arch.insn);
812 int len, err = -EIO;
813
814 /* Copy only available bytes, -EIO if nothing was read */
815 do {
816 if (offs >= i_size_read(uprobe->inode))
817 break;
818
819 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
820 err = __copy_insn(mapping, filp, insn, len, offs);
821 if (err)
822 break;
823
824 insn += len;
825 offs += len;
826 size -= len;
827 } while (size);
828
829 return err;
830 }
831
832 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
833 struct mm_struct *mm, unsigned long vaddr)
834 {
835 int ret = 0;
836
837 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
838 return ret;
839
840 /* TODO: move this into _register, until then we abuse this sem. */
841 down_write(&uprobe->consumer_rwsem);
842 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
843 goto out;
844
845 ret = copy_insn(uprobe, file);
846 if (ret)
847 goto out;
848
849 ret = -ENOTSUPP;
850 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
851 goto out;
852
853 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
854 if (ret)
855 goto out;
856
857 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
858 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
859
860 out:
861 up_write(&uprobe->consumer_rwsem);
862
863 return ret;
864 }
865
866 static inline bool consumer_filter(struct uprobe_consumer *uc,
867 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
868 {
869 return !uc->filter || uc->filter(uc, ctx, mm);
870 }
871
872 static bool filter_chain(struct uprobe *uprobe,
873 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
874 {
875 struct uprobe_consumer *uc;
876 bool ret = false;
877
878 down_read(&uprobe->consumer_rwsem);
879 for (uc = uprobe->consumers; uc; uc = uc->next) {
880 ret = consumer_filter(uc, ctx, mm);
881 if (ret)
882 break;
883 }
884 up_read(&uprobe->consumer_rwsem);
885
886 return ret;
887 }
888
889 static int
890 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
891 struct vm_area_struct *vma, unsigned long vaddr)
892 {
893 bool first_uprobe;
894 int ret;
895
896 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
897 if (ret)
898 return ret;
899
900 /*
901 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
902 * the task can hit this breakpoint right after __replace_page().
903 */
904 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
905 if (first_uprobe)
906 set_bit(MMF_HAS_UPROBES, &mm->flags);
907
908 ret = set_swbp(&uprobe->arch, mm, vaddr);
909 if (!ret)
910 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
911 else if (first_uprobe)
912 clear_bit(MMF_HAS_UPROBES, &mm->flags);
913
914 return ret;
915 }
916
917 static int
918 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
919 {
920 set_bit(MMF_RECALC_UPROBES, &mm->flags);
921 return set_orig_insn(&uprobe->arch, mm, vaddr);
922 }
923
924 static inline bool uprobe_is_active(struct uprobe *uprobe)
925 {
926 return !RB_EMPTY_NODE(&uprobe->rb_node);
927 }
928 /*
929 * There could be threads that have already hit the breakpoint. They
930 * will recheck the current insn and restart if find_uprobe() fails.
931 * See find_active_uprobe().
932 */
933 static void delete_uprobe(struct uprobe *uprobe)
934 {
935 if (WARN_ON(!uprobe_is_active(uprobe)))
936 return;
937
938 write_lock(&uprobes_treelock);
939 rb_erase(&uprobe->rb_node, &uprobes_tree);
940 write_unlock(&uprobes_treelock);
941 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
942 put_uprobe(uprobe);
943 }
944
945 struct map_info {
946 struct map_info *next;
947 struct mm_struct *mm;
948 unsigned long vaddr;
949 };
950
951 static inline struct map_info *free_map_info(struct map_info *info)
952 {
953 struct map_info *next = info->next;
954 kfree(info);
955 return next;
956 }
957
958 static struct map_info *
959 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
960 {
961 unsigned long pgoff = offset >> PAGE_SHIFT;
962 struct vm_area_struct *vma;
963 struct map_info *curr = NULL;
964 struct map_info *prev = NULL;
965 struct map_info *info;
966 int more = 0;
967
968 again:
969 i_mmap_lock_read(mapping);
970 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
971 if (!valid_vma(vma, is_register))
972 continue;
973
974 if (!prev && !more) {
975 /*
976 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
977 * reclaim. This is optimistic, no harm done if it fails.
978 */
979 prev = kmalloc(sizeof(struct map_info),
980 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
981 if (prev)
982 prev->next = NULL;
983 }
984 if (!prev) {
985 more++;
986 continue;
987 }
988
989 if (!mmget_not_zero(vma->vm_mm))
990 continue;
991
992 info = prev;
993 prev = prev->next;
994 info->next = curr;
995 curr = info;
996
997 info->mm = vma->vm_mm;
998 info->vaddr = offset_to_vaddr(vma, offset);
999 }
1000 i_mmap_unlock_read(mapping);
1001
1002 if (!more)
1003 goto out;
1004
1005 prev = curr;
1006 while (curr) {
1007 mmput(curr->mm);
1008 curr = curr->next;
1009 }
1010
1011 do {
1012 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
1013 if (!info) {
1014 curr = ERR_PTR(-ENOMEM);
1015 goto out;
1016 }
1017 info->next = prev;
1018 prev = info;
1019 } while (--more);
1020
1021 goto again;
1022 out:
1023 while (prev)
1024 prev = free_map_info(prev);
1025 return curr;
1026 }
1027
1028 static int
1029 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
1030 {
1031 bool is_register = !!new;
1032 struct map_info *info;
1033 int err = 0;
1034
1035 percpu_down_write(&dup_mmap_sem);
1036 info = build_map_info(uprobe->inode->i_mapping,
1037 uprobe->offset, is_register);
1038 if (IS_ERR(info)) {
1039 err = PTR_ERR(info);
1040 goto out;
1041 }
1042
1043 while (info) {
1044 struct mm_struct *mm = info->mm;
1045 struct vm_area_struct *vma;
1046
1047 if (err && is_register)
1048 goto free;
1049
1050 mmap_write_lock(mm);
1051 vma = find_vma(mm, info->vaddr);
1052 if (!vma || !valid_vma(vma, is_register) ||
1053 file_inode(vma->vm_file) != uprobe->inode)
1054 goto unlock;
1055
1056 if (vma->vm_start > info->vaddr ||
1057 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
1058 goto unlock;
1059
1060 if (is_register) {
1061 /* consult only the "caller", new consumer. */
1062 if (consumer_filter(new,
1063 UPROBE_FILTER_REGISTER, mm))
1064 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
1065 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
1066 if (!filter_chain(uprobe,
1067 UPROBE_FILTER_UNREGISTER, mm))
1068 err |= remove_breakpoint(uprobe, mm, info->vaddr);
1069 }
1070
1071 unlock:
1072 mmap_write_unlock(mm);
1073 free:
1074 mmput(mm);
1075 info = free_map_info(info);
1076 }
1077 out:
1078 percpu_up_write(&dup_mmap_sem);
1079 return err;
1080 }
1081
1082 static void
1083 __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
1084 {
1085 int err;
1086
1087 if (WARN_ON(!consumer_del(uprobe, uc)))
1088 return;
1089
1090 err = register_for_each_vma(uprobe, NULL);
1091 /* TODO : cant unregister? schedule a worker thread */
1092 if (!uprobe->consumers && !err)
1093 delete_uprobe(uprobe);
1094 }
1095
1096 /*
1097 * uprobe_unregister - unregister an already registered probe.
1098 * @inode: the file in which the probe has to be removed.
1099 * @offset: offset from the start of the file.
1100 * @uc: identify which probe if multiple probes are colocated.
1101 */
1102 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
1103 {
1104 struct uprobe *uprobe;
1105
1106 uprobe = find_uprobe(inode, offset);
1107 if (WARN_ON(!uprobe))
1108 return;
1109
1110 down_write(&uprobe->register_rwsem);
1111 __uprobe_unregister(uprobe, uc);
1112 up_write(&uprobe->register_rwsem);
1113 put_uprobe(uprobe);
1114 }
1115 EXPORT_SYMBOL_GPL(uprobe_unregister);
1116
1117 /*
1118 * __uprobe_register - register a probe
1119 * @inode: the file in which the probe has to be placed.
1120 * @offset: offset from the start of the file.
1121 * @uc: information on howto handle the probe..
1122 *
1123 * Apart from the access refcount, __uprobe_register() takes a creation
1124 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
1125 * inserted into the rbtree (i.e first consumer for a @inode:@offset
1126 * tuple). Creation refcount stops uprobe_unregister from freeing the
1127 * @uprobe even before the register operation is complete. Creation
1128 * refcount is released when the last @uc for the @uprobe
1129 * unregisters. Caller of __uprobe_register() is required to keep @inode
1130 * (and the containing mount) referenced.
1131 *
1132 * Return errno if it cannot successully install probes
1133 * else return 0 (success)
1134 */
1135 static int __uprobe_register(struct inode *inode, loff_t offset,
1136 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1137 {
1138 struct uprobe *uprobe;
1139 int ret;
1140
1141 /* Uprobe must have at least one set consumer */
1142 if (!uc->handler && !uc->ret_handler)
1143 return -EINVAL;
1144
1145 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
1146 if (!inode->i_mapping->a_ops->read_folio &&
1147 !shmem_mapping(inode->i_mapping))
1148 return -EIO;
1149 /* Racy, just to catch the obvious mistakes */
1150 if (offset > i_size_read(inode))
1151 return -EINVAL;
1152
1153 /*
1154 * This ensures that copy_from_page(), copy_to_page() and
1155 * __update_ref_ctr() can't cross page boundary.
1156 */
1157 if (!IS_ALIGNED(offset, UPROBE_SWBP_INSN_SIZE))
1158 return -EINVAL;
1159 if (!IS_ALIGNED(ref_ctr_offset, sizeof(short)))
1160 return -EINVAL;
1161
1162 retry:
1163 uprobe = alloc_uprobe(inode, offset, ref_ctr_offset);
1164 if (!uprobe)
1165 return -ENOMEM;
1166 if (IS_ERR(uprobe))
1167 return PTR_ERR(uprobe);
1168
1169 /*
1170 * We can race with uprobe_unregister()->delete_uprobe().
1171 * Check uprobe_is_active() and retry if it is false.
1172 */
1173 down_write(&uprobe->register_rwsem);
1174 ret = -EAGAIN;
1175 if (likely(uprobe_is_active(uprobe))) {
1176 consumer_add(uprobe, uc);
1177 ret = register_for_each_vma(uprobe, uc);
1178 if (ret)
1179 __uprobe_unregister(uprobe, uc);
1180 }
1181 up_write(&uprobe->register_rwsem);
1182 put_uprobe(uprobe);
1183
1184 if (unlikely(ret == -EAGAIN))
1185 goto retry;
1186 return ret;
1187 }
1188
1189 int uprobe_register(struct inode *inode, loff_t offset,
1190 struct uprobe_consumer *uc)
1191 {
1192 return __uprobe_register(inode, offset, 0, uc);
1193 }
1194 EXPORT_SYMBOL_GPL(uprobe_register);
1195
1196 int uprobe_register_refctr(struct inode *inode, loff_t offset,
1197 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1198 {
1199 return __uprobe_register(inode, offset, ref_ctr_offset, uc);
1200 }
1201 EXPORT_SYMBOL_GPL(uprobe_register_refctr);
1202
1203 /*
1204 * uprobe_apply - unregister an already registered probe.
1205 * @inode: the file in which the probe has to be removed.
1206 * @offset: offset from the start of the file.
1207 * @uc: consumer which wants to add more or remove some breakpoints
1208 * @add: add or remove the breakpoints
1209 */
1210 int uprobe_apply(struct inode *inode, loff_t offset,
1211 struct uprobe_consumer *uc, bool add)
1212 {
1213 struct uprobe *uprobe;
1214 struct uprobe_consumer *con;
1215 int ret = -ENOENT;
1216
1217 uprobe = find_uprobe(inode, offset);
1218 if (WARN_ON(!uprobe))
1219 return ret;
1220
1221 down_write(&uprobe->register_rwsem);
1222 for (con = uprobe->consumers; con && con != uc ; con = con->next)
1223 ;
1224 if (con)
1225 ret = register_for_each_vma(uprobe, add ? uc : NULL);
1226 up_write(&uprobe->register_rwsem);
1227 put_uprobe(uprobe);
1228
1229 return ret;
1230 }
1231
1232 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
1233 {
1234 VMA_ITERATOR(vmi, mm, 0);
1235 struct vm_area_struct *vma;
1236 int err = 0;
1237
1238 mmap_read_lock(mm);
1239 for_each_vma(vmi, vma) {
1240 unsigned long vaddr;
1241 loff_t offset;
1242
1243 if (!valid_vma(vma, false) ||
1244 file_inode(vma->vm_file) != uprobe->inode)
1245 continue;
1246
1247 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
1248 if (uprobe->offset < offset ||
1249 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
1250 continue;
1251
1252 vaddr = offset_to_vaddr(vma, uprobe->offset);
1253 err |= remove_breakpoint(uprobe, mm, vaddr);
1254 }
1255 mmap_read_unlock(mm);
1256
1257 return err;
1258 }
1259
1260 static struct rb_node *
1261 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1262 {
1263 struct rb_node *n = uprobes_tree.rb_node;
1264
1265 while (n) {
1266 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1267
1268 if (inode < u->inode) {
1269 n = n->rb_left;
1270 } else if (inode > u->inode) {
1271 n = n->rb_right;
1272 } else {
1273 if (max < u->offset)
1274 n = n->rb_left;
1275 else if (min > u->offset)
1276 n = n->rb_right;
1277 else
1278 break;
1279 }
1280 }
1281
1282 return n;
1283 }
1284
1285 /*
1286 * For a given range in vma, build a list of probes that need to be inserted.
1287 */
1288 static void build_probe_list(struct inode *inode,
1289 struct vm_area_struct *vma,
1290 unsigned long start, unsigned long end,
1291 struct list_head *head)
1292 {
1293 loff_t min, max;
1294 struct rb_node *n, *t;
1295 struct uprobe *u;
1296
1297 INIT_LIST_HEAD(head);
1298 min = vaddr_to_offset(vma, start);
1299 max = min + (end - start) - 1;
1300
1301 read_lock(&uprobes_treelock);
1302 n = find_node_in_range(inode, min, max);
1303 if (n) {
1304 for (t = n; t; t = rb_prev(t)) {
1305 u = rb_entry(t, struct uprobe, rb_node);
1306 if (u->inode != inode || u->offset < min)
1307 break;
1308 list_add(&u->pending_list, head);
1309 get_uprobe(u);
1310 }
1311 for (t = n; (t = rb_next(t)); ) {
1312 u = rb_entry(t, struct uprobe, rb_node);
1313 if (u->inode != inode || u->offset > max)
1314 break;
1315 list_add(&u->pending_list, head);
1316 get_uprobe(u);
1317 }
1318 }
1319 read_unlock(&uprobes_treelock);
1320 }
1321
1322 /* @vma contains reference counter, not the probed instruction. */
1323 static int delayed_ref_ctr_inc(struct vm_area_struct *vma)
1324 {
1325 struct list_head *pos, *q;
1326 struct delayed_uprobe *du;
1327 unsigned long vaddr;
1328 int ret = 0, err = 0;
1329
1330 mutex_lock(&delayed_uprobe_lock);
1331 list_for_each_safe(pos, q, &delayed_uprobe_list) {
1332 du = list_entry(pos, struct delayed_uprobe, list);
1333
1334 if (du->mm != vma->vm_mm ||
1335 !valid_ref_ctr_vma(du->uprobe, vma))
1336 continue;
1337
1338 vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
1339 ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
1340 if (ret) {
1341 update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1);
1342 if (!err)
1343 err = ret;
1344 }
1345 delayed_uprobe_delete(du);
1346 }
1347 mutex_unlock(&delayed_uprobe_lock);
1348 return err;
1349 }
1350
1351 /*
1352 * Called from mmap_region/vma_merge with mm->mmap_lock acquired.
1353 *
1354 * Currently we ignore all errors and always return 0, the callers
1355 * can't handle the failure anyway.
1356 */
1357 int uprobe_mmap(struct vm_area_struct *vma)
1358 {
1359 struct list_head tmp_list;
1360 struct uprobe *uprobe, *u;
1361 struct inode *inode;
1362
1363 if (no_uprobe_events())
1364 return 0;
1365
1366 if (vma->vm_file &&
1367 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
1368 test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
1369 delayed_ref_ctr_inc(vma);
1370
1371 if (!valid_vma(vma, true))
1372 return 0;
1373
1374 inode = file_inode(vma->vm_file);
1375 if (!inode)
1376 return 0;
1377
1378 mutex_lock(uprobes_mmap_hash(inode));
1379 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1380 /*
1381 * We can race with uprobe_unregister(), this uprobe can be already
1382 * removed. But in this case filter_chain() must return false, all
1383 * consumers have gone away.
1384 */
1385 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1386 if (!fatal_signal_pending(current) &&
1387 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1388 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1389 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1390 }
1391 put_uprobe(uprobe);
1392 }
1393 mutex_unlock(uprobes_mmap_hash(inode));
1394
1395 return 0;
1396 }
1397
1398 static bool
1399 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1400 {
1401 loff_t min, max;
1402 struct inode *inode;
1403 struct rb_node *n;
1404
1405 inode = file_inode(vma->vm_file);
1406
1407 min = vaddr_to_offset(vma, start);
1408 max = min + (end - start) - 1;
1409
1410 read_lock(&uprobes_treelock);
1411 n = find_node_in_range(inode, min, max);
1412 read_unlock(&uprobes_treelock);
1413
1414 return !!n;
1415 }
1416
1417 /*
1418 * Called in context of a munmap of a vma.
1419 */
1420 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1421 {
1422 if (no_uprobe_events() || !valid_vma(vma, false))
1423 return;
1424
1425 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1426 return;
1427
1428 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1429 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1430 return;
1431
1432 if (vma_has_uprobes(vma, start, end))
1433 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1434 }
1435
1436 /* Slot allocation for XOL */
1437 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1438 {
1439 struct vm_area_struct *vma;
1440 int ret;
1441
1442 if (mmap_write_lock_killable(mm))
1443 return -EINTR;
1444
1445 if (mm->uprobes_state.xol_area) {
1446 ret = -EALREADY;
1447 goto fail;
1448 }
1449
1450 if (!area->vaddr) {
1451 /* Try to map as high as possible, this is only a hint. */
1452 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1453 PAGE_SIZE, 0, 0);
1454 if (IS_ERR_VALUE(area->vaddr)) {
1455 ret = area->vaddr;
1456 goto fail;
1457 }
1458 }
1459
1460 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1461 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1462 &area->xol_mapping);
1463 if (IS_ERR(vma)) {
1464 ret = PTR_ERR(vma);
1465 goto fail;
1466 }
1467
1468 ret = 0;
1469 /* pairs with get_xol_area() */
1470 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1471 fail:
1472 mmap_write_unlock(mm);
1473
1474 return ret;
1475 }
1476
1477 void * __weak arch_uprobe_trampoline(unsigned long *psize)
1478 {
1479 static uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1480
1481 *psize = UPROBE_SWBP_INSN_SIZE;
1482 return &insn;
1483 }
1484
1485 static struct xol_area *__create_xol_area(unsigned long vaddr)
1486 {
1487 struct mm_struct *mm = current->mm;
1488 unsigned long insns_size;
1489 struct xol_area *area;
1490 void *insns;
1491
1492 area = kzalloc(sizeof(*area), GFP_KERNEL);
1493 if (unlikely(!area))
1494 goto out;
1495
1496 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1497 GFP_KERNEL);
1498 if (!area->bitmap)
1499 goto free_area;
1500
1501 area->xol_mapping.name = "[uprobes]";
1502 area->xol_mapping.pages = area->pages;
1503 area->pages[0] = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
1504 if (!area->pages[0])
1505 goto free_bitmap;
1506 area->pages[1] = NULL;
1507
1508 area->vaddr = vaddr;
1509 init_waitqueue_head(&area->wq);
1510 /* Reserve the 1st slot for get_trampoline_vaddr() */
1511 set_bit(0, area->bitmap);
1512 atomic_set(&area->slot_count, 1);
1513 insns = arch_uprobe_trampoline(&insns_size);
1514 arch_uprobe_copy_ixol(area->pages[0], 0, insns, insns_size);
1515
1516 if (!xol_add_vma(mm, area))
1517 return area;
1518
1519 __free_page(area->pages[0]);
1520 free_bitmap:
1521 kfree(area->bitmap);
1522 free_area:
1523 kfree(area);
1524 out:
1525 return NULL;
1526 }
1527
1528 /*
1529 * get_xol_area - Allocate process's xol_area if necessary.
1530 * This area will be used for storing instructions for execution out of line.
1531 *
1532 * Returns the allocated area or NULL.
1533 */
1534 static struct xol_area *get_xol_area(void)
1535 {
1536 struct mm_struct *mm = current->mm;
1537 struct xol_area *area;
1538
1539 if (!mm->uprobes_state.xol_area)
1540 __create_xol_area(0);
1541
1542 /* Pairs with xol_add_vma() smp_store_release() */
1543 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1544 return area;
1545 }
1546
1547 /*
1548 * uprobe_clear_state - Free the area allocated for slots.
1549 */
1550 void uprobe_clear_state(struct mm_struct *mm)
1551 {
1552 struct xol_area *area = mm->uprobes_state.xol_area;
1553
1554 mutex_lock(&delayed_uprobe_lock);
1555 delayed_uprobe_remove(NULL, mm);
1556 mutex_unlock(&delayed_uprobe_lock);
1557
1558 if (!area)
1559 return;
1560
1561 put_page(area->pages[0]);
1562 kfree(area->bitmap);
1563 kfree(area);
1564 }
1565
1566 void uprobe_start_dup_mmap(void)
1567 {
1568 percpu_down_read(&dup_mmap_sem);
1569 }
1570
1571 void uprobe_end_dup_mmap(void)
1572 {
1573 percpu_up_read(&dup_mmap_sem);
1574 }
1575
1576 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1577 {
1578 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1579 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1580 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1581 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1582 }
1583 }
1584
1585 /*
1586 * - search for a free slot.
1587 */
1588 static unsigned long xol_take_insn_slot(struct xol_area *area)
1589 {
1590 unsigned long slot_addr;
1591 int slot_nr;
1592
1593 do {
1594 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1595 if (slot_nr < UINSNS_PER_PAGE) {
1596 if (!test_and_set_bit(slot_nr, area->bitmap))
1597 break;
1598
1599 slot_nr = UINSNS_PER_PAGE;
1600 continue;
1601 }
1602 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1603 } while (slot_nr >= UINSNS_PER_PAGE);
1604
1605 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1606 atomic_inc(&area->slot_count);
1607
1608 return slot_addr;
1609 }
1610
1611 /*
1612 * xol_get_insn_slot - allocate a slot for xol.
1613 * Returns the allocated slot address or 0.
1614 */
1615 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1616 {
1617 struct xol_area *area;
1618 unsigned long xol_vaddr;
1619
1620 area = get_xol_area();
1621 if (!area)
1622 return 0;
1623
1624 xol_vaddr = xol_take_insn_slot(area);
1625 if (unlikely(!xol_vaddr))
1626 return 0;
1627
1628 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1629 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1630
1631 return xol_vaddr;
1632 }
1633
1634 /*
1635 * xol_free_insn_slot - If slot was earlier allocated by
1636 * @xol_get_insn_slot(), make the slot available for
1637 * subsequent requests.
1638 */
1639 static void xol_free_insn_slot(struct task_struct *tsk)
1640 {
1641 struct xol_area *area;
1642 unsigned long vma_end;
1643 unsigned long slot_addr;
1644
1645 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1646 return;
1647
1648 slot_addr = tsk->utask->xol_vaddr;
1649 if (unlikely(!slot_addr))
1650 return;
1651
1652 area = tsk->mm->uprobes_state.xol_area;
1653 vma_end = area->vaddr + PAGE_SIZE;
1654 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1655 unsigned long offset;
1656 int slot_nr;
1657
1658 offset = slot_addr - area->vaddr;
1659 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1660 if (slot_nr >= UINSNS_PER_PAGE)
1661 return;
1662
1663 clear_bit(slot_nr, area->bitmap);
1664 atomic_dec(&area->slot_count);
1665 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1666 if (waitqueue_active(&area->wq))
1667 wake_up(&area->wq);
1668
1669 tsk->utask->xol_vaddr = 0;
1670 }
1671 }
1672
1673 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1674 void *src, unsigned long len)
1675 {
1676 /* Initialize the slot */
1677 copy_to_page(page, vaddr, src, len);
1678
1679 /*
1680 * We probably need flush_icache_user_page() but it needs vma.
1681 * This should work on most of architectures by default. If
1682 * architecture needs to do something different it can define
1683 * its own version of the function.
1684 */
1685 flush_dcache_page(page);
1686 }
1687
1688 /**
1689 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1690 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1691 * instruction.
1692 * Return the address of the breakpoint instruction.
1693 */
1694 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1695 {
1696 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1697 }
1698
1699 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1700 {
1701 struct uprobe_task *utask = current->utask;
1702
1703 if (unlikely(utask && utask->active_uprobe))
1704 return utask->vaddr;
1705
1706 return instruction_pointer(regs);
1707 }
1708
1709 static struct return_instance *free_ret_instance(struct return_instance *ri)
1710 {
1711 struct return_instance *next = ri->next;
1712 put_uprobe(ri->uprobe);
1713 kfree(ri);
1714 return next;
1715 }
1716
1717 /*
1718 * Called with no locks held.
1719 * Called in context of an exiting or an exec-ing thread.
1720 */
1721 void uprobe_free_utask(struct task_struct *t)
1722 {
1723 struct uprobe_task *utask = t->utask;
1724 struct return_instance *ri;
1725
1726 if (!utask)
1727 return;
1728
1729 if (utask->active_uprobe)
1730 put_uprobe(utask->active_uprobe);
1731
1732 ri = utask->return_instances;
1733 while (ri)
1734 ri = free_ret_instance(ri);
1735
1736 xol_free_insn_slot(t);
1737 kfree(utask);
1738 t->utask = NULL;
1739 }
1740
1741 /*
1742 * Allocate a uprobe_task object for the task if necessary.
1743 * Called when the thread hits a breakpoint.
1744 *
1745 * Returns:
1746 * - pointer to new uprobe_task on success
1747 * - NULL otherwise
1748 */
1749 static struct uprobe_task *get_utask(void)
1750 {
1751 if (!current->utask)
1752 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1753 return current->utask;
1754 }
1755
1756 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1757 {
1758 struct uprobe_task *n_utask;
1759 struct return_instance **p, *o, *n;
1760
1761 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1762 if (!n_utask)
1763 return -ENOMEM;
1764 t->utask = n_utask;
1765
1766 p = &n_utask->return_instances;
1767 for (o = o_utask->return_instances; o; o = o->next) {
1768 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1769 if (!n)
1770 return -ENOMEM;
1771
1772 *n = *o;
1773 get_uprobe(n->uprobe);
1774 n->next = NULL;
1775
1776 *p = n;
1777 p = &n->next;
1778 n_utask->depth++;
1779 }
1780
1781 return 0;
1782 }
1783
1784 static void uprobe_warn(struct task_struct *t, const char *msg)
1785 {
1786 pr_warn("uprobe: %s:%d failed to %s\n",
1787 current->comm, current->pid, msg);
1788 }
1789
1790 static void dup_xol_work(struct callback_head *work)
1791 {
1792 if (current->flags & PF_EXITING)
1793 return;
1794
1795 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1796 !fatal_signal_pending(current))
1797 uprobe_warn(current, "dup xol area");
1798 }
1799
1800 /*
1801 * Called in context of a new clone/fork from copy_process.
1802 */
1803 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1804 {
1805 struct uprobe_task *utask = current->utask;
1806 struct mm_struct *mm = current->mm;
1807 struct xol_area *area;
1808
1809 t->utask = NULL;
1810
1811 if (!utask || !utask->return_instances)
1812 return;
1813
1814 if (mm == t->mm && !(flags & CLONE_VFORK))
1815 return;
1816
1817 if (dup_utask(t, utask))
1818 return uprobe_warn(t, "dup ret instances");
1819
1820 /* The task can fork() after dup_xol_work() fails */
1821 area = mm->uprobes_state.xol_area;
1822 if (!area)
1823 return uprobe_warn(t, "dup xol area");
1824
1825 if (mm == t->mm)
1826 return;
1827
1828 t->utask->dup_xol_addr = area->vaddr;
1829 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1830 task_work_add(t, &t->utask->dup_xol_work, TWA_RESUME);
1831 }
1832
1833 /*
1834 * Current area->vaddr notion assume the trampoline address is always
1835 * equal area->vaddr.
1836 *
1837 * Returns -1 in case the xol_area is not allocated.
1838 */
1839 unsigned long uprobe_get_trampoline_vaddr(void)
1840 {
1841 struct xol_area *area;
1842 unsigned long trampoline_vaddr = -1;
1843
1844 /* Pairs with xol_add_vma() smp_store_release() */
1845 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1846 if (area)
1847 trampoline_vaddr = area->vaddr;
1848
1849 return trampoline_vaddr;
1850 }
1851
1852 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1853 struct pt_regs *regs)
1854 {
1855 struct return_instance *ri = utask->return_instances;
1856 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1857
1858 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1859 ri = free_ret_instance(ri);
1860 utask->depth--;
1861 }
1862 utask->return_instances = ri;
1863 }
1864
1865 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1866 {
1867 struct return_instance *ri;
1868 struct uprobe_task *utask;
1869 unsigned long orig_ret_vaddr, trampoline_vaddr;
1870 bool chained;
1871
1872 if (!get_xol_area())
1873 return;
1874
1875 utask = get_utask();
1876 if (!utask)
1877 return;
1878
1879 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1880 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1881 " nestedness limit pid/tgid=%d/%d\n",
1882 current->pid, current->tgid);
1883 return;
1884 }
1885
1886 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1887 if (!ri)
1888 return;
1889
1890 trampoline_vaddr = uprobe_get_trampoline_vaddr();
1891 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1892 if (orig_ret_vaddr == -1)
1893 goto fail;
1894
1895 /* drop the entries invalidated by longjmp() */
1896 chained = (orig_ret_vaddr == trampoline_vaddr);
1897 cleanup_return_instances(utask, chained, regs);
1898
1899 /*
1900 * We don't want to keep trampoline address in stack, rather keep the
1901 * original return address of first caller thru all the consequent
1902 * instances. This also makes breakpoint unwrapping easier.
1903 */
1904 if (chained) {
1905 if (!utask->return_instances) {
1906 /*
1907 * This situation is not possible. Likely we have an
1908 * attack from user-space.
1909 */
1910 uprobe_warn(current, "handle tail call");
1911 goto fail;
1912 }
1913 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1914 }
1915
1916 ri->uprobe = get_uprobe(uprobe);
1917 ri->func = instruction_pointer(regs);
1918 ri->stack = user_stack_pointer(regs);
1919 ri->orig_ret_vaddr = orig_ret_vaddr;
1920 ri->chained = chained;
1921
1922 utask->depth++;
1923 ri->next = utask->return_instances;
1924 utask->return_instances = ri;
1925
1926 return;
1927 fail:
1928 kfree(ri);
1929 }
1930
1931 /* Prepare to single-step probed instruction out of line. */
1932 static int
1933 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1934 {
1935 struct uprobe_task *utask;
1936 unsigned long xol_vaddr;
1937 int err;
1938
1939 utask = get_utask();
1940 if (!utask)
1941 return -ENOMEM;
1942
1943 xol_vaddr = xol_get_insn_slot(uprobe);
1944 if (!xol_vaddr)
1945 return -ENOMEM;
1946
1947 utask->xol_vaddr = xol_vaddr;
1948 utask->vaddr = bp_vaddr;
1949
1950 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1951 if (unlikely(err)) {
1952 xol_free_insn_slot(current);
1953 return err;
1954 }
1955
1956 utask->active_uprobe = uprobe;
1957 utask->state = UTASK_SSTEP;
1958 return 0;
1959 }
1960
1961 /*
1962 * If we are singlestepping, then ensure this thread is not connected to
1963 * non-fatal signals until completion of singlestep. When xol insn itself
1964 * triggers the signal, restart the original insn even if the task is
1965 * already SIGKILL'ed (since coredump should report the correct ip). This
1966 * is even more important if the task has a handler for SIGSEGV/etc, The
1967 * _same_ instruction should be repeated again after return from the signal
1968 * handler, and SSTEP can never finish in this case.
1969 */
1970 bool uprobe_deny_signal(void)
1971 {
1972 struct task_struct *t = current;
1973 struct uprobe_task *utask = t->utask;
1974
1975 if (likely(!utask || !utask->active_uprobe))
1976 return false;
1977
1978 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1979
1980 if (task_sigpending(t)) {
1981 spin_lock_irq(&t->sighand->siglock);
1982 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1983 spin_unlock_irq(&t->sighand->siglock);
1984
1985 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1986 utask->state = UTASK_SSTEP_TRAPPED;
1987 set_tsk_thread_flag(t, TIF_UPROBE);
1988 }
1989 }
1990
1991 return true;
1992 }
1993
1994 static void mmf_recalc_uprobes(struct mm_struct *mm)
1995 {
1996 VMA_ITERATOR(vmi, mm, 0);
1997 struct vm_area_struct *vma;
1998
1999 for_each_vma(vmi, vma) {
2000 if (!valid_vma(vma, false))
2001 continue;
2002 /*
2003 * This is not strictly accurate, we can race with
2004 * uprobe_unregister() and see the already removed
2005 * uprobe if delete_uprobe() was not yet called.
2006 * Or this uprobe can be filtered out.
2007 */
2008 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
2009 return;
2010 }
2011
2012 clear_bit(MMF_HAS_UPROBES, &mm->flags);
2013 }
2014
2015 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
2016 {
2017 struct page *page;
2018 uprobe_opcode_t opcode;
2019 int result;
2020
2021 if (WARN_ON_ONCE(!IS_ALIGNED(vaddr, UPROBE_SWBP_INSN_SIZE)))
2022 return -EINVAL;
2023
2024 pagefault_disable();
2025 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
2026 pagefault_enable();
2027
2028 if (likely(result == 0))
2029 goto out;
2030
2031 /*
2032 * The NULL 'tsk' here ensures that any faults that occur here
2033 * will not be accounted to the task. 'mm' *is* current->mm,
2034 * but we treat this as a 'remote' access since it is
2035 * essentially a kernel access to the memory.
2036 */
2037 result = get_user_pages_remote(mm, vaddr, 1, FOLL_FORCE, &page, NULL);
2038 if (result < 0)
2039 return result;
2040
2041 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
2042 put_page(page);
2043 out:
2044 /* This needs to return true for any variant of the trap insn */
2045 return is_trap_insn(&opcode);
2046 }
2047
2048 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
2049 {
2050 struct mm_struct *mm = current->mm;
2051 struct uprobe *uprobe = NULL;
2052 struct vm_area_struct *vma;
2053
2054 mmap_read_lock(mm);
2055 vma = vma_lookup(mm, bp_vaddr);
2056 if (vma) {
2057 if (valid_vma(vma, false)) {
2058 struct inode *inode = file_inode(vma->vm_file);
2059 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
2060
2061 uprobe = find_uprobe(inode, offset);
2062 }
2063
2064 if (!uprobe)
2065 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
2066 } else {
2067 *is_swbp = -EFAULT;
2068 }
2069
2070 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
2071 mmf_recalc_uprobes(mm);
2072 mmap_read_unlock(mm);
2073
2074 return uprobe;
2075 }
2076
2077 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
2078 {
2079 struct uprobe_consumer *uc;
2080 int remove = UPROBE_HANDLER_REMOVE;
2081 bool need_prep = false; /* prepare return uprobe, when needed */
2082
2083 down_read(&uprobe->register_rwsem);
2084 current->utask->auprobe = &uprobe->arch;
2085 for (uc = uprobe->consumers; uc; uc = uc->next) {
2086 int rc = 0;
2087
2088 if (uc->handler) {
2089 rc = uc->handler(uc, regs);
2090 WARN(rc & ~UPROBE_HANDLER_MASK,
2091 "bad rc=0x%x from %ps()\n", rc, uc->handler);
2092 }
2093
2094 if (uc->ret_handler)
2095 need_prep = true;
2096
2097 remove &= rc;
2098 }
2099 current->utask->auprobe = NULL;
2100
2101 if (need_prep && !remove)
2102 prepare_uretprobe(uprobe, regs); /* put bp at return */
2103
2104 if (remove && uprobe->consumers) {
2105 WARN_ON(!uprobe_is_active(uprobe));
2106 unapply_uprobe(uprobe, current->mm);
2107 }
2108 up_read(&uprobe->register_rwsem);
2109 }
2110
2111 static void
2112 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
2113 {
2114 struct uprobe *uprobe = ri->uprobe;
2115 struct uprobe_consumer *uc;
2116
2117 down_read(&uprobe->register_rwsem);
2118 for (uc = uprobe->consumers; uc; uc = uc->next) {
2119 if (uc->ret_handler)
2120 uc->ret_handler(uc, ri->func, regs);
2121 }
2122 up_read(&uprobe->register_rwsem);
2123 }
2124
2125 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
2126 {
2127 bool chained;
2128
2129 do {
2130 chained = ri->chained;
2131 ri = ri->next; /* can't be NULL if chained */
2132 } while (chained);
2133
2134 return ri;
2135 }
2136
2137 void uprobe_handle_trampoline(struct pt_regs *regs)
2138 {
2139 struct uprobe_task *utask;
2140 struct return_instance *ri, *next;
2141 bool valid;
2142
2143 utask = current->utask;
2144 if (!utask)
2145 goto sigill;
2146
2147 ri = utask->return_instances;
2148 if (!ri)
2149 goto sigill;
2150
2151 do {
2152 /*
2153 * We should throw out the frames invalidated by longjmp().
2154 * If this chain is valid, then the next one should be alive
2155 * or NULL; the latter case means that nobody but ri->func
2156 * could hit this trampoline on return. TODO: sigaltstack().
2157 */
2158 next = find_next_ret_chain(ri);
2159 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
2160
2161 instruction_pointer_set(regs, ri->orig_ret_vaddr);
2162 do {
2163 /* pop current instance from the stack of pending return instances,
2164 * as it's not pending anymore: we just fixed up original
2165 * instruction pointer in regs and are about to call handlers;
2166 * this allows fixup_uretprobe_trampoline_entries() to properly fix up
2167 * captured stack traces from uretprobe handlers, in which pending
2168 * trampoline addresses on the stack are replaced with correct
2169 * original return addresses
2170 */
2171 utask->return_instances = ri->next;
2172 if (valid)
2173 handle_uretprobe_chain(ri, regs);
2174 ri = free_ret_instance(ri);
2175 utask->depth--;
2176 } while (ri != next);
2177 } while (!valid);
2178
2179 utask->return_instances = ri;
2180 return;
2181
2182 sigill:
2183 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
2184 force_sig(SIGILL);
2185
2186 }
2187
2188 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
2189 {
2190 return false;
2191 }
2192
2193 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
2194 struct pt_regs *regs)
2195 {
2196 return true;
2197 }
2198
2199 /*
2200 * Run handler and ask thread to singlestep.
2201 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2202 */
2203 static void handle_swbp(struct pt_regs *regs)
2204 {
2205 struct uprobe *uprobe;
2206 unsigned long bp_vaddr;
2207 int is_swbp;
2208
2209 bp_vaddr = uprobe_get_swbp_addr(regs);
2210 if (bp_vaddr == uprobe_get_trampoline_vaddr())
2211 return uprobe_handle_trampoline(regs);
2212
2213 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
2214 if (!uprobe) {
2215 if (is_swbp > 0) {
2216 /* No matching uprobe; signal SIGTRAP. */
2217 force_sig(SIGTRAP);
2218 } else {
2219 /*
2220 * Either we raced with uprobe_unregister() or we can't
2221 * access this memory. The latter is only possible if
2222 * another thread plays with our ->mm. In both cases
2223 * we can simply restart. If this vma was unmapped we
2224 * can pretend this insn was not executed yet and get
2225 * the (correct) SIGSEGV after restart.
2226 */
2227 instruction_pointer_set(regs, bp_vaddr);
2228 }
2229 return;
2230 }
2231
2232 /* change it in advance for ->handler() and restart */
2233 instruction_pointer_set(regs, bp_vaddr);
2234
2235 /*
2236 * TODO: move copy_insn/etc into _register and remove this hack.
2237 * After we hit the bp, _unregister + _register can install the
2238 * new and not-yet-analyzed uprobe at the same address, restart.
2239 */
2240 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
2241 goto out;
2242
2243 /*
2244 * Pairs with the smp_wmb() in prepare_uprobe().
2245 *
2246 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2247 * we must also see the stores to &uprobe->arch performed by the
2248 * prepare_uprobe() call.
2249 */
2250 smp_rmb();
2251
2252 /* Tracing handlers use ->utask to communicate with fetch methods */
2253 if (!get_utask())
2254 goto out;
2255
2256 if (arch_uprobe_ignore(&uprobe->arch, regs))
2257 goto out;
2258
2259 handler_chain(uprobe, regs);
2260
2261 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
2262 goto out;
2263
2264 if (!pre_ssout(uprobe, regs, bp_vaddr))
2265 return;
2266
2267 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2268 out:
2269 put_uprobe(uprobe);
2270 }
2271
2272 /*
2273 * Perform required fix-ups and disable singlestep.
2274 * Allow pending signals to take effect.
2275 */
2276 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
2277 {
2278 struct uprobe *uprobe;
2279 int err = 0;
2280
2281 uprobe = utask->active_uprobe;
2282 if (utask->state == UTASK_SSTEP_ACK)
2283 err = arch_uprobe_post_xol(&uprobe->arch, regs);
2284 else if (utask->state == UTASK_SSTEP_TRAPPED)
2285 arch_uprobe_abort_xol(&uprobe->arch, regs);
2286 else
2287 WARN_ON_ONCE(1);
2288
2289 put_uprobe(uprobe);
2290 utask->active_uprobe = NULL;
2291 utask->state = UTASK_RUNNING;
2292 xol_free_insn_slot(current);
2293
2294 spin_lock_irq(¤t->sighand->siglock);
2295 recalc_sigpending(); /* see uprobe_deny_signal() */
2296 spin_unlock_irq(¤t->sighand->siglock);
2297
2298 if (unlikely(err)) {
2299 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
2300 force_sig(SIGILL);
2301 }
2302 }
2303
2304 /*
2305 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2306 * allows the thread to return from interrupt. After that handle_swbp()
2307 * sets utask->active_uprobe.
2308 *
2309 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2310 * and allows the thread to return from interrupt.
2311 *
2312 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2313 * uprobe_notify_resume().
2314 */
2315 void uprobe_notify_resume(struct pt_regs *regs)
2316 {
2317 struct uprobe_task *utask;
2318
2319 clear_thread_flag(TIF_UPROBE);
2320
2321 utask = current->utask;
2322 if (utask && utask->active_uprobe)
2323 handle_singlestep(utask, regs);
2324 else
2325 handle_swbp(regs);
2326 }
2327
2328 /*
2329 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2330 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2331 */
2332 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2333 {
2334 if (!current->mm)
2335 return 0;
2336
2337 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2338 (!current->utask || !current->utask->return_instances))
2339 return 0;
2340
2341 set_thread_flag(TIF_UPROBE);
2342 return 1;
2343 }
2344
2345 /*
2346 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2347 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2348 */
2349 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2350 {
2351 struct uprobe_task *utask = current->utask;
2352
2353 if (!current->mm || !utask || !utask->active_uprobe)
2354 /* task is currently not uprobed */
2355 return 0;
2356
2357 utask->state = UTASK_SSTEP_ACK;
2358 set_thread_flag(TIF_UPROBE);
2359 return 1;
2360 }
2361
2362 static struct notifier_block uprobe_exception_nb = {
2363 .notifier_call = arch_uprobe_exception_notify,
2364 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2365 };
2366
2367 void __init uprobes_init(void)
2368 {
2369 int i;
2370
2371 for (i = 0; i < UPROBES_HASH_SZ; i++)
2372 mutex_init(&uprobes_mmap_mutex[i]);
2373
2374 BUG_ON(register_die_notifier(&uprobe_exception_nb));
2375 }
2376
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.