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> /* set_pte_at_notify */
22#include <linux/swap.h> /* try_to_free_swap */
23#include <linux/ptrace.h> /* user_enable_single_step */
24#include <linux/kdebug.h> /* notifier mechanism */
25#include "../../mm/internal.h" /* munlock_vma_page */
26#include <linux/percpu-rwsem.h>
27#include <linux/task_work.h>
28#include <linux/shmem_fs.h>
29#include <linux/khugepaged.h>
30
31#include <linux/uprobes.h>
32
33#define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
34#define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
35
36static struct rb_root uprobes_tree = RB_ROOT;
37/*
38 * allows us to skip the uprobe_mmap if there are no uprobe events active
39 * at this time. Probably a fine grained per inode count is better?
40 */
41#define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
42
43static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
44
45#define UPROBES_HASH_SZ 13
46/* serialize uprobe->pending_list */
47static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
48#define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
49
50DEFINE_STATIC_PERCPU_RWSEM(dup_mmap_sem);
51
52/* Have a copy of original instruction */
53#define UPROBE_COPY_INSN 0
54
55struct uprobe {
56 struct rb_node rb_node; /* node in the rb tree */
57 refcount_t ref;
58 struct rw_semaphore register_rwsem;
59 struct rw_semaphore consumer_rwsem;
60 struct list_head pending_list;
61 struct uprobe_consumer *consumers;
62 struct inode *inode; /* Also hold a ref to inode */
63 loff_t offset;
64 loff_t ref_ctr_offset;
65 unsigned long flags;
66
67 /*
68 * The generic code assumes that it has two members of unknown type
69 * owned by the arch-specific code:
70 *
71 * insn - copy_insn() saves the original instruction here for
72 * arch_uprobe_analyze_insn().
73 *
74 * ixol - potentially modified instruction to execute out of
75 * line, copied to xol_area by xol_get_insn_slot().
76 */
77 struct arch_uprobe arch;
78};
79
80struct delayed_uprobe {
81 struct list_head list;
82 struct uprobe *uprobe;
83 struct mm_struct *mm;
84};
85
86static DEFINE_MUTEX(delayed_uprobe_lock);
87static LIST_HEAD(delayed_uprobe_list);
88
89/*
90 * Execute out of line area: anonymous executable mapping installed
91 * by the probed task to execute the copy of the original instruction
92 * mangled by set_swbp().
93 *
94 * On a breakpoint hit, thread contests for a slot. It frees the
95 * slot after singlestep. Currently a fixed number of slots are
96 * allocated.
97 */
98struct xol_area {
99 wait_queue_head_t wq; /* if all slots are busy */
100 atomic_t slot_count; /* number of in-use slots */
101 unsigned long *bitmap; /* 0 = free slot */
102
103 struct vm_special_mapping xol_mapping;
104 struct page *pages[2];
105 /*
106 * We keep the vma's vm_start rather than a pointer to the vma
107 * itself. The probed process or a naughty kernel module could make
108 * the vma go away, and we must handle that reasonably gracefully.
109 */
110 unsigned long vaddr; /* Page(s) of instruction slots */
111};
112
113/*
114 * valid_vma: Verify if the specified vma is an executable vma
115 * Relax restrictions while unregistering: vm_flags might have
116 * changed after breakpoint was inserted.
117 * - is_register: indicates if we are in register context.
118 * - Return 1 if the specified virtual address is in an
119 * executable vma.
120 */
121static bool valid_vma(struct vm_area_struct *vma, bool is_register)
122{
123 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
124
125 if (is_register)
126 flags |= VM_WRITE;
127
128 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
129}
130
131static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
132{
133 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
134}
135
136static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
137{
138 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
139}
140
141/**
142 * __replace_page - replace page in vma by new page.
143 * based on replace_page in mm/ksm.c
144 *
145 * @vma: vma that holds the pte pointing to page
146 * @addr: address the old @page is mapped at
147 * @old_page: the page we are replacing by new_page
148 * @new_page: the modified page we replace page by
149 *
150 * If @new_page is NULL, only unmap @old_page.
151 *
152 * Returns 0 on success, negative error code otherwise.
153 */
154static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
155 struct page *old_page, struct page *new_page)
156{
157 struct mm_struct *mm = vma->vm_mm;
158 DEFINE_FOLIO_VMA_WALK(pvmw, page_folio(old_page), vma, addr, 0);
159 int err;
160 struct mmu_notifier_range range;
161
162 mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, addr,
163 addr + PAGE_SIZE);
164
165 if (new_page) {
166 err = mem_cgroup_charge(page_folio(new_page), vma->vm_mm,
167 GFP_KERNEL);
168 if (err)
169 return err;
170 }
171
172 /* For try_to_free_swap() below */
173 lock_page(old_page);
174
175 mmu_notifier_invalidate_range_start(&range);
176 err = -EAGAIN;
177 if (!page_vma_mapped_walk(&pvmw))
178 goto unlock;
179 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
180
181 if (new_page) {
182 get_page(new_page);
183 page_add_new_anon_rmap(new_page, vma, addr);
184 lru_cache_add_inactive_or_unevictable(new_page, vma);
185 } else
186 /* no new page, just dec_mm_counter for old_page */
187 dec_mm_counter(mm, MM_ANONPAGES);
188
189 if (!PageAnon(old_page)) {
190 dec_mm_counter(mm, mm_counter_file(old_page));
191 inc_mm_counter(mm, MM_ANONPAGES);
192 }
193
194 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
195 ptep_clear_flush_notify(vma, addr, pvmw.pte);
196 if (new_page)
197 set_pte_at_notify(mm, addr, pvmw.pte,
198 mk_pte(new_page, vma->vm_page_prot));
199
200 page_remove_rmap(old_page, vma, false);
201 if (!page_mapped(old_page))
202 try_to_free_swap(old_page);
203 page_vma_mapped_walk_done(&pvmw);
204 put_page(old_page);
205
206 err = 0;
207 unlock:
208 mmu_notifier_invalidate_range_end(&range);
209 unlock_page(old_page);
210 return err;
211}
212
213/**
214 * is_swbp_insn - check if instruction is breakpoint instruction.
215 * @insn: instruction to be checked.
216 * Default implementation of is_swbp_insn
217 * Returns true if @insn is a breakpoint instruction.
218 */
219bool __weak is_swbp_insn(uprobe_opcode_t *insn)
220{
221 return *insn == UPROBE_SWBP_INSN;
222}
223
224/**
225 * is_trap_insn - check if instruction is breakpoint instruction.
226 * @insn: instruction to be checked.
227 * Default implementation of is_trap_insn
228 * Returns true if @insn is a breakpoint instruction.
229 *
230 * This function is needed for the case where an architecture has multiple
231 * trap instructions (like powerpc).
232 */
233bool __weak is_trap_insn(uprobe_opcode_t *insn)
234{
235 return is_swbp_insn(insn);
236}
237
238static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
239{
240 void *kaddr = kmap_atomic(page);
241 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
242 kunmap_atomic(kaddr);
243}
244
245static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
246{
247 void *kaddr = kmap_atomic(page);
248 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
249 kunmap_atomic(kaddr);
250}
251
252static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
253{
254 uprobe_opcode_t old_opcode;
255 bool is_swbp;
256
257 /*
258 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
259 * We do not check if it is any other 'trap variant' which could
260 * be conditional trap instruction such as the one powerpc supports.
261 *
262 * The logic is that we do not care if the underlying instruction
263 * is a trap variant; uprobes always wins over any other (gdb)
264 * breakpoint.
265 */
266 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
267 is_swbp = is_swbp_insn(&old_opcode);
268
269 if (is_swbp_insn(new_opcode)) {
270 if (is_swbp) /* register: already installed? */
271 return 0;
272 } else {
273 if (!is_swbp) /* unregister: was it changed by us? */
274 return 0;
275 }
276
277 return 1;
278}
279
280static struct delayed_uprobe *
281delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm)
282{
283 struct delayed_uprobe *du;
284
285 list_for_each_entry(du, &delayed_uprobe_list, list)
286 if (du->uprobe == uprobe && du->mm == mm)
287 return du;
288 return NULL;
289}
290
291static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
292{
293 struct delayed_uprobe *du;
294
295 if (delayed_uprobe_check(uprobe, mm))
296 return 0;
297
298 du = kzalloc(sizeof(*du), GFP_KERNEL);
299 if (!du)
300 return -ENOMEM;
301
302 du->uprobe = uprobe;
303 du->mm = mm;
304 list_add(&du->list, &delayed_uprobe_list);
305 return 0;
306}
307
308static void delayed_uprobe_delete(struct delayed_uprobe *du)
309{
310 if (WARN_ON(!du))
311 return;
312 list_del(&du->list);
313 kfree(du);
314}
315
316static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm)
317{
318 struct list_head *pos, *q;
319 struct delayed_uprobe *du;
320
321 if (!uprobe && !mm)
322 return;
323
324 list_for_each_safe(pos, q, &delayed_uprobe_list) {
325 du = list_entry(pos, struct delayed_uprobe, list);
326
327 if (uprobe && du->uprobe != uprobe)
328 continue;
329 if (mm && du->mm != mm)
330 continue;
331
332 delayed_uprobe_delete(du);
333 }
334}
335
336static bool valid_ref_ctr_vma(struct uprobe *uprobe,
337 struct vm_area_struct *vma)
338{
339 unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
340
341 return uprobe->ref_ctr_offset &&
342 vma->vm_file &&
343 file_inode(vma->vm_file) == uprobe->inode &&
344 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
345 vma->vm_start <= vaddr &&
346 vma->vm_end > vaddr;
347}
348
349static struct vm_area_struct *
350find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm)
351{
352 struct vm_area_struct *tmp;
353
354 for (tmp = mm->mmap; tmp; tmp = tmp->vm_next)
355 if (valid_ref_ctr_vma(uprobe, tmp))
356 return tmp;
357
358 return NULL;
359}
360
361static int
362__update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d)
363{
364 void *kaddr;
365 struct page *page;
366 struct vm_area_struct *vma;
367 int ret;
368 short *ptr;
369
370 if (!vaddr || !d)
371 return -EINVAL;
372
373 ret = get_user_pages_remote(mm, vaddr, 1,
374 FOLL_WRITE, &page, &vma, NULL);
375 if (unlikely(ret <= 0)) {
376 /*
377 * We are asking for 1 page. If get_user_pages_remote() fails,
378 * it may return 0, in that case we have to return error.
379 */
380 return ret == 0 ? -EBUSY : ret;
381 }
382
383 kaddr = kmap_atomic(page);
384 ptr = kaddr + (vaddr & ~PAGE_MASK);
385
386 if (unlikely(*ptr + d < 0)) {
387 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
388 "curr val: %d, delta: %d\n", vaddr, *ptr, d);
389 ret = -EINVAL;
390 goto out;
391 }
392
393 *ptr += d;
394 ret = 0;
395out:
396 kunmap_atomic(kaddr);
397 put_page(page);
398 return ret;
399}
400
401static void update_ref_ctr_warn(struct uprobe *uprobe,
402 struct mm_struct *mm, short d)
403{
404 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
405 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
406 d > 0 ? "increment" : "decrement", uprobe->inode->i_ino,
407 (unsigned long long) uprobe->offset,
408 (unsigned long long) uprobe->ref_ctr_offset, mm);
409}
410
411static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
412 short d)
413{
414 struct vm_area_struct *rc_vma;
415 unsigned long rc_vaddr;
416 int ret = 0;
417
418 rc_vma = find_ref_ctr_vma(uprobe, mm);
419
420 if (rc_vma) {
421 rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset);
422 ret = __update_ref_ctr(mm, rc_vaddr, d);
423 if (ret)
424 update_ref_ctr_warn(uprobe, mm, d);
425
426 if (d > 0)
427 return ret;
428 }
429
430 mutex_lock(&delayed_uprobe_lock);
431 if (d > 0)
432 ret = delayed_uprobe_add(uprobe, mm);
433 else
434 delayed_uprobe_remove(uprobe, mm);
435 mutex_unlock(&delayed_uprobe_lock);
436
437 return ret;
438}
439
440/*
441 * NOTE:
442 * Expect the breakpoint instruction to be the smallest size instruction for
443 * the architecture. If an arch has variable length instruction and the
444 * breakpoint instruction is not of the smallest length instruction
445 * supported by that architecture then we need to modify is_trap_at_addr and
446 * uprobe_write_opcode accordingly. This would never be a problem for archs
447 * that have fixed length instructions.
448 *
449 * uprobe_write_opcode - write the opcode at a given virtual address.
450 * @auprobe: arch specific probepoint information.
451 * @mm: the probed process address space.
452 * @vaddr: the virtual address to store the opcode.
453 * @opcode: opcode to be written at @vaddr.
454 *
455 * Called with mm->mmap_lock held for write.
456 * Return 0 (success) or a negative errno.
457 */
458int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
459 unsigned long vaddr, uprobe_opcode_t opcode)
460{
461 struct uprobe *uprobe;
462 struct page *old_page, *new_page;
463 struct vm_area_struct *vma;
464 int ret, is_register, ref_ctr_updated = 0;
465 bool orig_page_huge = false;
466 unsigned int gup_flags = FOLL_FORCE;
467
468 is_register = is_swbp_insn(&opcode);
469 uprobe = container_of(auprobe, struct uprobe, arch);
470
471retry:
472 if (is_register)
473 gup_flags |= FOLL_SPLIT_PMD;
474 /* Read the page with vaddr into memory */
475 ret = get_user_pages_remote(mm, vaddr, 1, gup_flags,
476 &old_page, &vma, NULL);
477 if (ret <= 0)
478 return ret;
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, old_page, new_page);
541 if (new_page)
542 put_page(new_page);
543put_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);
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 */
569int __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 */
583int __weak
584set_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
590static struct uprobe *get_uprobe(struct uprobe *uprobe)
591{
592 refcount_inc(&uprobe->ref);
593 return uprobe;
594}
595
596static 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
611static __always_inline
612int 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
633struct __uprobe_key {
634 struct inode *inode;
635 loff_t offset;
636};
637
638static 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
644static 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
650static 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 */
668static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
669{
670 struct uprobe *uprobe;
671
672 spin_lock(&uprobes_treelock);
673 uprobe = __find_uprobe(inode, offset);
674 spin_unlock(&uprobes_treelock);
675
676 return uprobe;
677}
678
679static 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 */
700static struct uprobe *insert_uprobe(struct uprobe *uprobe)
701{
702 struct uprobe *u;
703
704 spin_lock(&uprobes_treelock);
705 u = __insert_uprobe(uprobe);
706 spin_unlock(&uprobes_treelock);
707
708 return u;
709}
710
711static void
712ref_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
721static 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
753static 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 */
766static 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
784static 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
806static 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
832static 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
866static 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
872static 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
889static int
890install_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
917static int
918remove_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
924static 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 */
933static void delete_uprobe(struct uprobe *uprobe)
934{
935 if (WARN_ON(!uprobe_is_active(uprobe)))
936 return;
937
938 spin_lock(&uprobes_treelock);
939 rb_erase(&uprobe->rb_node, &uprobes_tree);
940 spin_unlock(&uprobes_treelock);
941 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
942 put_uprobe(uprobe);
943}
944
945struct map_info {
946 struct map_info *next;
947 struct mm_struct *mm;
948 unsigned long vaddr;
949};
950
951static 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
958static struct map_info *
959build_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
1028static int
1029register_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
1082static 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 */
1102void 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}
1115EXPORT_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 */
1135static 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
1189int uprobe_register(struct inode *inode, loff_t offset,
1190 struct uprobe_consumer *uc)
1191{
1192 return __uprobe_register(inode, offset, 0, uc);
1193}
1194EXPORT_SYMBOL_GPL(uprobe_register);
1195
1196int 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}
1201EXPORT_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 */
1210int 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
1232static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
1233{
1234 struct vm_area_struct *vma;
1235 int err = 0;
1236
1237 mmap_read_lock(mm);
1238 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1239 unsigned long vaddr;
1240 loff_t offset;
1241
1242 if (!valid_vma(vma, false) ||
1243 file_inode(vma->vm_file) != uprobe->inode)
1244 continue;
1245
1246 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
1247 if (uprobe->offset < offset ||
1248 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
1249 continue;
1250
1251 vaddr = offset_to_vaddr(vma, uprobe->offset);
1252 err |= remove_breakpoint(uprobe, mm, vaddr);
1253 }
1254 mmap_read_unlock(mm);
1255
1256 return err;
1257}
1258
1259static struct rb_node *
1260find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1261{
1262 struct rb_node *n = uprobes_tree.rb_node;
1263
1264 while (n) {
1265 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1266
1267 if (inode < u->inode) {
1268 n = n->rb_left;
1269 } else if (inode > u->inode) {
1270 n = n->rb_right;
1271 } else {
1272 if (max < u->offset)
1273 n = n->rb_left;
1274 else if (min > u->offset)
1275 n = n->rb_right;
1276 else
1277 break;
1278 }
1279 }
1280
1281 return n;
1282}
1283
1284/*
1285 * For a given range in vma, build a list of probes that need to be inserted.
1286 */
1287static void build_probe_list(struct inode *inode,
1288 struct vm_area_struct *vma,
1289 unsigned long start, unsigned long end,
1290 struct list_head *head)
1291{
1292 loff_t min, max;
1293 struct rb_node *n, *t;
1294 struct uprobe *u;
1295
1296 INIT_LIST_HEAD(head);
1297 min = vaddr_to_offset(vma, start);
1298 max = min + (end - start) - 1;
1299
1300 spin_lock(&uprobes_treelock);
1301 n = find_node_in_range(inode, min, max);
1302 if (n) {
1303 for (t = n; t; t = rb_prev(t)) {
1304 u = rb_entry(t, struct uprobe, rb_node);
1305 if (u->inode != inode || u->offset < min)
1306 break;
1307 list_add(&u->pending_list, head);
1308 get_uprobe(u);
1309 }
1310 for (t = n; (t = rb_next(t)); ) {
1311 u = rb_entry(t, struct uprobe, rb_node);
1312 if (u->inode != inode || u->offset > max)
1313 break;
1314 list_add(&u->pending_list, head);
1315 get_uprobe(u);
1316 }
1317 }
1318 spin_unlock(&uprobes_treelock);
1319}
1320
1321/* @vma contains reference counter, not the probed instruction. */
1322static int delayed_ref_ctr_inc(struct vm_area_struct *vma)
1323{
1324 struct list_head *pos, *q;
1325 struct delayed_uprobe *du;
1326 unsigned long vaddr;
1327 int ret = 0, err = 0;
1328
1329 mutex_lock(&delayed_uprobe_lock);
1330 list_for_each_safe(pos, q, &delayed_uprobe_list) {
1331 du = list_entry(pos, struct delayed_uprobe, list);
1332
1333 if (du->mm != vma->vm_mm ||
1334 !valid_ref_ctr_vma(du->uprobe, vma))
1335 continue;
1336
1337 vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
1338 ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
1339 if (ret) {
1340 update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1);
1341 if (!err)
1342 err = ret;
1343 }
1344 delayed_uprobe_delete(du);
1345 }
1346 mutex_unlock(&delayed_uprobe_lock);
1347 return err;
1348}
1349
1350/*
1351 * Called from mmap_region/vma_adjust with mm->mmap_lock acquired.
1352 *
1353 * Currently we ignore all errors and always return 0, the callers
1354 * can't handle the failure anyway.
1355 */
1356int uprobe_mmap(struct vm_area_struct *vma)
1357{
1358 struct list_head tmp_list;
1359 struct uprobe *uprobe, *u;
1360 struct inode *inode;
1361
1362 if (no_uprobe_events())
1363 return 0;
1364
1365 if (vma->vm_file &&
1366 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
1367 test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
1368 delayed_ref_ctr_inc(vma);
1369
1370 if (!valid_vma(vma, true))
1371 return 0;
1372
1373 inode = file_inode(vma->vm_file);
1374 if (!inode)
1375 return 0;
1376
1377 mutex_lock(uprobes_mmap_hash(inode));
1378 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1379 /*
1380 * We can race with uprobe_unregister(), this uprobe can be already
1381 * removed. But in this case filter_chain() must return false, all
1382 * consumers have gone away.
1383 */
1384 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1385 if (!fatal_signal_pending(current) &&
1386 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1387 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1388 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1389 }
1390 put_uprobe(uprobe);
1391 }
1392 mutex_unlock(uprobes_mmap_hash(inode));
1393
1394 return 0;
1395}
1396
1397static bool
1398vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1399{
1400 loff_t min, max;
1401 struct inode *inode;
1402 struct rb_node *n;
1403
1404 inode = file_inode(vma->vm_file);
1405
1406 min = vaddr_to_offset(vma, start);
1407 max = min + (end - start) - 1;
1408
1409 spin_lock(&uprobes_treelock);
1410 n = find_node_in_range(inode, min, max);
1411 spin_unlock(&uprobes_treelock);
1412
1413 return !!n;
1414}
1415
1416/*
1417 * Called in context of a munmap of a vma.
1418 */
1419void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1420{
1421 if (no_uprobe_events() || !valid_vma(vma, false))
1422 return;
1423
1424 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1425 return;
1426
1427 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1428 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1429 return;
1430
1431 if (vma_has_uprobes(vma, start, end))
1432 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1433}
1434
1435/* Slot allocation for XOL */
1436static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1437{
1438 struct vm_area_struct *vma;
1439 int ret;
1440
1441 if (mmap_write_lock_killable(mm))
1442 return -EINTR;
1443
1444 if (mm->uprobes_state.xol_area) {
1445 ret = -EALREADY;
1446 goto fail;
1447 }
1448
1449 if (!area->vaddr) {
1450 /* Try to map as high as possible, this is only a hint. */
1451 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1452 PAGE_SIZE, 0, 0);
1453 if (IS_ERR_VALUE(area->vaddr)) {
1454 ret = area->vaddr;
1455 goto fail;
1456 }
1457 }
1458
1459 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1460 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1461 &area->xol_mapping);
1462 if (IS_ERR(vma)) {
1463 ret = PTR_ERR(vma);
1464 goto fail;
1465 }
1466
1467 ret = 0;
1468 /* pairs with get_xol_area() */
1469 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1470 fail:
1471 mmap_write_unlock(mm);
1472
1473 return ret;
1474}
1475
1476static struct xol_area *__create_xol_area(unsigned long vaddr)
1477{
1478 struct mm_struct *mm = current->mm;
1479 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1480 struct xol_area *area;
1481
1482 area = kmalloc(sizeof(*area), GFP_KERNEL);
1483 if (unlikely(!area))
1484 goto out;
1485
1486 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1487 GFP_KERNEL);
1488 if (!area->bitmap)
1489 goto free_area;
1490
1491 area->xol_mapping.name = "[uprobes]";
1492 area->xol_mapping.fault = NULL;
1493 area->xol_mapping.pages = area->pages;
1494 area->pages[0] = alloc_page(GFP_HIGHUSER);
1495 if (!area->pages[0])
1496 goto free_bitmap;
1497 area->pages[1] = NULL;
1498
1499 area->vaddr = vaddr;
1500 init_waitqueue_head(&area->wq);
1501 /* Reserve the 1st slot for get_trampoline_vaddr() */
1502 set_bit(0, area->bitmap);
1503 atomic_set(&area->slot_count, 1);
1504 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1505
1506 if (!xol_add_vma(mm, area))
1507 return area;
1508
1509 __free_page(area->pages[0]);
1510 free_bitmap:
1511 kfree(area->bitmap);
1512 free_area:
1513 kfree(area);
1514 out:
1515 return NULL;
1516}
1517
1518/*
1519 * get_xol_area - Allocate process's xol_area if necessary.
1520 * This area will be used for storing instructions for execution out of line.
1521 *
1522 * Returns the allocated area or NULL.
1523 */
1524static struct xol_area *get_xol_area(void)
1525{
1526 struct mm_struct *mm = current->mm;
1527 struct xol_area *area;
1528
1529 if (!mm->uprobes_state.xol_area)
1530 __create_xol_area(0);
1531
1532 /* Pairs with xol_add_vma() smp_store_release() */
1533 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1534 return area;
1535}
1536
1537/*
1538 * uprobe_clear_state - Free the area allocated for slots.
1539 */
1540void uprobe_clear_state(struct mm_struct *mm)
1541{
1542 struct xol_area *area = mm->uprobes_state.xol_area;
1543
1544 mutex_lock(&delayed_uprobe_lock);
1545 delayed_uprobe_remove(NULL, mm);
1546 mutex_unlock(&delayed_uprobe_lock);
1547
1548 if (!area)
1549 return;
1550
1551 put_page(area->pages[0]);
1552 kfree(area->bitmap);
1553 kfree(area);
1554}
1555
1556void uprobe_start_dup_mmap(void)
1557{
1558 percpu_down_read(&dup_mmap_sem);
1559}
1560
1561void uprobe_end_dup_mmap(void)
1562{
1563 percpu_up_read(&dup_mmap_sem);
1564}
1565
1566void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1567{
1568 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1569 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1570 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1571 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1572 }
1573}
1574
1575/*
1576 * - search for a free slot.
1577 */
1578static unsigned long xol_take_insn_slot(struct xol_area *area)
1579{
1580 unsigned long slot_addr;
1581 int slot_nr;
1582
1583 do {
1584 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1585 if (slot_nr < UINSNS_PER_PAGE) {
1586 if (!test_and_set_bit(slot_nr, area->bitmap))
1587 break;
1588
1589 slot_nr = UINSNS_PER_PAGE;
1590 continue;
1591 }
1592 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1593 } while (slot_nr >= UINSNS_PER_PAGE);
1594
1595 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1596 atomic_inc(&area->slot_count);
1597
1598 return slot_addr;
1599}
1600
1601/*
1602 * xol_get_insn_slot - allocate a slot for xol.
1603 * Returns the allocated slot address or 0.
1604 */
1605static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1606{
1607 struct xol_area *area;
1608 unsigned long xol_vaddr;
1609
1610 area = get_xol_area();
1611 if (!area)
1612 return 0;
1613
1614 xol_vaddr = xol_take_insn_slot(area);
1615 if (unlikely(!xol_vaddr))
1616 return 0;
1617
1618 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1619 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1620
1621 return xol_vaddr;
1622}
1623
1624/*
1625 * xol_free_insn_slot - If slot was earlier allocated by
1626 * @xol_get_insn_slot(), make the slot available for
1627 * subsequent requests.
1628 */
1629static void xol_free_insn_slot(struct task_struct *tsk)
1630{
1631 struct xol_area *area;
1632 unsigned long vma_end;
1633 unsigned long slot_addr;
1634
1635 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1636 return;
1637
1638 slot_addr = tsk->utask->xol_vaddr;
1639 if (unlikely(!slot_addr))
1640 return;
1641
1642 area = tsk->mm->uprobes_state.xol_area;
1643 vma_end = area->vaddr + PAGE_SIZE;
1644 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1645 unsigned long offset;
1646 int slot_nr;
1647
1648 offset = slot_addr - area->vaddr;
1649 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1650 if (slot_nr >= UINSNS_PER_PAGE)
1651 return;
1652
1653 clear_bit(slot_nr, area->bitmap);
1654 atomic_dec(&area->slot_count);
1655 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1656 if (waitqueue_active(&area->wq))
1657 wake_up(&area->wq);
1658
1659 tsk->utask->xol_vaddr = 0;
1660 }
1661}
1662
1663void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1664 void *src, unsigned long len)
1665{
1666 /* Initialize the slot */
1667 copy_to_page(page, vaddr, src, len);
1668
1669 /*
1670 * We probably need flush_icache_user_page() but it needs vma.
1671 * This should work on most of architectures by default. If
1672 * architecture needs to do something different it can define
1673 * its own version of the function.
1674 */
1675 flush_dcache_page(page);
1676}
1677
1678/**
1679 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1680 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1681 * instruction.
1682 * Return the address of the breakpoint instruction.
1683 */
1684unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1685{
1686 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1687}
1688
1689unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1690{
1691 struct uprobe_task *utask = current->utask;
1692
1693 if (unlikely(utask && utask->active_uprobe))
1694 return utask->vaddr;
1695
1696 return instruction_pointer(regs);
1697}
1698
1699static struct return_instance *free_ret_instance(struct return_instance *ri)
1700{
1701 struct return_instance *next = ri->next;
1702 put_uprobe(ri->uprobe);
1703 kfree(ri);
1704 return next;
1705}
1706
1707/*
1708 * Called with no locks held.
1709 * Called in context of an exiting or an exec-ing thread.
1710 */
1711void uprobe_free_utask(struct task_struct *t)
1712{
1713 struct uprobe_task *utask = t->utask;
1714 struct return_instance *ri;
1715
1716 if (!utask)
1717 return;
1718
1719 if (utask->active_uprobe)
1720 put_uprobe(utask->active_uprobe);
1721
1722 ri = utask->return_instances;
1723 while (ri)
1724 ri = free_ret_instance(ri);
1725
1726 xol_free_insn_slot(t);
1727 kfree(utask);
1728 t->utask = NULL;
1729}
1730
1731/*
1732 * Allocate a uprobe_task object for the task if necessary.
1733 * Called when the thread hits a breakpoint.
1734 *
1735 * Returns:
1736 * - pointer to new uprobe_task on success
1737 * - NULL otherwise
1738 */
1739static struct uprobe_task *get_utask(void)
1740{
1741 if (!current->utask)
1742 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1743 return current->utask;
1744}
1745
1746static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1747{
1748 struct uprobe_task *n_utask;
1749 struct return_instance **p, *o, *n;
1750
1751 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1752 if (!n_utask)
1753 return -ENOMEM;
1754 t->utask = n_utask;
1755
1756 p = &n_utask->return_instances;
1757 for (o = o_utask->return_instances; o; o = o->next) {
1758 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1759 if (!n)
1760 return -ENOMEM;
1761
1762 *n = *o;
1763 get_uprobe(n->uprobe);
1764 n->next = NULL;
1765
1766 *p = n;
1767 p = &n->next;
1768 n_utask->depth++;
1769 }
1770
1771 return 0;
1772}
1773
1774static void uprobe_warn(struct task_struct *t, const char *msg)
1775{
1776 pr_warn("uprobe: %s:%d failed to %s\n",
1777 current->comm, current->pid, msg);
1778}
1779
1780static void dup_xol_work(struct callback_head *work)
1781{
1782 if (current->flags & PF_EXITING)
1783 return;
1784
1785 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1786 !fatal_signal_pending(current))
1787 uprobe_warn(current, "dup xol area");
1788}
1789
1790/*
1791 * Called in context of a new clone/fork from copy_process.
1792 */
1793void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1794{
1795 struct uprobe_task *utask = current->utask;
1796 struct mm_struct *mm = current->mm;
1797 struct xol_area *area;
1798
1799 t->utask = NULL;
1800
1801 if (!utask || !utask->return_instances)
1802 return;
1803
1804 if (mm == t->mm && !(flags & CLONE_VFORK))
1805 return;
1806
1807 if (dup_utask(t, utask))
1808 return uprobe_warn(t, "dup ret instances");
1809
1810 /* The task can fork() after dup_xol_work() fails */
1811 area = mm->uprobes_state.xol_area;
1812 if (!area)
1813 return uprobe_warn(t, "dup xol area");
1814
1815 if (mm == t->mm)
1816 return;
1817
1818 t->utask->dup_xol_addr = area->vaddr;
1819 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1820 task_work_add(t, &t->utask->dup_xol_work, TWA_RESUME);
1821}
1822
1823/*
1824 * Current area->vaddr notion assume the trampoline address is always
1825 * equal area->vaddr.
1826 *
1827 * Returns -1 in case the xol_area is not allocated.
1828 */
1829static unsigned long get_trampoline_vaddr(void)
1830{
1831 struct xol_area *area;
1832 unsigned long trampoline_vaddr = -1;
1833
1834 /* Pairs with xol_add_vma() smp_store_release() */
1835 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1836 if (area)
1837 trampoline_vaddr = area->vaddr;
1838
1839 return trampoline_vaddr;
1840}
1841
1842static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1843 struct pt_regs *regs)
1844{
1845 struct return_instance *ri = utask->return_instances;
1846 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1847
1848 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1849 ri = free_ret_instance(ri);
1850 utask->depth--;
1851 }
1852 utask->return_instances = ri;
1853}
1854
1855static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1856{
1857 struct return_instance *ri;
1858 struct uprobe_task *utask;
1859 unsigned long orig_ret_vaddr, trampoline_vaddr;
1860 bool chained;
1861
1862 if (!get_xol_area())
1863 return;
1864
1865 utask = get_utask();
1866 if (!utask)
1867 return;
1868
1869 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1870 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1871 " nestedness limit pid/tgid=%d/%d\n",
1872 current->pid, current->tgid);
1873 return;
1874 }
1875
1876 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1877 if (!ri)
1878 return;
1879
1880 trampoline_vaddr = get_trampoline_vaddr();
1881 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1882 if (orig_ret_vaddr == -1)
1883 goto fail;
1884
1885 /* drop the entries invalidated by longjmp() */
1886 chained = (orig_ret_vaddr == trampoline_vaddr);
1887 cleanup_return_instances(utask, chained, regs);
1888
1889 /*
1890 * We don't want to keep trampoline address in stack, rather keep the
1891 * original return address of first caller thru all the consequent
1892 * instances. This also makes breakpoint unwrapping easier.
1893 */
1894 if (chained) {
1895 if (!utask->return_instances) {
1896 /*
1897 * This situation is not possible. Likely we have an
1898 * attack from user-space.
1899 */
1900 uprobe_warn(current, "handle tail call");
1901 goto fail;
1902 }
1903 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1904 }
1905
1906 ri->uprobe = get_uprobe(uprobe);
1907 ri->func = instruction_pointer(regs);
1908 ri->stack = user_stack_pointer(regs);
1909 ri->orig_ret_vaddr = orig_ret_vaddr;
1910 ri->chained = chained;
1911
1912 utask->depth++;
1913 ri->next = utask->return_instances;
1914 utask->return_instances = ri;
1915
1916 return;
1917 fail:
1918 kfree(ri);
1919}
1920
1921/* Prepare to single-step probed instruction out of line. */
1922static int
1923pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1924{
1925 struct uprobe_task *utask;
1926 unsigned long xol_vaddr;
1927 int err;
1928
1929 utask = get_utask();
1930 if (!utask)
1931 return -ENOMEM;
1932
1933 xol_vaddr = xol_get_insn_slot(uprobe);
1934 if (!xol_vaddr)
1935 return -ENOMEM;
1936
1937 utask->xol_vaddr = xol_vaddr;
1938 utask->vaddr = bp_vaddr;
1939
1940 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1941 if (unlikely(err)) {
1942 xol_free_insn_slot(current);
1943 return err;
1944 }
1945
1946 utask->active_uprobe = uprobe;
1947 utask->state = UTASK_SSTEP;
1948 return 0;
1949}
1950
1951/*
1952 * If we are singlestepping, then ensure this thread is not connected to
1953 * non-fatal signals until completion of singlestep. When xol insn itself
1954 * triggers the signal, restart the original insn even if the task is
1955 * already SIGKILL'ed (since coredump should report the correct ip). This
1956 * is even more important if the task has a handler for SIGSEGV/etc, The
1957 * _same_ instruction should be repeated again after return from the signal
1958 * handler, and SSTEP can never finish in this case.
1959 */
1960bool uprobe_deny_signal(void)
1961{
1962 struct task_struct *t = current;
1963 struct uprobe_task *utask = t->utask;
1964
1965 if (likely(!utask || !utask->active_uprobe))
1966 return false;
1967
1968 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1969
1970 if (task_sigpending(t)) {
1971 spin_lock_irq(&t->sighand->siglock);
1972 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1973 spin_unlock_irq(&t->sighand->siglock);
1974
1975 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1976 utask->state = UTASK_SSTEP_TRAPPED;
1977 set_tsk_thread_flag(t, TIF_UPROBE);
1978 }
1979 }
1980
1981 return true;
1982}
1983
1984static void mmf_recalc_uprobes(struct mm_struct *mm)
1985{
1986 struct vm_area_struct *vma;
1987
1988 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1989 if (!valid_vma(vma, false))
1990 continue;
1991 /*
1992 * This is not strictly accurate, we can race with
1993 * uprobe_unregister() and see the already removed
1994 * uprobe if delete_uprobe() was not yet called.
1995 * Or this uprobe can be filtered out.
1996 */
1997 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1998 return;
1999 }
2000
2001 clear_bit(MMF_HAS_UPROBES, &mm->flags);
2002}
2003
2004static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
2005{
2006 struct page *page;
2007 uprobe_opcode_t opcode;
2008 int result;
2009
2010 if (WARN_ON_ONCE(!IS_ALIGNED(vaddr, UPROBE_SWBP_INSN_SIZE)))
2011 return -EINVAL;
2012
2013 pagefault_disable();
2014 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
2015 pagefault_enable();
2016
2017 if (likely(result == 0))
2018 goto out;
2019
2020 /*
2021 * The NULL 'tsk' here ensures that any faults that occur here
2022 * will not be accounted to the task. 'mm' *is* current->mm,
2023 * but we treat this as a 'remote' access since it is
2024 * essentially a kernel access to the memory.
2025 */
2026 result = get_user_pages_remote(mm, vaddr, 1, FOLL_FORCE, &page,
2027 NULL, NULL);
2028 if (result < 0)
2029 return result;
2030
2031 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
2032 put_page(page);
2033 out:
2034 /* This needs to return true for any variant of the trap insn */
2035 return is_trap_insn(&opcode);
2036}
2037
2038static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
2039{
2040 struct mm_struct *mm = current->mm;
2041 struct uprobe *uprobe = NULL;
2042 struct vm_area_struct *vma;
2043
2044 mmap_read_lock(mm);
2045 vma = vma_lookup(mm, bp_vaddr);
2046 if (vma) {
2047 if (valid_vma(vma, false)) {
2048 struct inode *inode = file_inode(vma->vm_file);
2049 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
2050
2051 uprobe = find_uprobe(inode, offset);
2052 }
2053
2054 if (!uprobe)
2055 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
2056 } else {
2057 *is_swbp = -EFAULT;
2058 }
2059
2060 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
2061 mmf_recalc_uprobes(mm);
2062 mmap_read_unlock(mm);
2063
2064 return uprobe;
2065}
2066
2067static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
2068{
2069 struct uprobe_consumer *uc;
2070 int remove = UPROBE_HANDLER_REMOVE;
2071 bool need_prep = false; /* prepare return uprobe, when needed */
2072
2073 down_read(&uprobe->register_rwsem);
2074 for (uc = uprobe->consumers; uc; uc = uc->next) {
2075 int rc = 0;
2076
2077 if (uc->handler) {
2078 rc = uc->handler(uc, regs);
2079 WARN(rc & ~UPROBE_HANDLER_MASK,
2080 "bad rc=0x%x from %ps()\n", rc, uc->handler);
2081 }
2082
2083 if (uc->ret_handler)
2084 need_prep = true;
2085
2086 remove &= rc;
2087 }
2088
2089 if (need_prep && !remove)
2090 prepare_uretprobe(uprobe, regs); /* put bp at return */
2091
2092 if (remove && uprobe->consumers) {
2093 WARN_ON(!uprobe_is_active(uprobe));
2094 unapply_uprobe(uprobe, current->mm);
2095 }
2096 up_read(&uprobe->register_rwsem);
2097}
2098
2099static void
2100handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
2101{
2102 struct uprobe *uprobe = ri->uprobe;
2103 struct uprobe_consumer *uc;
2104
2105 down_read(&uprobe->register_rwsem);
2106 for (uc = uprobe->consumers; uc; uc = uc->next) {
2107 if (uc->ret_handler)
2108 uc->ret_handler(uc, ri->func, regs);
2109 }
2110 up_read(&uprobe->register_rwsem);
2111}
2112
2113static struct return_instance *find_next_ret_chain(struct return_instance *ri)
2114{
2115 bool chained;
2116
2117 do {
2118 chained = ri->chained;
2119 ri = ri->next; /* can't be NULL if chained */
2120 } while (chained);
2121
2122 return ri;
2123}
2124
2125static void handle_trampoline(struct pt_regs *regs)
2126{
2127 struct uprobe_task *utask;
2128 struct return_instance *ri, *next;
2129 bool valid;
2130
2131 utask = current->utask;
2132 if (!utask)
2133 goto sigill;
2134
2135 ri = utask->return_instances;
2136 if (!ri)
2137 goto sigill;
2138
2139 do {
2140 /*
2141 * We should throw out the frames invalidated by longjmp().
2142 * If this chain is valid, then the next one should be alive
2143 * or NULL; the latter case means that nobody but ri->func
2144 * could hit this trampoline on return. TODO: sigaltstack().
2145 */
2146 next = find_next_ret_chain(ri);
2147 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
2148
2149 instruction_pointer_set(regs, ri->orig_ret_vaddr);
2150 do {
2151 if (valid)
2152 handle_uretprobe_chain(ri, regs);
2153 ri = free_ret_instance(ri);
2154 utask->depth--;
2155 } while (ri != next);
2156 } while (!valid);
2157
2158 utask->return_instances = ri;
2159 return;
2160
2161 sigill:
2162 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
2163 force_sig(SIGILL);
2164
2165}
2166
2167bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
2168{
2169 return false;
2170}
2171
2172bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
2173 struct pt_regs *regs)
2174{
2175 return true;
2176}
2177
2178/*
2179 * Run handler and ask thread to singlestep.
2180 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2181 */
2182static void handle_swbp(struct pt_regs *regs)
2183{
2184 struct uprobe *uprobe;
2185 unsigned long bp_vaddr;
2186 int is_swbp;
2187
2188 bp_vaddr = uprobe_get_swbp_addr(regs);
2189 if (bp_vaddr == get_trampoline_vaddr())
2190 return handle_trampoline(regs);
2191
2192 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
2193 if (!uprobe) {
2194 if (is_swbp > 0) {
2195 /* No matching uprobe; signal SIGTRAP. */
2196 force_sig(SIGTRAP);
2197 } else {
2198 /*
2199 * Either we raced with uprobe_unregister() or we can't
2200 * access this memory. The latter is only possible if
2201 * another thread plays with our ->mm. In both cases
2202 * we can simply restart. If this vma was unmapped we
2203 * can pretend this insn was not executed yet and get
2204 * the (correct) SIGSEGV after restart.
2205 */
2206 instruction_pointer_set(regs, bp_vaddr);
2207 }
2208 return;
2209 }
2210
2211 /* change it in advance for ->handler() and restart */
2212 instruction_pointer_set(regs, bp_vaddr);
2213
2214 /*
2215 * TODO: move copy_insn/etc into _register and remove this hack.
2216 * After we hit the bp, _unregister + _register can install the
2217 * new and not-yet-analyzed uprobe at the same address, restart.
2218 */
2219 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
2220 goto out;
2221
2222 /*
2223 * Pairs with the smp_wmb() in prepare_uprobe().
2224 *
2225 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2226 * we must also see the stores to &uprobe->arch performed by the
2227 * prepare_uprobe() call.
2228 */
2229 smp_rmb();
2230
2231 /* Tracing handlers use ->utask to communicate with fetch methods */
2232 if (!get_utask())
2233 goto out;
2234
2235 if (arch_uprobe_ignore(&uprobe->arch, regs))
2236 goto out;
2237
2238 handler_chain(uprobe, regs);
2239
2240 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
2241 goto out;
2242
2243 if (!pre_ssout(uprobe, regs, bp_vaddr))
2244 return;
2245
2246 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2247out:
2248 put_uprobe(uprobe);
2249}
2250
2251/*
2252 * Perform required fix-ups and disable singlestep.
2253 * Allow pending signals to take effect.
2254 */
2255static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
2256{
2257 struct uprobe *uprobe;
2258 int err = 0;
2259
2260 uprobe = utask->active_uprobe;
2261 if (utask->state == UTASK_SSTEP_ACK)
2262 err = arch_uprobe_post_xol(&uprobe->arch, regs);
2263 else if (utask->state == UTASK_SSTEP_TRAPPED)
2264 arch_uprobe_abort_xol(&uprobe->arch, regs);
2265 else
2266 WARN_ON_ONCE(1);
2267
2268 put_uprobe(uprobe);
2269 utask->active_uprobe = NULL;
2270 utask->state = UTASK_RUNNING;
2271 xol_free_insn_slot(current);
2272
2273 spin_lock_irq(&current->sighand->siglock);
2274 recalc_sigpending(); /* see uprobe_deny_signal() */
2275 spin_unlock_irq(&current->sighand->siglock);
2276
2277 if (unlikely(err)) {
2278 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
2279 force_sig(SIGILL);
2280 }
2281}
2282
2283/*
2284 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2285 * allows the thread to return from interrupt. After that handle_swbp()
2286 * sets utask->active_uprobe.
2287 *
2288 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2289 * and allows the thread to return from interrupt.
2290 *
2291 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2292 * uprobe_notify_resume().
2293 */
2294void uprobe_notify_resume(struct pt_regs *regs)
2295{
2296 struct uprobe_task *utask;
2297
2298 clear_thread_flag(TIF_UPROBE);
2299
2300 utask = current->utask;
2301 if (utask && utask->active_uprobe)
2302 handle_singlestep(utask, regs);
2303 else
2304 handle_swbp(regs);
2305}
2306
2307/*
2308 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2309 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2310 */
2311int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2312{
2313 if (!current->mm)
2314 return 0;
2315
2316 if (!test_bit(MMF_HAS_UPROBES, &current->mm->flags) &&
2317 (!current->utask || !current->utask->return_instances))
2318 return 0;
2319
2320 set_thread_flag(TIF_UPROBE);
2321 return 1;
2322}
2323
2324/*
2325 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2326 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2327 */
2328int uprobe_post_sstep_notifier(struct pt_regs *regs)
2329{
2330 struct uprobe_task *utask = current->utask;
2331
2332 if (!current->mm || !utask || !utask->active_uprobe)
2333 /* task is currently not uprobed */
2334 return 0;
2335
2336 utask->state = UTASK_SSTEP_ACK;
2337 set_thread_flag(TIF_UPROBE);
2338 return 1;
2339}
2340
2341static struct notifier_block uprobe_exception_nb = {
2342 .notifier_call = arch_uprobe_exception_notify,
2343 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2344};
2345
2346void __init uprobes_init(void)
2347{
2348 int i;
2349
2350 for (i = 0; i < UPROBES_HASH_SZ; i++)
2351 mutex_init(&uprobes_mmap_mutex[i]);
2352
2353 BUG_ON(register_die_notifier(&uprobe_exception_nb));
2354}
2355

source code of linux/kernel/events/uprobes.c