1 // SPDX-License-Identifier: GPL-2.0+
3 * User-space Probes (UProbes)
5 * Copyright (C) IBM Corporation, 2008-2012
9 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra
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>
31 #include <linux/uprobes.h>
33 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
34 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
36 static struct rb_root uprobes_tree
= RB_ROOT
;
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?
41 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
43 static DEFINE_SPINLOCK(uprobes_treelock
); /* serialize rbtree access */
45 #define UPROBES_HASH_SZ 13
46 /* serialize uprobe->pending_list */
47 static struct mutex uprobes_mmap_mutex
[UPROBES_HASH_SZ
];
48 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
50 DEFINE_STATIC_PERCPU_RWSEM(dup_mmap_sem
);
52 /* Have a copy of original instruction */
53 #define UPROBE_COPY_INSN 0
56 struct rb_node rb_node
; /* node in the rb tree */
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 */
64 loff_t ref_ctr_offset
;
68 * The generic code assumes that it has two members of unknown type
69 * owned by the arch-specific code:
71 * insn - copy_insn() saves the original instruction here for
72 * arch_uprobe_analyze_insn().
74 * ixol - potentially modified instruction to execute out of
75 * line, copied to xol_area by xol_get_insn_slot().
77 struct arch_uprobe arch
;
80 struct delayed_uprobe
{
81 struct list_head list
;
82 struct uprobe
*uprobe
;
86 static DEFINE_MUTEX(delayed_uprobe_lock
);
87 static LIST_HEAD(delayed_uprobe_list
);
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().
94 * On a breakpoint hit, thread contests for a slot. It frees the
95 * slot after singlestep. Currently a fixed number of slots are
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 */
103 struct vm_special_mapping xol_mapping
;
104 struct page
*pages
[2];
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.
110 unsigned long vaddr
; /* Page(s) of instruction slots */
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
121 static bool valid_vma(struct vm_area_struct
*vma
, bool is_register
)
123 vm_flags_t flags
= VM_HUGETLB
| VM_MAYEXEC
| VM_MAYSHARE
;
128 return vma
->vm_file
&& (vma
->vm_flags
& flags
) == VM_MAYEXEC
;
131 static unsigned long offset_to_vaddr(struct vm_area_struct
*vma
, loff_t offset
)
133 return vma
->vm_start
+ offset
- ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
136 static loff_t
vaddr_to_offset(struct vm_area_struct
*vma
, unsigned long vaddr
)
138 return ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
) + (vaddr
- vma
->vm_start
);
142 * __replace_page - replace page in vma by new page.
143 * based on replace_page in mm/ksm.c
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
150 * If @new_page is NULL, only unmap @old_page.
152 * Returns 0 on success, negative error code otherwise.
154 static int __replace_page(struct vm_area_struct
*vma
, unsigned long addr
,
155 struct page
*old_page
, struct page
*new_page
)
157 struct mm_struct
*mm
= vma
->vm_mm
;
158 struct page_vma_mapped_walk pvmw
= {
159 .page
= compound_head(old_page
),
164 struct mmu_notifier_range range
;
166 mmu_notifier_range_init(&range
, MMU_NOTIFY_CLEAR
, 0, vma
, mm
, addr
,
170 err
= mem_cgroup_charge(new_page
, vma
->vm_mm
, GFP_KERNEL
);
175 /* For try_to_free_swap() and munlock_vma_page() below */
178 mmu_notifier_invalidate_range_start(&range
);
180 if (!page_vma_mapped_walk(&pvmw
))
182 VM_BUG_ON_PAGE(addr
!= pvmw
.address
, old_page
);
186 page_add_new_anon_rmap(new_page
, vma
, addr
, false);
187 lru_cache_add_active_or_unevictable(new_page
, vma
);
189 /* no new page, just dec_mm_counter for old_page */
190 dec_mm_counter(mm
, MM_ANONPAGES
);
192 if (!PageAnon(old_page
)) {
193 dec_mm_counter(mm
, mm_counter_file(old_page
));
194 inc_mm_counter(mm
, MM_ANONPAGES
);
197 flush_cache_page(vma
, addr
, pte_pfn(*pvmw
.pte
));
198 ptep_clear_flush_notify(vma
, addr
, pvmw
.pte
);
200 set_pte_at_notify(mm
, addr
, pvmw
.pte
,
201 mk_pte(new_page
, vma
->vm_page_prot
));
203 page_remove_rmap(old_page
, false);
204 if (!page_mapped(old_page
))
205 try_to_free_swap(old_page
);
206 page_vma_mapped_walk_done(&pvmw
);
208 if (vma
->vm_flags
& VM_LOCKED
)
209 munlock_vma_page(old_page
);
214 mmu_notifier_invalidate_range_end(&range
);
215 unlock_page(old_page
);
220 * is_swbp_insn - check if instruction is breakpoint instruction.
221 * @insn: instruction to be checked.
222 * Default implementation of is_swbp_insn
223 * Returns true if @insn is a breakpoint instruction.
225 bool __weak
is_swbp_insn(uprobe_opcode_t
*insn
)
227 return *insn
== UPROBE_SWBP_INSN
;
231 * is_trap_insn - check if instruction is breakpoint instruction.
232 * @insn: instruction to be checked.
233 * Default implementation of is_trap_insn
234 * Returns true if @insn is a breakpoint instruction.
236 * This function is needed for the case where an architecture has multiple
237 * trap instructions (like powerpc).
239 bool __weak
is_trap_insn(uprobe_opcode_t
*insn
)
241 return is_swbp_insn(insn
);
244 static void copy_from_page(struct page
*page
, unsigned long vaddr
, void *dst
, int len
)
246 void *kaddr
= kmap_atomic(page
);
247 memcpy(dst
, kaddr
+ (vaddr
& ~PAGE_MASK
), len
);
248 kunmap_atomic(kaddr
);
251 static void copy_to_page(struct page
*page
, unsigned long vaddr
, const void *src
, int len
)
253 void *kaddr
= kmap_atomic(page
);
254 memcpy(kaddr
+ (vaddr
& ~PAGE_MASK
), src
, len
);
255 kunmap_atomic(kaddr
);
258 static int verify_opcode(struct page
*page
, unsigned long vaddr
, uprobe_opcode_t
*new_opcode
)
260 uprobe_opcode_t old_opcode
;
264 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
265 * We do not check if it is any other 'trap variant' which could
266 * be conditional trap instruction such as the one powerpc supports.
268 * The logic is that we do not care if the underlying instruction
269 * is a trap variant; uprobes always wins over any other (gdb)
272 copy_from_page(page
, vaddr
, &old_opcode
, UPROBE_SWBP_INSN_SIZE
);
273 is_swbp
= is_swbp_insn(&old_opcode
);
275 if (is_swbp_insn(new_opcode
)) {
276 if (is_swbp
) /* register: already installed? */
279 if (!is_swbp
) /* unregister: was it changed by us? */
286 static struct delayed_uprobe
*
287 delayed_uprobe_check(struct uprobe
*uprobe
, struct mm_struct
*mm
)
289 struct delayed_uprobe
*du
;
291 list_for_each_entry(du
, &delayed_uprobe_list
, list
)
292 if (du
->uprobe
== uprobe
&& du
->mm
== mm
)
297 static int delayed_uprobe_add(struct uprobe
*uprobe
, struct mm_struct
*mm
)
299 struct delayed_uprobe
*du
;
301 if (delayed_uprobe_check(uprobe
, mm
))
304 du
= kzalloc(sizeof(*du
), GFP_KERNEL
);
310 list_add(&du
->list
, &delayed_uprobe_list
);
314 static void delayed_uprobe_delete(struct delayed_uprobe
*du
)
322 static void delayed_uprobe_remove(struct uprobe
*uprobe
, struct mm_struct
*mm
)
324 struct list_head
*pos
, *q
;
325 struct delayed_uprobe
*du
;
330 list_for_each_safe(pos
, q
, &delayed_uprobe_list
) {
331 du
= list_entry(pos
, struct delayed_uprobe
, list
);
333 if (uprobe
&& du
->uprobe
!= uprobe
)
335 if (mm
&& du
->mm
!= mm
)
338 delayed_uprobe_delete(du
);
342 static bool valid_ref_ctr_vma(struct uprobe
*uprobe
,
343 struct vm_area_struct
*vma
)
345 unsigned long vaddr
= offset_to_vaddr(vma
, uprobe
->ref_ctr_offset
);
347 return uprobe
->ref_ctr_offset
&&
349 file_inode(vma
->vm_file
) == uprobe
->inode
&&
350 (vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) == VM_WRITE
&&
351 vma
->vm_start
<= vaddr
&&
355 static struct vm_area_struct
*
356 find_ref_ctr_vma(struct uprobe
*uprobe
, struct mm_struct
*mm
)
358 struct vm_area_struct
*tmp
;
360 for (tmp
= mm
->mmap
; tmp
; tmp
= tmp
->vm_next
)
361 if (valid_ref_ctr_vma(uprobe
, tmp
))
368 __update_ref_ctr(struct mm_struct
*mm
, unsigned long vaddr
, short d
)
372 struct vm_area_struct
*vma
;
379 ret
= get_user_pages_remote(NULL
, mm
, vaddr
, 1,
380 FOLL_WRITE
, &page
, &vma
, NULL
);
381 if (unlikely(ret
<= 0)) {
383 * We are asking for 1 page. If get_user_pages_remote() fails,
384 * it may return 0, in that case we have to return error.
386 return ret
== 0 ? -EBUSY
: ret
;
389 kaddr
= kmap_atomic(page
);
390 ptr
= kaddr
+ (vaddr
& ~PAGE_MASK
);
392 if (unlikely(*ptr
+ d
< 0)) {
393 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
394 "curr val: %d, delta: %d\n", vaddr
, *ptr
, d
);
402 kunmap_atomic(kaddr
);
407 static void update_ref_ctr_warn(struct uprobe
*uprobe
,
408 struct mm_struct
*mm
, short d
)
410 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
411 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
412 d
> 0 ? "increment" : "decrement", uprobe
->inode
->i_ino
,
413 (unsigned long long) uprobe
->offset
,
414 (unsigned long long) uprobe
->ref_ctr_offset
, mm
);
417 static int update_ref_ctr(struct uprobe
*uprobe
, struct mm_struct
*mm
,
420 struct vm_area_struct
*rc_vma
;
421 unsigned long rc_vaddr
;
424 rc_vma
= find_ref_ctr_vma(uprobe
, mm
);
427 rc_vaddr
= offset_to_vaddr(rc_vma
, uprobe
->ref_ctr_offset
);
428 ret
= __update_ref_ctr(mm
, rc_vaddr
, d
);
430 update_ref_ctr_warn(uprobe
, mm
, d
);
436 mutex_lock(&delayed_uprobe_lock
);
438 ret
= delayed_uprobe_add(uprobe
, mm
);
440 delayed_uprobe_remove(uprobe
, mm
);
441 mutex_unlock(&delayed_uprobe_lock
);
448 * Expect the breakpoint instruction to be the smallest size instruction for
449 * the architecture. If an arch has variable length instruction and the
450 * breakpoint instruction is not of the smallest length instruction
451 * supported by that architecture then we need to modify is_trap_at_addr and
452 * uprobe_write_opcode accordingly. This would never be a problem for archs
453 * that have fixed length instructions.
455 * uprobe_write_opcode - write the opcode at a given virtual address.
456 * @mm: the probed process address space.
457 * @vaddr: the virtual address to store the opcode.
458 * @opcode: opcode to be written at @vaddr.
460 * Called with mm->mmap_lock held for write.
461 * Return 0 (success) or a negative errno.
463 int uprobe_write_opcode(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
,
464 unsigned long vaddr
, uprobe_opcode_t opcode
)
466 struct uprobe
*uprobe
;
467 struct page
*old_page
, *new_page
;
468 struct vm_area_struct
*vma
;
469 int ret
, is_register
, ref_ctr_updated
= 0;
470 bool orig_page_huge
= false;
471 unsigned int gup_flags
= FOLL_FORCE
;
473 is_register
= is_swbp_insn(&opcode
);
474 uprobe
= container_of(auprobe
, struct uprobe
, arch
);
478 gup_flags
|= FOLL_SPLIT_PMD
;
479 /* Read the page with vaddr into memory */
480 ret
= get_user_pages_remote(NULL
, mm
, vaddr
, 1, gup_flags
,
481 &old_page
, &vma
, NULL
);
485 ret
= verify_opcode(old_page
, vaddr
, &opcode
);
489 if (WARN(!is_register
&& PageCompound(old_page
),
490 "uprobe unregister should never work on compound page\n")) {
495 /* We are going to replace instruction, update ref_ctr. */
496 if (!ref_ctr_updated
&& uprobe
->ref_ctr_offset
) {
497 ret
= update_ref_ctr(uprobe
, mm
, is_register
? 1 : -1);
505 if (!is_register
&& !PageAnon(old_page
))
508 ret
= anon_vma_prepare(vma
);
513 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, vaddr
);
517 __SetPageUptodate(new_page
);
518 copy_highpage(new_page
, old_page
);
519 copy_to_page(new_page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
522 struct page
*orig_page
;
525 VM_BUG_ON_PAGE(!PageAnon(old_page
), old_page
);
527 index
= vaddr_to_offset(vma
, vaddr
& PAGE_MASK
) >> PAGE_SHIFT
;
528 orig_page
= find_get_page(vma
->vm_file
->f_inode
->i_mapping
,
532 if (PageUptodate(orig_page
) &&
533 pages_identical(new_page
, orig_page
)) {
534 /* let go new_page */
538 if (PageCompound(orig_page
))
539 orig_page_huge
= true;
545 ret
= __replace_page(vma
, vaddr
, old_page
, new_page
);
551 if (unlikely(ret
== -EAGAIN
))
554 /* Revert back reference counter if instruction update failed. */
555 if (ret
&& is_register
&& ref_ctr_updated
)
556 update_ref_ctr(uprobe
, mm
, -1);
558 /* try collapse pmd for compound page */
559 if (!ret
&& orig_page_huge
)
560 collapse_pte_mapped_thp(mm
, vaddr
);
566 * set_swbp - store breakpoint at a given address.
567 * @auprobe: arch specific probepoint information.
568 * @mm: the probed process address space.
569 * @vaddr: the virtual address to insert the opcode.
571 * For mm @mm, store the breakpoint instruction at @vaddr.
572 * Return 0 (success) or a negative errno.
574 int __weak
set_swbp(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
576 return uprobe_write_opcode(auprobe
, mm
, vaddr
, UPROBE_SWBP_INSN
);
580 * set_orig_insn - Restore the original instruction.
581 * @mm: the probed process address space.
582 * @auprobe: arch specific probepoint information.
583 * @vaddr: the virtual address to insert the opcode.
585 * For mm @mm, restore the original opcode (opcode) at @vaddr.
586 * Return 0 (success) or a negative errno.
589 set_orig_insn(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
591 return uprobe_write_opcode(auprobe
, mm
, vaddr
,
592 *(uprobe_opcode_t
*)&auprobe
->insn
);
595 static struct uprobe
*get_uprobe(struct uprobe
*uprobe
)
597 refcount_inc(&uprobe
->ref
);
601 static void put_uprobe(struct uprobe
*uprobe
)
603 if (refcount_dec_and_test(&uprobe
->ref
)) {
605 * If application munmap(exec_vma) before uprobe_unregister()
606 * gets called, we don't get a chance to remove uprobe from
607 * delayed_uprobe_list from remove_breakpoint(). Do it here.
609 mutex_lock(&delayed_uprobe_lock
);
610 delayed_uprobe_remove(uprobe
, NULL
);
611 mutex_unlock(&delayed_uprobe_lock
);
616 static int match_uprobe(struct uprobe
*l
, struct uprobe
*r
)
618 if (l
->inode
< r
->inode
)
621 if (l
->inode
> r
->inode
)
624 if (l
->offset
< r
->offset
)
627 if (l
->offset
> r
->offset
)
633 static struct uprobe
*__find_uprobe(struct inode
*inode
, loff_t offset
)
635 struct uprobe u
= { .inode
= inode
, .offset
= offset
};
636 struct rb_node
*n
= uprobes_tree
.rb_node
;
637 struct uprobe
*uprobe
;
641 uprobe
= rb_entry(n
, struct uprobe
, rb_node
);
642 match
= match_uprobe(&u
, uprobe
);
644 return get_uprobe(uprobe
);
655 * Find a uprobe corresponding to a given inode:offset
656 * Acquires uprobes_treelock
658 static struct uprobe
*find_uprobe(struct inode
*inode
, loff_t offset
)
660 struct uprobe
*uprobe
;
662 spin_lock(&uprobes_treelock
);
663 uprobe
= __find_uprobe(inode
, offset
);
664 spin_unlock(&uprobes_treelock
);
669 static struct uprobe
*__insert_uprobe(struct uprobe
*uprobe
)
671 struct rb_node
**p
= &uprobes_tree
.rb_node
;
672 struct rb_node
*parent
= NULL
;
678 u
= rb_entry(parent
, struct uprobe
, rb_node
);
679 match
= match_uprobe(uprobe
, u
);
681 return get_uprobe(u
);
684 p
= &parent
->rb_left
;
686 p
= &parent
->rb_right
;
691 rb_link_node(&uprobe
->rb_node
, parent
, p
);
692 rb_insert_color(&uprobe
->rb_node
, &uprobes_tree
);
693 /* get access + creation ref */
694 refcount_set(&uprobe
->ref
, 2);
700 * Acquire uprobes_treelock.
701 * Matching uprobe already exists in rbtree;
702 * increment (access refcount) and return the matching uprobe.
704 * No matching uprobe; insert the uprobe in rb_tree;
705 * get a double refcount (access + creation) and return NULL.
707 static struct uprobe
*insert_uprobe(struct uprobe
*uprobe
)
711 spin_lock(&uprobes_treelock
);
712 u
= __insert_uprobe(uprobe
);
713 spin_unlock(&uprobes_treelock
);
719 ref_ctr_mismatch_warn(struct uprobe
*cur_uprobe
, struct uprobe
*uprobe
)
721 pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx "
722 "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n",
723 uprobe
->inode
->i_ino
, (unsigned long long) uprobe
->offset
,
724 (unsigned long long) cur_uprobe
->ref_ctr_offset
,
725 (unsigned long long) uprobe
->ref_ctr_offset
);
728 static struct uprobe
*alloc_uprobe(struct inode
*inode
, loff_t offset
,
729 loff_t ref_ctr_offset
)
731 struct uprobe
*uprobe
, *cur_uprobe
;
733 uprobe
= kzalloc(sizeof(struct uprobe
), GFP_KERNEL
);
737 uprobe
->inode
= inode
;
738 uprobe
->offset
= offset
;
739 uprobe
->ref_ctr_offset
= ref_ctr_offset
;
740 init_rwsem(&uprobe
->register_rwsem
);
741 init_rwsem(&uprobe
->consumer_rwsem
);
743 /* add to uprobes_tree, sorted on inode:offset */
744 cur_uprobe
= insert_uprobe(uprobe
);
745 /* a uprobe exists for this inode:offset combination */
747 if (cur_uprobe
->ref_ctr_offset
!= uprobe
->ref_ctr_offset
) {
748 ref_ctr_mismatch_warn(cur_uprobe
, uprobe
);
749 put_uprobe(cur_uprobe
);
751 return ERR_PTR(-EINVAL
);
760 static void consumer_add(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
762 down_write(&uprobe
->consumer_rwsem
);
763 uc
->next
= uprobe
->consumers
;
764 uprobe
->consumers
= uc
;
765 up_write(&uprobe
->consumer_rwsem
);
769 * For uprobe @uprobe, delete the consumer @uc.
770 * Return true if the @uc is deleted successfully
773 static bool consumer_del(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
775 struct uprobe_consumer
**con
;
778 down_write(&uprobe
->consumer_rwsem
);
779 for (con
= &uprobe
->consumers
; *con
; con
= &(*con
)->next
) {
786 up_write(&uprobe
->consumer_rwsem
);
791 static int __copy_insn(struct address_space
*mapping
, struct file
*filp
,
792 void *insn
, int nbytes
, loff_t offset
)
796 * Ensure that the page that has the original instruction is populated
797 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
798 * see uprobe_register().
800 if (mapping
->a_ops
->readpage
)
801 page
= read_mapping_page(mapping
, offset
>> PAGE_SHIFT
, filp
);
803 page
= shmem_read_mapping_page(mapping
, offset
>> PAGE_SHIFT
);
805 return PTR_ERR(page
);
807 copy_from_page(page
, offset
, insn
, nbytes
);
813 static int copy_insn(struct uprobe
*uprobe
, struct file
*filp
)
815 struct address_space
*mapping
= uprobe
->inode
->i_mapping
;
816 loff_t offs
= uprobe
->offset
;
817 void *insn
= &uprobe
->arch
.insn
;
818 int size
= sizeof(uprobe
->arch
.insn
);
821 /* Copy only available bytes, -EIO if nothing was read */
823 if (offs
>= i_size_read(uprobe
->inode
))
826 len
= min_t(int, size
, PAGE_SIZE
- (offs
& ~PAGE_MASK
));
827 err
= __copy_insn(mapping
, filp
, insn
, len
, offs
);
839 static int prepare_uprobe(struct uprobe
*uprobe
, struct file
*file
,
840 struct mm_struct
*mm
, unsigned long vaddr
)
844 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
847 /* TODO: move this into _register, until then we abuse this sem. */
848 down_write(&uprobe
->consumer_rwsem
);
849 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
852 ret
= copy_insn(uprobe
, file
);
857 if (is_trap_insn((uprobe_opcode_t
*)&uprobe
->arch
.insn
))
860 ret
= arch_uprobe_analyze_insn(&uprobe
->arch
, mm
, vaddr
);
864 /* uprobe_write_opcode() assumes we don't cross page boundary */
865 BUG_ON((uprobe
->offset
& ~PAGE_MASK
) +
866 UPROBE_SWBP_INSN_SIZE
> PAGE_SIZE
);
868 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
869 set_bit(UPROBE_COPY_INSN
, &uprobe
->flags
);
872 up_write(&uprobe
->consumer_rwsem
);
877 static inline bool consumer_filter(struct uprobe_consumer
*uc
,
878 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
880 return !uc
->filter
|| uc
->filter(uc
, ctx
, mm
);
883 static bool filter_chain(struct uprobe
*uprobe
,
884 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
886 struct uprobe_consumer
*uc
;
889 down_read(&uprobe
->consumer_rwsem
);
890 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
891 ret
= consumer_filter(uc
, ctx
, mm
);
895 up_read(&uprobe
->consumer_rwsem
);
901 install_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
,
902 struct vm_area_struct
*vma
, unsigned long vaddr
)
907 ret
= prepare_uprobe(uprobe
, vma
->vm_file
, mm
, vaddr
);
912 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
913 * the task can hit this breakpoint right after __replace_page().
915 first_uprobe
= !test_bit(MMF_HAS_UPROBES
, &mm
->flags
);
917 set_bit(MMF_HAS_UPROBES
, &mm
->flags
);
919 ret
= set_swbp(&uprobe
->arch
, mm
, vaddr
);
921 clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
922 else if (first_uprobe
)
923 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
929 remove_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
931 set_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
932 return set_orig_insn(&uprobe
->arch
, mm
, vaddr
);
935 static inline bool uprobe_is_active(struct uprobe
*uprobe
)
937 return !RB_EMPTY_NODE(&uprobe
->rb_node
);
940 * There could be threads that have already hit the breakpoint. They
941 * will recheck the current insn and restart if find_uprobe() fails.
942 * See find_active_uprobe().
944 static void delete_uprobe(struct uprobe
*uprobe
)
946 if (WARN_ON(!uprobe_is_active(uprobe
)))
949 spin_lock(&uprobes_treelock
);
950 rb_erase(&uprobe
->rb_node
, &uprobes_tree
);
951 spin_unlock(&uprobes_treelock
);
952 RB_CLEAR_NODE(&uprobe
->rb_node
); /* for uprobe_is_active() */
957 struct map_info
*next
;
958 struct mm_struct
*mm
;
962 static inline struct map_info
*free_map_info(struct map_info
*info
)
964 struct map_info
*next
= info
->next
;
969 static struct map_info
*
970 build_map_info(struct address_space
*mapping
, loff_t offset
, bool is_register
)
972 unsigned long pgoff
= offset
>> PAGE_SHIFT
;
973 struct vm_area_struct
*vma
;
974 struct map_info
*curr
= NULL
;
975 struct map_info
*prev
= NULL
;
976 struct map_info
*info
;
980 i_mmap_lock_read(mapping
);
981 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
, pgoff
, pgoff
) {
982 if (!valid_vma(vma
, is_register
))
985 if (!prev
&& !more
) {
987 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
988 * reclaim. This is optimistic, no harm done if it fails.
990 prev
= kmalloc(sizeof(struct map_info
),
991 GFP_NOWAIT
| __GFP_NOMEMALLOC
| __GFP_NOWARN
);
1000 if (!mmget_not_zero(vma
->vm_mm
))
1008 info
->mm
= vma
->vm_mm
;
1009 info
->vaddr
= offset_to_vaddr(vma
, offset
);
1011 i_mmap_unlock_read(mapping
);
1023 info
= kmalloc(sizeof(struct map_info
), GFP_KERNEL
);
1025 curr
= ERR_PTR(-ENOMEM
);
1035 prev
= free_map_info(prev
);
1040 register_for_each_vma(struct uprobe
*uprobe
, struct uprobe_consumer
*new)
1042 bool is_register
= !!new;
1043 struct map_info
*info
;
1046 percpu_down_write(&dup_mmap_sem
);
1047 info
= build_map_info(uprobe
->inode
->i_mapping
,
1048 uprobe
->offset
, is_register
);
1050 err
= PTR_ERR(info
);
1055 struct mm_struct
*mm
= info
->mm
;
1056 struct vm_area_struct
*vma
;
1058 if (err
&& is_register
)
1061 mmap_write_lock(mm
);
1062 vma
= find_vma(mm
, info
->vaddr
);
1063 if (!vma
|| !valid_vma(vma
, is_register
) ||
1064 file_inode(vma
->vm_file
) != uprobe
->inode
)
1067 if (vma
->vm_start
> info
->vaddr
||
1068 vaddr_to_offset(vma
, info
->vaddr
) != uprobe
->offset
)
1072 /* consult only the "caller", new consumer. */
1073 if (consumer_filter(new,
1074 UPROBE_FILTER_REGISTER
, mm
))
1075 err
= install_breakpoint(uprobe
, mm
, vma
, info
->vaddr
);
1076 } else if (test_bit(MMF_HAS_UPROBES
, &mm
->flags
)) {
1077 if (!filter_chain(uprobe
,
1078 UPROBE_FILTER_UNREGISTER
, mm
))
1079 err
|= remove_breakpoint(uprobe
, mm
, info
->vaddr
);
1083 mmap_write_unlock(mm
);
1086 info
= free_map_info(info
);
1089 percpu_up_write(&dup_mmap_sem
);
1094 __uprobe_unregister(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
1098 if (WARN_ON(!consumer_del(uprobe
, uc
)))
1101 err
= register_for_each_vma(uprobe
, NULL
);
1102 /* TODO : cant unregister? schedule a worker thread */
1103 if (!uprobe
->consumers
&& !err
)
1104 delete_uprobe(uprobe
);
1108 * uprobe_unregister - unregister an already registered probe.
1109 * @inode: the file in which the probe has to be removed.
1110 * @offset: offset from the start of the file.
1111 * @uc: identify which probe if multiple probes are colocated.
1113 void uprobe_unregister(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
1115 struct uprobe
*uprobe
;
1117 uprobe
= find_uprobe(inode
, offset
);
1118 if (WARN_ON(!uprobe
))
1121 down_write(&uprobe
->register_rwsem
);
1122 __uprobe_unregister(uprobe
, uc
);
1123 up_write(&uprobe
->register_rwsem
);
1126 EXPORT_SYMBOL_GPL(uprobe_unregister
);
1129 * __uprobe_register - register a probe
1130 * @inode: the file in which the probe has to be placed.
1131 * @offset: offset from the start of the file.
1132 * @uc: information on howto handle the probe..
1134 * Apart from the access refcount, __uprobe_register() takes a creation
1135 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
1136 * inserted into the rbtree (i.e first consumer for a @inode:@offset
1137 * tuple). Creation refcount stops uprobe_unregister from freeing the
1138 * @uprobe even before the register operation is complete. Creation
1139 * refcount is released when the last @uc for the @uprobe
1140 * unregisters. Caller of __uprobe_register() is required to keep @inode
1141 * (and the containing mount) referenced.
1143 * Return errno if it cannot successully install probes
1144 * else return 0 (success)
1146 static int __uprobe_register(struct inode
*inode
, loff_t offset
,
1147 loff_t ref_ctr_offset
, struct uprobe_consumer
*uc
)
1149 struct uprobe
*uprobe
;
1152 /* Uprobe must have at least one set consumer */
1153 if (!uc
->handler
&& !uc
->ret_handler
)
1156 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
1157 if (!inode
->i_mapping
->a_ops
->readpage
&& !shmem_mapping(inode
->i_mapping
))
1159 /* Racy, just to catch the obvious mistakes */
1160 if (offset
> i_size_read(inode
))
1164 uprobe
= alloc_uprobe(inode
, offset
, ref_ctr_offset
);
1168 return PTR_ERR(uprobe
);
1171 * We can race with uprobe_unregister()->delete_uprobe().
1172 * Check uprobe_is_active() and retry if it is false.
1174 down_write(&uprobe
->register_rwsem
);
1176 if (likely(uprobe_is_active(uprobe
))) {
1177 consumer_add(uprobe
, uc
);
1178 ret
= register_for_each_vma(uprobe
, uc
);
1180 __uprobe_unregister(uprobe
, uc
);
1182 up_write(&uprobe
->register_rwsem
);
1185 if (unlikely(ret
== -EAGAIN
))
1190 int uprobe_register(struct inode
*inode
, loff_t offset
,
1191 struct uprobe_consumer
*uc
)
1193 return __uprobe_register(inode
, offset
, 0, uc
);
1195 EXPORT_SYMBOL_GPL(uprobe_register
);
1197 int uprobe_register_refctr(struct inode
*inode
, loff_t offset
,
1198 loff_t ref_ctr_offset
, struct uprobe_consumer
*uc
)
1200 return __uprobe_register(inode
, offset
, ref_ctr_offset
, uc
);
1202 EXPORT_SYMBOL_GPL(uprobe_register_refctr
);
1205 * uprobe_apply - unregister an already registered probe.
1206 * @inode: the file in which the probe has to be removed.
1207 * @offset: offset from the start of the file.
1208 * @uc: consumer which wants to add more or remove some breakpoints
1209 * @add: add or remove the breakpoints
1211 int uprobe_apply(struct inode
*inode
, loff_t offset
,
1212 struct uprobe_consumer
*uc
, bool add
)
1214 struct uprobe
*uprobe
;
1215 struct uprobe_consumer
*con
;
1218 uprobe
= find_uprobe(inode
, offset
);
1219 if (WARN_ON(!uprobe
))
1222 down_write(&uprobe
->register_rwsem
);
1223 for (con
= uprobe
->consumers
; con
&& con
!= uc
; con
= con
->next
)
1226 ret
= register_for_each_vma(uprobe
, add
? uc
: NULL
);
1227 up_write(&uprobe
->register_rwsem
);
1233 static int unapply_uprobe(struct uprobe
*uprobe
, struct mm_struct
*mm
)
1235 struct vm_area_struct
*vma
;
1239 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1240 unsigned long vaddr
;
1243 if (!valid_vma(vma
, false) ||
1244 file_inode(vma
->vm_file
) != uprobe
->inode
)
1247 offset
= (loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
;
1248 if (uprobe
->offset
< offset
||
1249 uprobe
->offset
>= offset
+ vma
->vm_end
- vma
->vm_start
)
1252 vaddr
= offset_to_vaddr(vma
, uprobe
->offset
);
1253 err
|= remove_breakpoint(uprobe
, mm
, vaddr
);
1255 mmap_read_unlock(mm
);
1260 static struct rb_node
*
1261 find_node_in_range(struct inode
*inode
, loff_t min
, loff_t max
)
1263 struct rb_node
*n
= uprobes_tree
.rb_node
;
1266 struct uprobe
*u
= rb_entry(n
, struct uprobe
, rb_node
);
1268 if (inode
< u
->inode
) {
1270 } else if (inode
> u
->inode
) {
1273 if (max
< u
->offset
)
1275 else if (min
> u
->offset
)
1286 * For a given range in vma, build a list of probes that need to be inserted.
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
)
1294 struct rb_node
*n
, *t
;
1297 INIT_LIST_HEAD(head
);
1298 min
= vaddr_to_offset(vma
, start
);
1299 max
= min
+ (end
- start
) - 1;
1301 spin_lock(&uprobes_treelock
);
1302 n
= find_node_in_range(inode
, min
, max
);
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
)
1308 list_add(&u
->pending_list
, head
);
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
)
1315 list_add(&u
->pending_list
, head
);
1319 spin_unlock(&uprobes_treelock
);
1322 /* @vma contains reference counter, not the probed instruction. */
1323 static int delayed_ref_ctr_inc(struct vm_area_struct
*vma
)
1325 struct list_head
*pos
, *q
;
1326 struct delayed_uprobe
*du
;
1327 unsigned long vaddr
;
1328 int ret
= 0, err
= 0;
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
);
1334 if (du
->mm
!= vma
->vm_mm
||
1335 !valid_ref_ctr_vma(du
->uprobe
, vma
))
1338 vaddr
= offset_to_vaddr(vma
, du
->uprobe
->ref_ctr_offset
);
1339 ret
= __update_ref_ctr(vma
->vm_mm
, vaddr
, 1);
1341 update_ref_ctr_warn(du
->uprobe
, vma
->vm_mm
, 1);
1345 delayed_uprobe_delete(du
);
1347 mutex_unlock(&delayed_uprobe_lock
);
1352 * Called from mmap_region/vma_adjust with mm->mmap_lock acquired.
1354 * Currently we ignore all errors and always return 0, the callers
1355 * can't handle the failure anyway.
1357 int uprobe_mmap(struct vm_area_struct
*vma
)
1359 struct list_head tmp_list
;
1360 struct uprobe
*uprobe
, *u
;
1361 struct inode
*inode
;
1363 if (no_uprobe_events())
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
);
1371 if (!valid_vma(vma
, true))
1374 inode
= file_inode(vma
->vm_file
);
1378 mutex_lock(uprobes_mmap_hash(inode
));
1379 build_probe_list(inode
, vma
, vma
->vm_start
, vma
->vm_end
, &tmp_list
);
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.
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
);
1393 mutex_unlock(uprobes_mmap_hash(inode
));
1399 vma_has_uprobes(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1402 struct inode
*inode
;
1405 inode
= file_inode(vma
->vm_file
);
1407 min
= vaddr_to_offset(vma
, start
);
1408 max
= min
+ (end
- start
) - 1;
1410 spin_lock(&uprobes_treelock
);
1411 n
= find_node_in_range(inode
, min
, max
);
1412 spin_unlock(&uprobes_treelock
);
1418 * Called in context of a munmap of a vma.
1420 void uprobe_munmap(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1422 if (no_uprobe_events() || !valid_vma(vma
, false))
1425 if (!atomic_read(&vma
->vm_mm
->mm_users
)) /* called by mmput() ? */
1428 if (!test_bit(MMF_HAS_UPROBES
, &vma
->vm_mm
->flags
) ||
1429 test_bit(MMF_RECALC_UPROBES
, &vma
->vm_mm
->flags
))
1432 if (vma_has_uprobes(vma
, start
, end
))
1433 set_bit(MMF_RECALC_UPROBES
, &vma
->vm_mm
->flags
);
1436 /* Slot allocation for XOL */
1437 static int xol_add_vma(struct mm_struct
*mm
, struct xol_area
*area
)
1439 struct vm_area_struct
*vma
;
1442 if (mmap_write_lock_killable(mm
))
1445 if (mm
->uprobes_state
.xol_area
) {
1451 /* Try to map as high as possible, this is only a hint. */
1452 area
->vaddr
= get_unmapped_area(NULL
, TASK_SIZE
- PAGE_SIZE
,
1454 if (IS_ERR_VALUE(area
->vaddr
)) {
1460 vma
= _install_special_mapping(mm
, area
->vaddr
, PAGE_SIZE
,
1461 VM_EXEC
|VM_MAYEXEC
|VM_DONTCOPY
|VM_IO
,
1462 &area
->xol_mapping
);
1469 /* pairs with get_xol_area() */
1470 smp_store_release(&mm
->uprobes_state
.xol_area
, area
); /* ^^^ */
1472 mmap_write_unlock(mm
);
1477 static struct xol_area
*__create_xol_area(unsigned long vaddr
)
1479 struct mm_struct
*mm
= current
->mm
;
1480 uprobe_opcode_t insn
= UPROBE_SWBP_INSN
;
1481 struct xol_area
*area
;
1483 area
= kmalloc(sizeof(*area
), GFP_KERNEL
);
1484 if (unlikely(!area
))
1487 area
->bitmap
= kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE
), sizeof(long),
1492 area
->xol_mapping
.name
= "[uprobes]";
1493 area
->xol_mapping
.fault
= NULL
;
1494 area
->xol_mapping
.pages
= area
->pages
;
1495 area
->pages
[0] = alloc_page(GFP_HIGHUSER
);
1496 if (!area
->pages
[0])
1498 area
->pages
[1] = NULL
;
1500 area
->vaddr
= vaddr
;
1501 init_waitqueue_head(&area
->wq
);
1502 /* Reserve the 1st slot for get_trampoline_vaddr() */
1503 set_bit(0, area
->bitmap
);
1504 atomic_set(&area
->slot_count
, 1);
1505 arch_uprobe_copy_ixol(area
->pages
[0], 0, &insn
, UPROBE_SWBP_INSN_SIZE
);
1507 if (!xol_add_vma(mm
, area
))
1510 __free_page(area
->pages
[0]);
1512 kfree(area
->bitmap
);
1520 * get_xol_area - Allocate process's xol_area if necessary.
1521 * This area will be used for storing instructions for execution out of line.
1523 * Returns the allocated area or NULL.
1525 static struct xol_area
*get_xol_area(void)
1527 struct mm_struct
*mm
= current
->mm
;
1528 struct xol_area
*area
;
1530 if (!mm
->uprobes_state
.xol_area
)
1531 __create_xol_area(0);
1533 /* Pairs with xol_add_vma() smp_store_release() */
1534 area
= READ_ONCE(mm
->uprobes_state
.xol_area
); /* ^^^ */
1539 * uprobe_clear_state - Free the area allocated for slots.
1541 void uprobe_clear_state(struct mm_struct
*mm
)
1543 struct xol_area
*area
= mm
->uprobes_state
.xol_area
;
1545 mutex_lock(&delayed_uprobe_lock
);
1546 delayed_uprobe_remove(NULL
, mm
);
1547 mutex_unlock(&delayed_uprobe_lock
);
1552 put_page(area
->pages
[0]);
1553 kfree(area
->bitmap
);
1557 void uprobe_start_dup_mmap(void)
1559 percpu_down_read(&dup_mmap_sem
);
1562 void uprobe_end_dup_mmap(void)
1564 percpu_up_read(&dup_mmap_sem
);
1567 void uprobe_dup_mmap(struct mm_struct
*oldmm
, struct mm_struct
*newmm
)
1569 if (test_bit(MMF_HAS_UPROBES
, &oldmm
->flags
)) {
1570 set_bit(MMF_HAS_UPROBES
, &newmm
->flags
);
1571 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1572 set_bit(MMF_RECALC_UPROBES
, &newmm
->flags
);
1577 * - search for a free slot.
1579 static unsigned long xol_take_insn_slot(struct xol_area
*area
)
1581 unsigned long slot_addr
;
1585 slot_nr
= find_first_zero_bit(area
->bitmap
, UINSNS_PER_PAGE
);
1586 if (slot_nr
< UINSNS_PER_PAGE
) {
1587 if (!test_and_set_bit(slot_nr
, area
->bitmap
))
1590 slot_nr
= UINSNS_PER_PAGE
;
1593 wait_event(area
->wq
, (atomic_read(&area
->slot_count
) < UINSNS_PER_PAGE
));
1594 } while (slot_nr
>= UINSNS_PER_PAGE
);
1596 slot_addr
= area
->vaddr
+ (slot_nr
* UPROBE_XOL_SLOT_BYTES
);
1597 atomic_inc(&area
->slot_count
);
1603 * xol_get_insn_slot - allocate a slot for xol.
1604 * Returns the allocated slot address or 0.
1606 static unsigned long xol_get_insn_slot(struct uprobe
*uprobe
)
1608 struct xol_area
*area
;
1609 unsigned long xol_vaddr
;
1611 area
= get_xol_area();
1615 xol_vaddr
= xol_take_insn_slot(area
);
1616 if (unlikely(!xol_vaddr
))
1619 arch_uprobe_copy_ixol(area
->pages
[0], xol_vaddr
,
1620 &uprobe
->arch
.ixol
, sizeof(uprobe
->arch
.ixol
));
1626 * xol_free_insn_slot - If slot was earlier allocated by
1627 * @xol_get_insn_slot(), make the slot available for
1628 * subsequent requests.
1630 static void xol_free_insn_slot(struct task_struct
*tsk
)
1632 struct xol_area
*area
;
1633 unsigned long vma_end
;
1634 unsigned long slot_addr
;
1636 if (!tsk
->mm
|| !tsk
->mm
->uprobes_state
.xol_area
|| !tsk
->utask
)
1639 slot_addr
= tsk
->utask
->xol_vaddr
;
1640 if (unlikely(!slot_addr
))
1643 area
= tsk
->mm
->uprobes_state
.xol_area
;
1644 vma_end
= area
->vaddr
+ PAGE_SIZE
;
1645 if (area
->vaddr
<= slot_addr
&& slot_addr
< vma_end
) {
1646 unsigned long offset
;
1649 offset
= slot_addr
- area
->vaddr
;
1650 slot_nr
= offset
/ UPROBE_XOL_SLOT_BYTES
;
1651 if (slot_nr
>= UINSNS_PER_PAGE
)
1654 clear_bit(slot_nr
, area
->bitmap
);
1655 atomic_dec(&area
->slot_count
);
1656 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1657 if (waitqueue_active(&area
->wq
))
1660 tsk
->utask
->xol_vaddr
= 0;
1664 void __weak
arch_uprobe_copy_ixol(struct page
*page
, unsigned long vaddr
,
1665 void *src
, unsigned long len
)
1667 /* Initialize the slot */
1668 copy_to_page(page
, vaddr
, src
, len
);
1671 * We probably need flush_icache_user_page() but it needs vma.
1672 * This should work on most of architectures by default. If
1673 * architecture needs to do something different it can define
1674 * its own version of the function.
1676 flush_dcache_page(page
);
1680 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1681 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1683 * Return the address of the breakpoint instruction.
1685 unsigned long __weak
uprobe_get_swbp_addr(struct pt_regs
*regs
)
1687 return instruction_pointer(regs
) - UPROBE_SWBP_INSN_SIZE
;
1690 unsigned long uprobe_get_trap_addr(struct pt_regs
*regs
)
1692 struct uprobe_task
*utask
= current
->utask
;
1694 if (unlikely(utask
&& utask
->active_uprobe
))
1695 return utask
->vaddr
;
1697 return instruction_pointer(regs
);
1700 static struct return_instance
*free_ret_instance(struct return_instance
*ri
)
1702 struct return_instance
*next
= ri
->next
;
1703 put_uprobe(ri
->uprobe
);
1709 * Called with no locks held.
1710 * Called in context of an exiting or an exec-ing thread.
1712 void uprobe_free_utask(struct task_struct
*t
)
1714 struct uprobe_task
*utask
= t
->utask
;
1715 struct return_instance
*ri
;
1720 if (utask
->active_uprobe
)
1721 put_uprobe(utask
->active_uprobe
);
1723 ri
= utask
->return_instances
;
1725 ri
= free_ret_instance(ri
);
1727 xol_free_insn_slot(t
);
1733 * Allocate a uprobe_task object for the task if if necessary.
1734 * Called when the thread hits a breakpoint.
1737 * - pointer to new uprobe_task on success
1740 static struct uprobe_task
*get_utask(void)
1742 if (!current
->utask
)
1743 current
->utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1744 return current
->utask
;
1747 static int dup_utask(struct task_struct
*t
, struct uprobe_task
*o_utask
)
1749 struct uprobe_task
*n_utask
;
1750 struct return_instance
**p
, *o
, *n
;
1752 n_utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1757 p
= &n_utask
->return_instances
;
1758 for (o
= o_utask
->return_instances
; o
; o
= o
->next
) {
1759 n
= kmalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1764 get_uprobe(n
->uprobe
);
1775 static void uprobe_warn(struct task_struct
*t
, const char *msg
)
1777 pr_warn("uprobe: %s:%d failed to %s\n",
1778 current
->comm
, current
->pid
, msg
);
1781 static void dup_xol_work(struct callback_head
*work
)
1783 if (current
->flags
& PF_EXITING
)
1786 if (!__create_xol_area(current
->utask
->dup_xol_addr
) &&
1787 !fatal_signal_pending(current
))
1788 uprobe_warn(current
, "dup xol area");
1792 * Called in context of a new clone/fork from copy_process.
1794 void uprobe_copy_process(struct task_struct
*t
, unsigned long flags
)
1796 struct uprobe_task
*utask
= current
->utask
;
1797 struct mm_struct
*mm
= current
->mm
;
1798 struct xol_area
*area
;
1802 if (!utask
|| !utask
->return_instances
)
1805 if (mm
== t
->mm
&& !(flags
& CLONE_VFORK
))
1808 if (dup_utask(t
, utask
))
1809 return uprobe_warn(t
, "dup ret instances");
1811 /* The task can fork() after dup_xol_work() fails */
1812 area
= mm
->uprobes_state
.xol_area
;
1814 return uprobe_warn(t
, "dup xol area");
1819 t
->utask
->dup_xol_addr
= area
->vaddr
;
1820 init_task_work(&t
->utask
->dup_xol_work
, dup_xol_work
);
1821 task_work_add(t
, &t
->utask
->dup_xol_work
, true);
1825 * Current area->vaddr notion assume the trampoline address is always
1826 * equal area->vaddr.
1828 * Returns -1 in case the xol_area is not allocated.
1830 static unsigned long get_trampoline_vaddr(void)
1832 struct xol_area
*area
;
1833 unsigned long trampoline_vaddr
= -1;
1835 /* Pairs with xol_add_vma() smp_store_release() */
1836 area
= READ_ONCE(current
->mm
->uprobes_state
.xol_area
); /* ^^^ */
1838 trampoline_vaddr
= area
->vaddr
;
1840 return trampoline_vaddr
;
1843 static void cleanup_return_instances(struct uprobe_task
*utask
, bool chained
,
1844 struct pt_regs
*regs
)
1846 struct return_instance
*ri
= utask
->return_instances
;
1847 enum rp_check ctx
= chained
? RP_CHECK_CHAIN_CALL
: RP_CHECK_CALL
;
1849 while (ri
&& !arch_uretprobe_is_alive(ri
, ctx
, regs
)) {
1850 ri
= free_ret_instance(ri
);
1853 utask
->return_instances
= ri
;
1856 static void prepare_uretprobe(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1858 struct return_instance
*ri
;
1859 struct uprobe_task
*utask
;
1860 unsigned long orig_ret_vaddr
, trampoline_vaddr
;
1863 if (!get_xol_area())
1866 utask
= get_utask();
1870 if (utask
->depth
>= MAX_URETPROBE_DEPTH
) {
1871 printk_ratelimited(KERN_INFO
"uprobe: omit uretprobe due to"
1872 " nestedness limit pid/tgid=%d/%d\n",
1873 current
->pid
, current
->tgid
);
1877 ri
= kmalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1881 trampoline_vaddr
= get_trampoline_vaddr();
1882 orig_ret_vaddr
= arch_uretprobe_hijack_return_addr(trampoline_vaddr
, regs
);
1883 if (orig_ret_vaddr
== -1)
1886 /* drop the entries invalidated by longjmp() */
1887 chained
= (orig_ret_vaddr
== trampoline_vaddr
);
1888 cleanup_return_instances(utask
, chained
, regs
);
1891 * We don't want to keep trampoline address in stack, rather keep the
1892 * original return address of first caller thru all the consequent
1893 * instances. This also makes breakpoint unwrapping easier.
1896 if (!utask
->return_instances
) {
1898 * This situation is not possible. Likely we have an
1899 * attack from user-space.
1901 uprobe_warn(current
, "handle tail call");
1904 orig_ret_vaddr
= utask
->return_instances
->orig_ret_vaddr
;
1907 ri
->uprobe
= get_uprobe(uprobe
);
1908 ri
->func
= instruction_pointer(regs
);
1909 ri
->stack
= user_stack_pointer(regs
);
1910 ri
->orig_ret_vaddr
= orig_ret_vaddr
;
1911 ri
->chained
= chained
;
1914 ri
->next
= utask
->return_instances
;
1915 utask
->return_instances
= ri
;
1922 /* Prepare to single-step probed instruction out of line. */
1924 pre_ssout(struct uprobe
*uprobe
, struct pt_regs
*regs
, unsigned long bp_vaddr
)
1926 struct uprobe_task
*utask
;
1927 unsigned long xol_vaddr
;
1930 utask
= get_utask();
1934 xol_vaddr
= xol_get_insn_slot(uprobe
);
1938 utask
->xol_vaddr
= xol_vaddr
;
1939 utask
->vaddr
= bp_vaddr
;
1941 err
= arch_uprobe_pre_xol(&uprobe
->arch
, regs
);
1942 if (unlikely(err
)) {
1943 xol_free_insn_slot(current
);
1947 utask
->active_uprobe
= uprobe
;
1948 utask
->state
= UTASK_SSTEP
;
1953 * If we are singlestepping, then ensure this thread is not connected to
1954 * non-fatal signals until completion of singlestep. When xol insn itself
1955 * triggers the signal, restart the original insn even if the task is
1956 * already SIGKILL'ed (since coredump should report the correct ip). This
1957 * is even more important if the task has a handler for SIGSEGV/etc, The
1958 * _same_ instruction should be repeated again after return from the signal
1959 * handler, and SSTEP can never finish in this case.
1961 bool uprobe_deny_signal(void)
1963 struct task_struct
*t
= current
;
1964 struct uprobe_task
*utask
= t
->utask
;
1966 if (likely(!utask
|| !utask
->active_uprobe
))
1969 WARN_ON_ONCE(utask
->state
!= UTASK_SSTEP
);
1971 if (signal_pending(t
)) {
1972 spin_lock_irq(&t
->sighand
->siglock
);
1973 clear_tsk_thread_flag(t
, TIF_SIGPENDING
);
1974 spin_unlock_irq(&t
->sighand
->siglock
);
1976 if (__fatal_signal_pending(t
) || arch_uprobe_xol_was_trapped(t
)) {
1977 utask
->state
= UTASK_SSTEP_TRAPPED
;
1978 set_tsk_thread_flag(t
, TIF_UPROBE
);
1985 static void mmf_recalc_uprobes(struct mm_struct
*mm
)
1987 struct vm_area_struct
*vma
;
1989 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1990 if (!valid_vma(vma
, false))
1993 * This is not strictly accurate, we can race with
1994 * uprobe_unregister() and see the already removed
1995 * uprobe if delete_uprobe() was not yet called.
1996 * Or this uprobe can be filtered out.
1998 if (vma_has_uprobes(vma
, vma
->vm_start
, vma
->vm_end
))
2002 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
2005 static int is_trap_at_addr(struct mm_struct
*mm
, unsigned long vaddr
)
2008 uprobe_opcode_t opcode
;
2011 pagefault_disable();
2012 result
= __get_user(opcode
, (uprobe_opcode_t __user
*)vaddr
);
2015 if (likely(result
== 0))
2019 * The NULL 'tsk' here ensures that any faults that occur here
2020 * will not be accounted to the task. 'mm' *is* current->mm,
2021 * but we treat this as a 'remote' access since it is
2022 * essentially a kernel access to the memory.
2024 result
= get_user_pages_remote(NULL
, mm
, vaddr
, 1, FOLL_FORCE
, &page
,
2029 copy_from_page(page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
2032 /* This needs to return true for any variant of the trap insn */
2033 return is_trap_insn(&opcode
);
2036 static struct uprobe
*find_active_uprobe(unsigned long bp_vaddr
, int *is_swbp
)
2038 struct mm_struct
*mm
= current
->mm
;
2039 struct uprobe
*uprobe
= NULL
;
2040 struct vm_area_struct
*vma
;
2043 vma
= find_vma(mm
, bp_vaddr
);
2044 if (vma
&& vma
->vm_start
<= bp_vaddr
) {
2045 if (valid_vma(vma
, false)) {
2046 struct inode
*inode
= file_inode(vma
->vm_file
);
2047 loff_t offset
= vaddr_to_offset(vma
, bp_vaddr
);
2049 uprobe
= find_uprobe(inode
, offset
);
2053 *is_swbp
= is_trap_at_addr(mm
, bp_vaddr
);
2058 if (!uprobe
&& test_and_clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
))
2059 mmf_recalc_uprobes(mm
);
2060 mmap_read_unlock(mm
);
2065 static void handler_chain(struct uprobe
*uprobe
, struct pt_regs
*regs
)
2067 struct uprobe_consumer
*uc
;
2068 int remove
= UPROBE_HANDLER_REMOVE
;
2069 bool need_prep
= false; /* prepare return uprobe, when needed */
2071 down_read(&uprobe
->register_rwsem
);
2072 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
2076 rc
= uc
->handler(uc
, regs
);
2077 WARN(rc
& ~UPROBE_HANDLER_MASK
,
2078 "bad rc=0x%x from %ps()\n", rc
, uc
->handler
);
2081 if (uc
->ret_handler
)
2087 if (need_prep
&& !remove
)
2088 prepare_uretprobe(uprobe
, regs
); /* put bp at return */
2090 if (remove
&& uprobe
->consumers
) {
2091 WARN_ON(!uprobe_is_active(uprobe
));
2092 unapply_uprobe(uprobe
, current
->mm
);
2094 up_read(&uprobe
->register_rwsem
);
2098 handle_uretprobe_chain(struct return_instance
*ri
, struct pt_regs
*regs
)
2100 struct uprobe
*uprobe
= ri
->uprobe
;
2101 struct uprobe_consumer
*uc
;
2103 down_read(&uprobe
->register_rwsem
);
2104 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
2105 if (uc
->ret_handler
)
2106 uc
->ret_handler(uc
, ri
->func
, regs
);
2108 up_read(&uprobe
->register_rwsem
);
2111 static struct return_instance
*find_next_ret_chain(struct return_instance
*ri
)
2116 chained
= ri
->chained
;
2117 ri
= ri
->next
; /* can't be NULL if chained */
2123 static void handle_trampoline(struct pt_regs
*regs
)
2125 struct uprobe_task
*utask
;
2126 struct return_instance
*ri
, *next
;
2129 utask
= current
->utask
;
2133 ri
= utask
->return_instances
;
2139 * We should throw out the frames invalidated by longjmp().
2140 * If this chain is valid, then the next one should be alive
2141 * or NULL; the latter case means that nobody but ri->func
2142 * could hit this trampoline on return. TODO: sigaltstack().
2144 next
= find_next_ret_chain(ri
);
2145 valid
= !next
|| arch_uretprobe_is_alive(next
, RP_CHECK_RET
, regs
);
2147 instruction_pointer_set(regs
, ri
->orig_ret_vaddr
);
2150 handle_uretprobe_chain(ri
, regs
);
2151 ri
= free_ret_instance(ri
);
2153 } while (ri
!= next
);
2156 utask
->return_instances
= ri
;
2160 uprobe_warn(current
, "handle uretprobe, sending SIGILL.");
2165 bool __weak
arch_uprobe_ignore(struct arch_uprobe
*aup
, struct pt_regs
*regs
)
2170 bool __weak
arch_uretprobe_is_alive(struct return_instance
*ret
, enum rp_check ctx
,
2171 struct pt_regs
*regs
)
2177 * Run handler and ask thread to singlestep.
2178 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2180 static void handle_swbp(struct pt_regs
*regs
)
2182 struct uprobe
*uprobe
;
2183 unsigned long bp_vaddr
;
2184 int uninitialized_var(is_swbp
);
2186 bp_vaddr
= uprobe_get_swbp_addr(regs
);
2187 if (bp_vaddr
== get_trampoline_vaddr())
2188 return handle_trampoline(regs
);
2190 uprobe
= find_active_uprobe(bp_vaddr
, &is_swbp
);
2193 /* No matching uprobe; signal SIGTRAP. */
2194 send_sig(SIGTRAP
, current
, 0);
2197 * Either we raced with uprobe_unregister() or we can't
2198 * access this memory. The latter is only possible if
2199 * another thread plays with our ->mm. In both cases
2200 * we can simply restart. If this vma was unmapped we
2201 * can pretend this insn was not executed yet and get
2202 * the (correct) SIGSEGV after restart.
2204 instruction_pointer_set(regs
, bp_vaddr
);
2209 /* change it in advance for ->handler() and restart */
2210 instruction_pointer_set(regs
, bp_vaddr
);
2213 * TODO: move copy_insn/etc into _register and remove this hack.
2214 * After we hit the bp, _unregister + _register can install the
2215 * new and not-yet-analyzed uprobe at the same address, restart.
2217 if (unlikely(!test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
)))
2221 * Pairs with the smp_wmb() in prepare_uprobe().
2223 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2224 * we must also see the stores to &uprobe->arch performed by the
2225 * prepare_uprobe() call.
2229 /* Tracing handlers use ->utask to communicate with fetch methods */
2233 if (arch_uprobe_ignore(&uprobe
->arch
, regs
))
2236 handler_chain(uprobe
, regs
);
2238 if (arch_uprobe_skip_sstep(&uprobe
->arch
, regs
))
2241 if (!pre_ssout(uprobe
, regs
, bp_vaddr
))
2244 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2250 * Perform required fix-ups and disable singlestep.
2251 * Allow pending signals to take effect.
2253 static void handle_singlestep(struct uprobe_task
*utask
, struct pt_regs
*regs
)
2255 struct uprobe
*uprobe
;
2258 uprobe
= utask
->active_uprobe
;
2259 if (utask
->state
== UTASK_SSTEP_ACK
)
2260 err
= arch_uprobe_post_xol(&uprobe
->arch
, regs
);
2261 else if (utask
->state
== UTASK_SSTEP_TRAPPED
)
2262 arch_uprobe_abort_xol(&uprobe
->arch
, regs
);
2267 utask
->active_uprobe
= NULL
;
2268 utask
->state
= UTASK_RUNNING
;
2269 xol_free_insn_slot(current
);
2271 spin_lock_irq(¤t
->sighand
->siglock
);
2272 recalc_sigpending(); /* see uprobe_deny_signal() */
2273 spin_unlock_irq(¤t
->sighand
->siglock
);
2275 if (unlikely(err
)) {
2276 uprobe_warn(current
, "execute the probed insn, sending SIGILL.");
2282 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2283 * allows the thread to return from interrupt. After that handle_swbp()
2284 * sets utask->active_uprobe.
2286 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2287 * and allows the thread to return from interrupt.
2289 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2290 * uprobe_notify_resume().
2292 void uprobe_notify_resume(struct pt_regs
*regs
)
2294 struct uprobe_task
*utask
;
2296 clear_thread_flag(TIF_UPROBE
);
2298 utask
= current
->utask
;
2299 if (utask
&& utask
->active_uprobe
)
2300 handle_singlestep(utask
, regs
);
2306 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2307 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2309 int uprobe_pre_sstep_notifier(struct pt_regs
*regs
)
2314 if (!test_bit(MMF_HAS_UPROBES
, ¤t
->mm
->flags
) &&
2315 (!current
->utask
|| !current
->utask
->return_instances
))
2318 set_thread_flag(TIF_UPROBE
);
2323 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2324 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2326 int uprobe_post_sstep_notifier(struct pt_regs
*regs
)
2328 struct uprobe_task
*utask
= current
->utask
;
2330 if (!current
->mm
|| !utask
|| !utask
->active_uprobe
)
2331 /* task is currently not uprobed */
2334 utask
->state
= UTASK_SSTEP_ACK
;
2335 set_thread_flag(TIF_UPROBE
);
2339 static struct notifier_block uprobe_exception_nb
= {
2340 .notifier_call
= arch_uprobe_exception_notify
,
2341 .priority
= INT_MAX
-1, /* notified after kprobes, kgdb */
2344 void __init
uprobes_init(void)
2348 for (i
= 0; i
< UPROBES_HASH_SZ
; i
++)
2349 mutex_init(&uprobes_mmap_mutex
[i
]);
2351 BUG_ON(register_die_notifier(&uprobe_exception_nb
));