2 * User-space Probes (UProbes)
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright (C) IBM Corporation, 2008-2012
22 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
25 #include <linux/kernel.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h> /* read_mapping_page */
28 #include <linux/slab.h>
29 #include <linux/sched.h>
30 #include <linux/export.h>
31 #include <linux/rmap.h> /* anon_vma_prepare */
32 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
33 #include <linux/swap.h> /* try_to_free_swap */
34 #include <linux/ptrace.h> /* user_enable_single_step */
35 #include <linux/kdebug.h> /* notifier mechanism */
36 #include "../../mm/internal.h" /* munlock_vma_page */
37 #include <linux/percpu-rwsem.h>
38 #include <linux/task_work.h>
40 #include <linux/uprobes.h>
42 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
43 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
45 static struct rb_root uprobes_tree
= RB_ROOT
;
47 * allows us to skip the uprobe_mmap if there are no uprobe events active
48 * at this time. Probably a fine grained per inode count is better?
50 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
52 static DEFINE_SPINLOCK(uprobes_treelock
); /* serialize rbtree access */
54 #define UPROBES_HASH_SZ 13
55 /* serialize uprobe->pending_list */
56 static struct mutex uprobes_mmap_mutex
[UPROBES_HASH_SZ
];
57 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
59 static struct percpu_rw_semaphore dup_mmap_sem
;
61 /* Have a copy of original instruction */
62 #define UPROBE_COPY_INSN 0
63 /* Can skip singlestep */
64 #define UPROBE_SKIP_SSTEP 1
67 struct rb_node rb_node
; /* node in the rb tree */
69 struct rw_semaphore register_rwsem
;
70 struct rw_semaphore consumer_rwsem
;
71 struct list_head pending_list
;
72 struct uprobe_consumer
*consumers
;
73 struct inode
*inode
; /* Also hold a ref to inode */
76 struct arch_uprobe arch
;
79 struct return_instance
{
80 struct uprobe
*uprobe
;
82 unsigned long orig_ret_vaddr
; /* original return address */
83 bool chained
; /* true, if instance is nested */
85 struct return_instance
*next
; /* keep as stack */
89 * valid_vma: Verify if the specified vma is an executable vma
90 * Relax restrictions while unregistering: vm_flags might have
91 * changed after breakpoint was inserted.
92 * - is_register: indicates if we are in register context.
93 * - Return 1 if the specified virtual address is in an
96 static bool valid_vma(struct vm_area_struct
*vma
, bool is_register
)
98 vm_flags_t flags
= VM_HUGETLB
| VM_MAYEXEC
| VM_SHARED
;
103 return vma
->vm_file
&& (vma
->vm_flags
& flags
) == VM_MAYEXEC
;
106 static unsigned long offset_to_vaddr(struct vm_area_struct
*vma
, loff_t offset
)
108 return vma
->vm_start
+ offset
- ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
111 static loff_t
vaddr_to_offset(struct vm_area_struct
*vma
, unsigned long vaddr
)
113 return ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
) + (vaddr
- vma
->vm_start
);
117 * __replace_page - replace page in vma by new page.
118 * based on replace_page in mm/ksm.c
120 * @vma: vma that holds the pte pointing to page
121 * @addr: address the old @page is mapped at
122 * @page: the cowed page we are replacing by kpage
123 * @kpage: the modified page we replace page by
125 * Returns 0 on success, -EFAULT on failure.
127 static int __replace_page(struct vm_area_struct
*vma
, unsigned long addr
,
128 struct page
*page
, struct page
*kpage
)
130 struct mm_struct
*mm
= vma
->vm_mm
;
134 /* For mmu_notifiers */
135 const unsigned long mmun_start
= addr
;
136 const unsigned long mmun_end
= addr
+ PAGE_SIZE
;
138 /* For try_to_free_swap() and munlock_vma_page() below */
141 mmu_notifier_invalidate_range_start(mm
, mmun_start
, mmun_end
);
143 ptep
= page_check_address(page
, mm
, addr
, &ptl
, 0);
148 page_add_new_anon_rmap(kpage
, vma
, addr
);
150 if (!PageAnon(page
)) {
151 dec_mm_counter(mm
, MM_FILEPAGES
);
152 inc_mm_counter(mm
, MM_ANONPAGES
);
155 flush_cache_page(vma
, addr
, pte_pfn(*ptep
));
156 ptep_clear_flush(vma
, addr
, ptep
);
157 set_pte_at_notify(mm
, addr
, ptep
, mk_pte(kpage
, vma
->vm_page_prot
));
159 page_remove_rmap(page
);
160 if (!page_mapped(page
))
161 try_to_free_swap(page
);
162 pte_unmap_unlock(ptep
, ptl
);
164 if (vma
->vm_flags
& VM_LOCKED
)
165 munlock_vma_page(page
);
170 mmu_notifier_invalidate_range_end(mm
, mmun_start
, mmun_end
);
176 * is_swbp_insn - check if instruction is breakpoint instruction.
177 * @insn: instruction to be checked.
178 * Default implementation of is_swbp_insn
179 * Returns true if @insn is a breakpoint instruction.
181 bool __weak
is_swbp_insn(uprobe_opcode_t
*insn
)
183 return *insn
== UPROBE_SWBP_INSN
;
187 * is_trap_insn - check if instruction is breakpoint instruction.
188 * @insn: instruction to be checked.
189 * Default implementation of is_trap_insn
190 * Returns true if @insn is a breakpoint instruction.
192 * This function is needed for the case where an architecture has multiple
193 * trap instructions (like powerpc).
195 bool __weak
is_trap_insn(uprobe_opcode_t
*insn
)
197 return is_swbp_insn(insn
);
200 static void copy_from_page(struct page
*page
, unsigned long vaddr
, void *dst
, int len
)
202 void *kaddr
= kmap_atomic(page
);
203 memcpy(dst
, kaddr
+ (vaddr
& ~PAGE_MASK
), len
);
204 kunmap_atomic(kaddr
);
207 static void copy_to_page(struct page
*page
, unsigned long vaddr
, const void *src
, int len
)
209 void *kaddr
= kmap_atomic(page
);
210 memcpy(kaddr
+ (vaddr
& ~PAGE_MASK
), src
, len
);
211 kunmap_atomic(kaddr
);
214 static int verify_opcode(struct page
*page
, unsigned long vaddr
, uprobe_opcode_t
*new_opcode
)
216 uprobe_opcode_t old_opcode
;
220 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
221 * We do not check if it is any other 'trap variant' which could
222 * be conditional trap instruction such as the one powerpc supports.
224 * The logic is that we do not care if the underlying instruction
225 * is a trap variant; uprobes always wins over any other (gdb)
228 copy_from_page(page
, vaddr
, &old_opcode
, UPROBE_SWBP_INSN_SIZE
);
229 is_swbp
= is_swbp_insn(&old_opcode
);
231 if (is_swbp_insn(new_opcode
)) {
232 if (is_swbp
) /* register: already installed? */
235 if (!is_swbp
) /* unregister: was it changed by us? */
244 * Expect the breakpoint instruction to be the smallest size instruction for
245 * the architecture. If an arch has variable length instruction and the
246 * breakpoint instruction is not of the smallest length instruction
247 * supported by that architecture then we need to modify is_trap_at_addr and
248 * uprobe_write_opcode accordingly. This would never be a problem for archs
249 * that have fixed length instructions.
253 * uprobe_write_opcode - write the opcode at a given virtual address.
254 * @mm: the probed process address space.
255 * @vaddr: the virtual address to store the opcode.
256 * @opcode: opcode to be written at @vaddr.
258 * Called with mm->mmap_sem held (for read and with a reference to
261 * For mm @mm, write the opcode at @vaddr.
262 * Return 0 (success) or a negative errno.
264 int uprobe_write_opcode(struct mm_struct
*mm
, unsigned long vaddr
,
265 uprobe_opcode_t opcode
)
267 struct page
*old_page
, *new_page
;
268 struct vm_area_struct
*vma
;
272 /* Read the page with vaddr into memory */
273 ret
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &old_page
, &vma
);
277 ret
= verify_opcode(old_page
, vaddr
, &opcode
);
282 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, vaddr
);
286 __SetPageUptodate(new_page
);
288 copy_highpage(new_page
, old_page
);
289 copy_to_page(new_page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
291 ret
= anon_vma_prepare(vma
);
295 ret
= __replace_page(vma
, vaddr
, old_page
, new_page
);
298 page_cache_release(new_page
);
302 if (unlikely(ret
== -EAGAIN
))
308 * set_swbp - store breakpoint at a given address.
309 * @auprobe: arch specific probepoint information.
310 * @mm: the probed process address space.
311 * @vaddr: the virtual address to insert the opcode.
313 * For mm @mm, store the breakpoint instruction at @vaddr.
314 * Return 0 (success) or a negative errno.
316 int __weak
set_swbp(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
318 return uprobe_write_opcode(mm
, vaddr
, UPROBE_SWBP_INSN
);
322 * set_orig_insn - Restore the original instruction.
323 * @mm: the probed process address space.
324 * @auprobe: arch specific probepoint information.
325 * @vaddr: the virtual address to insert the opcode.
327 * For mm @mm, restore the original opcode (opcode) at @vaddr.
328 * Return 0 (success) or a negative errno.
331 set_orig_insn(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
333 return uprobe_write_opcode(mm
, vaddr
, *(uprobe_opcode_t
*)auprobe
->insn
);
336 static int match_uprobe(struct uprobe
*l
, struct uprobe
*r
)
338 if (l
->inode
< r
->inode
)
341 if (l
->inode
> r
->inode
)
344 if (l
->offset
< r
->offset
)
347 if (l
->offset
> r
->offset
)
353 static struct uprobe
*__find_uprobe(struct inode
*inode
, loff_t offset
)
355 struct uprobe u
= { .inode
= inode
, .offset
= offset
};
356 struct rb_node
*n
= uprobes_tree
.rb_node
;
357 struct uprobe
*uprobe
;
361 uprobe
= rb_entry(n
, struct uprobe
, rb_node
);
362 match
= match_uprobe(&u
, uprobe
);
364 atomic_inc(&uprobe
->ref
);
377 * Find a uprobe corresponding to a given inode:offset
378 * Acquires uprobes_treelock
380 static struct uprobe
*find_uprobe(struct inode
*inode
, loff_t offset
)
382 struct uprobe
*uprobe
;
384 spin_lock(&uprobes_treelock
);
385 uprobe
= __find_uprobe(inode
, offset
);
386 spin_unlock(&uprobes_treelock
);
391 static struct uprobe
*__insert_uprobe(struct uprobe
*uprobe
)
393 struct rb_node
**p
= &uprobes_tree
.rb_node
;
394 struct rb_node
*parent
= NULL
;
400 u
= rb_entry(parent
, struct uprobe
, rb_node
);
401 match
= match_uprobe(uprobe
, u
);
408 p
= &parent
->rb_left
;
410 p
= &parent
->rb_right
;
415 rb_link_node(&uprobe
->rb_node
, parent
, p
);
416 rb_insert_color(&uprobe
->rb_node
, &uprobes_tree
);
417 /* get access + creation ref */
418 atomic_set(&uprobe
->ref
, 2);
424 * Acquire uprobes_treelock.
425 * Matching uprobe already exists in rbtree;
426 * increment (access refcount) and return the matching uprobe.
428 * No matching uprobe; insert the uprobe in rb_tree;
429 * get a double refcount (access + creation) and return NULL.
431 static struct uprobe
*insert_uprobe(struct uprobe
*uprobe
)
435 spin_lock(&uprobes_treelock
);
436 u
= __insert_uprobe(uprobe
);
437 spin_unlock(&uprobes_treelock
);
442 static void put_uprobe(struct uprobe
*uprobe
)
444 if (atomic_dec_and_test(&uprobe
->ref
))
448 static struct uprobe
*alloc_uprobe(struct inode
*inode
, loff_t offset
)
450 struct uprobe
*uprobe
, *cur_uprobe
;
452 uprobe
= kzalloc(sizeof(struct uprobe
), GFP_KERNEL
);
456 uprobe
->inode
= igrab(inode
);
457 uprobe
->offset
= offset
;
458 init_rwsem(&uprobe
->register_rwsem
);
459 init_rwsem(&uprobe
->consumer_rwsem
);
460 /* For now assume that the instruction need not be single-stepped */
461 __set_bit(UPROBE_SKIP_SSTEP
, &uprobe
->flags
);
463 /* add to uprobes_tree, sorted on inode:offset */
464 cur_uprobe
= insert_uprobe(uprobe
);
466 /* a uprobe exists for this inode:offset combination */
476 static void consumer_add(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
478 down_write(&uprobe
->consumer_rwsem
);
479 uc
->next
= uprobe
->consumers
;
480 uprobe
->consumers
= uc
;
481 up_write(&uprobe
->consumer_rwsem
);
485 * For uprobe @uprobe, delete the consumer @uc.
486 * Return true if the @uc is deleted successfully
489 static bool consumer_del(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
491 struct uprobe_consumer
**con
;
494 down_write(&uprobe
->consumer_rwsem
);
495 for (con
= &uprobe
->consumers
; *con
; con
= &(*con
)->next
) {
502 up_write(&uprobe
->consumer_rwsem
);
507 static int __copy_insn(struct address_space
*mapping
, struct file
*filp
,
508 void *insn
, int nbytes
, loff_t offset
)
512 if (!mapping
->a_ops
->readpage
)
515 * Ensure that the page that has the original instruction is
516 * populated and in page-cache.
518 page
= read_mapping_page(mapping
, offset
>> PAGE_CACHE_SHIFT
, filp
);
520 return PTR_ERR(page
);
522 copy_from_page(page
, offset
, insn
, nbytes
);
523 page_cache_release(page
);
528 static int copy_insn(struct uprobe
*uprobe
, struct file
*filp
)
530 struct address_space
*mapping
= uprobe
->inode
->i_mapping
;
531 loff_t offs
= uprobe
->offset
;
532 void *insn
= uprobe
->arch
.insn
;
533 int size
= MAX_UINSN_BYTES
;
536 /* Copy only available bytes, -EIO if nothing was read */
538 if (offs
>= i_size_read(uprobe
->inode
))
541 len
= min_t(int, size
, PAGE_SIZE
- (offs
& ~PAGE_MASK
));
542 err
= __copy_insn(mapping
, filp
, insn
, len
, offs
);
554 static int prepare_uprobe(struct uprobe
*uprobe
, struct file
*file
,
555 struct mm_struct
*mm
, unsigned long vaddr
)
559 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
562 /* TODO: move this into _register, until then we abuse this sem. */
563 down_write(&uprobe
->consumer_rwsem
);
564 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
567 ret
= copy_insn(uprobe
, file
);
572 if (is_trap_insn((uprobe_opcode_t
*)uprobe
->arch
.insn
))
575 ret
= arch_uprobe_analyze_insn(&uprobe
->arch
, mm
, vaddr
);
579 /* uprobe_write_opcode() assumes we don't cross page boundary */
580 BUG_ON((uprobe
->offset
& ~PAGE_MASK
) +
581 UPROBE_SWBP_INSN_SIZE
> PAGE_SIZE
);
583 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
584 set_bit(UPROBE_COPY_INSN
, &uprobe
->flags
);
587 up_write(&uprobe
->consumer_rwsem
);
592 static inline bool consumer_filter(struct uprobe_consumer
*uc
,
593 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
595 return !uc
->filter
|| uc
->filter(uc
, ctx
, mm
);
598 static bool filter_chain(struct uprobe
*uprobe
,
599 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
601 struct uprobe_consumer
*uc
;
604 down_read(&uprobe
->consumer_rwsem
);
605 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
606 ret
= consumer_filter(uc
, ctx
, mm
);
610 up_read(&uprobe
->consumer_rwsem
);
616 install_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
,
617 struct vm_area_struct
*vma
, unsigned long vaddr
)
622 ret
= prepare_uprobe(uprobe
, vma
->vm_file
, mm
, vaddr
);
627 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
628 * the task can hit this breakpoint right after __replace_page().
630 first_uprobe
= !test_bit(MMF_HAS_UPROBES
, &mm
->flags
);
632 set_bit(MMF_HAS_UPROBES
, &mm
->flags
);
634 ret
= set_swbp(&uprobe
->arch
, mm
, vaddr
);
636 clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
637 else if (first_uprobe
)
638 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
644 remove_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
646 set_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
647 return set_orig_insn(&uprobe
->arch
, mm
, vaddr
);
650 static inline bool uprobe_is_active(struct uprobe
*uprobe
)
652 return !RB_EMPTY_NODE(&uprobe
->rb_node
);
655 * There could be threads that have already hit the breakpoint. They
656 * will recheck the current insn and restart if find_uprobe() fails.
657 * See find_active_uprobe().
659 static void delete_uprobe(struct uprobe
*uprobe
)
661 if (WARN_ON(!uprobe_is_active(uprobe
)))
664 spin_lock(&uprobes_treelock
);
665 rb_erase(&uprobe
->rb_node
, &uprobes_tree
);
666 spin_unlock(&uprobes_treelock
);
667 RB_CLEAR_NODE(&uprobe
->rb_node
); /* for uprobe_is_active() */
673 struct map_info
*next
;
674 struct mm_struct
*mm
;
678 static inline struct map_info
*free_map_info(struct map_info
*info
)
680 struct map_info
*next
= info
->next
;
685 static struct map_info
*
686 build_map_info(struct address_space
*mapping
, loff_t offset
, bool is_register
)
688 unsigned long pgoff
= offset
>> PAGE_SHIFT
;
689 struct vm_area_struct
*vma
;
690 struct map_info
*curr
= NULL
;
691 struct map_info
*prev
= NULL
;
692 struct map_info
*info
;
696 mutex_lock(&mapping
->i_mmap_mutex
);
697 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
, pgoff
, pgoff
) {
698 if (!valid_vma(vma
, is_register
))
701 if (!prev
&& !more
) {
703 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
704 * reclaim. This is optimistic, no harm done if it fails.
706 prev
= kmalloc(sizeof(struct map_info
),
707 GFP_NOWAIT
| __GFP_NOMEMALLOC
| __GFP_NOWARN
);
716 if (!atomic_inc_not_zero(&vma
->vm_mm
->mm_users
))
724 info
->mm
= vma
->vm_mm
;
725 info
->vaddr
= offset_to_vaddr(vma
, offset
);
727 mutex_unlock(&mapping
->i_mmap_mutex
);
739 info
= kmalloc(sizeof(struct map_info
), GFP_KERNEL
);
741 curr
= ERR_PTR(-ENOMEM
);
751 prev
= free_map_info(prev
);
756 register_for_each_vma(struct uprobe
*uprobe
, struct uprobe_consumer
*new)
758 bool is_register
= !!new;
759 struct map_info
*info
;
762 percpu_down_write(&dup_mmap_sem
);
763 info
= build_map_info(uprobe
->inode
->i_mapping
,
764 uprobe
->offset
, is_register
);
771 struct mm_struct
*mm
= info
->mm
;
772 struct vm_area_struct
*vma
;
774 if (err
&& is_register
)
777 down_write(&mm
->mmap_sem
);
778 vma
= find_vma(mm
, info
->vaddr
);
779 if (!vma
|| !valid_vma(vma
, is_register
) ||
780 file_inode(vma
->vm_file
) != uprobe
->inode
)
783 if (vma
->vm_start
> info
->vaddr
||
784 vaddr_to_offset(vma
, info
->vaddr
) != uprobe
->offset
)
788 /* consult only the "caller", new consumer. */
789 if (consumer_filter(new,
790 UPROBE_FILTER_REGISTER
, mm
))
791 err
= install_breakpoint(uprobe
, mm
, vma
, info
->vaddr
);
792 } else if (test_bit(MMF_HAS_UPROBES
, &mm
->flags
)) {
793 if (!filter_chain(uprobe
,
794 UPROBE_FILTER_UNREGISTER
, mm
))
795 err
|= remove_breakpoint(uprobe
, mm
, info
->vaddr
);
799 up_write(&mm
->mmap_sem
);
802 info
= free_map_info(info
);
805 percpu_up_write(&dup_mmap_sem
);
809 static int __uprobe_register(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
811 consumer_add(uprobe
, uc
);
812 return register_for_each_vma(uprobe
, uc
);
815 static void __uprobe_unregister(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
819 if (!consumer_del(uprobe
, uc
)) /* WARN? */
822 err
= register_for_each_vma(uprobe
, NULL
);
823 /* TODO : cant unregister? schedule a worker thread */
824 if (!uprobe
->consumers
&& !err
)
825 delete_uprobe(uprobe
);
829 * uprobe_register - register a probe
830 * @inode: the file in which the probe has to be placed.
831 * @offset: offset from the start of the file.
832 * @uc: information on howto handle the probe..
834 * Apart from the access refcount, uprobe_register() takes a creation
835 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
836 * inserted into the rbtree (i.e first consumer for a @inode:@offset
837 * tuple). Creation refcount stops uprobe_unregister from freeing the
838 * @uprobe even before the register operation is complete. Creation
839 * refcount is released when the last @uc for the @uprobe
842 * Return errno if it cannot successully install probes
843 * else return 0 (success)
845 int uprobe_register(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
847 struct uprobe
*uprobe
;
850 /* Uprobe must have at least one set consumer */
851 if (!uc
->handler
&& !uc
->ret_handler
)
854 /* Racy, just to catch the obvious mistakes */
855 if (offset
> i_size_read(inode
))
859 uprobe
= alloc_uprobe(inode
, offset
);
863 * We can race with uprobe_unregister()->delete_uprobe().
864 * Check uprobe_is_active() and retry if it is false.
866 down_write(&uprobe
->register_rwsem
);
868 if (likely(uprobe_is_active(uprobe
))) {
869 ret
= __uprobe_register(uprobe
, uc
);
871 __uprobe_unregister(uprobe
, uc
);
873 up_write(&uprobe
->register_rwsem
);
876 if (unlikely(ret
== -EAGAIN
))
880 EXPORT_SYMBOL_GPL(uprobe_register
);
883 * uprobe_apply - unregister a already registered probe.
884 * @inode: the file in which the probe has to be removed.
885 * @offset: offset from the start of the file.
886 * @uc: consumer which wants to add more or remove some breakpoints
887 * @add: add or remove the breakpoints
889 int uprobe_apply(struct inode
*inode
, loff_t offset
,
890 struct uprobe_consumer
*uc
, bool add
)
892 struct uprobe
*uprobe
;
893 struct uprobe_consumer
*con
;
896 uprobe
= find_uprobe(inode
, offset
);
900 down_write(&uprobe
->register_rwsem
);
901 for (con
= uprobe
->consumers
; con
&& con
!= uc
; con
= con
->next
)
904 ret
= register_for_each_vma(uprobe
, add
? uc
: NULL
);
905 up_write(&uprobe
->register_rwsem
);
912 * uprobe_unregister - unregister a already registered probe.
913 * @inode: the file in which the probe has to be removed.
914 * @offset: offset from the start of the file.
915 * @uc: identify which probe if multiple probes are colocated.
917 void uprobe_unregister(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
919 struct uprobe
*uprobe
;
921 uprobe
= find_uprobe(inode
, offset
);
925 down_write(&uprobe
->register_rwsem
);
926 __uprobe_unregister(uprobe
, uc
);
927 up_write(&uprobe
->register_rwsem
);
930 EXPORT_SYMBOL_GPL(uprobe_unregister
);
932 static int unapply_uprobe(struct uprobe
*uprobe
, struct mm_struct
*mm
)
934 struct vm_area_struct
*vma
;
937 down_read(&mm
->mmap_sem
);
938 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
942 if (!valid_vma(vma
, false) ||
943 file_inode(vma
->vm_file
) != uprobe
->inode
)
946 offset
= (loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
;
947 if (uprobe
->offset
< offset
||
948 uprobe
->offset
>= offset
+ vma
->vm_end
- vma
->vm_start
)
951 vaddr
= offset_to_vaddr(vma
, uprobe
->offset
);
952 err
|= remove_breakpoint(uprobe
, mm
, vaddr
);
954 up_read(&mm
->mmap_sem
);
959 static struct rb_node
*
960 find_node_in_range(struct inode
*inode
, loff_t min
, loff_t max
)
962 struct rb_node
*n
= uprobes_tree
.rb_node
;
965 struct uprobe
*u
= rb_entry(n
, struct uprobe
, rb_node
);
967 if (inode
< u
->inode
) {
969 } else if (inode
> u
->inode
) {
974 else if (min
> u
->offset
)
985 * For a given range in vma, build a list of probes that need to be inserted.
987 static void build_probe_list(struct inode
*inode
,
988 struct vm_area_struct
*vma
,
989 unsigned long start
, unsigned long end
,
990 struct list_head
*head
)
993 struct rb_node
*n
, *t
;
996 INIT_LIST_HEAD(head
);
997 min
= vaddr_to_offset(vma
, start
);
998 max
= min
+ (end
- start
) - 1;
1000 spin_lock(&uprobes_treelock
);
1001 n
= find_node_in_range(inode
, min
, max
);
1003 for (t
= n
; t
; t
= rb_prev(t
)) {
1004 u
= rb_entry(t
, struct uprobe
, rb_node
);
1005 if (u
->inode
!= inode
|| u
->offset
< min
)
1007 list_add(&u
->pending_list
, head
);
1008 atomic_inc(&u
->ref
);
1010 for (t
= n
; (t
= rb_next(t
)); ) {
1011 u
= rb_entry(t
, struct uprobe
, rb_node
);
1012 if (u
->inode
!= inode
|| u
->offset
> max
)
1014 list_add(&u
->pending_list
, head
);
1015 atomic_inc(&u
->ref
);
1018 spin_unlock(&uprobes_treelock
);
1022 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1024 * Currently we ignore all errors and always return 0, the callers
1025 * can't handle the failure anyway.
1027 int uprobe_mmap(struct vm_area_struct
*vma
)
1029 struct list_head tmp_list
;
1030 struct uprobe
*uprobe
, *u
;
1031 struct inode
*inode
;
1033 if (no_uprobe_events() || !valid_vma(vma
, true))
1036 inode
= file_inode(vma
->vm_file
);
1040 mutex_lock(uprobes_mmap_hash(inode
));
1041 build_probe_list(inode
, vma
, vma
->vm_start
, vma
->vm_end
, &tmp_list
);
1043 * We can race with uprobe_unregister(), this uprobe can be already
1044 * removed. But in this case filter_chain() must return false, all
1045 * consumers have gone away.
1047 list_for_each_entry_safe(uprobe
, u
, &tmp_list
, pending_list
) {
1048 if (!fatal_signal_pending(current
) &&
1049 filter_chain(uprobe
, UPROBE_FILTER_MMAP
, vma
->vm_mm
)) {
1050 unsigned long vaddr
= offset_to_vaddr(vma
, uprobe
->offset
);
1051 install_breakpoint(uprobe
, vma
->vm_mm
, vma
, vaddr
);
1055 mutex_unlock(uprobes_mmap_hash(inode
));
1061 vma_has_uprobes(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1064 struct inode
*inode
;
1067 inode
= file_inode(vma
->vm_file
);
1069 min
= vaddr_to_offset(vma
, start
);
1070 max
= min
+ (end
- start
) - 1;
1072 spin_lock(&uprobes_treelock
);
1073 n
= find_node_in_range(inode
, min
, max
);
1074 spin_unlock(&uprobes_treelock
);
1080 * Called in context of a munmap of a vma.
1082 void uprobe_munmap(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1084 if (no_uprobe_events() || !valid_vma(vma
, false))
1087 if (!atomic_read(&vma
->vm_mm
->mm_users
)) /* called by mmput() ? */
1090 if (!test_bit(MMF_HAS_UPROBES
, &vma
->vm_mm
->flags
) ||
1091 test_bit(MMF_RECALC_UPROBES
, &vma
->vm_mm
->flags
))
1094 if (vma_has_uprobes(vma
, start
, end
))
1095 set_bit(MMF_RECALC_UPROBES
, &vma
->vm_mm
->flags
);
1098 /* Slot allocation for XOL */
1099 static int xol_add_vma(struct mm_struct
*mm
, struct xol_area
*area
)
1101 int ret
= -EALREADY
;
1103 down_write(&mm
->mmap_sem
);
1104 if (mm
->uprobes_state
.xol_area
)
1108 /* Try to map as high as possible, this is only a hint. */
1109 area
->vaddr
= get_unmapped_area(NULL
, TASK_SIZE
- PAGE_SIZE
,
1111 if (area
->vaddr
& ~PAGE_MASK
) {
1117 ret
= install_special_mapping(mm
, area
->vaddr
, PAGE_SIZE
,
1118 VM_EXEC
|VM_MAYEXEC
|VM_DONTCOPY
|VM_IO
, &area
->page
);
1122 smp_wmb(); /* pairs with get_xol_area() */
1123 mm
->uprobes_state
.xol_area
= area
;
1125 up_write(&mm
->mmap_sem
);
1130 static struct xol_area
*__create_xol_area(unsigned long vaddr
)
1132 struct mm_struct
*mm
= current
->mm
;
1133 uprobe_opcode_t insn
= UPROBE_SWBP_INSN
;
1134 struct xol_area
*area
;
1136 area
= kmalloc(sizeof(*area
), GFP_KERNEL
);
1137 if (unlikely(!area
))
1140 area
->bitmap
= kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE
) * sizeof(long), GFP_KERNEL
);
1144 area
->page
= alloc_page(GFP_HIGHUSER
);
1148 area
->vaddr
= vaddr
;
1149 init_waitqueue_head(&area
->wq
);
1150 /* Reserve the 1st slot for get_trampoline_vaddr() */
1151 set_bit(0, area
->bitmap
);
1152 atomic_set(&area
->slot_count
, 1);
1153 copy_to_page(area
->page
, 0, &insn
, UPROBE_SWBP_INSN_SIZE
);
1155 if (!xol_add_vma(mm
, area
))
1158 __free_page(area
->page
);
1160 kfree(area
->bitmap
);
1168 * get_xol_area - Allocate process's xol_area if necessary.
1169 * This area will be used for storing instructions for execution out of line.
1171 * Returns the allocated area or NULL.
1173 static struct xol_area
*get_xol_area(void)
1175 struct mm_struct
*mm
= current
->mm
;
1176 struct xol_area
*area
;
1178 if (!mm
->uprobes_state
.xol_area
)
1179 __create_xol_area(0);
1181 area
= mm
->uprobes_state
.xol_area
;
1182 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1187 * uprobe_clear_state - Free the area allocated for slots.
1189 void uprobe_clear_state(struct mm_struct
*mm
)
1191 struct xol_area
*area
= mm
->uprobes_state
.xol_area
;
1196 put_page(area
->page
);
1197 kfree(area
->bitmap
);
1201 void uprobe_start_dup_mmap(void)
1203 percpu_down_read(&dup_mmap_sem
);
1206 void uprobe_end_dup_mmap(void)
1208 percpu_up_read(&dup_mmap_sem
);
1211 void uprobe_dup_mmap(struct mm_struct
*oldmm
, struct mm_struct
*newmm
)
1213 newmm
->uprobes_state
.xol_area
= NULL
;
1215 if (test_bit(MMF_HAS_UPROBES
, &oldmm
->flags
)) {
1216 set_bit(MMF_HAS_UPROBES
, &newmm
->flags
);
1217 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1218 set_bit(MMF_RECALC_UPROBES
, &newmm
->flags
);
1223 * - search for a free slot.
1225 static unsigned long xol_take_insn_slot(struct xol_area
*area
)
1227 unsigned long slot_addr
;
1231 slot_nr
= find_first_zero_bit(area
->bitmap
, UINSNS_PER_PAGE
);
1232 if (slot_nr
< UINSNS_PER_PAGE
) {
1233 if (!test_and_set_bit(slot_nr
, area
->bitmap
))
1236 slot_nr
= UINSNS_PER_PAGE
;
1239 wait_event(area
->wq
, (atomic_read(&area
->slot_count
) < UINSNS_PER_PAGE
));
1240 } while (slot_nr
>= UINSNS_PER_PAGE
);
1242 slot_addr
= area
->vaddr
+ (slot_nr
* UPROBE_XOL_SLOT_BYTES
);
1243 atomic_inc(&area
->slot_count
);
1249 * xol_get_insn_slot - allocate a slot for xol.
1250 * Returns the allocated slot address or 0.
1252 static unsigned long xol_get_insn_slot(struct uprobe
*uprobe
)
1254 struct xol_area
*area
;
1255 unsigned long xol_vaddr
;
1257 area
= get_xol_area();
1261 xol_vaddr
= xol_take_insn_slot(area
);
1262 if (unlikely(!xol_vaddr
))
1265 /* Initialize the slot */
1266 copy_to_page(area
->page
, xol_vaddr
,
1267 uprobe
->arch
.ixol
, sizeof(uprobe
->arch
.ixol
));
1269 * We probably need flush_icache_user_range() but it needs vma.
1270 * This should work on supported architectures too.
1272 flush_dcache_page(area
->page
);
1278 * xol_free_insn_slot - If slot was earlier allocated by
1279 * @xol_get_insn_slot(), make the slot available for
1280 * subsequent requests.
1282 static void xol_free_insn_slot(struct task_struct
*tsk
)
1284 struct xol_area
*area
;
1285 unsigned long vma_end
;
1286 unsigned long slot_addr
;
1288 if (!tsk
->mm
|| !tsk
->mm
->uprobes_state
.xol_area
|| !tsk
->utask
)
1291 slot_addr
= tsk
->utask
->xol_vaddr
;
1292 if (unlikely(!slot_addr
))
1295 area
= tsk
->mm
->uprobes_state
.xol_area
;
1296 vma_end
= area
->vaddr
+ PAGE_SIZE
;
1297 if (area
->vaddr
<= slot_addr
&& slot_addr
< vma_end
) {
1298 unsigned long offset
;
1301 offset
= slot_addr
- area
->vaddr
;
1302 slot_nr
= offset
/ UPROBE_XOL_SLOT_BYTES
;
1303 if (slot_nr
>= UINSNS_PER_PAGE
)
1306 clear_bit(slot_nr
, area
->bitmap
);
1307 atomic_dec(&area
->slot_count
);
1308 if (waitqueue_active(&area
->wq
))
1311 tsk
->utask
->xol_vaddr
= 0;
1316 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1317 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1319 * Return the address of the breakpoint instruction.
1321 unsigned long __weak
uprobe_get_swbp_addr(struct pt_regs
*regs
)
1323 return instruction_pointer(regs
) - UPROBE_SWBP_INSN_SIZE
;
1327 * Called with no locks held.
1328 * Called in context of a exiting or a exec-ing thread.
1330 void uprobe_free_utask(struct task_struct
*t
)
1332 struct uprobe_task
*utask
= t
->utask
;
1333 struct return_instance
*ri
, *tmp
;
1338 if (utask
->active_uprobe
)
1339 put_uprobe(utask
->active_uprobe
);
1341 ri
= utask
->return_instances
;
1346 put_uprobe(tmp
->uprobe
);
1350 xol_free_insn_slot(t
);
1356 * Allocate a uprobe_task object for the task if if necessary.
1357 * Called when the thread hits a breakpoint.
1360 * - pointer to new uprobe_task on success
1363 static struct uprobe_task
*get_utask(void)
1365 if (!current
->utask
)
1366 current
->utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1367 return current
->utask
;
1370 static int dup_utask(struct task_struct
*t
, struct uprobe_task
*o_utask
)
1372 struct uprobe_task
*n_utask
;
1373 struct return_instance
**p
, *o
, *n
;
1375 n_utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1380 p
= &n_utask
->return_instances
;
1381 for (o
= o_utask
->return_instances
; o
; o
= o
->next
) {
1382 n
= kmalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1387 atomic_inc(&n
->uprobe
->ref
);
1398 static void uprobe_warn(struct task_struct
*t
, const char *msg
)
1400 pr_warn("uprobe: %s:%d failed to %s\n",
1401 current
->comm
, current
->pid
, msg
);
1404 static void dup_xol_work(struct callback_head
*work
)
1406 if (current
->flags
& PF_EXITING
)
1409 if (!__create_xol_area(current
->utask
->dup_xol_addr
))
1410 uprobe_warn(current
, "dup xol area");
1414 * Called in context of a new clone/fork from copy_process.
1416 void uprobe_copy_process(struct task_struct
*t
, unsigned long flags
)
1418 struct uprobe_task
*utask
= current
->utask
;
1419 struct mm_struct
*mm
= current
->mm
;
1420 struct xol_area
*area
;
1424 if (!utask
|| !utask
->return_instances
)
1427 if (mm
== t
->mm
&& !(flags
& CLONE_VFORK
))
1430 if (dup_utask(t
, utask
))
1431 return uprobe_warn(t
, "dup ret instances");
1433 /* The task can fork() after dup_xol_work() fails */
1434 area
= mm
->uprobes_state
.xol_area
;
1436 return uprobe_warn(t
, "dup xol area");
1441 t
->utask
->dup_xol_addr
= area
->vaddr
;
1442 init_task_work(&t
->utask
->dup_xol_work
, dup_xol_work
);
1443 task_work_add(t
, &t
->utask
->dup_xol_work
, true);
1447 * Current area->vaddr notion assume the trampoline address is always
1448 * equal area->vaddr.
1450 * Returns -1 in case the xol_area is not allocated.
1452 static unsigned long get_trampoline_vaddr(void)
1454 struct xol_area
*area
;
1455 unsigned long trampoline_vaddr
= -1;
1457 area
= current
->mm
->uprobes_state
.xol_area
;
1458 smp_read_barrier_depends();
1460 trampoline_vaddr
= area
->vaddr
;
1462 return trampoline_vaddr
;
1465 static void prepare_uretprobe(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1467 struct return_instance
*ri
;
1468 struct uprobe_task
*utask
;
1469 unsigned long orig_ret_vaddr
, trampoline_vaddr
;
1470 bool chained
= false;
1472 if (!get_xol_area())
1475 utask
= get_utask();
1479 if (utask
->depth
>= MAX_URETPROBE_DEPTH
) {
1480 printk_ratelimited(KERN_INFO
"uprobe: omit uretprobe due to"
1481 " nestedness limit pid/tgid=%d/%d\n",
1482 current
->pid
, current
->tgid
);
1486 ri
= kzalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1490 trampoline_vaddr
= get_trampoline_vaddr();
1491 orig_ret_vaddr
= arch_uretprobe_hijack_return_addr(trampoline_vaddr
, regs
);
1492 if (orig_ret_vaddr
== -1)
1496 * We don't want to keep trampoline address in stack, rather keep the
1497 * original return address of first caller thru all the consequent
1498 * instances. This also makes breakpoint unwrapping easier.
1500 if (orig_ret_vaddr
== trampoline_vaddr
) {
1501 if (!utask
->return_instances
) {
1503 * This situation is not possible. Likely we have an
1504 * attack from user-space.
1506 pr_warn("uprobe: unable to set uretprobe pid/tgid=%d/%d\n",
1507 current
->pid
, current
->tgid
);
1512 orig_ret_vaddr
= utask
->return_instances
->orig_ret_vaddr
;
1515 atomic_inc(&uprobe
->ref
);
1516 ri
->uprobe
= uprobe
;
1517 ri
->func
= instruction_pointer(regs
);
1518 ri
->orig_ret_vaddr
= orig_ret_vaddr
;
1519 ri
->chained
= chained
;
1523 /* add instance to the stack */
1524 ri
->next
= utask
->return_instances
;
1525 utask
->return_instances
= ri
;
1533 /* Prepare to single-step probed instruction out of line. */
1535 pre_ssout(struct uprobe
*uprobe
, struct pt_regs
*regs
, unsigned long bp_vaddr
)
1537 struct uprobe_task
*utask
;
1538 unsigned long xol_vaddr
;
1541 utask
= get_utask();
1545 xol_vaddr
= xol_get_insn_slot(uprobe
);
1549 utask
->xol_vaddr
= xol_vaddr
;
1550 utask
->vaddr
= bp_vaddr
;
1552 err
= arch_uprobe_pre_xol(&uprobe
->arch
, regs
);
1553 if (unlikely(err
)) {
1554 xol_free_insn_slot(current
);
1558 utask
->active_uprobe
= uprobe
;
1559 utask
->state
= UTASK_SSTEP
;
1564 * If we are singlestepping, then ensure this thread is not connected to
1565 * non-fatal signals until completion of singlestep. When xol insn itself
1566 * triggers the signal, restart the original insn even if the task is
1567 * already SIGKILL'ed (since coredump should report the correct ip). This
1568 * is even more important if the task has a handler for SIGSEGV/etc, The
1569 * _same_ instruction should be repeated again after return from the signal
1570 * handler, and SSTEP can never finish in this case.
1572 bool uprobe_deny_signal(void)
1574 struct task_struct
*t
= current
;
1575 struct uprobe_task
*utask
= t
->utask
;
1577 if (likely(!utask
|| !utask
->active_uprobe
))
1580 WARN_ON_ONCE(utask
->state
!= UTASK_SSTEP
);
1582 if (signal_pending(t
)) {
1583 spin_lock_irq(&t
->sighand
->siglock
);
1584 clear_tsk_thread_flag(t
, TIF_SIGPENDING
);
1585 spin_unlock_irq(&t
->sighand
->siglock
);
1587 if (__fatal_signal_pending(t
) || arch_uprobe_xol_was_trapped(t
)) {
1588 utask
->state
= UTASK_SSTEP_TRAPPED
;
1589 set_tsk_thread_flag(t
, TIF_UPROBE
);
1590 set_tsk_thread_flag(t
, TIF_NOTIFY_RESUME
);
1598 * Avoid singlestepping the original instruction if the original instruction
1599 * is a NOP or can be emulated.
1601 static bool can_skip_sstep(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1603 if (test_bit(UPROBE_SKIP_SSTEP
, &uprobe
->flags
)) {
1604 if (arch_uprobe_skip_sstep(&uprobe
->arch
, regs
))
1606 clear_bit(UPROBE_SKIP_SSTEP
, &uprobe
->flags
);
1611 static void mmf_recalc_uprobes(struct mm_struct
*mm
)
1613 struct vm_area_struct
*vma
;
1615 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1616 if (!valid_vma(vma
, false))
1619 * This is not strictly accurate, we can race with
1620 * uprobe_unregister() and see the already removed
1621 * uprobe if delete_uprobe() was not yet called.
1622 * Or this uprobe can be filtered out.
1624 if (vma_has_uprobes(vma
, vma
->vm_start
, vma
->vm_end
))
1628 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
1631 static int is_trap_at_addr(struct mm_struct
*mm
, unsigned long vaddr
)
1634 uprobe_opcode_t opcode
;
1637 pagefault_disable();
1638 result
= __copy_from_user_inatomic(&opcode
, (void __user
*)vaddr
,
1642 if (likely(result
== 0))
1645 result
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &page
, NULL
);
1649 copy_from_page(page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
1652 /* This needs to return true for any variant of the trap insn */
1653 return is_trap_insn(&opcode
);
1656 static struct uprobe
*find_active_uprobe(unsigned long bp_vaddr
, int *is_swbp
)
1658 struct mm_struct
*mm
= current
->mm
;
1659 struct uprobe
*uprobe
= NULL
;
1660 struct vm_area_struct
*vma
;
1662 down_read(&mm
->mmap_sem
);
1663 vma
= find_vma(mm
, bp_vaddr
);
1664 if (vma
&& vma
->vm_start
<= bp_vaddr
) {
1665 if (valid_vma(vma
, false)) {
1666 struct inode
*inode
= file_inode(vma
->vm_file
);
1667 loff_t offset
= vaddr_to_offset(vma
, bp_vaddr
);
1669 uprobe
= find_uprobe(inode
, offset
);
1673 *is_swbp
= is_trap_at_addr(mm
, bp_vaddr
);
1678 if (!uprobe
&& test_and_clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
))
1679 mmf_recalc_uprobes(mm
);
1680 up_read(&mm
->mmap_sem
);
1685 static void handler_chain(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1687 struct uprobe_consumer
*uc
;
1688 int remove
= UPROBE_HANDLER_REMOVE
;
1689 bool need_prep
= false; /* prepare return uprobe, when needed */
1691 down_read(&uprobe
->register_rwsem
);
1692 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
1696 rc
= uc
->handler(uc
, regs
);
1697 WARN(rc
& ~UPROBE_HANDLER_MASK
,
1698 "bad rc=0x%x from %pf()\n", rc
, uc
->handler
);
1701 if (uc
->ret_handler
)
1707 if (need_prep
&& !remove
)
1708 prepare_uretprobe(uprobe
, regs
); /* put bp at return */
1710 if (remove
&& uprobe
->consumers
) {
1711 WARN_ON(!uprobe_is_active(uprobe
));
1712 unapply_uprobe(uprobe
, current
->mm
);
1714 up_read(&uprobe
->register_rwsem
);
1718 handle_uretprobe_chain(struct return_instance
*ri
, struct pt_regs
*regs
)
1720 struct uprobe
*uprobe
= ri
->uprobe
;
1721 struct uprobe_consumer
*uc
;
1723 down_read(&uprobe
->register_rwsem
);
1724 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
1725 if (uc
->ret_handler
)
1726 uc
->ret_handler(uc
, ri
->func
, regs
);
1728 up_read(&uprobe
->register_rwsem
);
1731 static bool handle_trampoline(struct pt_regs
*regs
)
1733 struct uprobe_task
*utask
;
1734 struct return_instance
*ri
, *tmp
;
1737 utask
= current
->utask
;
1741 ri
= utask
->return_instances
;
1746 * TODO: we should throw out return_instance's invalidated by
1747 * longjmp(), currently we assume that the probed function always
1750 instruction_pointer_set(regs
, ri
->orig_ret_vaddr
);
1753 handle_uretprobe_chain(ri
, regs
);
1755 chained
= ri
->chained
;
1756 put_uprobe(ri
->uprobe
);
1768 utask
->return_instances
= ri
;
1774 * Run handler and ask thread to singlestep.
1775 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1777 static void handle_swbp(struct pt_regs
*regs
)
1779 struct uprobe
*uprobe
;
1780 unsigned long bp_vaddr
;
1781 int uninitialized_var(is_swbp
);
1783 bp_vaddr
= uprobe_get_swbp_addr(regs
);
1784 if (bp_vaddr
== get_trampoline_vaddr()) {
1785 if (handle_trampoline(regs
))
1788 pr_warn("uprobe: unable to handle uretprobe pid/tgid=%d/%d\n",
1789 current
->pid
, current
->tgid
);
1792 uprobe
= find_active_uprobe(bp_vaddr
, &is_swbp
);
1795 /* No matching uprobe; signal SIGTRAP. */
1796 send_sig(SIGTRAP
, current
, 0);
1799 * Either we raced with uprobe_unregister() or we can't
1800 * access this memory. The latter is only possible if
1801 * another thread plays with our ->mm. In both cases
1802 * we can simply restart. If this vma was unmapped we
1803 * can pretend this insn was not executed yet and get
1804 * the (correct) SIGSEGV after restart.
1806 instruction_pointer_set(regs
, bp_vaddr
);
1811 /* change it in advance for ->handler() and restart */
1812 instruction_pointer_set(regs
, bp_vaddr
);
1815 * TODO: move copy_insn/etc into _register and remove this hack.
1816 * After we hit the bp, _unregister + _register can install the
1817 * new and not-yet-analyzed uprobe at the same address, restart.
1819 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1820 if (unlikely(!test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
)))
1823 handler_chain(uprobe
, regs
);
1824 if (can_skip_sstep(uprobe
, regs
))
1827 if (!pre_ssout(uprobe
, regs
, bp_vaddr
))
1830 /* can_skip_sstep() succeeded, or restart if can't singlestep */
1836 * Perform required fix-ups and disable singlestep.
1837 * Allow pending signals to take effect.
1839 static void handle_singlestep(struct uprobe_task
*utask
, struct pt_regs
*regs
)
1841 struct uprobe
*uprobe
;
1843 uprobe
= utask
->active_uprobe
;
1844 if (utask
->state
== UTASK_SSTEP_ACK
)
1845 arch_uprobe_post_xol(&uprobe
->arch
, regs
);
1846 else if (utask
->state
== UTASK_SSTEP_TRAPPED
)
1847 arch_uprobe_abort_xol(&uprobe
->arch
, regs
);
1852 utask
->active_uprobe
= NULL
;
1853 utask
->state
= UTASK_RUNNING
;
1854 xol_free_insn_slot(current
);
1856 spin_lock_irq(¤t
->sighand
->siglock
);
1857 recalc_sigpending(); /* see uprobe_deny_signal() */
1858 spin_unlock_irq(¤t
->sighand
->siglock
);
1862 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1863 * allows the thread to return from interrupt. After that handle_swbp()
1864 * sets utask->active_uprobe.
1866 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1867 * and allows the thread to return from interrupt.
1869 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1870 * uprobe_notify_resume().
1872 void uprobe_notify_resume(struct pt_regs
*regs
)
1874 struct uprobe_task
*utask
;
1876 clear_thread_flag(TIF_UPROBE
);
1878 utask
= current
->utask
;
1879 if (utask
&& utask
->active_uprobe
)
1880 handle_singlestep(utask
, regs
);
1886 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1887 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1889 int uprobe_pre_sstep_notifier(struct pt_regs
*regs
)
1894 if (!test_bit(MMF_HAS_UPROBES
, ¤t
->mm
->flags
) &&
1895 (!current
->utask
|| !current
->utask
->return_instances
))
1898 set_thread_flag(TIF_UPROBE
);
1903 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1904 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1906 int uprobe_post_sstep_notifier(struct pt_regs
*regs
)
1908 struct uprobe_task
*utask
= current
->utask
;
1910 if (!current
->mm
|| !utask
|| !utask
->active_uprobe
)
1911 /* task is currently not uprobed */
1914 utask
->state
= UTASK_SSTEP_ACK
;
1915 set_thread_flag(TIF_UPROBE
);
1919 static struct notifier_block uprobe_exception_nb
= {
1920 .notifier_call
= arch_uprobe_exception_notify
,
1921 .priority
= INT_MAX
-1, /* notified after kprobes, kgdb */
1924 static int __init
init_uprobes(void)
1928 for (i
= 0; i
< UPROBES_HASH_SZ
; i
++)
1929 mutex_init(&uprobes_mmap_mutex
[i
]);
1931 if (percpu_init_rwsem(&dup_mmap_sem
))
1934 return register_die_notifier(&uprobe_exception_nb
);
1936 __initcall(init_uprobes
);