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
65 struct rb_node rb_node
; /* node in the rb tree */
67 struct rw_semaphore register_rwsem
;
68 struct rw_semaphore consumer_rwsem
;
69 struct list_head pending_list
;
70 struct uprobe_consumer
*consumers
;
71 struct inode
*inode
; /* Also hold a ref to inode */
76 * The generic code assumes that it has two members of unknown type
77 * owned by the arch-specific code:
79 * insn - copy_insn() saves the original instruction here for
80 * arch_uprobe_analyze_insn().
82 * ixol - potentially modified instruction to execute out of
83 * line, copied to xol_area by xol_get_insn_slot().
85 struct arch_uprobe arch
;
88 struct return_instance
{
89 struct uprobe
*uprobe
;
91 unsigned long orig_ret_vaddr
; /* original return address */
92 bool chained
; /* true, if instance is nested */
94 struct return_instance
*next
; /* keep as stack */
98 * Execute out of line area: anonymous executable mapping installed
99 * by the probed task to execute the copy of the original instruction
100 * mangled by set_swbp().
102 * On a breakpoint hit, thread contests for a slot. It frees the
103 * slot after singlestep. Currently a fixed number of slots are
107 wait_queue_head_t wq
; /* if all slots are busy */
108 atomic_t slot_count
; /* number of in-use slots */
109 unsigned long *bitmap
; /* 0 = free slot */
113 * We keep the vma's vm_start rather than a pointer to the vma
114 * itself. The probed process or a naughty kernel module could make
115 * the vma go away, and we must handle that reasonably gracefully.
117 unsigned long vaddr
; /* Page(s) of instruction slots */
121 * valid_vma: Verify if the specified vma is an executable vma
122 * Relax restrictions while unregistering: vm_flags might have
123 * changed after breakpoint was inserted.
124 * - is_register: indicates if we are in register context.
125 * - Return 1 if the specified virtual address is in an
128 static bool valid_vma(struct vm_area_struct
*vma
, bool is_register
)
130 vm_flags_t flags
= VM_HUGETLB
| VM_MAYEXEC
| VM_SHARED
;
135 return vma
->vm_file
&& (vma
->vm_flags
& flags
) == VM_MAYEXEC
;
138 static unsigned long offset_to_vaddr(struct vm_area_struct
*vma
, loff_t offset
)
140 return vma
->vm_start
+ offset
- ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
143 static loff_t
vaddr_to_offset(struct vm_area_struct
*vma
, unsigned long vaddr
)
145 return ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
) + (vaddr
- vma
->vm_start
);
149 * __replace_page - replace page in vma by new page.
150 * based on replace_page in mm/ksm.c
152 * @vma: vma that holds the pte pointing to page
153 * @addr: address the old @page is mapped at
154 * @page: the cowed page we are replacing by kpage
155 * @kpage: the modified page we replace page by
157 * Returns 0 on success, -EFAULT on failure.
159 static int __replace_page(struct vm_area_struct
*vma
, unsigned long addr
,
160 struct page
*page
, struct page
*kpage
)
162 struct mm_struct
*mm
= vma
->vm_mm
;
166 /* For mmu_notifiers */
167 const unsigned long mmun_start
= addr
;
168 const unsigned long mmun_end
= addr
+ PAGE_SIZE
;
170 /* For try_to_free_swap() and munlock_vma_page() below */
173 mmu_notifier_invalidate_range_start(mm
, mmun_start
, mmun_end
);
175 ptep
= page_check_address(page
, mm
, addr
, &ptl
, 0);
180 page_add_new_anon_rmap(kpage
, vma
, addr
);
182 if (!PageAnon(page
)) {
183 dec_mm_counter(mm
, MM_FILEPAGES
);
184 inc_mm_counter(mm
, MM_ANONPAGES
);
187 flush_cache_page(vma
, addr
, pte_pfn(*ptep
));
188 ptep_clear_flush(vma
, addr
, ptep
);
189 set_pte_at_notify(mm
, addr
, ptep
, mk_pte(kpage
, vma
->vm_page_prot
));
191 page_remove_rmap(page
);
192 if (!page_mapped(page
))
193 try_to_free_swap(page
);
194 pte_unmap_unlock(ptep
, ptl
);
196 if (vma
->vm_flags
& VM_LOCKED
)
197 munlock_vma_page(page
);
202 mmu_notifier_invalidate_range_end(mm
, mmun_start
, mmun_end
);
208 * is_swbp_insn - check if instruction is breakpoint instruction.
209 * @insn: instruction to be checked.
210 * Default implementation of is_swbp_insn
211 * Returns true if @insn is a breakpoint instruction.
213 bool __weak
is_swbp_insn(uprobe_opcode_t
*insn
)
215 return *insn
== UPROBE_SWBP_INSN
;
219 * is_trap_insn - check if instruction is breakpoint instruction.
220 * @insn: instruction to be checked.
221 * Default implementation of is_trap_insn
222 * Returns true if @insn is a breakpoint instruction.
224 * This function is needed for the case where an architecture has multiple
225 * trap instructions (like powerpc).
227 bool __weak
is_trap_insn(uprobe_opcode_t
*insn
)
229 return is_swbp_insn(insn
);
232 static void copy_from_page(struct page
*page
, unsigned long vaddr
, void *dst
, int len
)
234 void *kaddr
= kmap_atomic(page
);
235 memcpy(dst
, kaddr
+ (vaddr
& ~PAGE_MASK
), len
);
236 kunmap_atomic(kaddr
);
239 static void copy_to_page(struct page
*page
, unsigned long vaddr
, const void *src
, int len
)
241 void *kaddr
= kmap_atomic(page
);
242 memcpy(kaddr
+ (vaddr
& ~PAGE_MASK
), src
, len
);
243 kunmap_atomic(kaddr
);
246 static int verify_opcode(struct page
*page
, unsigned long vaddr
, uprobe_opcode_t
*new_opcode
)
248 uprobe_opcode_t old_opcode
;
252 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
253 * We do not check if it is any other 'trap variant' which could
254 * be conditional trap instruction such as the one powerpc supports.
256 * The logic is that we do not care if the underlying instruction
257 * is a trap variant; uprobes always wins over any other (gdb)
260 copy_from_page(page
, vaddr
, &old_opcode
, UPROBE_SWBP_INSN_SIZE
);
261 is_swbp
= is_swbp_insn(&old_opcode
);
263 if (is_swbp_insn(new_opcode
)) {
264 if (is_swbp
) /* register: already installed? */
267 if (!is_swbp
) /* unregister: was it changed by us? */
276 * Expect the breakpoint instruction to be the smallest size instruction for
277 * the architecture. If an arch has variable length instruction and the
278 * breakpoint instruction is not of the smallest length instruction
279 * supported by that architecture then we need to modify is_trap_at_addr and
280 * uprobe_write_opcode accordingly. This would never be a problem for archs
281 * that have fixed length instructions.
285 * uprobe_write_opcode - write the opcode at a given virtual address.
286 * @mm: the probed process address space.
287 * @vaddr: the virtual address to store the opcode.
288 * @opcode: opcode to be written at @vaddr.
290 * Called with mm->mmap_sem held (for read and with a reference to
293 * For mm @mm, write the opcode at @vaddr.
294 * Return 0 (success) or a negative errno.
296 int uprobe_write_opcode(struct mm_struct
*mm
, unsigned long vaddr
,
297 uprobe_opcode_t opcode
)
299 struct page
*old_page
, *new_page
;
300 struct vm_area_struct
*vma
;
304 /* Read the page with vaddr into memory */
305 ret
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &old_page
, &vma
);
309 ret
= verify_opcode(old_page
, vaddr
, &opcode
);
314 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, vaddr
);
318 __SetPageUptodate(new_page
);
320 copy_highpage(new_page
, old_page
);
321 copy_to_page(new_page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
323 ret
= anon_vma_prepare(vma
);
327 ret
= __replace_page(vma
, vaddr
, old_page
, new_page
);
330 page_cache_release(new_page
);
334 if (unlikely(ret
== -EAGAIN
))
340 * set_swbp - store breakpoint at a given address.
341 * @auprobe: arch specific probepoint information.
342 * @mm: the probed process address space.
343 * @vaddr: the virtual address to insert the opcode.
345 * For mm @mm, store the breakpoint instruction at @vaddr.
346 * Return 0 (success) or a negative errno.
348 int __weak
set_swbp(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
350 return uprobe_write_opcode(mm
, vaddr
, UPROBE_SWBP_INSN
);
354 * set_orig_insn - Restore the original instruction.
355 * @mm: the probed process address space.
356 * @auprobe: arch specific probepoint information.
357 * @vaddr: the virtual address to insert the opcode.
359 * For mm @mm, restore the original opcode (opcode) at @vaddr.
360 * Return 0 (success) or a negative errno.
363 set_orig_insn(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
365 return uprobe_write_opcode(mm
, vaddr
, *(uprobe_opcode_t
*)&auprobe
->insn
);
368 static int match_uprobe(struct uprobe
*l
, struct uprobe
*r
)
370 if (l
->inode
< r
->inode
)
373 if (l
->inode
> r
->inode
)
376 if (l
->offset
< r
->offset
)
379 if (l
->offset
> r
->offset
)
385 static struct uprobe
*__find_uprobe(struct inode
*inode
, loff_t offset
)
387 struct uprobe u
= { .inode
= inode
, .offset
= offset
};
388 struct rb_node
*n
= uprobes_tree
.rb_node
;
389 struct uprobe
*uprobe
;
393 uprobe
= rb_entry(n
, struct uprobe
, rb_node
);
394 match
= match_uprobe(&u
, uprobe
);
396 atomic_inc(&uprobe
->ref
);
409 * Find a uprobe corresponding to a given inode:offset
410 * Acquires uprobes_treelock
412 static struct uprobe
*find_uprobe(struct inode
*inode
, loff_t offset
)
414 struct uprobe
*uprobe
;
416 spin_lock(&uprobes_treelock
);
417 uprobe
= __find_uprobe(inode
, offset
);
418 spin_unlock(&uprobes_treelock
);
423 static struct uprobe
*__insert_uprobe(struct uprobe
*uprobe
)
425 struct rb_node
**p
= &uprobes_tree
.rb_node
;
426 struct rb_node
*parent
= NULL
;
432 u
= rb_entry(parent
, struct uprobe
, rb_node
);
433 match
= match_uprobe(uprobe
, u
);
440 p
= &parent
->rb_left
;
442 p
= &parent
->rb_right
;
447 rb_link_node(&uprobe
->rb_node
, parent
, p
);
448 rb_insert_color(&uprobe
->rb_node
, &uprobes_tree
);
449 /* get access + creation ref */
450 atomic_set(&uprobe
->ref
, 2);
456 * Acquire uprobes_treelock.
457 * Matching uprobe already exists in rbtree;
458 * increment (access refcount) and return the matching uprobe.
460 * No matching uprobe; insert the uprobe in rb_tree;
461 * get a double refcount (access + creation) and return NULL.
463 static struct uprobe
*insert_uprobe(struct uprobe
*uprobe
)
467 spin_lock(&uprobes_treelock
);
468 u
= __insert_uprobe(uprobe
);
469 spin_unlock(&uprobes_treelock
);
474 static void put_uprobe(struct uprobe
*uprobe
)
476 if (atomic_dec_and_test(&uprobe
->ref
))
480 static struct uprobe
*alloc_uprobe(struct inode
*inode
, loff_t offset
)
482 struct uprobe
*uprobe
, *cur_uprobe
;
484 uprobe
= kzalloc(sizeof(struct uprobe
), GFP_KERNEL
);
488 uprobe
->inode
= igrab(inode
);
489 uprobe
->offset
= offset
;
490 init_rwsem(&uprobe
->register_rwsem
);
491 init_rwsem(&uprobe
->consumer_rwsem
);
493 /* add to uprobes_tree, sorted on inode:offset */
494 cur_uprobe
= insert_uprobe(uprobe
);
495 /* a uprobe exists for this inode:offset combination */
505 static void consumer_add(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
507 down_write(&uprobe
->consumer_rwsem
);
508 uc
->next
= uprobe
->consumers
;
509 uprobe
->consumers
= uc
;
510 up_write(&uprobe
->consumer_rwsem
);
514 * For uprobe @uprobe, delete the consumer @uc.
515 * Return true if the @uc is deleted successfully
518 static bool consumer_del(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
520 struct uprobe_consumer
**con
;
523 down_write(&uprobe
->consumer_rwsem
);
524 for (con
= &uprobe
->consumers
; *con
; con
= &(*con
)->next
) {
531 up_write(&uprobe
->consumer_rwsem
);
536 static int __copy_insn(struct address_space
*mapping
, struct file
*filp
,
537 void *insn
, int nbytes
, loff_t offset
)
541 if (!mapping
->a_ops
->readpage
)
544 * Ensure that the page that has the original instruction is
545 * populated and in page-cache.
547 page
= read_mapping_page(mapping
, offset
>> PAGE_CACHE_SHIFT
, filp
);
549 return PTR_ERR(page
);
551 copy_from_page(page
, offset
, insn
, nbytes
);
552 page_cache_release(page
);
557 static int copy_insn(struct uprobe
*uprobe
, struct file
*filp
)
559 struct address_space
*mapping
= uprobe
->inode
->i_mapping
;
560 loff_t offs
= uprobe
->offset
;
561 void *insn
= &uprobe
->arch
.insn
;
562 int size
= sizeof(uprobe
->arch
.insn
);
565 /* Copy only available bytes, -EIO if nothing was read */
567 if (offs
>= i_size_read(uprobe
->inode
))
570 len
= min_t(int, size
, PAGE_SIZE
- (offs
& ~PAGE_MASK
));
571 err
= __copy_insn(mapping
, filp
, insn
, len
, offs
);
583 static int prepare_uprobe(struct uprobe
*uprobe
, struct file
*file
,
584 struct mm_struct
*mm
, unsigned long vaddr
)
588 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
591 /* TODO: move this into _register, until then we abuse this sem. */
592 down_write(&uprobe
->consumer_rwsem
);
593 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
596 ret
= copy_insn(uprobe
, file
);
601 if (is_trap_insn((uprobe_opcode_t
*)&uprobe
->arch
.insn
))
604 ret
= arch_uprobe_analyze_insn(&uprobe
->arch
, mm
, vaddr
);
608 /* uprobe_write_opcode() assumes we don't cross page boundary */
609 BUG_ON((uprobe
->offset
& ~PAGE_MASK
) +
610 UPROBE_SWBP_INSN_SIZE
> PAGE_SIZE
);
612 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
613 set_bit(UPROBE_COPY_INSN
, &uprobe
->flags
);
616 up_write(&uprobe
->consumer_rwsem
);
621 static inline bool consumer_filter(struct uprobe_consumer
*uc
,
622 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
624 return !uc
->filter
|| uc
->filter(uc
, ctx
, mm
);
627 static bool filter_chain(struct uprobe
*uprobe
,
628 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
630 struct uprobe_consumer
*uc
;
633 down_read(&uprobe
->consumer_rwsem
);
634 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
635 ret
= consumer_filter(uc
, ctx
, mm
);
639 up_read(&uprobe
->consumer_rwsem
);
645 install_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
,
646 struct vm_area_struct
*vma
, unsigned long vaddr
)
651 ret
= prepare_uprobe(uprobe
, vma
->vm_file
, mm
, vaddr
);
656 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
657 * the task can hit this breakpoint right after __replace_page().
659 first_uprobe
= !test_bit(MMF_HAS_UPROBES
, &mm
->flags
);
661 set_bit(MMF_HAS_UPROBES
, &mm
->flags
);
663 ret
= set_swbp(&uprobe
->arch
, mm
, vaddr
);
665 clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
666 else if (first_uprobe
)
667 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
673 remove_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
675 set_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
676 return set_orig_insn(&uprobe
->arch
, mm
, vaddr
);
679 static inline bool uprobe_is_active(struct uprobe
*uprobe
)
681 return !RB_EMPTY_NODE(&uprobe
->rb_node
);
684 * There could be threads that have already hit the breakpoint. They
685 * will recheck the current insn and restart if find_uprobe() fails.
686 * See find_active_uprobe().
688 static void delete_uprobe(struct uprobe
*uprobe
)
690 if (WARN_ON(!uprobe_is_active(uprobe
)))
693 spin_lock(&uprobes_treelock
);
694 rb_erase(&uprobe
->rb_node
, &uprobes_tree
);
695 spin_unlock(&uprobes_treelock
);
696 RB_CLEAR_NODE(&uprobe
->rb_node
); /* for uprobe_is_active() */
702 struct map_info
*next
;
703 struct mm_struct
*mm
;
707 static inline struct map_info
*free_map_info(struct map_info
*info
)
709 struct map_info
*next
= info
->next
;
714 static struct map_info
*
715 build_map_info(struct address_space
*mapping
, loff_t offset
, bool is_register
)
717 unsigned long pgoff
= offset
>> PAGE_SHIFT
;
718 struct vm_area_struct
*vma
;
719 struct map_info
*curr
= NULL
;
720 struct map_info
*prev
= NULL
;
721 struct map_info
*info
;
725 mutex_lock(&mapping
->i_mmap_mutex
);
726 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
, pgoff
, pgoff
) {
727 if (!valid_vma(vma
, is_register
))
730 if (!prev
&& !more
) {
732 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
733 * reclaim. This is optimistic, no harm done if it fails.
735 prev
= kmalloc(sizeof(struct map_info
),
736 GFP_NOWAIT
| __GFP_NOMEMALLOC
| __GFP_NOWARN
);
745 if (!atomic_inc_not_zero(&vma
->vm_mm
->mm_users
))
753 info
->mm
= vma
->vm_mm
;
754 info
->vaddr
= offset_to_vaddr(vma
, offset
);
756 mutex_unlock(&mapping
->i_mmap_mutex
);
768 info
= kmalloc(sizeof(struct map_info
), GFP_KERNEL
);
770 curr
= ERR_PTR(-ENOMEM
);
780 prev
= free_map_info(prev
);
785 register_for_each_vma(struct uprobe
*uprobe
, struct uprobe_consumer
*new)
787 bool is_register
= !!new;
788 struct map_info
*info
;
791 percpu_down_write(&dup_mmap_sem
);
792 info
= build_map_info(uprobe
->inode
->i_mapping
,
793 uprobe
->offset
, is_register
);
800 struct mm_struct
*mm
= info
->mm
;
801 struct vm_area_struct
*vma
;
803 if (err
&& is_register
)
806 down_write(&mm
->mmap_sem
);
807 vma
= find_vma(mm
, info
->vaddr
);
808 if (!vma
|| !valid_vma(vma
, is_register
) ||
809 file_inode(vma
->vm_file
) != uprobe
->inode
)
812 if (vma
->vm_start
> info
->vaddr
||
813 vaddr_to_offset(vma
, info
->vaddr
) != uprobe
->offset
)
817 /* consult only the "caller", new consumer. */
818 if (consumer_filter(new,
819 UPROBE_FILTER_REGISTER
, mm
))
820 err
= install_breakpoint(uprobe
, mm
, vma
, info
->vaddr
);
821 } else if (test_bit(MMF_HAS_UPROBES
, &mm
->flags
)) {
822 if (!filter_chain(uprobe
,
823 UPROBE_FILTER_UNREGISTER
, mm
))
824 err
|= remove_breakpoint(uprobe
, mm
, info
->vaddr
);
828 up_write(&mm
->mmap_sem
);
831 info
= free_map_info(info
);
834 percpu_up_write(&dup_mmap_sem
);
838 static int __uprobe_register(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
840 consumer_add(uprobe
, uc
);
841 return register_for_each_vma(uprobe
, uc
);
844 static void __uprobe_unregister(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
848 if (!consumer_del(uprobe
, uc
)) /* WARN? */
851 err
= register_for_each_vma(uprobe
, NULL
);
852 /* TODO : cant unregister? schedule a worker thread */
853 if (!uprobe
->consumers
&& !err
)
854 delete_uprobe(uprobe
);
858 * uprobe_register - register a probe
859 * @inode: the file in which the probe has to be placed.
860 * @offset: offset from the start of the file.
861 * @uc: information on howto handle the probe..
863 * Apart from the access refcount, uprobe_register() takes a creation
864 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
865 * inserted into the rbtree (i.e first consumer for a @inode:@offset
866 * tuple). Creation refcount stops uprobe_unregister from freeing the
867 * @uprobe even before the register operation is complete. Creation
868 * refcount is released when the last @uc for the @uprobe
871 * Return errno if it cannot successully install probes
872 * else return 0 (success)
874 int uprobe_register(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
876 struct uprobe
*uprobe
;
879 /* Uprobe must have at least one set consumer */
880 if (!uc
->handler
&& !uc
->ret_handler
)
883 /* Racy, just to catch the obvious mistakes */
884 if (offset
> i_size_read(inode
))
888 uprobe
= alloc_uprobe(inode
, offset
);
892 * We can race with uprobe_unregister()->delete_uprobe().
893 * Check uprobe_is_active() and retry if it is false.
895 down_write(&uprobe
->register_rwsem
);
897 if (likely(uprobe_is_active(uprobe
))) {
898 ret
= __uprobe_register(uprobe
, uc
);
900 __uprobe_unregister(uprobe
, uc
);
902 up_write(&uprobe
->register_rwsem
);
905 if (unlikely(ret
== -EAGAIN
))
909 EXPORT_SYMBOL_GPL(uprobe_register
);
912 * uprobe_apply - 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: consumer which wants to add more or remove some breakpoints
916 * @add: add or remove the breakpoints
918 int uprobe_apply(struct inode
*inode
, loff_t offset
,
919 struct uprobe_consumer
*uc
, bool add
)
921 struct uprobe
*uprobe
;
922 struct uprobe_consumer
*con
;
925 uprobe
= find_uprobe(inode
, offset
);
929 down_write(&uprobe
->register_rwsem
);
930 for (con
= uprobe
->consumers
; con
&& con
!= uc
; con
= con
->next
)
933 ret
= register_for_each_vma(uprobe
, add
? uc
: NULL
);
934 up_write(&uprobe
->register_rwsem
);
941 * uprobe_unregister - unregister a already registered probe.
942 * @inode: the file in which the probe has to be removed.
943 * @offset: offset from the start of the file.
944 * @uc: identify which probe if multiple probes are colocated.
946 void uprobe_unregister(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
948 struct uprobe
*uprobe
;
950 uprobe
= find_uprobe(inode
, offset
);
954 down_write(&uprobe
->register_rwsem
);
955 __uprobe_unregister(uprobe
, uc
);
956 up_write(&uprobe
->register_rwsem
);
959 EXPORT_SYMBOL_GPL(uprobe_unregister
);
961 static int unapply_uprobe(struct uprobe
*uprobe
, struct mm_struct
*mm
)
963 struct vm_area_struct
*vma
;
966 down_read(&mm
->mmap_sem
);
967 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
971 if (!valid_vma(vma
, false) ||
972 file_inode(vma
->vm_file
) != uprobe
->inode
)
975 offset
= (loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
;
976 if (uprobe
->offset
< offset
||
977 uprobe
->offset
>= offset
+ vma
->vm_end
- vma
->vm_start
)
980 vaddr
= offset_to_vaddr(vma
, uprobe
->offset
);
981 err
|= remove_breakpoint(uprobe
, mm
, vaddr
);
983 up_read(&mm
->mmap_sem
);
988 static struct rb_node
*
989 find_node_in_range(struct inode
*inode
, loff_t min
, loff_t max
)
991 struct rb_node
*n
= uprobes_tree
.rb_node
;
994 struct uprobe
*u
= rb_entry(n
, struct uprobe
, rb_node
);
996 if (inode
< u
->inode
) {
998 } else if (inode
> u
->inode
) {
1001 if (max
< u
->offset
)
1003 else if (min
> u
->offset
)
1014 * For a given range in vma, build a list of probes that need to be inserted.
1016 static void build_probe_list(struct inode
*inode
,
1017 struct vm_area_struct
*vma
,
1018 unsigned long start
, unsigned long end
,
1019 struct list_head
*head
)
1022 struct rb_node
*n
, *t
;
1025 INIT_LIST_HEAD(head
);
1026 min
= vaddr_to_offset(vma
, start
);
1027 max
= min
+ (end
- start
) - 1;
1029 spin_lock(&uprobes_treelock
);
1030 n
= find_node_in_range(inode
, min
, max
);
1032 for (t
= n
; t
; t
= rb_prev(t
)) {
1033 u
= rb_entry(t
, struct uprobe
, rb_node
);
1034 if (u
->inode
!= inode
|| u
->offset
< min
)
1036 list_add(&u
->pending_list
, head
);
1037 atomic_inc(&u
->ref
);
1039 for (t
= n
; (t
= rb_next(t
)); ) {
1040 u
= rb_entry(t
, struct uprobe
, rb_node
);
1041 if (u
->inode
!= inode
|| u
->offset
> max
)
1043 list_add(&u
->pending_list
, head
);
1044 atomic_inc(&u
->ref
);
1047 spin_unlock(&uprobes_treelock
);
1051 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1053 * Currently we ignore all errors and always return 0, the callers
1054 * can't handle the failure anyway.
1056 int uprobe_mmap(struct vm_area_struct
*vma
)
1058 struct list_head tmp_list
;
1059 struct uprobe
*uprobe
, *u
;
1060 struct inode
*inode
;
1062 if (no_uprobe_events() || !valid_vma(vma
, true))
1065 inode
= file_inode(vma
->vm_file
);
1069 mutex_lock(uprobes_mmap_hash(inode
));
1070 build_probe_list(inode
, vma
, vma
->vm_start
, vma
->vm_end
, &tmp_list
);
1072 * We can race with uprobe_unregister(), this uprobe can be already
1073 * removed. But in this case filter_chain() must return false, all
1074 * consumers have gone away.
1076 list_for_each_entry_safe(uprobe
, u
, &tmp_list
, pending_list
) {
1077 if (!fatal_signal_pending(current
) &&
1078 filter_chain(uprobe
, UPROBE_FILTER_MMAP
, vma
->vm_mm
)) {
1079 unsigned long vaddr
= offset_to_vaddr(vma
, uprobe
->offset
);
1080 install_breakpoint(uprobe
, vma
->vm_mm
, vma
, vaddr
);
1084 mutex_unlock(uprobes_mmap_hash(inode
));
1090 vma_has_uprobes(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1093 struct inode
*inode
;
1096 inode
= file_inode(vma
->vm_file
);
1098 min
= vaddr_to_offset(vma
, start
);
1099 max
= min
+ (end
- start
) - 1;
1101 spin_lock(&uprobes_treelock
);
1102 n
= find_node_in_range(inode
, min
, max
);
1103 spin_unlock(&uprobes_treelock
);
1109 * Called in context of a munmap of a vma.
1111 void uprobe_munmap(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1113 if (no_uprobe_events() || !valid_vma(vma
, false))
1116 if (!atomic_read(&vma
->vm_mm
->mm_users
)) /* called by mmput() ? */
1119 if (!test_bit(MMF_HAS_UPROBES
, &vma
->vm_mm
->flags
) ||
1120 test_bit(MMF_RECALC_UPROBES
, &vma
->vm_mm
->flags
))
1123 if (vma_has_uprobes(vma
, start
, end
))
1124 set_bit(MMF_RECALC_UPROBES
, &vma
->vm_mm
->flags
);
1127 /* Slot allocation for XOL */
1128 static int xol_add_vma(struct mm_struct
*mm
, struct xol_area
*area
)
1130 int ret
= -EALREADY
;
1132 down_write(&mm
->mmap_sem
);
1133 if (mm
->uprobes_state
.xol_area
)
1137 /* Try to map as high as possible, this is only a hint. */
1138 area
->vaddr
= get_unmapped_area(NULL
, TASK_SIZE
- PAGE_SIZE
,
1140 if (area
->vaddr
& ~PAGE_MASK
) {
1146 ret
= install_special_mapping(mm
, area
->vaddr
, PAGE_SIZE
,
1147 VM_EXEC
|VM_MAYEXEC
|VM_DONTCOPY
|VM_IO
, &area
->page
);
1151 smp_wmb(); /* pairs with get_xol_area() */
1152 mm
->uprobes_state
.xol_area
= area
;
1154 up_write(&mm
->mmap_sem
);
1159 static struct xol_area
*__create_xol_area(unsigned long vaddr
)
1161 struct mm_struct
*mm
= current
->mm
;
1162 uprobe_opcode_t insn
= UPROBE_SWBP_INSN
;
1163 struct xol_area
*area
;
1165 area
= kmalloc(sizeof(*area
), GFP_KERNEL
);
1166 if (unlikely(!area
))
1169 area
->bitmap
= kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE
) * sizeof(long), GFP_KERNEL
);
1173 area
->page
= alloc_page(GFP_HIGHUSER
);
1177 area
->vaddr
= vaddr
;
1178 init_waitqueue_head(&area
->wq
);
1179 /* Reserve the 1st slot for get_trampoline_vaddr() */
1180 set_bit(0, area
->bitmap
);
1181 atomic_set(&area
->slot_count
, 1);
1182 copy_to_page(area
->page
, 0, &insn
, UPROBE_SWBP_INSN_SIZE
);
1184 if (!xol_add_vma(mm
, area
))
1187 __free_page(area
->page
);
1189 kfree(area
->bitmap
);
1197 * get_xol_area - Allocate process's xol_area if necessary.
1198 * This area will be used for storing instructions for execution out of line.
1200 * Returns the allocated area or NULL.
1202 static struct xol_area
*get_xol_area(void)
1204 struct mm_struct
*mm
= current
->mm
;
1205 struct xol_area
*area
;
1207 if (!mm
->uprobes_state
.xol_area
)
1208 __create_xol_area(0);
1210 area
= mm
->uprobes_state
.xol_area
;
1211 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1216 * uprobe_clear_state - Free the area allocated for slots.
1218 void uprobe_clear_state(struct mm_struct
*mm
)
1220 struct xol_area
*area
= mm
->uprobes_state
.xol_area
;
1225 put_page(area
->page
);
1226 kfree(area
->bitmap
);
1230 void uprobe_start_dup_mmap(void)
1232 percpu_down_read(&dup_mmap_sem
);
1235 void uprobe_end_dup_mmap(void)
1237 percpu_up_read(&dup_mmap_sem
);
1240 void uprobe_dup_mmap(struct mm_struct
*oldmm
, struct mm_struct
*newmm
)
1242 newmm
->uprobes_state
.xol_area
= NULL
;
1244 if (test_bit(MMF_HAS_UPROBES
, &oldmm
->flags
)) {
1245 set_bit(MMF_HAS_UPROBES
, &newmm
->flags
);
1246 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1247 set_bit(MMF_RECALC_UPROBES
, &newmm
->flags
);
1252 * - search for a free slot.
1254 static unsigned long xol_take_insn_slot(struct xol_area
*area
)
1256 unsigned long slot_addr
;
1260 slot_nr
= find_first_zero_bit(area
->bitmap
, UINSNS_PER_PAGE
);
1261 if (slot_nr
< UINSNS_PER_PAGE
) {
1262 if (!test_and_set_bit(slot_nr
, area
->bitmap
))
1265 slot_nr
= UINSNS_PER_PAGE
;
1268 wait_event(area
->wq
, (atomic_read(&area
->slot_count
) < UINSNS_PER_PAGE
));
1269 } while (slot_nr
>= UINSNS_PER_PAGE
);
1271 slot_addr
= area
->vaddr
+ (slot_nr
* UPROBE_XOL_SLOT_BYTES
);
1272 atomic_inc(&area
->slot_count
);
1278 * xol_get_insn_slot - allocate a slot for xol.
1279 * Returns the allocated slot address or 0.
1281 static unsigned long xol_get_insn_slot(struct uprobe
*uprobe
)
1283 struct xol_area
*area
;
1284 unsigned long xol_vaddr
;
1286 area
= get_xol_area();
1290 xol_vaddr
= xol_take_insn_slot(area
);
1291 if (unlikely(!xol_vaddr
))
1294 arch_uprobe_copy_ixol(area
->page
, xol_vaddr
,
1295 &uprobe
->arch
.ixol
, sizeof(uprobe
->arch
.ixol
));
1301 * xol_free_insn_slot - If slot was earlier allocated by
1302 * @xol_get_insn_slot(), make the slot available for
1303 * subsequent requests.
1305 static void xol_free_insn_slot(struct task_struct
*tsk
)
1307 struct xol_area
*area
;
1308 unsigned long vma_end
;
1309 unsigned long slot_addr
;
1311 if (!tsk
->mm
|| !tsk
->mm
->uprobes_state
.xol_area
|| !tsk
->utask
)
1314 slot_addr
= tsk
->utask
->xol_vaddr
;
1315 if (unlikely(!slot_addr
))
1318 area
= tsk
->mm
->uprobes_state
.xol_area
;
1319 vma_end
= area
->vaddr
+ PAGE_SIZE
;
1320 if (area
->vaddr
<= slot_addr
&& slot_addr
< vma_end
) {
1321 unsigned long offset
;
1324 offset
= slot_addr
- area
->vaddr
;
1325 slot_nr
= offset
/ UPROBE_XOL_SLOT_BYTES
;
1326 if (slot_nr
>= UINSNS_PER_PAGE
)
1329 clear_bit(slot_nr
, area
->bitmap
);
1330 atomic_dec(&area
->slot_count
);
1331 if (waitqueue_active(&area
->wq
))
1334 tsk
->utask
->xol_vaddr
= 0;
1338 void __weak
arch_uprobe_copy_ixol(struct page
*page
, unsigned long vaddr
,
1339 void *src
, unsigned long len
)
1341 /* Initialize the slot */
1342 copy_to_page(page
, vaddr
, src
, len
);
1345 * We probably need flush_icache_user_range() but it needs vma.
1346 * This should work on most of architectures by default. If
1347 * architecture needs to do something different it can define
1348 * its own version of the function.
1350 flush_dcache_page(page
);
1354 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1355 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1357 * Return the address of the breakpoint instruction.
1359 unsigned long __weak
uprobe_get_swbp_addr(struct pt_regs
*regs
)
1361 return instruction_pointer(regs
) - UPROBE_SWBP_INSN_SIZE
;
1365 * Called with no locks held.
1366 * Called in context of a exiting or a exec-ing thread.
1368 void uprobe_free_utask(struct task_struct
*t
)
1370 struct uprobe_task
*utask
= t
->utask
;
1371 struct return_instance
*ri
, *tmp
;
1376 if (utask
->active_uprobe
)
1377 put_uprobe(utask
->active_uprobe
);
1379 ri
= utask
->return_instances
;
1384 put_uprobe(tmp
->uprobe
);
1388 xol_free_insn_slot(t
);
1394 * Allocate a uprobe_task object for the task if if necessary.
1395 * Called when the thread hits a breakpoint.
1398 * - pointer to new uprobe_task on success
1401 static struct uprobe_task
*get_utask(void)
1403 if (!current
->utask
)
1404 current
->utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1405 return current
->utask
;
1408 static int dup_utask(struct task_struct
*t
, struct uprobe_task
*o_utask
)
1410 struct uprobe_task
*n_utask
;
1411 struct return_instance
**p
, *o
, *n
;
1413 n_utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1418 p
= &n_utask
->return_instances
;
1419 for (o
= o_utask
->return_instances
; o
; o
= o
->next
) {
1420 n
= kmalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1425 atomic_inc(&n
->uprobe
->ref
);
1436 static void uprobe_warn(struct task_struct
*t
, const char *msg
)
1438 pr_warn("uprobe: %s:%d failed to %s\n",
1439 current
->comm
, current
->pid
, msg
);
1442 static void dup_xol_work(struct callback_head
*work
)
1444 if (current
->flags
& PF_EXITING
)
1447 if (!__create_xol_area(current
->utask
->dup_xol_addr
))
1448 uprobe_warn(current
, "dup xol area");
1452 * Called in context of a new clone/fork from copy_process.
1454 void uprobe_copy_process(struct task_struct
*t
, unsigned long flags
)
1456 struct uprobe_task
*utask
= current
->utask
;
1457 struct mm_struct
*mm
= current
->mm
;
1458 struct xol_area
*area
;
1462 if (!utask
|| !utask
->return_instances
)
1465 if (mm
== t
->mm
&& !(flags
& CLONE_VFORK
))
1468 if (dup_utask(t
, utask
))
1469 return uprobe_warn(t
, "dup ret instances");
1471 /* The task can fork() after dup_xol_work() fails */
1472 area
= mm
->uprobes_state
.xol_area
;
1474 return uprobe_warn(t
, "dup xol area");
1479 t
->utask
->dup_xol_addr
= area
->vaddr
;
1480 init_task_work(&t
->utask
->dup_xol_work
, dup_xol_work
);
1481 task_work_add(t
, &t
->utask
->dup_xol_work
, true);
1485 * Current area->vaddr notion assume the trampoline address is always
1486 * equal area->vaddr.
1488 * Returns -1 in case the xol_area is not allocated.
1490 static unsigned long get_trampoline_vaddr(void)
1492 struct xol_area
*area
;
1493 unsigned long trampoline_vaddr
= -1;
1495 area
= current
->mm
->uprobes_state
.xol_area
;
1496 smp_read_barrier_depends();
1498 trampoline_vaddr
= area
->vaddr
;
1500 return trampoline_vaddr
;
1503 static void prepare_uretprobe(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1505 struct return_instance
*ri
;
1506 struct uprobe_task
*utask
;
1507 unsigned long orig_ret_vaddr
, trampoline_vaddr
;
1508 bool chained
= false;
1510 if (!get_xol_area())
1513 utask
= get_utask();
1517 if (utask
->depth
>= MAX_URETPROBE_DEPTH
) {
1518 printk_ratelimited(KERN_INFO
"uprobe: omit uretprobe due to"
1519 " nestedness limit pid/tgid=%d/%d\n",
1520 current
->pid
, current
->tgid
);
1524 ri
= kzalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1528 trampoline_vaddr
= get_trampoline_vaddr();
1529 orig_ret_vaddr
= arch_uretprobe_hijack_return_addr(trampoline_vaddr
, regs
);
1530 if (orig_ret_vaddr
== -1)
1534 * We don't want to keep trampoline address in stack, rather keep the
1535 * original return address of first caller thru all the consequent
1536 * instances. This also makes breakpoint unwrapping easier.
1538 if (orig_ret_vaddr
== trampoline_vaddr
) {
1539 if (!utask
->return_instances
) {
1541 * This situation is not possible. Likely we have an
1542 * attack from user-space.
1544 pr_warn("uprobe: unable to set uretprobe pid/tgid=%d/%d\n",
1545 current
->pid
, current
->tgid
);
1550 orig_ret_vaddr
= utask
->return_instances
->orig_ret_vaddr
;
1553 atomic_inc(&uprobe
->ref
);
1554 ri
->uprobe
= uprobe
;
1555 ri
->func
= instruction_pointer(regs
);
1556 ri
->orig_ret_vaddr
= orig_ret_vaddr
;
1557 ri
->chained
= chained
;
1561 /* add instance to the stack */
1562 ri
->next
= utask
->return_instances
;
1563 utask
->return_instances
= ri
;
1571 /* Prepare to single-step probed instruction out of line. */
1573 pre_ssout(struct uprobe
*uprobe
, struct pt_regs
*regs
, unsigned long bp_vaddr
)
1575 struct uprobe_task
*utask
;
1576 unsigned long xol_vaddr
;
1579 utask
= get_utask();
1583 xol_vaddr
= xol_get_insn_slot(uprobe
);
1587 utask
->xol_vaddr
= xol_vaddr
;
1588 utask
->vaddr
= bp_vaddr
;
1590 err
= arch_uprobe_pre_xol(&uprobe
->arch
, regs
);
1591 if (unlikely(err
)) {
1592 xol_free_insn_slot(current
);
1596 utask
->active_uprobe
= uprobe
;
1597 utask
->state
= UTASK_SSTEP
;
1602 * If we are singlestepping, then ensure this thread is not connected to
1603 * non-fatal signals until completion of singlestep. When xol insn itself
1604 * triggers the signal, restart the original insn even if the task is
1605 * already SIGKILL'ed (since coredump should report the correct ip). This
1606 * is even more important if the task has a handler for SIGSEGV/etc, The
1607 * _same_ instruction should be repeated again after return from the signal
1608 * handler, and SSTEP can never finish in this case.
1610 bool uprobe_deny_signal(void)
1612 struct task_struct
*t
= current
;
1613 struct uprobe_task
*utask
= t
->utask
;
1615 if (likely(!utask
|| !utask
->active_uprobe
))
1618 WARN_ON_ONCE(utask
->state
!= UTASK_SSTEP
);
1620 if (signal_pending(t
)) {
1621 spin_lock_irq(&t
->sighand
->siglock
);
1622 clear_tsk_thread_flag(t
, TIF_SIGPENDING
);
1623 spin_unlock_irq(&t
->sighand
->siglock
);
1625 if (__fatal_signal_pending(t
) || arch_uprobe_xol_was_trapped(t
)) {
1626 utask
->state
= UTASK_SSTEP_TRAPPED
;
1627 set_tsk_thread_flag(t
, TIF_UPROBE
);
1628 set_tsk_thread_flag(t
, TIF_NOTIFY_RESUME
);
1635 static void mmf_recalc_uprobes(struct mm_struct
*mm
)
1637 struct vm_area_struct
*vma
;
1639 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1640 if (!valid_vma(vma
, false))
1643 * This is not strictly accurate, we can race with
1644 * uprobe_unregister() and see the already removed
1645 * uprobe if delete_uprobe() was not yet called.
1646 * Or this uprobe can be filtered out.
1648 if (vma_has_uprobes(vma
, vma
->vm_start
, vma
->vm_end
))
1652 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
1655 static int is_trap_at_addr(struct mm_struct
*mm
, unsigned long vaddr
)
1658 uprobe_opcode_t opcode
;
1661 pagefault_disable();
1662 result
= __copy_from_user_inatomic(&opcode
, (void __user
*)vaddr
,
1666 if (likely(result
== 0))
1669 result
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &page
, NULL
);
1673 copy_from_page(page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
1676 /* This needs to return true for any variant of the trap insn */
1677 return is_trap_insn(&opcode
);
1680 static struct uprobe
*find_active_uprobe(unsigned long bp_vaddr
, int *is_swbp
)
1682 struct mm_struct
*mm
= current
->mm
;
1683 struct uprobe
*uprobe
= NULL
;
1684 struct vm_area_struct
*vma
;
1686 down_read(&mm
->mmap_sem
);
1687 vma
= find_vma(mm
, bp_vaddr
);
1688 if (vma
&& vma
->vm_start
<= bp_vaddr
) {
1689 if (valid_vma(vma
, false)) {
1690 struct inode
*inode
= file_inode(vma
->vm_file
);
1691 loff_t offset
= vaddr_to_offset(vma
, bp_vaddr
);
1693 uprobe
= find_uprobe(inode
, offset
);
1697 *is_swbp
= is_trap_at_addr(mm
, bp_vaddr
);
1702 if (!uprobe
&& test_and_clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
))
1703 mmf_recalc_uprobes(mm
);
1704 up_read(&mm
->mmap_sem
);
1709 static void handler_chain(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1711 struct uprobe_consumer
*uc
;
1712 int remove
= UPROBE_HANDLER_REMOVE
;
1713 bool need_prep
= false; /* prepare return uprobe, when needed */
1715 down_read(&uprobe
->register_rwsem
);
1716 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
1720 rc
= uc
->handler(uc
, regs
);
1721 WARN(rc
& ~UPROBE_HANDLER_MASK
,
1722 "bad rc=0x%x from %pf()\n", rc
, uc
->handler
);
1725 if (uc
->ret_handler
)
1731 if (need_prep
&& !remove
)
1732 prepare_uretprobe(uprobe
, regs
); /* put bp at return */
1734 if (remove
&& uprobe
->consumers
) {
1735 WARN_ON(!uprobe_is_active(uprobe
));
1736 unapply_uprobe(uprobe
, current
->mm
);
1738 up_read(&uprobe
->register_rwsem
);
1742 handle_uretprobe_chain(struct return_instance
*ri
, struct pt_regs
*regs
)
1744 struct uprobe
*uprobe
= ri
->uprobe
;
1745 struct uprobe_consumer
*uc
;
1747 down_read(&uprobe
->register_rwsem
);
1748 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
1749 if (uc
->ret_handler
)
1750 uc
->ret_handler(uc
, ri
->func
, regs
);
1752 up_read(&uprobe
->register_rwsem
);
1755 static bool handle_trampoline(struct pt_regs
*regs
)
1757 struct uprobe_task
*utask
;
1758 struct return_instance
*ri
, *tmp
;
1761 utask
= current
->utask
;
1765 ri
= utask
->return_instances
;
1770 * TODO: we should throw out return_instance's invalidated by
1771 * longjmp(), currently we assume that the probed function always
1774 instruction_pointer_set(regs
, ri
->orig_ret_vaddr
);
1777 handle_uretprobe_chain(ri
, regs
);
1779 chained
= ri
->chained
;
1780 put_uprobe(ri
->uprobe
);
1792 utask
->return_instances
= ri
;
1797 bool __weak
arch_uprobe_ignore(struct arch_uprobe
*aup
, struct pt_regs
*regs
)
1803 * Run handler and ask thread to singlestep.
1804 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1806 static void handle_swbp(struct pt_regs
*regs
)
1808 struct uprobe
*uprobe
;
1809 unsigned long bp_vaddr
;
1810 int uninitialized_var(is_swbp
);
1812 bp_vaddr
= uprobe_get_swbp_addr(regs
);
1813 if (bp_vaddr
== get_trampoline_vaddr()) {
1814 if (handle_trampoline(regs
))
1817 pr_warn("uprobe: unable to handle uretprobe pid/tgid=%d/%d\n",
1818 current
->pid
, current
->tgid
);
1821 uprobe
= find_active_uprobe(bp_vaddr
, &is_swbp
);
1824 /* No matching uprobe; signal SIGTRAP. */
1825 send_sig(SIGTRAP
, current
, 0);
1828 * Either we raced with uprobe_unregister() or we can't
1829 * access this memory. The latter is only possible if
1830 * another thread plays with our ->mm. In both cases
1831 * we can simply restart. If this vma was unmapped we
1832 * can pretend this insn was not executed yet and get
1833 * the (correct) SIGSEGV after restart.
1835 instruction_pointer_set(regs
, bp_vaddr
);
1840 /* change it in advance for ->handler() and restart */
1841 instruction_pointer_set(regs
, bp_vaddr
);
1844 * TODO: move copy_insn/etc into _register and remove this hack.
1845 * After we hit the bp, _unregister + _register can install the
1846 * new and not-yet-analyzed uprobe at the same address, restart.
1848 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1849 if (unlikely(!test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
)))
1852 /* Tracing handlers use ->utask to communicate with fetch methods */
1856 if (arch_uprobe_ignore(&uprobe
->arch
, regs
))
1859 handler_chain(uprobe
, regs
);
1861 if (arch_uprobe_skip_sstep(&uprobe
->arch
, regs
))
1864 if (!pre_ssout(uprobe
, regs
, bp_vaddr
))
1867 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1873 * Perform required fix-ups and disable singlestep.
1874 * Allow pending signals to take effect.
1876 static void handle_singlestep(struct uprobe_task
*utask
, struct pt_regs
*regs
)
1878 struct uprobe
*uprobe
;
1881 uprobe
= utask
->active_uprobe
;
1882 if (utask
->state
== UTASK_SSTEP_ACK
)
1883 err
= arch_uprobe_post_xol(&uprobe
->arch
, regs
);
1884 else if (utask
->state
== UTASK_SSTEP_TRAPPED
)
1885 arch_uprobe_abort_xol(&uprobe
->arch
, regs
);
1890 utask
->active_uprobe
= NULL
;
1891 utask
->state
= UTASK_RUNNING
;
1892 xol_free_insn_slot(current
);
1894 spin_lock_irq(¤t
->sighand
->siglock
);
1895 recalc_sigpending(); /* see uprobe_deny_signal() */
1896 spin_unlock_irq(¤t
->sighand
->siglock
);
1898 if (unlikely(err
)) {
1899 uprobe_warn(current
, "execute the probed insn, sending SIGILL.");
1900 force_sig_info(SIGILL
, SEND_SIG_FORCED
, current
);
1905 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1906 * allows the thread to return from interrupt. After that handle_swbp()
1907 * sets utask->active_uprobe.
1909 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1910 * and allows the thread to return from interrupt.
1912 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1913 * uprobe_notify_resume().
1915 void uprobe_notify_resume(struct pt_regs
*regs
)
1917 struct uprobe_task
*utask
;
1919 clear_thread_flag(TIF_UPROBE
);
1921 utask
= current
->utask
;
1922 if (utask
&& utask
->active_uprobe
)
1923 handle_singlestep(utask
, regs
);
1929 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1930 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1932 int uprobe_pre_sstep_notifier(struct pt_regs
*regs
)
1937 if (!test_bit(MMF_HAS_UPROBES
, ¤t
->mm
->flags
) &&
1938 (!current
->utask
|| !current
->utask
->return_instances
))
1941 set_thread_flag(TIF_UPROBE
);
1946 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1947 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1949 int uprobe_post_sstep_notifier(struct pt_regs
*regs
)
1951 struct uprobe_task
*utask
= current
->utask
;
1953 if (!current
->mm
|| !utask
|| !utask
->active_uprobe
)
1954 /* task is currently not uprobed */
1957 utask
->state
= UTASK_SSTEP_ACK
;
1958 set_thread_flag(TIF_UPROBE
);
1962 static struct notifier_block uprobe_exception_nb
= {
1963 .notifier_call
= arch_uprobe_exception_notify
,
1964 .priority
= INT_MAX
-1, /* notified after kprobes, kgdb */
1967 static int __init
init_uprobes(void)
1971 for (i
= 0; i
< UPROBES_HASH_SZ
; i
++)
1972 mutex_init(&uprobes_mmap_mutex
[i
]);
1974 if (percpu_init_rwsem(&dup_mmap_sem
))
1977 return register_die_notifier(&uprobe_exception_nb
);
1979 __initcall(init_uprobes
);