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 */
78 * The generic code assumes that it has two members of unknown type
79 * owned by the arch-specific code:
81 * insn - copy_insn() saves the original instruction here for
82 * arch_uprobe_analyze_insn().
84 * ixol - potentially modified instruction to execute out of
85 * line, copied to xol_area by xol_get_insn_slot().
87 struct arch_uprobe arch
;
90 struct return_instance
{
91 struct uprobe
*uprobe
;
93 unsigned long orig_ret_vaddr
; /* original return address */
94 bool chained
; /* true, if instance is nested */
96 struct return_instance
*next
; /* keep as stack */
100 * Execute out of line area: anonymous executable mapping installed
101 * by the probed task to execute the copy of the original instruction
102 * mangled by set_swbp().
104 * On a breakpoint hit, thread contests for a slot. It frees the
105 * slot after singlestep. Currently a fixed number of slots are
109 wait_queue_head_t wq
; /* if all slots are busy */
110 atomic_t slot_count
; /* number of in-use slots */
111 unsigned long *bitmap
; /* 0 = free slot */
115 * We keep the vma's vm_start rather than a pointer to the vma
116 * itself. The probed process or a naughty kernel module could make
117 * the vma go away, and we must handle that reasonably gracefully.
119 unsigned long vaddr
; /* Page(s) of instruction slots */
123 * valid_vma: Verify if the specified vma is an executable vma
124 * Relax restrictions while unregistering: vm_flags might have
125 * changed after breakpoint was inserted.
126 * - is_register: indicates if we are in register context.
127 * - Return 1 if the specified virtual address is in an
130 static bool valid_vma(struct vm_area_struct
*vma
, bool is_register
)
132 vm_flags_t flags
= VM_HUGETLB
| VM_MAYEXEC
| VM_SHARED
;
137 return vma
->vm_file
&& (vma
->vm_flags
& flags
) == VM_MAYEXEC
;
140 static unsigned long offset_to_vaddr(struct vm_area_struct
*vma
, loff_t offset
)
142 return vma
->vm_start
+ offset
- ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
145 static loff_t
vaddr_to_offset(struct vm_area_struct
*vma
, unsigned long vaddr
)
147 return ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
) + (vaddr
- vma
->vm_start
);
151 * __replace_page - replace page in vma by new page.
152 * based on replace_page in mm/ksm.c
154 * @vma: vma that holds the pte pointing to page
155 * @addr: address the old @page is mapped at
156 * @page: the cowed page we are replacing by kpage
157 * @kpage: the modified page we replace page by
159 * Returns 0 on success, -EFAULT on failure.
161 static int __replace_page(struct vm_area_struct
*vma
, unsigned long addr
,
162 struct page
*page
, struct page
*kpage
)
164 struct mm_struct
*mm
= vma
->vm_mm
;
168 /* For mmu_notifiers */
169 const unsigned long mmun_start
= addr
;
170 const unsigned long mmun_end
= addr
+ PAGE_SIZE
;
172 /* For try_to_free_swap() and munlock_vma_page() below */
175 mmu_notifier_invalidate_range_start(mm
, mmun_start
, mmun_end
);
177 ptep
= page_check_address(page
, mm
, addr
, &ptl
, 0);
182 page_add_new_anon_rmap(kpage
, vma
, addr
);
184 if (!PageAnon(page
)) {
185 dec_mm_counter(mm
, MM_FILEPAGES
);
186 inc_mm_counter(mm
, MM_ANONPAGES
);
189 flush_cache_page(vma
, addr
, pte_pfn(*ptep
));
190 ptep_clear_flush(vma
, addr
, ptep
);
191 set_pte_at_notify(mm
, addr
, ptep
, mk_pte(kpage
, vma
->vm_page_prot
));
193 page_remove_rmap(page
);
194 if (!page_mapped(page
))
195 try_to_free_swap(page
);
196 pte_unmap_unlock(ptep
, ptl
);
198 if (vma
->vm_flags
& VM_LOCKED
)
199 munlock_vma_page(page
);
204 mmu_notifier_invalidate_range_end(mm
, mmun_start
, mmun_end
);
210 * is_swbp_insn - check if instruction is breakpoint instruction.
211 * @insn: instruction to be checked.
212 * Default implementation of is_swbp_insn
213 * Returns true if @insn is a breakpoint instruction.
215 bool __weak
is_swbp_insn(uprobe_opcode_t
*insn
)
217 return *insn
== UPROBE_SWBP_INSN
;
221 * is_trap_insn - check if instruction is breakpoint instruction.
222 * @insn: instruction to be checked.
223 * Default implementation of is_trap_insn
224 * Returns true if @insn is a breakpoint instruction.
226 * This function is needed for the case where an architecture has multiple
227 * trap instructions (like powerpc).
229 bool __weak
is_trap_insn(uprobe_opcode_t
*insn
)
231 return is_swbp_insn(insn
);
234 static void copy_from_page(struct page
*page
, unsigned long vaddr
, void *dst
, int len
)
236 void *kaddr
= kmap_atomic(page
);
237 memcpy(dst
, kaddr
+ (vaddr
& ~PAGE_MASK
), len
);
238 kunmap_atomic(kaddr
);
241 static void copy_to_page(struct page
*page
, unsigned long vaddr
, const void *src
, int len
)
243 void *kaddr
= kmap_atomic(page
);
244 memcpy(kaddr
+ (vaddr
& ~PAGE_MASK
), src
, len
);
245 kunmap_atomic(kaddr
);
248 static int verify_opcode(struct page
*page
, unsigned long vaddr
, uprobe_opcode_t
*new_opcode
)
250 uprobe_opcode_t old_opcode
;
254 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
255 * We do not check if it is any other 'trap variant' which could
256 * be conditional trap instruction such as the one powerpc supports.
258 * The logic is that we do not care if the underlying instruction
259 * is a trap variant; uprobes always wins over any other (gdb)
262 copy_from_page(page
, vaddr
, &old_opcode
, UPROBE_SWBP_INSN_SIZE
);
263 is_swbp
= is_swbp_insn(&old_opcode
);
265 if (is_swbp_insn(new_opcode
)) {
266 if (is_swbp
) /* register: already installed? */
269 if (!is_swbp
) /* unregister: was it changed by us? */
278 * Expect the breakpoint instruction to be the smallest size instruction for
279 * the architecture. If an arch has variable length instruction and the
280 * breakpoint instruction is not of the smallest length instruction
281 * supported by that architecture then we need to modify is_trap_at_addr and
282 * uprobe_write_opcode accordingly. This would never be a problem for archs
283 * that have fixed length instructions.
287 * uprobe_write_opcode - write the opcode at a given virtual address.
288 * @mm: the probed process address space.
289 * @vaddr: the virtual address to store the opcode.
290 * @opcode: opcode to be written at @vaddr.
292 * Called with mm->mmap_sem held (for read and with a reference to
295 * For mm @mm, write the opcode at @vaddr.
296 * Return 0 (success) or a negative errno.
298 int uprobe_write_opcode(struct mm_struct
*mm
, unsigned long vaddr
,
299 uprobe_opcode_t opcode
)
301 struct page
*old_page
, *new_page
;
302 struct vm_area_struct
*vma
;
306 /* Read the page with vaddr into memory */
307 ret
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &old_page
, &vma
);
311 ret
= verify_opcode(old_page
, vaddr
, &opcode
);
316 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, vaddr
);
320 __SetPageUptodate(new_page
);
322 copy_highpage(new_page
, old_page
);
323 copy_to_page(new_page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
325 ret
= anon_vma_prepare(vma
);
329 ret
= __replace_page(vma
, vaddr
, old_page
, new_page
);
332 page_cache_release(new_page
);
336 if (unlikely(ret
== -EAGAIN
))
342 * set_swbp - store breakpoint at a given address.
343 * @auprobe: arch specific probepoint information.
344 * @mm: the probed process address space.
345 * @vaddr: the virtual address to insert the opcode.
347 * For mm @mm, store the breakpoint instruction at @vaddr.
348 * Return 0 (success) or a negative errno.
350 int __weak
set_swbp(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
352 return uprobe_write_opcode(mm
, vaddr
, UPROBE_SWBP_INSN
);
356 * set_orig_insn - Restore the original instruction.
357 * @mm: the probed process address space.
358 * @auprobe: arch specific probepoint information.
359 * @vaddr: the virtual address to insert the opcode.
361 * For mm @mm, restore the original opcode (opcode) at @vaddr.
362 * Return 0 (success) or a negative errno.
365 set_orig_insn(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
367 return uprobe_write_opcode(mm
, vaddr
, *(uprobe_opcode_t
*)&auprobe
->insn
);
370 static int match_uprobe(struct uprobe
*l
, struct uprobe
*r
)
372 if (l
->inode
< r
->inode
)
375 if (l
->inode
> r
->inode
)
378 if (l
->offset
< r
->offset
)
381 if (l
->offset
> r
->offset
)
387 static struct uprobe
*__find_uprobe(struct inode
*inode
, loff_t offset
)
389 struct uprobe u
= { .inode
= inode
, .offset
= offset
};
390 struct rb_node
*n
= uprobes_tree
.rb_node
;
391 struct uprobe
*uprobe
;
395 uprobe
= rb_entry(n
, struct uprobe
, rb_node
);
396 match
= match_uprobe(&u
, uprobe
);
398 atomic_inc(&uprobe
->ref
);
411 * Find a uprobe corresponding to a given inode:offset
412 * Acquires uprobes_treelock
414 static struct uprobe
*find_uprobe(struct inode
*inode
, loff_t offset
)
416 struct uprobe
*uprobe
;
418 spin_lock(&uprobes_treelock
);
419 uprobe
= __find_uprobe(inode
, offset
);
420 spin_unlock(&uprobes_treelock
);
425 static struct uprobe
*__insert_uprobe(struct uprobe
*uprobe
)
427 struct rb_node
**p
= &uprobes_tree
.rb_node
;
428 struct rb_node
*parent
= NULL
;
434 u
= rb_entry(parent
, struct uprobe
, rb_node
);
435 match
= match_uprobe(uprobe
, u
);
442 p
= &parent
->rb_left
;
444 p
= &parent
->rb_right
;
449 rb_link_node(&uprobe
->rb_node
, parent
, p
);
450 rb_insert_color(&uprobe
->rb_node
, &uprobes_tree
);
451 /* get access + creation ref */
452 atomic_set(&uprobe
->ref
, 2);
458 * Acquire uprobes_treelock.
459 * Matching uprobe already exists in rbtree;
460 * increment (access refcount) and return the matching uprobe.
462 * No matching uprobe; insert the uprobe in rb_tree;
463 * get a double refcount (access + creation) and return NULL.
465 static struct uprobe
*insert_uprobe(struct uprobe
*uprobe
)
469 spin_lock(&uprobes_treelock
);
470 u
= __insert_uprobe(uprobe
);
471 spin_unlock(&uprobes_treelock
);
476 static void put_uprobe(struct uprobe
*uprobe
)
478 if (atomic_dec_and_test(&uprobe
->ref
))
482 static struct uprobe
*alloc_uprobe(struct inode
*inode
, loff_t offset
)
484 struct uprobe
*uprobe
, *cur_uprobe
;
486 uprobe
= kzalloc(sizeof(struct uprobe
), GFP_KERNEL
);
490 uprobe
->inode
= igrab(inode
);
491 uprobe
->offset
= offset
;
492 init_rwsem(&uprobe
->register_rwsem
);
493 init_rwsem(&uprobe
->consumer_rwsem
);
494 /* For now assume that the instruction need not be single-stepped */
495 __set_bit(UPROBE_SKIP_SSTEP
, &uprobe
->flags
);
497 /* add to uprobes_tree, sorted on inode:offset */
498 cur_uprobe
= insert_uprobe(uprobe
);
500 /* a uprobe exists for this inode:offset combination */
510 static void consumer_add(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
512 down_write(&uprobe
->consumer_rwsem
);
513 uc
->next
= uprobe
->consumers
;
514 uprobe
->consumers
= uc
;
515 up_write(&uprobe
->consumer_rwsem
);
519 * For uprobe @uprobe, delete the consumer @uc.
520 * Return true if the @uc is deleted successfully
523 static bool consumer_del(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
525 struct uprobe_consumer
**con
;
528 down_write(&uprobe
->consumer_rwsem
);
529 for (con
= &uprobe
->consumers
; *con
; con
= &(*con
)->next
) {
536 up_write(&uprobe
->consumer_rwsem
);
541 static int __copy_insn(struct address_space
*mapping
, struct file
*filp
,
542 void *insn
, int nbytes
, loff_t offset
)
546 if (!mapping
->a_ops
->readpage
)
549 * Ensure that the page that has the original instruction is
550 * populated and in page-cache.
552 page
= read_mapping_page(mapping
, offset
>> PAGE_CACHE_SHIFT
, filp
);
554 return PTR_ERR(page
);
556 copy_from_page(page
, offset
, insn
, nbytes
);
557 page_cache_release(page
);
562 static int copy_insn(struct uprobe
*uprobe
, struct file
*filp
)
564 struct address_space
*mapping
= uprobe
->inode
->i_mapping
;
565 loff_t offs
= uprobe
->offset
;
566 void *insn
= &uprobe
->arch
.insn
;
567 int size
= sizeof(uprobe
->arch
.insn
);
570 /* Copy only available bytes, -EIO if nothing was read */
572 if (offs
>= i_size_read(uprobe
->inode
))
575 len
= min_t(int, size
, PAGE_SIZE
- (offs
& ~PAGE_MASK
));
576 err
= __copy_insn(mapping
, filp
, insn
, len
, offs
);
588 static int prepare_uprobe(struct uprobe
*uprobe
, struct file
*file
,
589 struct mm_struct
*mm
, unsigned long vaddr
)
593 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
596 /* TODO: move this into _register, until then we abuse this sem. */
597 down_write(&uprobe
->consumer_rwsem
);
598 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
601 ret
= copy_insn(uprobe
, file
);
606 if (is_trap_insn((uprobe_opcode_t
*)&uprobe
->arch
.insn
))
609 ret
= arch_uprobe_analyze_insn(&uprobe
->arch
, mm
, vaddr
);
613 /* uprobe_write_opcode() assumes we don't cross page boundary */
614 BUG_ON((uprobe
->offset
& ~PAGE_MASK
) +
615 UPROBE_SWBP_INSN_SIZE
> PAGE_SIZE
);
617 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
618 set_bit(UPROBE_COPY_INSN
, &uprobe
->flags
);
621 up_write(&uprobe
->consumer_rwsem
);
626 static inline bool consumer_filter(struct uprobe_consumer
*uc
,
627 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
629 return !uc
->filter
|| uc
->filter(uc
, ctx
, mm
);
632 static bool filter_chain(struct uprobe
*uprobe
,
633 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
635 struct uprobe_consumer
*uc
;
638 down_read(&uprobe
->consumer_rwsem
);
639 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
640 ret
= consumer_filter(uc
, ctx
, mm
);
644 up_read(&uprobe
->consumer_rwsem
);
650 install_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
,
651 struct vm_area_struct
*vma
, unsigned long vaddr
)
656 ret
= prepare_uprobe(uprobe
, vma
->vm_file
, mm
, vaddr
);
661 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
662 * the task can hit this breakpoint right after __replace_page().
664 first_uprobe
= !test_bit(MMF_HAS_UPROBES
, &mm
->flags
);
666 set_bit(MMF_HAS_UPROBES
, &mm
->flags
);
668 ret
= set_swbp(&uprobe
->arch
, mm
, vaddr
);
670 clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
671 else if (first_uprobe
)
672 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
678 remove_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
680 set_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
681 return set_orig_insn(&uprobe
->arch
, mm
, vaddr
);
684 static inline bool uprobe_is_active(struct uprobe
*uprobe
)
686 return !RB_EMPTY_NODE(&uprobe
->rb_node
);
689 * There could be threads that have already hit the breakpoint. They
690 * will recheck the current insn and restart if find_uprobe() fails.
691 * See find_active_uprobe().
693 static void delete_uprobe(struct uprobe
*uprobe
)
695 if (WARN_ON(!uprobe_is_active(uprobe
)))
698 spin_lock(&uprobes_treelock
);
699 rb_erase(&uprobe
->rb_node
, &uprobes_tree
);
700 spin_unlock(&uprobes_treelock
);
701 RB_CLEAR_NODE(&uprobe
->rb_node
); /* for uprobe_is_active() */
707 struct map_info
*next
;
708 struct mm_struct
*mm
;
712 static inline struct map_info
*free_map_info(struct map_info
*info
)
714 struct map_info
*next
= info
->next
;
719 static struct map_info
*
720 build_map_info(struct address_space
*mapping
, loff_t offset
, bool is_register
)
722 unsigned long pgoff
= offset
>> PAGE_SHIFT
;
723 struct vm_area_struct
*vma
;
724 struct map_info
*curr
= NULL
;
725 struct map_info
*prev
= NULL
;
726 struct map_info
*info
;
730 mutex_lock(&mapping
->i_mmap_mutex
);
731 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
, pgoff
, pgoff
) {
732 if (!valid_vma(vma
, is_register
))
735 if (!prev
&& !more
) {
737 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
738 * reclaim. This is optimistic, no harm done if it fails.
740 prev
= kmalloc(sizeof(struct map_info
),
741 GFP_NOWAIT
| __GFP_NOMEMALLOC
| __GFP_NOWARN
);
750 if (!atomic_inc_not_zero(&vma
->vm_mm
->mm_users
))
758 info
->mm
= vma
->vm_mm
;
759 info
->vaddr
= offset_to_vaddr(vma
, offset
);
761 mutex_unlock(&mapping
->i_mmap_mutex
);
773 info
= kmalloc(sizeof(struct map_info
), GFP_KERNEL
);
775 curr
= ERR_PTR(-ENOMEM
);
785 prev
= free_map_info(prev
);
790 register_for_each_vma(struct uprobe
*uprobe
, struct uprobe_consumer
*new)
792 bool is_register
= !!new;
793 struct map_info
*info
;
796 percpu_down_write(&dup_mmap_sem
);
797 info
= build_map_info(uprobe
->inode
->i_mapping
,
798 uprobe
->offset
, is_register
);
805 struct mm_struct
*mm
= info
->mm
;
806 struct vm_area_struct
*vma
;
808 if (err
&& is_register
)
811 down_write(&mm
->mmap_sem
);
812 vma
= find_vma(mm
, info
->vaddr
);
813 if (!vma
|| !valid_vma(vma
, is_register
) ||
814 file_inode(vma
->vm_file
) != uprobe
->inode
)
817 if (vma
->vm_start
> info
->vaddr
||
818 vaddr_to_offset(vma
, info
->vaddr
) != uprobe
->offset
)
822 /* consult only the "caller", new consumer. */
823 if (consumer_filter(new,
824 UPROBE_FILTER_REGISTER
, mm
))
825 err
= install_breakpoint(uprobe
, mm
, vma
, info
->vaddr
);
826 } else if (test_bit(MMF_HAS_UPROBES
, &mm
->flags
)) {
827 if (!filter_chain(uprobe
,
828 UPROBE_FILTER_UNREGISTER
, mm
))
829 err
|= remove_breakpoint(uprobe
, mm
, info
->vaddr
);
833 up_write(&mm
->mmap_sem
);
836 info
= free_map_info(info
);
839 percpu_up_write(&dup_mmap_sem
);
843 static int __uprobe_register(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
845 consumer_add(uprobe
, uc
);
846 return register_for_each_vma(uprobe
, uc
);
849 static void __uprobe_unregister(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
853 if (!consumer_del(uprobe
, uc
)) /* WARN? */
856 err
= register_for_each_vma(uprobe
, NULL
);
857 /* TODO : cant unregister? schedule a worker thread */
858 if (!uprobe
->consumers
&& !err
)
859 delete_uprobe(uprobe
);
863 * uprobe_register - register a probe
864 * @inode: the file in which the probe has to be placed.
865 * @offset: offset from the start of the file.
866 * @uc: information on howto handle the probe..
868 * Apart from the access refcount, uprobe_register() takes a creation
869 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
870 * inserted into the rbtree (i.e first consumer for a @inode:@offset
871 * tuple). Creation refcount stops uprobe_unregister from freeing the
872 * @uprobe even before the register operation is complete. Creation
873 * refcount is released when the last @uc for the @uprobe
876 * Return errno if it cannot successully install probes
877 * else return 0 (success)
879 int uprobe_register(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
881 struct uprobe
*uprobe
;
884 /* Uprobe must have at least one set consumer */
885 if (!uc
->handler
&& !uc
->ret_handler
)
888 /* Racy, just to catch the obvious mistakes */
889 if (offset
> i_size_read(inode
))
893 uprobe
= alloc_uprobe(inode
, offset
);
897 * We can race with uprobe_unregister()->delete_uprobe().
898 * Check uprobe_is_active() and retry if it is false.
900 down_write(&uprobe
->register_rwsem
);
902 if (likely(uprobe_is_active(uprobe
))) {
903 ret
= __uprobe_register(uprobe
, uc
);
905 __uprobe_unregister(uprobe
, uc
);
907 up_write(&uprobe
->register_rwsem
);
910 if (unlikely(ret
== -EAGAIN
))
914 EXPORT_SYMBOL_GPL(uprobe_register
);
917 * uprobe_apply - unregister a already registered probe.
918 * @inode: the file in which the probe has to be removed.
919 * @offset: offset from the start of the file.
920 * @uc: consumer which wants to add more or remove some breakpoints
921 * @add: add or remove the breakpoints
923 int uprobe_apply(struct inode
*inode
, loff_t offset
,
924 struct uprobe_consumer
*uc
, bool add
)
926 struct uprobe
*uprobe
;
927 struct uprobe_consumer
*con
;
930 uprobe
= find_uprobe(inode
, offset
);
934 down_write(&uprobe
->register_rwsem
);
935 for (con
= uprobe
->consumers
; con
&& con
!= uc
; con
= con
->next
)
938 ret
= register_for_each_vma(uprobe
, add
? uc
: NULL
);
939 up_write(&uprobe
->register_rwsem
);
946 * uprobe_unregister - unregister a already registered probe.
947 * @inode: the file in which the probe has to be removed.
948 * @offset: offset from the start of the file.
949 * @uc: identify which probe if multiple probes are colocated.
951 void uprobe_unregister(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
953 struct uprobe
*uprobe
;
955 uprobe
= find_uprobe(inode
, offset
);
959 down_write(&uprobe
->register_rwsem
);
960 __uprobe_unregister(uprobe
, uc
);
961 up_write(&uprobe
->register_rwsem
);
964 EXPORT_SYMBOL_GPL(uprobe_unregister
);
966 static int unapply_uprobe(struct uprobe
*uprobe
, struct mm_struct
*mm
)
968 struct vm_area_struct
*vma
;
971 down_read(&mm
->mmap_sem
);
972 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
976 if (!valid_vma(vma
, false) ||
977 file_inode(vma
->vm_file
) != uprobe
->inode
)
980 offset
= (loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
;
981 if (uprobe
->offset
< offset
||
982 uprobe
->offset
>= offset
+ vma
->vm_end
- vma
->vm_start
)
985 vaddr
= offset_to_vaddr(vma
, uprobe
->offset
);
986 err
|= remove_breakpoint(uprobe
, mm
, vaddr
);
988 up_read(&mm
->mmap_sem
);
993 static struct rb_node
*
994 find_node_in_range(struct inode
*inode
, loff_t min
, loff_t max
)
996 struct rb_node
*n
= uprobes_tree
.rb_node
;
999 struct uprobe
*u
= rb_entry(n
, struct uprobe
, rb_node
);
1001 if (inode
< u
->inode
) {
1003 } else if (inode
> u
->inode
) {
1006 if (max
< u
->offset
)
1008 else if (min
> u
->offset
)
1019 * For a given range in vma, build a list of probes that need to be inserted.
1021 static void build_probe_list(struct inode
*inode
,
1022 struct vm_area_struct
*vma
,
1023 unsigned long start
, unsigned long end
,
1024 struct list_head
*head
)
1027 struct rb_node
*n
, *t
;
1030 INIT_LIST_HEAD(head
);
1031 min
= vaddr_to_offset(vma
, start
);
1032 max
= min
+ (end
- start
) - 1;
1034 spin_lock(&uprobes_treelock
);
1035 n
= find_node_in_range(inode
, min
, max
);
1037 for (t
= n
; t
; t
= rb_prev(t
)) {
1038 u
= rb_entry(t
, struct uprobe
, rb_node
);
1039 if (u
->inode
!= inode
|| u
->offset
< min
)
1041 list_add(&u
->pending_list
, head
);
1042 atomic_inc(&u
->ref
);
1044 for (t
= n
; (t
= rb_next(t
)); ) {
1045 u
= rb_entry(t
, struct uprobe
, rb_node
);
1046 if (u
->inode
!= inode
|| u
->offset
> max
)
1048 list_add(&u
->pending_list
, head
);
1049 atomic_inc(&u
->ref
);
1052 spin_unlock(&uprobes_treelock
);
1056 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1058 * Currently we ignore all errors and always return 0, the callers
1059 * can't handle the failure anyway.
1061 int uprobe_mmap(struct vm_area_struct
*vma
)
1063 struct list_head tmp_list
;
1064 struct uprobe
*uprobe
, *u
;
1065 struct inode
*inode
;
1067 if (no_uprobe_events() || !valid_vma(vma
, true))
1070 inode
= file_inode(vma
->vm_file
);
1074 mutex_lock(uprobes_mmap_hash(inode
));
1075 build_probe_list(inode
, vma
, vma
->vm_start
, vma
->vm_end
, &tmp_list
);
1077 * We can race with uprobe_unregister(), this uprobe can be already
1078 * removed. But in this case filter_chain() must return false, all
1079 * consumers have gone away.
1081 list_for_each_entry_safe(uprobe
, u
, &tmp_list
, pending_list
) {
1082 if (!fatal_signal_pending(current
) &&
1083 filter_chain(uprobe
, UPROBE_FILTER_MMAP
, vma
->vm_mm
)) {
1084 unsigned long vaddr
= offset_to_vaddr(vma
, uprobe
->offset
);
1085 install_breakpoint(uprobe
, vma
->vm_mm
, vma
, vaddr
);
1089 mutex_unlock(uprobes_mmap_hash(inode
));
1095 vma_has_uprobes(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1098 struct inode
*inode
;
1101 inode
= file_inode(vma
->vm_file
);
1103 min
= vaddr_to_offset(vma
, start
);
1104 max
= min
+ (end
- start
) - 1;
1106 spin_lock(&uprobes_treelock
);
1107 n
= find_node_in_range(inode
, min
, max
);
1108 spin_unlock(&uprobes_treelock
);
1114 * Called in context of a munmap of a vma.
1116 void uprobe_munmap(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1118 if (no_uprobe_events() || !valid_vma(vma
, false))
1121 if (!atomic_read(&vma
->vm_mm
->mm_users
)) /* called by mmput() ? */
1124 if (!test_bit(MMF_HAS_UPROBES
, &vma
->vm_mm
->flags
) ||
1125 test_bit(MMF_RECALC_UPROBES
, &vma
->vm_mm
->flags
))
1128 if (vma_has_uprobes(vma
, start
, end
))
1129 set_bit(MMF_RECALC_UPROBES
, &vma
->vm_mm
->flags
);
1132 /* Slot allocation for XOL */
1133 static int xol_add_vma(struct mm_struct
*mm
, struct xol_area
*area
)
1135 int ret
= -EALREADY
;
1137 down_write(&mm
->mmap_sem
);
1138 if (mm
->uprobes_state
.xol_area
)
1142 /* Try to map as high as possible, this is only a hint. */
1143 area
->vaddr
= get_unmapped_area(NULL
, TASK_SIZE
- PAGE_SIZE
,
1145 if (area
->vaddr
& ~PAGE_MASK
) {
1151 ret
= install_special_mapping(mm
, area
->vaddr
, PAGE_SIZE
,
1152 VM_EXEC
|VM_MAYEXEC
|VM_DONTCOPY
|VM_IO
, &area
->page
);
1156 smp_wmb(); /* pairs with get_xol_area() */
1157 mm
->uprobes_state
.xol_area
= area
;
1159 up_write(&mm
->mmap_sem
);
1164 static struct xol_area
*__create_xol_area(unsigned long vaddr
)
1166 struct mm_struct
*mm
= current
->mm
;
1167 uprobe_opcode_t insn
= UPROBE_SWBP_INSN
;
1168 struct xol_area
*area
;
1170 area
= kmalloc(sizeof(*area
), GFP_KERNEL
);
1171 if (unlikely(!area
))
1174 area
->bitmap
= kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE
) * sizeof(long), GFP_KERNEL
);
1178 area
->page
= alloc_page(GFP_HIGHUSER
);
1182 area
->vaddr
= vaddr
;
1183 init_waitqueue_head(&area
->wq
);
1184 /* Reserve the 1st slot for get_trampoline_vaddr() */
1185 set_bit(0, area
->bitmap
);
1186 atomic_set(&area
->slot_count
, 1);
1187 copy_to_page(area
->page
, 0, &insn
, UPROBE_SWBP_INSN_SIZE
);
1189 if (!xol_add_vma(mm
, area
))
1192 __free_page(area
->page
);
1194 kfree(area
->bitmap
);
1202 * get_xol_area - Allocate process's xol_area if necessary.
1203 * This area will be used for storing instructions for execution out of line.
1205 * Returns the allocated area or NULL.
1207 static struct xol_area
*get_xol_area(void)
1209 struct mm_struct
*mm
= current
->mm
;
1210 struct xol_area
*area
;
1212 if (!mm
->uprobes_state
.xol_area
)
1213 __create_xol_area(0);
1215 area
= mm
->uprobes_state
.xol_area
;
1216 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1221 * uprobe_clear_state - Free the area allocated for slots.
1223 void uprobe_clear_state(struct mm_struct
*mm
)
1225 struct xol_area
*area
= mm
->uprobes_state
.xol_area
;
1230 put_page(area
->page
);
1231 kfree(area
->bitmap
);
1235 void uprobe_start_dup_mmap(void)
1237 percpu_down_read(&dup_mmap_sem
);
1240 void uprobe_end_dup_mmap(void)
1242 percpu_up_read(&dup_mmap_sem
);
1245 void uprobe_dup_mmap(struct mm_struct
*oldmm
, struct mm_struct
*newmm
)
1247 newmm
->uprobes_state
.xol_area
= NULL
;
1249 if (test_bit(MMF_HAS_UPROBES
, &oldmm
->flags
)) {
1250 set_bit(MMF_HAS_UPROBES
, &newmm
->flags
);
1251 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1252 set_bit(MMF_RECALC_UPROBES
, &newmm
->flags
);
1257 * - search for a free slot.
1259 static unsigned long xol_take_insn_slot(struct xol_area
*area
)
1261 unsigned long slot_addr
;
1265 slot_nr
= find_first_zero_bit(area
->bitmap
, UINSNS_PER_PAGE
);
1266 if (slot_nr
< UINSNS_PER_PAGE
) {
1267 if (!test_and_set_bit(slot_nr
, area
->bitmap
))
1270 slot_nr
= UINSNS_PER_PAGE
;
1273 wait_event(area
->wq
, (atomic_read(&area
->slot_count
) < UINSNS_PER_PAGE
));
1274 } while (slot_nr
>= UINSNS_PER_PAGE
);
1276 slot_addr
= area
->vaddr
+ (slot_nr
* UPROBE_XOL_SLOT_BYTES
);
1277 atomic_inc(&area
->slot_count
);
1283 * xol_get_insn_slot - allocate a slot for xol.
1284 * Returns the allocated slot address or 0.
1286 static unsigned long xol_get_insn_slot(struct uprobe
*uprobe
)
1288 struct xol_area
*area
;
1289 unsigned long xol_vaddr
;
1291 area
= get_xol_area();
1295 xol_vaddr
= xol_take_insn_slot(area
);
1296 if (unlikely(!xol_vaddr
))
1299 /* Initialize the slot */
1300 copy_to_page(area
->page
, xol_vaddr
,
1301 &uprobe
->arch
.ixol
, sizeof(uprobe
->arch
.ixol
));
1303 * We probably need flush_icache_user_range() but it needs vma.
1304 * This should work on supported architectures too.
1306 flush_dcache_page(area
->page
);
1312 * xol_free_insn_slot - If slot was earlier allocated by
1313 * @xol_get_insn_slot(), make the slot available for
1314 * subsequent requests.
1316 static void xol_free_insn_slot(struct task_struct
*tsk
)
1318 struct xol_area
*area
;
1319 unsigned long vma_end
;
1320 unsigned long slot_addr
;
1322 if (!tsk
->mm
|| !tsk
->mm
->uprobes_state
.xol_area
|| !tsk
->utask
)
1325 slot_addr
= tsk
->utask
->xol_vaddr
;
1326 if (unlikely(!slot_addr
))
1329 area
= tsk
->mm
->uprobes_state
.xol_area
;
1330 vma_end
= area
->vaddr
+ PAGE_SIZE
;
1331 if (area
->vaddr
<= slot_addr
&& slot_addr
< vma_end
) {
1332 unsigned long offset
;
1335 offset
= slot_addr
- area
->vaddr
;
1336 slot_nr
= offset
/ UPROBE_XOL_SLOT_BYTES
;
1337 if (slot_nr
>= UINSNS_PER_PAGE
)
1340 clear_bit(slot_nr
, area
->bitmap
);
1341 atomic_dec(&area
->slot_count
);
1342 if (waitqueue_active(&area
->wq
))
1345 tsk
->utask
->xol_vaddr
= 0;
1350 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1351 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1353 * Return the address of the breakpoint instruction.
1355 unsigned long __weak
uprobe_get_swbp_addr(struct pt_regs
*regs
)
1357 return instruction_pointer(regs
) - UPROBE_SWBP_INSN_SIZE
;
1361 * Called with no locks held.
1362 * Called in context of a exiting or a exec-ing thread.
1364 void uprobe_free_utask(struct task_struct
*t
)
1366 struct uprobe_task
*utask
= t
->utask
;
1367 struct return_instance
*ri
, *tmp
;
1372 if (utask
->active_uprobe
)
1373 put_uprobe(utask
->active_uprobe
);
1375 ri
= utask
->return_instances
;
1380 put_uprobe(tmp
->uprobe
);
1384 xol_free_insn_slot(t
);
1390 * Allocate a uprobe_task object for the task if if necessary.
1391 * Called when the thread hits a breakpoint.
1394 * - pointer to new uprobe_task on success
1397 static struct uprobe_task
*get_utask(void)
1399 if (!current
->utask
)
1400 current
->utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1401 return current
->utask
;
1404 static int dup_utask(struct task_struct
*t
, struct uprobe_task
*o_utask
)
1406 struct uprobe_task
*n_utask
;
1407 struct return_instance
**p
, *o
, *n
;
1409 n_utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1414 p
= &n_utask
->return_instances
;
1415 for (o
= o_utask
->return_instances
; o
; o
= o
->next
) {
1416 n
= kmalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1421 atomic_inc(&n
->uprobe
->ref
);
1432 static void uprobe_warn(struct task_struct
*t
, const char *msg
)
1434 pr_warn("uprobe: %s:%d failed to %s\n",
1435 current
->comm
, current
->pid
, msg
);
1438 static void dup_xol_work(struct callback_head
*work
)
1440 if (current
->flags
& PF_EXITING
)
1443 if (!__create_xol_area(current
->utask
->dup_xol_addr
))
1444 uprobe_warn(current
, "dup xol area");
1448 * Called in context of a new clone/fork from copy_process.
1450 void uprobe_copy_process(struct task_struct
*t
, unsigned long flags
)
1452 struct uprobe_task
*utask
= current
->utask
;
1453 struct mm_struct
*mm
= current
->mm
;
1454 struct xol_area
*area
;
1458 if (!utask
|| !utask
->return_instances
)
1461 if (mm
== t
->mm
&& !(flags
& CLONE_VFORK
))
1464 if (dup_utask(t
, utask
))
1465 return uprobe_warn(t
, "dup ret instances");
1467 /* The task can fork() after dup_xol_work() fails */
1468 area
= mm
->uprobes_state
.xol_area
;
1470 return uprobe_warn(t
, "dup xol area");
1475 t
->utask
->dup_xol_addr
= area
->vaddr
;
1476 init_task_work(&t
->utask
->dup_xol_work
, dup_xol_work
);
1477 task_work_add(t
, &t
->utask
->dup_xol_work
, true);
1481 * Current area->vaddr notion assume the trampoline address is always
1482 * equal area->vaddr.
1484 * Returns -1 in case the xol_area is not allocated.
1486 static unsigned long get_trampoline_vaddr(void)
1488 struct xol_area
*area
;
1489 unsigned long trampoline_vaddr
= -1;
1491 area
= current
->mm
->uprobes_state
.xol_area
;
1492 smp_read_barrier_depends();
1494 trampoline_vaddr
= area
->vaddr
;
1496 return trampoline_vaddr
;
1499 static void prepare_uretprobe(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1501 struct return_instance
*ri
;
1502 struct uprobe_task
*utask
;
1503 unsigned long orig_ret_vaddr
, trampoline_vaddr
;
1504 bool chained
= false;
1506 if (!get_xol_area())
1509 utask
= get_utask();
1513 if (utask
->depth
>= MAX_URETPROBE_DEPTH
) {
1514 printk_ratelimited(KERN_INFO
"uprobe: omit uretprobe due to"
1515 " nestedness limit pid/tgid=%d/%d\n",
1516 current
->pid
, current
->tgid
);
1520 ri
= kzalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1524 trampoline_vaddr
= get_trampoline_vaddr();
1525 orig_ret_vaddr
= arch_uretprobe_hijack_return_addr(trampoline_vaddr
, regs
);
1526 if (orig_ret_vaddr
== -1)
1530 * We don't want to keep trampoline address in stack, rather keep the
1531 * original return address of first caller thru all the consequent
1532 * instances. This also makes breakpoint unwrapping easier.
1534 if (orig_ret_vaddr
== trampoline_vaddr
) {
1535 if (!utask
->return_instances
) {
1537 * This situation is not possible. Likely we have an
1538 * attack from user-space.
1540 pr_warn("uprobe: unable to set uretprobe pid/tgid=%d/%d\n",
1541 current
->pid
, current
->tgid
);
1546 orig_ret_vaddr
= utask
->return_instances
->orig_ret_vaddr
;
1549 atomic_inc(&uprobe
->ref
);
1550 ri
->uprobe
= uprobe
;
1551 ri
->func
= instruction_pointer(regs
);
1552 ri
->orig_ret_vaddr
= orig_ret_vaddr
;
1553 ri
->chained
= chained
;
1557 /* add instance to the stack */
1558 ri
->next
= utask
->return_instances
;
1559 utask
->return_instances
= ri
;
1567 /* Prepare to single-step probed instruction out of line. */
1569 pre_ssout(struct uprobe
*uprobe
, struct pt_regs
*regs
, unsigned long bp_vaddr
)
1571 struct uprobe_task
*utask
;
1572 unsigned long xol_vaddr
;
1575 utask
= get_utask();
1579 xol_vaddr
= xol_get_insn_slot(uprobe
);
1583 utask
->xol_vaddr
= xol_vaddr
;
1584 utask
->vaddr
= bp_vaddr
;
1586 err
= arch_uprobe_pre_xol(&uprobe
->arch
, regs
);
1587 if (unlikely(err
)) {
1588 xol_free_insn_slot(current
);
1592 utask
->active_uprobe
= uprobe
;
1593 utask
->state
= UTASK_SSTEP
;
1598 * If we are singlestepping, then ensure this thread is not connected to
1599 * non-fatal signals until completion of singlestep. When xol insn itself
1600 * triggers the signal, restart the original insn even if the task is
1601 * already SIGKILL'ed (since coredump should report the correct ip). This
1602 * is even more important if the task has a handler for SIGSEGV/etc, The
1603 * _same_ instruction should be repeated again after return from the signal
1604 * handler, and SSTEP can never finish in this case.
1606 bool uprobe_deny_signal(void)
1608 struct task_struct
*t
= current
;
1609 struct uprobe_task
*utask
= t
->utask
;
1611 if (likely(!utask
|| !utask
->active_uprobe
))
1614 WARN_ON_ONCE(utask
->state
!= UTASK_SSTEP
);
1616 if (signal_pending(t
)) {
1617 spin_lock_irq(&t
->sighand
->siglock
);
1618 clear_tsk_thread_flag(t
, TIF_SIGPENDING
);
1619 spin_unlock_irq(&t
->sighand
->siglock
);
1621 if (__fatal_signal_pending(t
) || arch_uprobe_xol_was_trapped(t
)) {
1622 utask
->state
= UTASK_SSTEP_TRAPPED
;
1623 set_tsk_thread_flag(t
, TIF_UPROBE
);
1624 set_tsk_thread_flag(t
, TIF_NOTIFY_RESUME
);
1632 * Avoid singlestepping the original instruction if the original instruction
1633 * is a NOP or can be emulated.
1635 static bool can_skip_sstep(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1637 if (test_bit(UPROBE_SKIP_SSTEP
, &uprobe
->flags
)) {
1638 if (arch_uprobe_skip_sstep(&uprobe
->arch
, regs
))
1640 clear_bit(UPROBE_SKIP_SSTEP
, &uprobe
->flags
);
1645 static void mmf_recalc_uprobes(struct mm_struct
*mm
)
1647 struct vm_area_struct
*vma
;
1649 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1650 if (!valid_vma(vma
, false))
1653 * This is not strictly accurate, we can race with
1654 * uprobe_unregister() and see the already removed
1655 * uprobe if delete_uprobe() was not yet called.
1656 * Or this uprobe can be filtered out.
1658 if (vma_has_uprobes(vma
, vma
->vm_start
, vma
->vm_end
))
1662 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
1665 static int is_trap_at_addr(struct mm_struct
*mm
, unsigned long vaddr
)
1668 uprobe_opcode_t opcode
;
1671 pagefault_disable();
1672 result
= __copy_from_user_inatomic(&opcode
, (void __user
*)vaddr
,
1676 if (likely(result
== 0))
1679 result
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &page
, NULL
);
1683 copy_from_page(page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
1686 /* This needs to return true for any variant of the trap insn */
1687 return is_trap_insn(&opcode
);
1690 static struct uprobe
*find_active_uprobe(unsigned long bp_vaddr
, int *is_swbp
)
1692 struct mm_struct
*mm
= current
->mm
;
1693 struct uprobe
*uprobe
= NULL
;
1694 struct vm_area_struct
*vma
;
1696 down_read(&mm
->mmap_sem
);
1697 vma
= find_vma(mm
, bp_vaddr
);
1698 if (vma
&& vma
->vm_start
<= bp_vaddr
) {
1699 if (valid_vma(vma
, false)) {
1700 struct inode
*inode
= file_inode(vma
->vm_file
);
1701 loff_t offset
= vaddr_to_offset(vma
, bp_vaddr
);
1703 uprobe
= find_uprobe(inode
, offset
);
1707 *is_swbp
= is_trap_at_addr(mm
, bp_vaddr
);
1712 if (!uprobe
&& test_and_clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
))
1713 mmf_recalc_uprobes(mm
);
1714 up_read(&mm
->mmap_sem
);
1719 static void handler_chain(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1721 struct uprobe_consumer
*uc
;
1722 int remove
= UPROBE_HANDLER_REMOVE
;
1723 bool need_prep
= false; /* prepare return uprobe, when needed */
1725 down_read(&uprobe
->register_rwsem
);
1726 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
1730 rc
= uc
->handler(uc
, regs
);
1731 WARN(rc
& ~UPROBE_HANDLER_MASK
,
1732 "bad rc=0x%x from %pf()\n", rc
, uc
->handler
);
1735 if (uc
->ret_handler
)
1741 if (need_prep
&& !remove
)
1742 prepare_uretprobe(uprobe
, regs
); /* put bp at return */
1744 if (remove
&& uprobe
->consumers
) {
1745 WARN_ON(!uprobe_is_active(uprobe
));
1746 unapply_uprobe(uprobe
, current
->mm
);
1748 up_read(&uprobe
->register_rwsem
);
1752 handle_uretprobe_chain(struct return_instance
*ri
, struct pt_regs
*regs
)
1754 struct uprobe
*uprobe
= ri
->uprobe
;
1755 struct uprobe_consumer
*uc
;
1757 down_read(&uprobe
->register_rwsem
);
1758 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
1759 if (uc
->ret_handler
)
1760 uc
->ret_handler(uc
, ri
->func
, regs
);
1762 up_read(&uprobe
->register_rwsem
);
1765 static bool handle_trampoline(struct pt_regs
*regs
)
1767 struct uprobe_task
*utask
;
1768 struct return_instance
*ri
, *tmp
;
1771 utask
= current
->utask
;
1775 ri
= utask
->return_instances
;
1780 * TODO: we should throw out return_instance's invalidated by
1781 * longjmp(), currently we assume that the probed function always
1784 instruction_pointer_set(regs
, ri
->orig_ret_vaddr
);
1787 handle_uretprobe_chain(ri
, regs
);
1789 chained
= ri
->chained
;
1790 put_uprobe(ri
->uprobe
);
1802 utask
->return_instances
= ri
;
1807 bool __weak
arch_uprobe_ignore(struct arch_uprobe
*aup
, struct pt_regs
*regs
)
1813 * Run handler and ask thread to singlestep.
1814 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1816 static void handle_swbp(struct pt_regs
*regs
)
1818 struct uprobe
*uprobe
;
1819 unsigned long bp_vaddr
;
1820 int uninitialized_var(is_swbp
);
1822 bp_vaddr
= uprobe_get_swbp_addr(regs
);
1823 if (bp_vaddr
== get_trampoline_vaddr()) {
1824 if (handle_trampoline(regs
))
1827 pr_warn("uprobe: unable to handle uretprobe pid/tgid=%d/%d\n",
1828 current
->pid
, current
->tgid
);
1831 uprobe
= find_active_uprobe(bp_vaddr
, &is_swbp
);
1834 /* No matching uprobe; signal SIGTRAP. */
1835 send_sig(SIGTRAP
, current
, 0);
1838 * Either we raced with uprobe_unregister() or we can't
1839 * access this memory. The latter is only possible if
1840 * another thread plays with our ->mm. In both cases
1841 * we can simply restart. If this vma was unmapped we
1842 * can pretend this insn was not executed yet and get
1843 * the (correct) SIGSEGV after restart.
1845 instruction_pointer_set(regs
, bp_vaddr
);
1850 /* change it in advance for ->handler() and restart */
1851 instruction_pointer_set(regs
, bp_vaddr
);
1854 * TODO: move copy_insn/etc into _register and remove this hack.
1855 * After we hit the bp, _unregister + _register can install the
1856 * new and not-yet-analyzed uprobe at the same address, restart.
1858 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1859 if (unlikely(!test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
)))
1862 /* Tracing handlers use ->utask to communicate with fetch methods */
1866 if (arch_uprobe_ignore(&uprobe
->arch
, regs
))
1869 handler_chain(uprobe
, regs
);
1871 if (can_skip_sstep(uprobe
, regs
))
1874 if (!pre_ssout(uprobe
, regs
, bp_vaddr
))
1877 /* can_skip_sstep() succeeded, or restart if can't singlestep */
1883 * Perform required fix-ups and disable singlestep.
1884 * Allow pending signals to take effect.
1886 static void handle_singlestep(struct uprobe_task
*utask
, struct pt_regs
*regs
)
1888 struct uprobe
*uprobe
;
1890 uprobe
= utask
->active_uprobe
;
1891 if (utask
->state
== UTASK_SSTEP_ACK
)
1892 arch_uprobe_post_xol(&uprobe
->arch
, regs
);
1893 else if (utask
->state
== UTASK_SSTEP_TRAPPED
)
1894 arch_uprobe_abort_xol(&uprobe
->arch
, regs
);
1899 utask
->active_uprobe
= NULL
;
1900 utask
->state
= UTASK_RUNNING
;
1901 xol_free_insn_slot(current
);
1903 spin_lock_irq(¤t
->sighand
->siglock
);
1904 recalc_sigpending(); /* see uprobe_deny_signal() */
1905 spin_unlock_irq(¤t
->sighand
->siglock
);
1909 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1910 * allows the thread to return from interrupt. After that handle_swbp()
1911 * sets utask->active_uprobe.
1913 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1914 * and allows the thread to return from interrupt.
1916 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1917 * uprobe_notify_resume().
1919 void uprobe_notify_resume(struct pt_regs
*regs
)
1921 struct uprobe_task
*utask
;
1923 clear_thread_flag(TIF_UPROBE
);
1925 utask
= current
->utask
;
1926 if (utask
&& utask
->active_uprobe
)
1927 handle_singlestep(utask
, regs
);
1933 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1934 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1936 int uprobe_pre_sstep_notifier(struct pt_regs
*regs
)
1941 if (!test_bit(MMF_HAS_UPROBES
, ¤t
->mm
->flags
) &&
1942 (!current
->utask
|| !current
->utask
->return_instances
))
1945 set_thread_flag(TIF_UPROBE
);
1950 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1951 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1953 int uprobe_post_sstep_notifier(struct pt_regs
*regs
)
1955 struct uprobe_task
*utask
= current
->utask
;
1957 if (!current
->mm
|| !utask
|| !utask
->active_uprobe
)
1958 /* task is currently not uprobed */
1961 utask
->state
= UTASK_SSTEP_ACK
;
1962 set_thread_flag(TIF_UPROBE
);
1966 static struct notifier_block uprobe_exception_nb
= {
1967 .notifier_call
= arch_uprobe_exception_notify
,
1968 .priority
= INT_MAX
-1, /* notified after kprobes, kgdb */
1971 static int __init
init_uprobes(void)
1975 for (i
= 0; i
< UPROBES_HASH_SZ
; i
++)
1976 mutex_init(&uprobes_mmap_mutex
[i
]);
1978 if (percpu_init_rwsem(&dup_mmap_sem
))
1981 return register_die_notifier(&uprobe_exception_nb
);
1983 __initcall(init_uprobes
);