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
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>
39 #include <linux/shmem_fs.h>
41 #include <linux/uprobes.h>
43 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
44 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
46 static struct rb_root uprobes_tree
= RB_ROOT
;
48 * allows us to skip the uprobe_mmap if there are no uprobe events active
49 * at this time. Probably a fine grained per inode count is better?
51 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
53 static DEFINE_SPINLOCK(uprobes_treelock
); /* serialize rbtree access */
55 #define UPROBES_HASH_SZ 13
56 /* serialize uprobe->pending_list */
57 static struct mutex uprobes_mmap_mutex
[UPROBES_HASH_SZ
];
58 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
60 static struct percpu_rw_semaphore dup_mmap_sem
;
62 /* Have a copy of original instruction */
63 #define UPROBE_COPY_INSN 0
66 struct rb_node rb_node
; /* node in the rb tree */
68 struct rw_semaphore register_rwsem
;
69 struct rw_semaphore consumer_rwsem
;
70 struct list_head pending_list
;
71 struct uprobe_consumer
*consumers
;
72 struct inode
*inode
; /* Also hold a ref to inode */
77 * The generic code assumes that it has two members of unknown type
78 * owned by the arch-specific code:
80 * insn - copy_insn() saves the original instruction here for
81 * arch_uprobe_analyze_insn().
83 * ixol - potentially modified instruction to execute out of
84 * line, copied to xol_area by xol_get_insn_slot().
86 struct arch_uprobe arch
;
90 * Execute out of line area: anonymous executable mapping installed
91 * by the probed task to execute the copy of the original instruction
92 * mangled by set_swbp().
94 * On a breakpoint hit, thread contests for a slot. It frees the
95 * slot after singlestep. Currently a fixed number of slots are
99 wait_queue_head_t wq
; /* if all slots are busy */
100 atomic_t slot_count
; /* number of in-use slots */
101 unsigned long *bitmap
; /* 0 = free slot */
103 struct vm_special_mapping xol_mapping
;
104 struct page
*pages
[2];
106 * We keep the vma's vm_start rather than a pointer to the vma
107 * itself. The probed process or a naughty kernel module could make
108 * the vma go away, and we must handle that reasonably gracefully.
110 unsigned long vaddr
; /* Page(s) of instruction slots */
114 * valid_vma: Verify if the specified vma is an executable vma
115 * Relax restrictions while unregistering: vm_flags might have
116 * changed after breakpoint was inserted.
117 * - is_register: indicates if we are in register context.
118 * - Return 1 if the specified virtual address is in an
121 static bool valid_vma(struct vm_area_struct
*vma
, bool is_register
)
123 vm_flags_t flags
= VM_HUGETLB
| VM_MAYEXEC
| VM_MAYSHARE
;
128 return vma
->vm_file
&& (vma
->vm_flags
& flags
) == VM_MAYEXEC
;
131 static unsigned long offset_to_vaddr(struct vm_area_struct
*vma
, loff_t offset
)
133 return vma
->vm_start
+ offset
- ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
136 static loff_t
vaddr_to_offset(struct vm_area_struct
*vma
, unsigned long vaddr
)
138 return ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
) + (vaddr
- vma
->vm_start
);
142 * __replace_page - replace page in vma by new page.
143 * based on replace_page in mm/ksm.c
145 * @vma: vma that holds the pte pointing to page
146 * @addr: address the old @page is mapped at
147 * @page: the cowed page we are replacing by kpage
148 * @kpage: the modified page we replace page by
150 * Returns 0 on success, -EFAULT on failure.
152 static int __replace_page(struct vm_area_struct
*vma
, unsigned long addr
,
153 struct page
*page
, struct page
*kpage
)
155 struct mm_struct
*mm
= vma
->vm_mm
;
159 /* For mmu_notifiers */
160 const unsigned long mmun_start
= addr
;
161 const unsigned long mmun_end
= addr
+ PAGE_SIZE
;
162 struct mem_cgroup
*memcg
;
164 err
= mem_cgroup_try_charge(kpage
, vma
->vm_mm
, GFP_KERNEL
, &memcg
,
169 /* For try_to_free_swap() and munlock_vma_page() below */
172 mmu_notifier_invalidate_range_start(mm
, mmun_start
, mmun_end
);
174 ptep
= page_check_address(page
, mm
, addr
, &ptl
, 0);
179 page_add_new_anon_rmap(kpage
, vma
, addr
, false);
180 mem_cgroup_commit_charge(kpage
, memcg
, false, false);
181 lru_cache_add_active_or_unevictable(kpage
, vma
);
183 if (!PageAnon(page
)) {
184 dec_mm_counter(mm
, mm_counter_file(page
));
185 inc_mm_counter(mm
, MM_ANONPAGES
);
188 flush_cache_page(vma
, addr
, pte_pfn(*ptep
));
189 ptep_clear_flush_notify(vma
, addr
, ptep
);
190 set_pte_at_notify(mm
, addr
, ptep
, mk_pte(kpage
, vma
->vm_page_prot
));
192 page_remove_rmap(page
, false);
193 if (!page_mapped(page
))
194 try_to_free_swap(page
);
195 pte_unmap_unlock(ptep
, ptl
);
197 if (vma
->vm_flags
& VM_LOCKED
)
198 munlock_vma_page(page
);
203 mem_cgroup_cancel_charge(kpage
, memcg
, false);
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.
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 write.
291 * Return 0 (success) or a negative errno.
293 int uprobe_write_opcode(struct mm_struct
*mm
, unsigned long vaddr
,
294 uprobe_opcode_t opcode
)
296 struct page
*old_page
, *new_page
;
297 struct vm_area_struct
*vma
;
301 /* Read the page with vaddr into memory */
302 ret
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &old_page
, &vma
);
306 ret
= verify_opcode(old_page
, vaddr
, &opcode
);
310 ret
= anon_vma_prepare(vma
);
315 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, vaddr
);
319 __SetPageUptodate(new_page
);
320 copy_highpage(new_page
, old_page
);
321 copy_to_page(new_page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
323 ret
= __replace_page(vma
, vaddr
, old_page
, new_page
);
324 page_cache_release(new_page
);
328 if (unlikely(ret
== -EAGAIN
))
334 * set_swbp - store breakpoint at a given address.
335 * @auprobe: arch specific probepoint information.
336 * @mm: the probed process address space.
337 * @vaddr: the virtual address to insert the opcode.
339 * For mm @mm, store the breakpoint instruction at @vaddr.
340 * Return 0 (success) or a negative errno.
342 int __weak
set_swbp(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
344 return uprobe_write_opcode(mm
, vaddr
, UPROBE_SWBP_INSN
);
348 * set_orig_insn - Restore the original instruction.
349 * @mm: the probed process address space.
350 * @auprobe: arch specific probepoint information.
351 * @vaddr: the virtual address to insert the opcode.
353 * For mm @mm, restore the original opcode (opcode) at @vaddr.
354 * Return 0 (success) or a negative errno.
357 set_orig_insn(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
359 return uprobe_write_opcode(mm
, vaddr
, *(uprobe_opcode_t
*)&auprobe
->insn
);
362 static struct uprobe
*get_uprobe(struct uprobe
*uprobe
)
364 atomic_inc(&uprobe
->ref
);
368 static void put_uprobe(struct uprobe
*uprobe
)
370 if (atomic_dec_and_test(&uprobe
->ref
))
374 static int match_uprobe(struct uprobe
*l
, struct uprobe
*r
)
376 if (l
->inode
< r
->inode
)
379 if (l
->inode
> r
->inode
)
382 if (l
->offset
< r
->offset
)
385 if (l
->offset
> r
->offset
)
391 static struct uprobe
*__find_uprobe(struct inode
*inode
, loff_t offset
)
393 struct uprobe u
= { .inode
= inode
, .offset
= offset
};
394 struct rb_node
*n
= uprobes_tree
.rb_node
;
395 struct uprobe
*uprobe
;
399 uprobe
= rb_entry(n
, struct uprobe
, rb_node
);
400 match
= match_uprobe(&u
, uprobe
);
402 return get_uprobe(uprobe
);
413 * Find a uprobe corresponding to a given inode:offset
414 * Acquires uprobes_treelock
416 static struct uprobe
*find_uprobe(struct inode
*inode
, loff_t offset
)
418 struct uprobe
*uprobe
;
420 spin_lock(&uprobes_treelock
);
421 uprobe
= __find_uprobe(inode
, offset
);
422 spin_unlock(&uprobes_treelock
);
427 static struct uprobe
*__insert_uprobe(struct uprobe
*uprobe
)
429 struct rb_node
**p
= &uprobes_tree
.rb_node
;
430 struct rb_node
*parent
= NULL
;
436 u
= rb_entry(parent
, struct uprobe
, rb_node
);
437 match
= match_uprobe(uprobe
, u
);
439 return get_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 struct uprobe
*alloc_uprobe(struct inode
*inode
, loff_t offset
)
478 struct uprobe
*uprobe
, *cur_uprobe
;
480 uprobe
= kzalloc(sizeof(struct uprobe
), GFP_KERNEL
);
484 uprobe
->inode
= igrab(inode
);
485 uprobe
->offset
= offset
;
486 init_rwsem(&uprobe
->register_rwsem
);
487 init_rwsem(&uprobe
->consumer_rwsem
);
489 /* add to uprobes_tree, sorted on inode:offset */
490 cur_uprobe
= insert_uprobe(uprobe
);
491 /* a uprobe exists for this inode:offset combination */
501 static void consumer_add(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
503 down_write(&uprobe
->consumer_rwsem
);
504 uc
->next
= uprobe
->consumers
;
505 uprobe
->consumers
= uc
;
506 up_write(&uprobe
->consumer_rwsem
);
510 * For uprobe @uprobe, delete the consumer @uc.
511 * Return true if the @uc is deleted successfully
514 static bool consumer_del(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
516 struct uprobe_consumer
**con
;
519 down_write(&uprobe
->consumer_rwsem
);
520 for (con
= &uprobe
->consumers
; *con
; con
= &(*con
)->next
) {
527 up_write(&uprobe
->consumer_rwsem
);
532 static int __copy_insn(struct address_space
*mapping
, struct file
*filp
,
533 void *insn
, int nbytes
, loff_t offset
)
537 * Ensure that the page that has the original instruction is populated
538 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
539 * see uprobe_register().
541 if (mapping
->a_ops
->readpage
)
542 page
= read_mapping_page(mapping
, offset
>> PAGE_CACHE_SHIFT
, filp
);
544 page
= shmem_read_mapping_page(mapping
, offset
>> PAGE_CACHE_SHIFT
);
546 return PTR_ERR(page
);
548 copy_from_page(page
, offset
, insn
, nbytes
);
549 page_cache_release(page
);
554 static int copy_insn(struct uprobe
*uprobe
, struct file
*filp
)
556 struct address_space
*mapping
= uprobe
->inode
->i_mapping
;
557 loff_t offs
= uprobe
->offset
;
558 void *insn
= &uprobe
->arch
.insn
;
559 int size
= sizeof(uprobe
->arch
.insn
);
562 /* Copy only available bytes, -EIO if nothing was read */
564 if (offs
>= i_size_read(uprobe
->inode
))
567 len
= min_t(int, size
, PAGE_SIZE
- (offs
& ~PAGE_MASK
));
568 err
= __copy_insn(mapping
, filp
, insn
, len
, offs
);
580 static int prepare_uprobe(struct uprobe
*uprobe
, struct file
*file
,
581 struct mm_struct
*mm
, unsigned long vaddr
)
585 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
588 /* TODO: move this into _register, until then we abuse this sem. */
589 down_write(&uprobe
->consumer_rwsem
);
590 if (test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
))
593 ret
= copy_insn(uprobe
, file
);
598 if (is_trap_insn((uprobe_opcode_t
*)&uprobe
->arch
.insn
))
601 ret
= arch_uprobe_analyze_insn(&uprobe
->arch
, mm
, vaddr
);
605 /* uprobe_write_opcode() assumes we don't cross page boundary */
606 BUG_ON((uprobe
->offset
& ~PAGE_MASK
) +
607 UPROBE_SWBP_INSN_SIZE
> PAGE_SIZE
);
609 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
610 set_bit(UPROBE_COPY_INSN
, &uprobe
->flags
);
613 up_write(&uprobe
->consumer_rwsem
);
618 static inline bool consumer_filter(struct uprobe_consumer
*uc
,
619 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
621 return !uc
->filter
|| uc
->filter(uc
, ctx
, mm
);
624 static bool filter_chain(struct uprobe
*uprobe
,
625 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
627 struct uprobe_consumer
*uc
;
630 down_read(&uprobe
->consumer_rwsem
);
631 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
632 ret
= consumer_filter(uc
, ctx
, mm
);
636 up_read(&uprobe
->consumer_rwsem
);
642 install_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
,
643 struct vm_area_struct
*vma
, unsigned long vaddr
)
648 ret
= prepare_uprobe(uprobe
, vma
->vm_file
, mm
, vaddr
);
653 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
654 * the task can hit this breakpoint right after __replace_page().
656 first_uprobe
= !test_bit(MMF_HAS_UPROBES
, &mm
->flags
);
658 set_bit(MMF_HAS_UPROBES
, &mm
->flags
);
660 ret
= set_swbp(&uprobe
->arch
, mm
, vaddr
);
662 clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
663 else if (first_uprobe
)
664 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
670 remove_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
672 set_bit(MMF_RECALC_UPROBES
, &mm
->flags
);
673 return set_orig_insn(&uprobe
->arch
, mm
, vaddr
);
676 static inline bool uprobe_is_active(struct uprobe
*uprobe
)
678 return !RB_EMPTY_NODE(&uprobe
->rb_node
);
681 * There could be threads that have already hit the breakpoint. They
682 * will recheck the current insn and restart if find_uprobe() fails.
683 * See find_active_uprobe().
685 static void delete_uprobe(struct uprobe
*uprobe
)
687 if (WARN_ON(!uprobe_is_active(uprobe
)))
690 spin_lock(&uprobes_treelock
);
691 rb_erase(&uprobe
->rb_node
, &uprobes_tree
);
692 spin_unlock(&uprobes_treelock
);
693 RB_CLEAR_NODE(&uprobe
->rb_node
); /* for uprobe_is_active() */
699 struct map_info
*next
;
700 struct mm_struct
*mm
;
704 static inline struct map_info
*free_map_info(struct map_info
*info
)
706 struct map_info
*next
= info
->next
;
711 static struct map_info
*
712 build_map_info(struct address_space
*mapping
, loff_t offset
, bool is_register
)
714 unsigned long pgoff
= offset
>> PAGE_SHIFT
;
715 struct vm_area_struct
*vma
;
716 struct map_info
*curr
= NULL
;
717 struct map_info
*prev
= NULL
;
718 struct map_info
*info
;
722 i_mmap_lock_read(mapping
);
723 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
, pgoff
, pgoff
) {
724 if (!valid_vma(vma
, is_register
))
727 if (!prev
&& !more
) {
729 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
730 * reclaim. This is optimistic, no harm done if it fails.
732 prev
= kmalloc(sizeof(struct map_info
),
733 GFP_NOWAIT
| __GFP_NOMEMALLOC
| __GFP_NOWARN
);
742 if (!atomic_inc_not_zero(&vma
->vm_mm
->mm_users
))
750 info
->mm
= vma
->vm_mm
;
751 info
->vaddr
= offset_to_vaddr(vma
, offset
);
753 i_mmap_unlock_read(mapping
);
765 info
= kmalloc(sizeof(struct map_info
), GFP_KERNEL
);
767 curr
= ERR_PTR(-ENOMEM
);
777 prev
= free_map_info(prev
);
782 register_for_each_vma(struct uprobe
*uprobe
, struct uprobe_consumer
*new)
784 bool is_register
= !!new;
785 struct map_info
*info
;
788 percpu_down_write(&dup_mmap_sem
);
789 info
= build_map_info(uprobe
->inode
->i_mapping
,
790 uprobe
->offset
, is_register
);
797 struct mm_struct
*mm
= info
->mm
;
798 struct vm_area_struct
*vma
;
800 if (err
&& is_register
)
803 down_write(&mm
->mmap_sem
);
804 vma
= find_vma(mm
, info
->vaddr
);
805 if (!vma
|| !valid_vma(vma
, is_register
) ||
806 file_inode(vma
->vm_file
) != uprobe
->inode
)
809 if (vma
->vm_start
> info
->vaddr
||
810 vaddr_to_offset(vma
, info
->vaddr
) != uprobe
->offset
)
814 /* consult only the "caller", new consumer. */
815 if (consumer_filter(new,
816 UPROBE_FILTER_REGISTER
, mm
))
817 err
= install_breakpoint(uprobe
, mm
, vma
, info
->vaddr
);
818 } else if (test_bit(MMF_HAS_UPROBES
, &mm
->flags
)) {
819 if (!filter_chain(uprobe
,
820 UPROBE_FILTER_UNREGISTER
, mm
))
821 err
|= remove_breakpoint(uprobe
, mm
, info
->vaddr
);
825 up_write(&mm
->mmap_sem
);
828 info
= free_map_info(info
);
831 percpu_up_write(&dup_mmap_sem
);
835 static int __uprobe_register(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
837 consumer_add(uprobe
, uc
);
838 return register_for_each_vma(uprobe
, uc
);
841 static void __uprobe_unregister(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
845 if (WARN_ON(!consumer_del(uprobe
, uc
)))
848 err
= register_for_each_vma(uprobe
, NULL
);
849 /* TODO : cant unregister? schedule a worker thread */
850 if (!uprobe
->consumers
&& !err
)
851 delete_uprobe(uprobe
);
855 * uprobe_register - register a probe
856 * @inode: the file in which the probe has to be placed.
857 * @offset: offset from the start of the file.
858 * @uc: information on howto handle the probe..
860 * Apart from the access refcount, uprobe_register() takes a creation
861 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
862 * inserted into the rbtree (i.e first consumer for a @inode:@offset
863 * tuple). Creation refcount stops uprobe_unregister from freeing the
864 * @uprobe even before the register operation is complete. Creation
865 * refcount is released when the last @uc for the @uprobe
868 * Return errno if it cannot successully install probes
869 * else return 0 (success)
871 int uprobe_register(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
873 struct uprobe
*uprobe
;
876 /* Uprobe must have at least one set consumer */
877 if (!uc
->handler
&& !uc
->ret_handler
)
880 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
881 if (!inode
->i_mapping
->a_ops
->readpage
&& !shmem_mapping(inode
->i_mapping
))
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
);
926 if (WARN_ON(!uprobe
))
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
);
951 if (WARN_ON(!uprobe
))
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
);
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
);
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 struct vm_area_struct
*vma
;
1133 down_write(&mm
->mmap_sem
);
1134 if (mm
->uprobes_state
.xol_area
) {
1140 /* Try to map as high as possible, this is only a hint. */
1141 area
->vaddr
= get_unmapped_area(NULL
, TASK_SIZE
- PAGE_SIZE
,
1143 if (area
->vaddr
& ~PAGE_MASK
) {
1149 vma
= _install_special_mapping(mm
, area
->vaddr
, PAGE_SIZE
,
1150 VM_EXEC
|VM_MAYEXEC
|VM_DONTCOPY
|VM_IO
,
1151 &area
->xol_mapping
);
1158 smp_wmb(); /* pairs with get_xol_area() */
1159 mm
->uprobes_state
.xol_area
= area
;
1161 up_write(&mm
->mmap_sem
);
1166 static struct xol_area
*__create_xol_area(unsigned long vaddr
)
1168 struct mm_struct
*mm
= current
->mm
;
1169 uprobe_opcode_t insn
= UPROBE_SWBP_INSN
;
1170 struct xol_area
*area
;
1172 area
= kmalloc(sizeof(*area
), GFP_KERNEL
);
1173 if (unlikely(!area
))
1176 area
->bitmap
= kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE
) * sizeof(long), GFP_KERNEL
);
1180 area
->xol_mapping
.name
= "[uprobes]";
1181 area
->xol_mapping
.pages
= area
->pages
;
1182 area
->pages
[0] = alloc_page(GFP_HIGHUSER
);
1183 if (!area
->pages
[0])
1185 area
->pages
[1] = NULL
;
1187 area
->vaddr
= vaddr
;
1188 init_waitqueue_head(&area
->wq
);
1189 /* Reserve the 1st slot for get_trampoline_vaddr() */
1190 set_bit(0, area
->bitmap
);
1191 atomic_set(&area
->slot_count
, 1);
1192 copy_to_page(area
->pages
[0], 0, &insn
, UPROBE_SWBP_INSN_SIZE
);
1194 if (!xol_add_vma(mm
, area
))
1197 __free_page(area
->pages
[0]);
1199 kfree(area
->bitmap
);
1207 * get_xol_area - Allocate process's xol_area if necessary.
1208 * This area will be used for storing instructions for execution out of line.
1210 * Returns the allocated area or NULL.
1212 static struct xol_area
*get_xol_area(void)
1214 struct mm_struct
*mm
= current
->mm
;
1215 struct xol_area
*area
;
1217 if (!mm
->uprobes_state
.xol_area
)
1218 __create_xol_area(0);
1220 area
= mm
->uprobes_state
.xol_area
;
1221 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1226 * uprobe_clear_state - Free the area allocated for slots.
1228 void uprobe_clear_state(struct mm_struct
*mm
)
1230 struct xol_area
*area
= mm
->uprobes_state
.xol_area
;
1235 put_page(area
->pages
[0]);
1236 kfree(area
->bitmap
);
1240 void uprobe_start_dup_mmap(void)
1242 percpu_down_read(&dup_mmap_sem
);
1245 void uprobe_end_dup_mmap(void)
1247 percpu_up_read(&dup_mmap_sem
);
1250 void uprobe_dup_mmap(struct mm_struct
*oldmm
, struct mm_struct
*newmm
)
1252 newmm
->uprobes_state
.xol_area
= NULL
;
1254 if (test_bit(MMF_HAS_UPROBES
, &oldmm
->flags
)) {
1255 set_bit(MMF_HAS_UPROBES
, &newmm
->flags
);
1256 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1257 set_bit(MMF_RECALC_UPROBES
, &newmm
->flags
);
1262 * - search for a free slot.
1264 static unsigned long xol_take_insn_slot(struct xol_area
*area
)
1266 unsigned long slot_addr
;
1270 slot_nr
= find_first_zero_bit(area
->bitmap
, UINSNS_PER_PAGE
);
1271 if (slot_nr
< UINSNS_PER_PAGE
) {
1272 if (!test_and_set_bit(slot_nr
, area
->bitmap
))
1275 slot_nr
= UINSNS_PER_PAGE
;
1278 wait_event(area
->wq
, (atomic_read(&area
->slot_count
) < UINSNS_PER_PAGE
));
1279 } while (slot_nr
>= UINSNS_PER_PAGE
);
1281 slot_addr
= area
->vaddr
+ (slot_nr
* UPROBE_XOL_SLOT_BYTES
);
1282 atomic_inc(&area
->slot_count
);
1288 * xol_get_insn_slot - allocate a slot for xol.
1289 * Returns the allocated slot address or 0.
1291 static unsigned long xol_get_insn_slot(struct uprobe
*uprobe
)
1293 struct xol_area
*area
;
1294 unsigned long xol_vaddr
;
1296 area
= get_xol_area();
1300 xol_vaddr
= xol_take_insn_slot(area
);
1301 if (unlikely(!xol_vaddr
))
1304 arch_uprobe_copy_ixol(area
->pages
[0], xol_vaddr
,
1305 &uprobe
->arch
.ixol
, sizeof(uprobe
->arch
.ixol
));
1311 * xol_free_insn_slot - If slot was earlier allocated by
1312 * @xol_get_insn_slot(), make the slot available for
1313 * subsequent requests.
1315 static void xol_free_insn_slot(struct task_struct
*tsk
)
1317 struct xol_area
*area
;
1318 unsigned long vma_end
;
1319 unsigned long slot_addr
;
1321 if (!tsk
->mm
|| !tsk
->mm
->uprobes_state
.xol_area
|| !tsk
->utask
)
1324 slot_addr
= tsk
->utask
->xol_vaddr
;
1325 if (unlikely(!slot_addr
))
1328 area
= tsk
->mm
->uprobes_state
.xol_area
;
1329 vma_end
= area
->vaddr
+ PAGE_SIZE
;
1330 if (area
->vaddr
<= slot_addr
&& slot_addr
< vma_end
) {
1331 unsigned long offset
;
1334 offset
= slot_addr
- area
->vaddr
;
1335 slot_nr
= offset
/ UPROBE_XOL_SLOT_BYTES
;
1336 if (slot_nr
>= UINSNS_PER_PAGE
)
1339 clear_bit(slot_nr
, area
->bitmap
);
1340 atomic_dec(&area
->slot_count
);
1341 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1342 if (waitqueue_active(&area
->wq
))
1345 tsk
->utask
->xol_vaddr
= 0;
1349 void __weak
arch_uprobe_copy_ixol(struct page
*page
, unsigned long vaddr
,
1350 void *src
, unsigned long len
)
1352 /* Initialize the slot */
1353 copy_to_page(page
, vaddr
, src
, len
);
1356 * We probably need flush_icache_user_range() but it needs vma.
1357 * This should work on most of architectures by default. If
1358 * architecture needs to do something different it can define
1359 * its own version of the function.
1361 flush_dcache_page(page
);
1365 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1366 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1368 * Return the address of the breakpoint instruction.
1370 unsigned long __weak
uprobe_get_swbp_addr(struct pt_regs
*regs
)
1372 return instruction_pointer(regs
) - UPROBE_SWBP_INSN_SIZE
;
1375 unsigned long uprobe_get_trap_addr(struct pt_regs
*regs
)
1377 struct uprobe_task
*utask
= current
->utask
;
1379 if (unlikely(utask
&& utask
->active_uprobe
))
1380 return utask
->vaddr
;
1382 return instruction_pointer(regs
);
1385 static struct return_instance
*free_ret_instance(struct return_instance
*ri
)
1387 struct return_instance
*next
= ri
->next
;
1388 put_uprobe(ri
->uprobe
);
1394 * Called with no locks held.
1395 * Called in context of a exiting or a exec-ing thread.
1397 void uprobe_free_utask(struct task_struct
*t
)
1399 struct uprobe_task
*utask
= t
->utask
;
1400 struct return_instance
*ri
;
1405 if (utask
->active_uprobe
)
1406 put_uprobe(utask
->active_uprobe
);
1408 ri
= utask
->return_instances
;
1410 ri
= free_ret_instance(ri
);
1412 xol_free_insn_slot(t
);
1418 * Allocate a uprobe_task object for the task if if necessary.
1419 * Called when the thread hits a breakpoint.
1422 * - pointer to new uprobe_task on success
1425 static struct uprobe_task
*get_utask(void)
1427 if (!current
->utask
)
1428 current
->utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1429 return current
->utask
;
1432 static int dup_utask(struct task_struct
*t
, struct uprobe_task
*o_utask
)
1434 struct uprobe_task
*n_utask
;
1435 struct return_instance
**p
, *o
, *n
;
1437 n_utask
= kzalloc(sizeof(struct uprobe_task
), GFP_KERNEL
);
1442 p
= &n_utask
->return_instances
;
1443 for (o
= o_utask
->return_instances
; o
; o
= o
->next
) {
1444 n
= kmalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1449 get_uprobe(n
->uprobe
);
1460 static void uprobe_warn(struct task_struct
*t
, const char *msg
)
1462 pr_warn("uprobe: %s:%d failed to %s\n",
1463 current
->comm
, current
->pid
, msg
);
1466 static void dup_xol_work(struct callback_head
*work
)
1468 if (current
->flags
& PF_EXITING
)
1471 if (!__create_xol_area(current
->utask
->dup_xol_addr
))
1472 uprobe_warn(current
, "dup xol area");
1476 * Called in context of a new clone/fork from copy_process.
1478 void uprobe_copy_process(struct task_struct
*t
, unsigned long flags
)
1480 struct uprobe_task
*utask
= current
->utask
;
1481 struct mm_struct
*mm
= current
->mm
;
1482 struct xol_area
*area
;
1486 if (!utask
|| !utask
->return_instances
)
1489 if (mm
== t
->mm
&& !(flags
& CLONE_VFORK
))
1492 if (dup_utask(t
, utask
))
1493 return uprobe_warn(t
, "dup ret instances");
1495 /* The task can fork() after dup_xol_work() fails */
1496 area
= mm
->uprobes_state
.xol_area
;
1498 return uprobe_warn(t
, "dup xol area");
1503 t
->utask
->dup_xol_addr
= area
->vaddr
;
1504 init_task_work(&t
->utask
->dup_xol_work
, dup_xol_work
);
1505 task_work_add(t
, &t
->utask
->dup_xol_work
, true);
1509 * Current area->vaddr notion assume the trampoline address is always
1510 * equal area->vaddr.
1512 * Returns -1 in case the xol_area is not allocated.
1514 static unsigned long get_trampoline_vaddr(void)
1516 struct xol_area
*area
;
1517 unsigned long trampoline_vaddr
= -1;
1519 area
= current
->mm
->uprobes_state
.xol_area
;
1520 smp_read_barrier_depends();
1522 trampoline_vaddr
= area
->vaddr
;
1524 return trampoline_vaddr
;
1527 static void cleanup_return_instances(struct uprobe_task
*utask
, bool chained
,
1528 struct pt_regs
*regs
)
1530 struct return_instance
*ri
= utask
->return_instances
;
1531 enum rp_check ctx
= chained
? RP_CHECK_CHAIN_CALL
: RP_CHECK_CALL
;
1533 while (ri
&& !arch_uretprobe_is_alive(ri
, ctx
, regs
)) {
1534 ri
= free_ret_instance(ri
);
1537 utask
->return_instances
= ri
;
1540 static void prepare_uretprobe(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1542 struct return_instance
*ri
;
1543 struct uprobe_task
*utask
;
1544 unsigned long orig_ret_vaddr
, trampoline_vaddr
;
1547 if (!get_xol_area())
1550 utask
= get_utask();
1554 if (utask
->depth
>= MAX_URETPROBE_DEPTH
) {
1555 printk_ratelimited(KERN_INFO
"uprobe: omit uretprobe due to"
1556 " nestedness limit pid/tgid=%d/%d\n",
1557 current
->pid
, current
->tgid
);
1561 ri
= kmalloc(sizeof(struct return_instance
), GFP_KERNEL
);
1565 trampoline_vaddr
= get_trampoline_vaddr();
1566 orig_ret_vaddr
= arch_uretprobe_hijack_return_addr(trampoline_vaddr
, regs
);
1567 if (orig_ret_vaddr
== -1)
1570 /* drop the entries invalidated by longjmp() */
1571 chained
= (orig_ret_vaddr
== trampoline_vaddr
);
1572 cleanup_return_instances(utask
, chained
, regs
);
1575 * We don't want to keep trampoline address in stack, rather keep the
1576 * original return address of first caller thru all the consequent
1577 * instances. This also makes breakpoint unwrapping easier.
1580 if (!utask
->return_instances
) {
1582 * This situation is not possible. Likely we have an
1583 * attack from user-space.
1585 uprobe_warn(current
, "handle tail call");
1588 orig_ret_vaddr
= utask
->return_instances
->orig_ret_vaddr
;
1591 ri
->uprobe
= get_uprobe(uprobe
);
1592 ri
->func
= instruction_pointer(regs
);
1593 ri
->stack
= user_stack_pointer(regs
);
1594 ri
->orig_ret_vaddr
= orig_ret_vaddr
;
1595 ri
->chained
= chained
;
1598 ri
->next
= utask
->return_instances
;
1599 utask
->return_instances
= ri
;
1606 /* Prepare to single-step probed instruction out of line. */
1608 pre_ssout(struct uprobe
*uprobe
, struct pt_regs
*regs
, unsigned long bp_vaddr
)
1610 struct uprobe_task
*utask
;
1611 unsigned long xol_vaddr
;
1614 utask
= get_utask();
1618 xol_vaddr
= xol_get_insn_slot(uprobe
);
1622 utask
->xol_vaddr
= xol_vaddr
;
1623 utask
->vaddr
= bp_vaddr
;
1625 err
= arch_uprobe_pre_xol(&uprobe
->arch
, regs
);
1626 if (unlikely(err
)) {
1627 xol_free_insn_slot(current
);
1631 utask
->active_uprobe
= uprobe
;
1632 utask
->state
= UTASK_SSTEP
;
1637 * If we are singlestepping, then ensure this thread is not connected to
1638 * non-fatal signals until completion of singlestep. When xol insn itself
1639 * triggers the signal, restart the original insn even if the task is
1640 * already SIGKILL'ed (since coredump should report the correct ip). This
1641 * is even more important if the task has a handler for SIGSEGV/etc, The
1642 * _same_ instruction should be repeated again after return from the signal
1643 * handler, and SSTEP can never finish in this case.
1645 bool uprobe_deny_signal(void)
1647 struct task_struct
*t
= current
;
1648 struct uprobe_task
*utask
= t
->utask
;
1650 if (likely(!utask
|| !utask
->active_uprobe
))
1653 WARN_ON_ONCE(utask
->state
!= UTASK_SSTEP
);
1655 if (signal_pending(t
)) {
1656 spin_lock_irq(&t
->sighand
->siglock
);
1657 clear_tsk_thread_flag(t
, TIF_SIGPENDING
);
1658 spin_unlock_irq(&t
->sighand
->siglock
);
1660 if (__fatal_signal_pending(t
) || arch_uprobe_xol_was_trapped(t
)) {
1661 utask
->state
= UTASK_SSTEP_TRAPPED
;
1662 set_tsk_thread_flag(t
, TIF_UPROBE
);
1669 static void mmf_recalc_uprobes(struct mm_struct
*mm
)
1671 struct vm_area_struct
*vma
;
1673 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1674 if (!valid_vma(vma
, false))
1677 * This is not strictly accurate, we can race with
1678 * uprobe_unregister() and see the already removed
1679 * uprobe if delete_uprobe() was not yet called.
1680 * Or this uprobe can be filtered out.
1682 if (vma_has_uprobes(vma
, vma
->vm_start
, vma
->vm_end
))
1686 clear_bit(MMF_HAS_UPROBES
, &mm
->flags
);
1689 static int is_trap_at_addr(struct mm_struct
*mm
, unsigned long vaddr
)
1692 uprobe_opcode_t opcode
;
1695 pagefault_disable();
1696 result
= __copy_from_user_inatomic(&opcode
, (void __user
*)vaddr
,
1700 if (likely(result
== 0))
1703 result
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &page
, NULL
);
1707 copy_from_page(page
, vaddr
, &opcode
, UPROBE_SWBP_INSN_SIZE
);
1710 /* This needs to return true for any variant of the trap insn */
1711 return is_trap_insn(&opcode
);
1714 static struct uprobe
*find_active_uprobe(unsigned long bp_vaddr
, int *is_swbp
)
1716 struct mm_struct
*mm
= current
->mm
;
1717 struct uprobe
*uprobe
= NULL
;
1718 struct vm_area_struct
*vma
;
1720 down_read(&mm
->mmap_sem
);
1721 vma
= find_vma(mm
, bp_vaddr
);
1722 if (vma
&& vma
->vm_start
<= bp_vaddr
) {
1723 if (valid_vma(vma
, false)) {
1724 struct inode
*inode
= file_inode(vma
->vm_file
);
1725 loff_t offset
= vaddr_to_offset(vma
, bp_vaddr
);
1727 uprobe
= find_uprobe(inode
, offset
);
1731 *is_swbp
= is_trap_at_addr(mm
, bp_vaddr
);
1736 if (!uprobe
&& test_and_clear_bit(MMF_RECALC_UPROBES
, &mm
->flags
))
1737 mmf_recalc_uprobes(mm
);
1738 up_read(&mm
->mmap_sem
);
1743 static void handler_chain(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1745 struct uprobe_consumer
*uc
;
1746 int remove
= UPROBE_HANDLER_REMOVE
;
1747 bool need_prep
= false; /* prepare return uprobe, when needed */
1749 down_read(&uprobe
->register_rwsem
);
1750 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
1754 rc
= uc
->handler(uc
, regs
);
1755 WARN(rc
& ~UPROBE_HANDLER_MASK
,
1756 "bad rc=0x%x from %pf()\n", rc
, uc
->handler
);
1759 if (uc
->ret_handler
)
1765 if (need_prep
&& !remove
)
1766 prepare_uretprobe(uprobe
, regs
); /* put bp at return */
1768 if (remove
&& uprobe
->consumers
) {
1769 WARN_ON(!uprobe_is_active(uprobe
));
1770 unapply_uprobe(uprobe
, current
->mm
);
1772 up_read(&uprobe
->register_rwsem
);
1776 handle_uretprobe_chain(struct return_instance
*ri
, struct pt_regs
*regs
)
1778 struct uprobe
*uprobe
= ri
->uprobe
;
1779 struct uprobe_consumer
*uc
;
1781 down_read(&uprobe
->register_rwsem
);
1782 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
1783 if (uc
->ret_handler
)
1784 uc
->ret_handler(uc
, ri
->func
, regs
);
1786 up_read(&uprobe
->register_rwsem
);
1789 static struct return_instance
*find_next_ret_chain(struct return_instance
*ri
)
1794 chained
= ri
->chained
;
1795 ri
= ri
->next
; /* can't be NULL if chained */
1801 static void handle_trampoline(struct pt_regs
*regs
)
1803 struct uprobe_task
*utask
;
1804 struct return_instance
*ri
, *next
;
1807 utask
= current
->utask
;
1811 ri
= utask
->return_instances
;
1817 * We should throw out the frames invalidated by longjmp().
1818 * If this chain is valid, then the next one should be alive
1819 * or NULL; the latter case means that nobody but ri->func
1820 * could hit this trampoline on return. TODO: sigaltstack().
1822 next
= find_next_ret_chain(ri
);
1823 valid
= !next
|| arch_uretprobe_is_alive(next
, RP_CHECK_RET
, regs
);
1825 instruction_pointer_set(regs
, ri
->orig_ret_vaddr
);
1828 handle_uretprobe_chain(ri
, regs
);
1829 ri
= free_ret_instance(ri
);
1831 } while (ri
!= next
);
1834 utask
->return_instances
= ri
;
1838 uprobe_warn(current
, "handle uretprobe, sending SIGILL.");
1839 force_sig_info(SIGILL
, SEND_SIG_FORCED
, current
);
1843 bool __weak
arch_uprobe_ignore(struct arch_uprobe
*aup
, struct pt_regs
*regs
)
1848 bool __weak
arch_uretprobe_is_alive(struct return_instance
*ret
, enum rp_check ctx
,
1849 struct pt_regs
*regs
)
1855 * Run handler and ask thread to singlestep.
1856 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1858 static void handle_swbp(struct pt_regs
*regs
)
1860 struct uprobe
*uprobe
;
1861 unsigned long bp_vaddr
;
1862 int uninitialized_var(is_swbp
);
1864 bp_vaddr
= uprobe_get_swbp_addr(regs
);
1865 if (bp_vaddr
== get_trampoline_vaddr())
1866 return handle_trampoline(regs
);
1868 uprobe
= find_active_uprobe(bp_vaddr
, &is_swbp
);
1871 /* No matching uprobe; signal SIGTRAP. */
1872 send_sig(SIGTRAP
, current
, 0);
1875 * Either we raced with uprobe_unregister() or we can't
1876 * access this memory. The latter is only possible if
1877 * another thread plays with our ->mm. In both cases
1878 * we can simply restart. If this vma was unmapped we
1879 * can pretend this insn was not executed yet and get
1880 * the (correct) SIGSEGV after restart.
1882 instruction_pointer_set(regs
, bp_vaddr
);
1887 /* change it in advance for ->handler() and restart */
1888 instruction_pointer_set(regs
, bp_vaddr
);
1891 * TODO: move copy_insn/etc into _register and remove this hack.
1892 * After we hit the bp, _unregister + _register can install the
1893 * new and not-yet-analyzed uprobe at the same address, restart.
1895 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1896 if (unlikely(!test_bit(UPROBE_COPY_INSN
, &uprobe
->flags
)))
1899 /* Tracing handlers use ->utask to communicate with fetch methods */
1903 if (arch_uprobe_ignore(&uprobe
->arch
, regs
))
1906 handler_chain(uprobe
, regs
);
1908 if (arch_uprobe_skip_sstep(&uprobe
->arch
, regs
))
1911 if (!pre_ssout(uprobe
, regs
, bp_vaddr
))
1914 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1920 * Perform required fix-ups and disable singlestep.
1921 * Allow pending signals to take effect.
1923 static void handle_singlestep(struct uprobe_task
*utask
, struct pt_regs
*regs
)
1925 struct uprobe
*uprobe
;
1928 uprobe
= utask
->active_uprobe
;
1929 if (utask
->state
== UTASK_SSTEP_ACK
)
1930 err
= arch_uprobe_post_xol(&uprobe
->arch
, regs
);
1931 else if (utask
->state
== UTASK_SSTEP_TRAPPED
)
1932 arch_uprobe_abort_xol(&uprobe
->arch
, regs
);
1937 utask
->active_uprobe
= NULL
;
1938 utask
->state
= UTASK_RUNNING
;
1939 xol_free_insn_slot(current
);
1941 spin_lock_irq(¤t
->sighand
->siglock
);
1942 recalc_sigpending(); /* see uprobe_deny_signal() */
1943 spin_unlock_irq(¤t
->sighand
->siglock
);
1945 if (unlikely(err
)) {
1946 uprobe_warn(current
, "execute the probed insn, sending SIGILL.");
1947 force_sig_info(SIGILL
, SEND_SIG_FORCED
, current
);
1952 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1953 * allows the thread to return from interrupt. After that handle_swbp()
1954 * sets utask->active_uprobe.
1956 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1957 * and allows the thread to return from interrupt.
1959 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1960 * uprobe_notify_resume().
1962 void uprobe_notify_resume(struct pt_regs
*regs
)
1964 struct uprobe_task
*utask
;
1966 clear_thread_flag(TIF_UPROBE
);
1968 utask
= current
->utask
;
1969 if (utask
&& utask
->active_uprobe
)
1970 handle_singlestep(utask
, regs
);
1976 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1977 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1979 int uprobe_pre_sstep_notifier(struct pt_regs
*regs
)
1984 if (!test_bit(MMF_HAS_UPROBES
, ¤t
->mm
->flags
) &&
1985 (!current
->utask
|| !current
->utask
->return_instances
))
1988 set_thread_flag(TIF_UPROBE
);
1993 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1994 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1996 int uprobe_post_sstep_notifier(struct pt_regs
*regs
)
1998 struct uprobe_task
*utask
= current
->utask
;
2000 if (!current
->mm
|| !utask
|| !utask
->active_uprobe
)
2001 /* task is currently not uprobed */
2004 utask
->state
= UTASK_SSTEP_ACK
;
2005 set_thread_flag(TIF_UPROBE
);
2009 static struct notifier_block uprobe_exception_nb
= {
2010 .notifier_call
= arch_uprobe_exception_notify
,
2011 .priority
= INT_MAX
-1, /* notified after kprobes, kgdb */
2014 static int __init
init_uprobes(void)
2018 for (i
= 0; i
< UPROBES_HASH_SZ
; i
++)
2019 mutex_init(&uprobes_mmap_mutex
[i
]);
2021 if (percpu_init_rwsem(&dup_mmap_sem
))
2024 return register_die_notifier(&uprobe_exception_nb
);
2026 __initcall(init_uprobes
);