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/rmap.h> /* anon_vma_prepare */
31 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
32 #include <linux/swap.h> /* try_to_free_swap */
33 #include <linux/ptrace.h> /* user_enable_single_step */
34 #include <linux/kdebug.h> /* notifier mechanism */
36 #include <linux/uprobes.h>
38 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
39 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
41 static struct rb_root uprobes_tree
= RB_ROOT
;
43 static DEFINE_SPINLOCK(uprobes_treelock
); /* serialize rbtree access */
45 #define UPROBES_HASH_SZ 13
48 * We need separate register/unregister and mmap/munmap lock hashes because
49 * of mmap_sem nesting.
51 * uprobe_register() needs to install probes on (potentially) all processes
52 * and thus needs to acquire multiple mmap_sems (consequtively, not
53 * concurrently), whereas uprobe_mmap() is called while holding mmap_sem
54 * for the particular process doing the mmap.
56 * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem
57 * because of lock order against i_mmap_mutex. This means there's a hole in
58 * the register vma iteration where a mmap() can happen.
60 * Thus uprobe_register() can race with uprobe_mmap() and we can try and
61 * install a probe where one is already installed.
64 /* serialize (un)register */
65 static struct mutex uprobes_mutex
[UPROBES_HASH_SZ
];
67 #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
69 /* serialize uprobe->pending_list */
70 static struct mutex uprobes_mmap_mutex
[UPROBES_HASH_SZ
];
71 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
74 * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe
75 * events active at this time. Probably a fine grained per inode count is
78 static atomic_t uprobe_events
= ATOMIC_INIT(0);
81 struct rb_node rb_node
; /* node in the rb tree */
83 struct rw_semaphore consumer_rwsem
;
84 struct list_head pending_list
;
85 struct uprobe_consumer
*consumers
;
86 struct inode
*inode
; /* Also hold a ref to inode */
89 struct arch_uprobe arch
;
93 * valid_vma: Verify if the specified vma is an executable vma
94 * Relax restrictions while unregistering: vm_flags might have
95 * changed after breakpoint was inserted.
96 * - is_register: indicates if we are in register context.
97 * - Return 1 if the specified virtual address is in an
100 static bool valid_vma(struct vm_area_struct
*vma
, bool is_register
)
108 if ((vma
->vm_flags
& (VM_HUGETLB
|VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
))
109 == (VM_READ
|VM_EXEC
))
115 static loff_t
vma_address(struct vm_area_struct
*vma
, loff_t offset
)
119 vaddr
= vma
->vm_start
+ offset
;
120 vaddr
-= vma
->vm_pgoff
<< PAGE_SHIFT
;
126 * __replace_page - replace page in vma by new page.
127 * based on replace_page in mm/ksm.c
129 * @vma: vma that holds the pte pointing to page
130 * @addr: address the old @page is mapped at
131 * @page: the cowed page we are replacing by kpage
132 * @kpage: the modified page we replace page by
134 * Returns 0 on success, -EFAULT on failure.
136 static int __replace_page(struct vm_area_struct
*vma
, unsigned long addr
,
137 struct page
*page
, struct page
*kpage
)
139 struct mm_struct
*mm
= vma
->vm_mm
;
143 ptep
= page_check_address(page
, mm
, addr
, &ptl
, 0);
148 page_add_new_anon_rmap(kpage
, vma
, addr
);
150 if (!PageAnon(page
)) {
151 dec_mm_counter(mm
, MM_FILEPAGES
);
152 inc_mm_counter(mm
, MM_ANONPAGES
);
155 flush_cache_page(vma
, addr
, pte_pfn(*ptep
));
156 ptep_clear_flush(vma
, addr
, ptep
);
157 set_pte_at_notify(mm
, addr
, ptep
, mk_pte(kpage
, vma
->vm_page_prot
));
159 page_remove_rmap(page
);
160 if (!page_mapped(page
))
161 try_to_free_swap(page
);
163 pte_unmap_unlock(ptep
, ptl
);
169 * is_swbp_insn - check if instruction is breakpoint instruction.
170 * @insn: instruction to be checked.
171 * Default implementation of is_swbp_insn
172 * Returns true if @insn is a breakpoint instruction.
174 bool __weak
is_swbp_insn(uprobe_opcode_t
*insn
)
176 return *insn
== UPROBE_SWBP_INSN
;
181 * Expect the breakpoint instruction to be the smallest size instruction for
182 * the architecture. If an arch has variable length instruction and the
183 * breakpoint instruction is not of the smallest length instruction
184 * supported by that architecture then we need to modify read_opcode /
185 * write_opcode accordingly. This would never be a problem for archs that
186 * have fixed length instructions.
190 * write_opcode - write the opcode at a given virtual address.
191 * @auprobe: arch breakpointing information.
192 * @mm: the probed process address space.
193 * @vaddr: the virtual address to store the opcode.
194 * @opcode: opcode to be written at @vaddr.
196 * Called with mm->mmap_sem held (for read and with a reference to
199 * For mm @mm, write the opcode at @vaddr.
200 * Return 0 (success) or a negative errno.
202 static int write_opcode(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
,
203 unsigned long vaddr
, uprobe_opcode_t opcode
)
205 struct page
*old_page
, *new_page
;
206 void *vaddr_old
, *vaddr_new
;
207 struct vm_area_struct
*vma
;
211 /* Read the page with vaddr into memory */
212 ret
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 0, &old_page
, &vma
);
217 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, vaddr
);
221 __SetPageUptodate(new_page
);
224 * lock page will serialize against do_wp_page()'s
225 * PageAnon() handling
228 /* copy the page now that we've got it stable */
229 vaddr_old
= kmap_atomic(old_page
);
230 vaddr_new
= kmap_atomic(new_page
);
232 memcpy(vaddr_new
, vaddr_old
, PAGE_SIZE
);
233 memcpy(vaddr_new
+ (vaddr
& ~PAGE_MASK
), &opcode
, UPROBE_SWBP_INSN_SIZE
);
235 kunmap_atomic(vaddr_new
);
236 kunmap_atomic(vaddr_old
);
238 ret
= anon_vma_prepare(vma
);
243 ret
= __replace_page(vma
, vaddr
, old_page
, new_page
);
244 unlock_page(new_page
);
247 unlock_page(old_page
);
248 page_cache_release(new_page
);
253 if (unlikely(ret
== -EAGAIN
))
259 * read_opcode - read the opcode at a given virtual address.
260 * @mm: the probed process address space.
261 * @vaddr: the virtual address to read the opcode.
262 * @opcode: location to store the read opcode.
264 * Called with mm->mmap_sem held (for read and with a reference to
267 * For mm @mm, read the opcode at @vaddr and store it in @opcode.
268 * Return 0 (success) or a negative errno.
270 static int read_opcode(struct mm_struct
*mm
, unsigned long vaddr
, uprobe_opcode_t
*opcode
)
276 ret
= get_user_pages(NULL
, mm
, vaddr
, 1, 0, 1, &page
, NULL
);
281 vaddr_new
= kmap_atomic(page
);
283 memcpy(opcode
, vaddr_new
+ vaddr
, UPROBE_SWBP_INSN_SIZE
);
284 kunmap_atomic(vaddr_new
);
292 static int is_swbp_at_addr(struct mm_struct
*mm
, unsigned long vaddr
)
294 uprobe_opcode_t opcode
;
297 if (current
->mm
== mm
) {
299 result
= __copy_from_user_inatomic(&opcode
, (void __user
*)vaddr
,
303 if (likely(result
== 0))
307 result
= read_opcode(mm
, vaddr
, &opcode
);
311 if (is_swbp_insn(&opcode
))
318 * set_swbp - store breakpoint at a given address.
319 * @auprobe: arch specific probepoint information.
320 * @mm: the probed process address space.
321 * @vaddr: the virtual address to insert the opcode.
323 * For mm @mm, store the breakpoint instruction at @vaddr.
324 * Return 0 (success) or a negative errno.
326 int __weak
set_swbp(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
330 * See the comment near uprobes_hash().
332 result
= is_swbp_at_addr(mm
, vaddr
);
339 return write_opcode(auprobe
, mm
, vaddr
, UPROBE_SWBP_INSN
);
343 * set_orig_insn - Restore the original instruction.
344 * @mm: the probed process address space.
345 * @auprobe: arch specific probepoint information.
346 * @vaddr: the virtual address to insert the opcode.
347 * @verify: if true, verify existance of breakpoint instruction.
349 * For mm @mm, restore the original opcode (opcode) at @vaddr.
350 * Return 0 (success) or a negative errno.
353 set_orig_insn(struct arch_uprobe
*auprobe
, struct mm_struct
*mm
, unsigned long vaddr
, bool verify
)
358 result
= is_swbp_at_addr(mm
, vaddr
);
365 return write_opcode(auprobe
, mm
, vaddr
, *(uprobe_opcode_t
*)auprobe
->insn
);
368 static int match_uprobe(struct uprobe
*l
, struct uprobe
*r
)
370 if (l
->inode
< r
->inode
)
373 if (l
->inode
> r
->inode
)
376 if (l
->offset
< r
->offset
)
379 if (l
->offset
> r
->offset
)
385 static struct uprobe
*__find_uprobe(struct inode
*inode
, loff_t offset
)
387 struct uprobe u
= { .inode
= inode
, .offset
= offset
};
388 struct rb_node
*n
= uprobes_tree
.rb_node
;
389 struct uprobe
*uprobe
;
393 uprobe
= rb_entry(n
, struct uprobe
, rb_node
);
394 match
= match_uprobe(&u
, uprobe
);
396 atomic_inc(&uprobe
->ref
);
409 * Find a uprobe corresponding to a given inode:offset
410 * Acquires uprobes_treelock
412 static struct uprobe
*find_uprobe(struct inode
*inode
, loff_t offset
)
414 struct uprobe
*uprobe
;
417 spin_lock_irqsave(&uprobes_treelock
, flags
);
418 uprobe
= __find_uprobe(inode
, offset
);
419 spin_unlock_irqrestore(&uprobes_treelock
, flags
);
424 static struct uprobe
*__insert_uprobe(struct uprobe
*uprobe
)
426 struct rb_node
**p
= &uprobes_tree
.rb_node
;
427 struct rb_node
*parent
= NULL
;
433 u
= rb_entry(parent
, struct uprobe
, rb_node
);
434 match
= match_uprobe(uprobe
, u
);
441 p
= &parent
->rb_left
;
443 p
= &parent
->rb_right
;
448 rb_link_node(&uprobe
->rb_node
, parent
, p
);
449 rb_insert_color(&uprobe
->rb_node
, &uprobes_tree
);
450 /* get access + creation ref */
451 atomic_set(&uprobe
->ref
, 2);
457 * Acquire uprobes_treelock.
458 * Matching uprobe already exists in rbtree;
459 * increment (access refcount) and return the matching uprobe.
461 * No matching uprobe; insert the uprobe in rb_tree;
462 * get a double refcount (access + creation) and return NULL.
464 static struct uprobe
*insert_uprobe(struct uprobe
*uprobe
)
469 spin_lock_irqsave(&uprobes_treelock
, flags
);
470 u
= __insert_uprobe(uprobe
);
471 spin_unlock_irqrestore(&uprobes_treelock
, flags
);
473 /* For now assume that the instruction need not be single-stepped */
474 uprobe
->flags
|= UPROBE_SKIP_SSTEP
;
479 static void put_uprobe(struct uprobe
*uprobe
)
481 if (atomic_dec_and_test(&uprobe
->ref
))
485 static struct uprobe
*alloc_uprobe(struct inode
*inode
, loff_t offset
)
487 struct uprobe
*uprobe
, *cur_uprobe
;
489 uprobe
= kzalloc(sizeof(struct uprobe
), GFP_KERNEL
);
493 uprobe
->inode
= igrab(inode
);
494 uprobe
->offset
= offset
;
495 init_rwsem(&uprobe
->consumer_rwsem
);
497 /* add to uprobes_tree, sorted on inode:offset */
498 cur_uprobe
= insert_uprobe(uprobe
);
500 /* a uprobe exists for this inode:offset combination */
506 atomic_inc(&uprobe_events
);
512 static void handler_chain(struct uprobe
*uprobe
, struct pt_regs
*regs
)
514 struct uprobe_consumer
*uc
;
516 if (!(uprobe
->flags
& UPROBE_RUN_HANDLER
))
519 down_read(&uprobe
->consumer_rwsem
);
520 for (uc
= uprobe
->consumers
; uc
; uc
= uc
->next
) {
521 if (!uc
->filter
|| uc
->filter(uc
, current
))
522 uc
->handler(uc
, regs
);
524 up_read(&uprobe
->consumer_rwsem
);
527 /* Returns the previous consumer */
528 static struct uprobe_consumer
*
529 consumer_add(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
531 down_write(&uprobe
->consumer_rwsem
);
532 uc
->next
= uprobe
->consumers
;
533 uprobe
->consumers
= uc
;
534 up_write(&uprobe
->consumer_rwsem
);
540 * For uprobe @uprobe, delete the consumer @uc.
541 * Return true if the @uc is deleted successfully
544 static bool consumer_del(struct uprobe
*uprobe
, struct uprobe_consumer
*uc
)
546 struct uprobe_consumer
**con
;
549 down_write(&uprobe
->consumer_rwsem
);
550 for (con
= &uprobe
->consumers
; *con
; con
= &(*con
)->next
) {
557 up_write(&uprobe
->consumer_rwsem
);
563 __copy_insn(struct address_space
*mapping
, struct file
*filp
, char *insn
,
564 unsigned long nbytes
, loff_t offset
)
574 if (!mapping
->a_ops
->readpage
)
577 idx
= offset
>> PAGE_CACHE_SHIFT
;
578 off
= offset
& ~PAGE_MASK
;
581 * Ensure that the page that has the original instruction is
582 * populated and in page-cache.
584 page
= read_mapping_page(mapping
, idx
, filp
);
586 return PTR_ERR(page
);
588 vaddr
= kmap_atomic(page
);
589 memcpy(insn
, vaddr
+ off
, nbytes
);
590 kunmap_atomic(vaddr
);
591 page_cache_release(page
);
596 static int copy_insn(struct uprobe
*uprobe
, struct file
*filp
)
598 struct address_space
*mapping
;
599 unsigned long nbytes
;
602 nbytes
= PAGE_SIZE
- (uprobe
->offset
& ~PAGE_MASK
);
603 mapping
= uprobe
->inode
->i_mapping
;
605 /* Instruction at end of binary; copy only available bytes */
606 if (uprobe
->offset
+ MAX_UINSN_BYTES
> uprobe
->inode
->i_size
)
607 bytes
= uprobe
->inode
->i_size
- uprobe
->offset
;
609 bytes
= MAX_UINSN_BYTES
;
611 /* Instruction at the page-boundary; copy bytes in second page */
612 if (nbytes
< bytes
) {
613 int err
= __copy_insn(mapping
, filp
, uprobe
->arch
.insn
+ nbytes
,
614 bytes
- nbytes
, uprobe
->offset
+ nbytes
);
619 return __copy_insn(mapping
, filp
, uprobe
->arch
.insn
, bytes
, uprobe
->offset
);
623 * How mm->uprobes_state.count gets updated
624 * uprobe_mmap() increments the count if
625 * - it successfully adds a breakpoint.
626 * - it cannot add a breakpoint, but sees that there is a underlying
627 * breakpoint (via a is_swbp_at_addr()).
629 * uprobe_munmap() decrements the count if
630 * - it sees a underlying breakpoint, (via is_swbp_at_addr)
631 * (Subsequent uprobe_unregister wouldnt find the breakpoint
632 * unless a uprobe_mmap kicks in, since the old vma would be
633 * dropped just after uprobe_munmap.)
635 * uprobe_register increments the count if:
636 * - it successfully adds a breakpoint.
638 * uprobe_unregister decrements the count if:
639 * - it sees a underlying breakpoint and removes successfully.
640 * (via is_swbp_at_addr)
641 * (Subsequent uprobe_munmap wouldnt find the breakpoint
642 * since there is no underlying breakpoint after the
643 * breakpoint removal.)
646 install_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
,
647 struct vm_area_struct
*vma
, unsigned long vaddr
)
652 * If probe is being deleted, unregister thread could be done with
653 * the vma-rmap-walk through. Adding a probe now can be fatal since
654 * nobody will be able to cleanup. Also we could be from fork or
655 * mremap path, where the probe might have already been inserted.
656 * Hence behave as if probe already existed.
658 if (!uprobe
->consumers
)
661 if (!(uprobe
->flags
& UPROBE_COPY_INSN
)) {
662 ret
= copy_insn(uprobe
, vma
->vm_file
);
666 if (is_swbp_insn((uprobe_opcode_t
*)uprobe
->arch
.insn
))
669 ret
= arch_uprobe_analyze_insn(&uprobe
->arch
, mm
, vaddr
);
673 /* write_opcode() assumes we don't cross page boundary */
674 BUG_ON((uprobe
->offset
& ~PAGE_MASK
) +
675 UPROBE_SWBP_INSN_SIZE
> PAGE_SIZE
);
677 uprobe
->flags
|= UPROBE_COPY_INSN
;
681 * Ideally, should be updating the probe count after the breakpoint
682 * has been successfully inserted. However a thread could hit the
683 * breakpoint we just inserted even before the probe count is
684 * incremented. If this is the first breakpoint placed, breakpoint
685 * notifier might ignore uprobes and pass the trap to the thread.
686 * Hence increment before and decrement on failure.
688 atomic_inc(&mm
->uprobes_state
.count
);
689 ret
= set_swbp(&uprobe
->arch
, mm
, vaddr
);
691 atomic_dec(&mm
->uprobes_state
.count
);
697 remove_breakpoint(struct uprobe
*uprobe
, struct mm_struct
*mm
, unsigned long vaddr
)
699 if (!set_orig_insn(&uprobe
->arch
, mm
, vaddr
, true))
700 atomic_dec(&mm
->uprobes_state
.count
);
704 * There could be threads that have already hit the breakpoint. They
705 * will recheck the current insn and restart if find_uprobe() fails.
706 * See find_active_uprobe().
708 static void delete_uprobe(struct uprobe
*uprobe
)
712 spin_lock_irqsave(&uprobes_treelock
, flags
);
713 rb_erase(&uprobe
->rb_node
, &uprobes_tree
);
714 spin_unlock_irqrestore(&uprobes_treelock
, flags
);
717 atomic_dec(&uprobe_events
);
721 struct map_info
*next
;
722 struct mm_struct
*mm
;
726 static inline struct map_info
*free_map_info(struct map_info
*info
)
728 struct map_info
*next
= info
->next
;
733 static struct map_info
*
734 build_map_info(struct address_space
*mapping
, loff_t offset
, bool is_register
)
736 unsigned long pgoff
= offset
>> PAGE_SHIFT
;
737 struct prio_tree_iter iter
;
738 struct vm_area_struct
*vma
;
739 struct map_info
*curr
= NULL
;
740 struct map_info
*prev
= NULL
;
741 struct map_info
*info
;
745 mutex_lock(&mapping
->i_mmap_mutex
);
746 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
747 if (!valid_vma(vma
, is_register
))
750 if (!prev
&& !more
) {
752 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
753 * reclaim. This is optimistic, no harm done if it fails.
755 prev
= kmalloc(sizeof(struct map_info
),
756 GFP_NOWAIT
| __GFP_NOMEMALLOC
| __GFP_NOWARN
);
765 if (!atomic_inc_not_zero(&vma
->vm_mm
->mm_users
))
773 info
->mm
= vma
->vm_mm
;
774 info
->vaddr
= vma_address(vma
, offset
);
776 mutex_unlock(&mapping
->i_mmap_mutex
);
788 info
= kmalloc(sizeof(struct map_info
), GFP_KERNEL
);
790 curr
= ERR_PTR(-ENOMEM
);
800 prev
= free_map_info(prev
);
804 static int register_for_each_vma(struct uprobe
*uprobe
, bool is_register
)
806 struct map_info
*info
;
809 info
= build_map_info(uprobe
->inode
->i_mapping
,
810 uprobe
->offset
, is_register
);
812 return PTR_ERR(info
);
815 struct mm_struct
*mm
= info
->mm
;
816 struct vm_area_struct
*vma
;
821 down_write(&mm
->mmap_sem
);
822 vma
= find_vma(mm
, (unsigned long)info
->vaddr
);
823 if (!vma
|| !valid_vma(vma
, is_register
))
826 if (vma
->vm_file
->f_mapping
->host
!= uprobe
->inode
||
827 vma_address(vma
, uprobe
->offset
) != info
->vaddr
)
831 err
= install_breakpoint(uprobe
, mm
, vma
, info
->vaddr
);
833 * We can race against uprobe_mmap(), see the
834 * comment near uprobe_hash().
839 remove_breakpoint(uprobe
, mm
, info
->vaddr
);
842 up_write(&mm
->mmap_sem
);
845 info
= free_map_info(info
);
851 static int __uprobe_register(struct uprobe
*uprobe
)
853 return register_for_each_vma(uprobe
, true);
856 static void __uprobe_unregister(struct uprobe
*uprobe
)
858 if (!register_for_each_vma(uprobe
, false))
859 delete_uprobe(uprobe
);
861 /* TODO : cant unregister? schedule a worker thread */
865 * uprobe_register - register a probe
866 * @inode: the file in which the probe has to be placed.
867 * @offset: offset from the start of the file.
868 * @uc: information on howto handle the probe..
870 * Apart from the access refcount, uprobe_register() takes a creation
871 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
872 * inserted into the rbtree (i.e first consumer for a @inode:@offset
873 * tuple). Creation refcount stops uprobe_unregister from freeing the
874 * @uprobe even before the register operation is complete. Creation
875 * refcount is released when the last @uc for the @uprobe
878 * Return errno if it cannot successully install probes
879 * else return 0 (success)
881 int uprobe_register(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
883 struct uprobe
*uprobe
;
886 if (!inode
|| !uc
|| uc
->next
)
889 if (offset
> i_size_read(inode
))
893 mutex_lock(uprobes_hash(inode
));
894 uprobe
= alloc_uprobe(inode
, offset
);
896 if (uprobe
&& !consumer_add(uprobe
, uc
)) {
897 ret
= __uprobe_register(uprobe
);
899 uprobe
->consumers
= NULL
;
900 __uprobe_unregister(uprobe
);
902 uprobe
->flags
|= UPROBE_RUN_HANDLER
;
906 mutex_unlock(uprobes_hash(inode
));
913 * uprobe_unregister - unregister a already registered probe.
914 * @inode: the file in which the probe has to be removed.
915 * @offset: offset from the start of the file.
916 * @uc: identify which probe if multiple probes are colocated.
918 void uprobe_unregister(struct inode
*inode
, loff_t offset
, struct uprobe_consumer
*uc
)
920 struct uprobe
*uprobe
;
925 uprobe
= find_uprobe(inode
, offset
);
929 mutex_lock(uprobes_hash(inode
));
931 if (consumer_del(uprobe
, uc
)) {
932 if (!uprobe
->consumers
) {
933 __uprobe_unregister(uprobe
);
934 uprobe
->flags
&= ~UPROBE_RUN_HANDLER
;
938 mutex_unlock(uprobes_hash(inode
));
944 * Of all the nodes that correspond to the given inode, return the node
945 * with the least offset.
947 static struct rb_node
*find_least_offset_node(struct inode
*inode
)
949 struct uprobe u
= { .inode
= inode
, .offset
= 0};
950 struct rb_node
*n
= uprobes_tree
.rb_node
;
951 struct rb_node
*close_node
= NULL
;
952 struct uprobe
*uprobe
;
956 uprobe
= rb_entry(n
, struct uprobe
, rb_node
);
957 match
= match_uprobe(&u
, uprobe
);
959 if (uprobe
->inode
== inode
)
975 * For a given inode, build a list of probes that need to be inserted.
977 static void build_probe_list(struct inode
*inode
, struct list_head
*head
)
979 struct uprobe
*uprobe
;
983 spin_lock_irqsave(&uprobes_treelock
, flags
);
985 n
= find_least_offset_node(inode
);
987 for (; n
; n
= rb_next(n
)) {
988 uprobe
= rb_entry(n
, struct uprobe
, rb_node
);
989 if (uprobe
->inode
!= inode
)
992 list_add(&uprobe
->pending_list
, head
);
993 atomic_inc(&uprobe
->ref
);
996 spin_unlock_irqrestore(&uprobes_treelock
, flags
);
1000 * Called from mmap_region.
1001 * called with mm->mmap_sem acquired.
1003 * Return -ve no if we fail to insert probes and we cannot
1005 * Return 0 otherwise. i.e:
1007 * - successful insertion of probes
1008 * - (or) no possible probes to be inserted.
1009 * - (or) insertion of probes failed but we can bail-out.
1011 int uprobe_mmap(struct vm_area_struct
*vma
)
1013 struct list_head tmp_list
;
1014 struct uprobe
*uprobe
;
1015 struct inode
*inode
;
1018 if (!atomic_read(&uprobe_events
) || !valid_vma(vma
, true))
1021 inode
= vma
->vm_file
->f_mapping
->host
;
1025 INIT_LIST_HEAD(&tmp_list
);
1026 mutex_lock(uprobes_mmap_hash(inode
));
1027 build_probe_list(inode
, &tmp_list
);
1032 list_for_each_entry(uprobe
, &tmp_list
, pending_list
) {
1034 loff_t vaddr
= vma_address(vma
, uprobe
->offset
);
1036 if (vaddr
< vma
->vm_start
|| vaddr
>= vma
->vm_end
) {
1041 ret
= install_breakpoint(uprobe
, vma
->vm_mm
, vma
, vaddr
);
1043 * We can race against uprobe_register(), see the
1044 * comment near uprobe_hash().
1046 if (ret
== -EEXIST
) {
1049 if (!is_swbp_at_addr(vma
->vm_mm
, vaddr
))
1053 * Unable to insert a breakpoint, but
1054 * breakpoint lies underneath. Increment the
1057 atomic_inc(&vma
->vm_mm
->uprobes_state
.count
);
1066 mutex_unlock(uprobes_mmap_hash(inode
));
1069 atomic_sub(count
, &vma
->vm_mm
->uprobes_state
.count
);
1075 * Called in context of a munmap of a vma.
1077 void uprobe_munmap(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
1079 struct list_head tmp_list
;
1080 struct uprobe
*uprobe
;
1081 struct inode
*inode
;
1083 if (!atomic_read(&uprobe_events
) || !valid_vma(vma
, false))
1086 if (!atomic_read(&vma
->vm_mm
->uprobes_state
.count
))
1089 inode
= vma
->vm_file
->f_mapping
->host
;
1093 INIT_LIST_HEAD(&tmp_list
);
1094 mutex_lock(uprobes_mmap_hash(inode
));
1095 build_probe_list(inode
, &tmp_list
);
1097 list_for_each_entry(uprobe
, &tmp_list
, pending_list
) {
1098 loff_t vaddr
= vma_address(vma
, uprobe
->offset
);
1100 if (vaddr
>= start
&& vaddr
< end
) {
1102 * An unregister could have removed the probe before
1103 * unmap. So check before we decrement the count.
1105 if (is_swbp_at_addr(vma
->vm_mm
, vaddr
) == 1)
1106 atomic_dec(&vma
->vm_mm
->uprobes_state
.count
);
1110 mutex_unlock(uprobes_mmap_hash(inode
));
1113 /* Slot allocation for XOL */
1114 static int xol_add_vma(struct xol_area
*area
)
1116 struct mm_struct
*mm
;
1119 area
->page
= alloc_page(GFP_HIGHUSER
);
1126 down_write(&mm
->mmap_sem
);
1127 if (mm
->uprobes_state
.xol_area
)
1132 /* Try to map as high as possible, this is only a hint. */
1133 area
->vaddr
= get_unmapped_area(NULL
, TASK_SIZE
- PAGE_SIZE
, PAGE_SIZE
, 0, 0);
1134 if (area
->vaddr
& ~PAGE_MASK
) {
1139 ret
= install_special_mapping(mm
, area
->vaddr
, PAGE_SIZE
,
1140 VM_EXEC
|VM_MAYEXEC
|VM_DONTCOPY
|VM_IO
, &area
->page
);
1144 smp_wmb(); /* pairs with get_xol_area() */
1145 mm
->uprobes_state
.xol_area
= area
;
1149 up_write(&mm
->mmap_sem
);
1151 __free_page(area
->page
);
1156 static struct xol_area
*get_xol_area(struct mm_struct
*mm
)
1158 struct xol_area
*area
;
1160 area
= mm
->uprobes_state
.xol_area
;
1161 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1167 * xol_alloc_area - Allocate process's xol_area.
1168 * This area will be used for storing instructions for execution out of
1171 * Returns the allocated area or NULL.
1173 static struct xol_area
*xol_alloc_area(void)
1175 struct xol_area
*area
;
1177 area
= kzalloc(sizeof(*area
), GFP_KERNEL
);
1178 if (unlikely(!area
))
1181 area
->bitmap
= kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE
) * sizeof(long), GFP_KERNEL
);
1186 init_waitqueue_head(&area
->wq
);
1187 if (!xol_add_vma(area
))
1191 kfree(area
->bitmap
);
1194 return get_xol_area(current
->mm
);
1198 * uprobe_clear_state - Free the area allocated for slots.
1200 void uprobe_clear_state(struct mm_struct
*mm
)
1202 struct xol_area
*area
= mm
->uprobes_state
.xol_area
;
1207 put_page(area
->page
);
1208 kfree(area
->bitmap
);
1213 * uprobe_reset_state - Free the area allocated for slots.
1215 void uprobe_reset_state(struct mm_struct
*mm
)
1217 mm
->uprobes_state
.xol_area
= NULL
;
1218 atomic_set(&mm
->uprobes_state
.count
, 0);
1222 * - search for a free slot.
1224 static unsigned long xol_take_insn_slot(struct xol_area
*area
)
1226 unsigned long slot_addr
;
1230 slot_nr
= find_first_zero_bit(area
->bitmap
, UINSNS_PER_PAGE
);
1231 if (slot_nr
< UINSNS_PER_PAGE
) {
1232 if (!test_and_set_bit(slot_nr
, area
->bitmap
))
1235 slot_nr
= UINSNS_PER_PAGE
;
1238 wait_event(area
->wq
, (atomic_read(&area
->slot_count
) < UINSNS_PER_PAGE
));
1239 } while (slot_nr
>= UINSNS_PER_PAGE
);
1241 slot_addr
= area
->vaddr
+ (slot_nr
* UPROBE_XOL_SLOT_BYTES
);
1242 atomic_inc(&area
->slot_count
);
1248 * xol_get_insn_slot - If was not allocated a slot, then
1250 * Returns the allocated slot address or 0.
1252 static unsigned long xol_get_insn_slot(struct uprobe
*uprobe
, unsigned long slot_addr
)
1254 struct xol_area
*area
;
1255 unsigned long offset
;
1258 area
= get_xol_area(current
->mm
);
1260 area
= xol_alloc_area();
1264 current
->utask
->xol_vaddr
= xol_take_insn_slot(area
);
1267 * Initialize the slot if xol_vaddr points to valid
1270 if (unlikely(!current
->utask
->xol_vaddr
))
1273 current
->utask
->vaddr
= slot_addr
;
1274 offset
= current
->utask
->xol_vaddr
& ~PAGE_MASK
;
1275 vaddr
= kmap_atomic(area
->page
);
1276 memcpy(vaddr
+ offset
, uprobe
->arch
.insn
, MAX_UINSN_BYTES
);
1277 kunmap_atomic(vaddr
);
1279 return current
->utask
->xol_vaddr
;
1283 * xol_free_insn_slot - If slot was earlier allocated by
1284 * @xol_get_insn_slot(), make the slot available for
1285 * subsequent requests.
1287 static void xol_free_insn_slot(struct task_struct
*tsk
)
1289 struct xol_area
*area
;
1290 unsigned long vma_end
;
1291 unsigned long slot_addr
;
1293 if (!tsk
->mm
|| !tsk
->mm
->uprobes_state
.xol_area
|| !tsk
->utask
)
1296 slot_addr
= tsk
->utask
->xol_vaddr
;
1298 if (unlikely(!slot_addr
|| IS_ERR_VALUE(slot_addr
)))
1301 area
= tsk
->mm
->uprobes_state
.xol_area
;
1302 vma_end
= area
->vaddr
+ PAGE_SIZE
;
1303 if (area
->vaddr
<= slot_addr
&& slot_addr
< vma_end
) {
1304 unsigned long offset
;
1307 offset
= slot_addr
- area
->vaddr
;
1308 slot_nr
= offset
/ UPROBE_XOL_SLOT_BYTES
;
1309 if (slot_nr
>= UINSNS_PER_PAGE
)
1312 clear_bit(slot_nr
, area
->bitmap
);
1313 atomic_dec(&area
->slot_count
);
1314 if (waitqueue_active(&area
->wq
))
1317 tsk
->utask
->xol_vaddr
= 0;
1322 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1323 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1325 * Return the address of the breakpoint instruction.
1327 unsigned long __weak
uprobe_get_swbp_addr(struct pt_regs
*regs
)
1329 return instruction_pointer(regs
) - UPROBE_SWBP_INSN_SIZE
;
1333 * Called with no locks held.
1334 * Called in context of a exiting or a exec-ing thread.
1336 void uprobe_free_utask(struct task_struct
*t
)
1338 struct uprobe_task
*utask
= t
->utask
;
1343 if (utask
->active_uprobe
)
1344 put_uprobe(utask
->active_uprobe
);
1346 xol_free_insn_slot(t
);
1352 * Called in context of a new clone/fork from copy_process.
1354 void uprobe_copy_process(struct task_struct
*t
)
1360 * Allocate a uprobe_task object for the task.
1361 * Called when the thread hits a breakpoint for the first time.
1364 * - pointer to new uprobe_task on success
1367 static struct uprobe_task
*add_utask(void)
1369 struct uprobe_task
*utask
;
1371 utask
= kzalloc(sizeof *utask
, GFP_KERNEL
);
1372 if (unlikely(!utask
))
1375 current
->utask
= utask
;
1379 /* Prepare to single-step probed instruction out of line. */
1381 pre_ssout(struct uprobe
*uprobe
, struct pt_regs
*regs
, unsigned long vaddr
)
1383 if (xol_get_insn_slot(uprobe
, vaddr
) && !arch_uprobe_pre_xol(&uprobe
->arch
, regs
))
1390 * If we are singlestepping, then ensure this thread is not connected to
1391 * non-fatal signals until completion of singlestep. When xol insn itself
1392 * triggers the signal, restart the original insn even if the task is
1393 * already SIGKILL'ed (since coredump should report the correct ip). This
1394 * is even more important if the task has a handler for SIGSEGV/etc, The
1395 * _same_ instruction should be repeated again after return from the signal
1396 * handler, and SSTEP can never finish in this case.
1398 bool uprobe_deny_signal(void)
1400 struct task_struct
*t
= current
;
1401 struct uprobe_task
*utask
= t
->utask
;
1403 if (likely(!utask
|| !utask
->active_uprobe
))
1406 WARN_ON_ONCE(utask
->state
!= UTASK_SSTEP
);
1408 if (signal_pending(t
)) {
1409 spin_lock_irq(&t
->sighand
->siglock
);
1410 clear_tsk_thread_flag(t
, TIF_SIGPENDING
);
1411 spin_unlock_irq(&t
->sighand
->siglock
);
1413 if (__fatal_signal_pending(t
) || arch_uprobe_xol_was_trapped(t
)) {
1414 utask
->state
= UTASK_SSTEP_TRAPPED
;
1415 set_tsk_thread_flag(t
, TIF_UPROBE
);
1416 set_tsk_thread_flag(t
, TIF_NOTIFY_RESUME
);
1424 * Avoid singlestepping the original instruction if the original instruction
1425 * is a NOP or can be emulated.
1427 static bool can_skip_sstep(struct uprobe
*uprobe
, struct pt_regs
*regs
)
1429 if (arch_uprobe_skip_sstep(&uprobe
->arch
, regs
))
1432 uprobe
->flags
&= ~UPROBE_SKIP_SSTEP
;
1436 static struct uprobe
*find_active_uprobe(unsigned long bp_vaddr
, int *is_swbp
)
1438 struct mm_struct
*mm
= current
->mm
;
1439 struct uprobe
*uprobe
= NULL
;
1440 struct vm_area_struct
*vma
;
1442 down_read(&mm
->mmap_sem
);
1443 vma
= find_vma(mm
, bp_vaddr
);
1444 if (vma
&& vma
->vm_start
<= bp_vaddr
) {
1445 if (valid_vma(vma
, false)) {
1446 struct inode
*inode
;
1449 inode
= vma
->vm_file
->f_mapping
->host
;
1450 offset
= bp_vaddr
- vma
->vm_start
;
1451 offset
+= (vma
->vm_pgoff
<< PAGE_SHIFT
);
1452 uprobe
= find_uprobe(inode
, offset
);
1456 *is_swbp
= is_swbp_at_addr(mm
, bp_vaddr
);
1460 up_read(&mm
->mmap_sem
);
1466 * Run handler and ask thread to singlestep.
1467 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1469 static void handle_swbp(struct pt_regs
*regs
)
1471 struct uprobe_task
*utask
;
1472 struct uprobe
*uprobe
;
1473 unsigned long bp_vaddr
;
1474 int uninitialized_var(is_swbp
);
1476 bp_vaddr
= uprobe_get_swbp_addr(regs
);
1477 uprobe
= find_active_uprobe(bp_vaddr
, &is_swbp
);
1481 /* No matching uprobe; signal SIGTRAP. */
1482 send_sig(SIGTRAP
, current
, 0);
1485 * Either we raced with uprobe_unregister() or we can't
1486 * access this memory. The latter is only possible if
1487 * another thread plays with our ->mm. In both cases
1488 * we can simply restart. If this vma was unmapped we
1489 * can pretend this insn was not executed yet and get
1490 * the (correct) SIGSEGV after restart.
1492 instruction_pointer_set(regs
, bp_vaddr
);
1497 utask
= current
->utask
;
1499 utask
= add_utask();
1500 /* Cannot allocate; re-execute the instruction. */
1504 utask
->active_uprobe
= uprobe
;
1505 handler_chain(uprobe
, regs
);
1506 if (uprobe
->flags
& UPROBE_SKIP_SSTEP
&& can_skip_sstep(uprobe
, regs
))
1509 utask
->state
= UTASK_SSTEP
;
1510 if (!pre_ssout(uprobe
, regs
, bp_vaddr
)) {
1511 user_enable_single_step(current
);
1517 utask
->active_uprobe
= NULL
;
1518 utask
->state
= UTASK_RUNNING
;
1521 if (!(uprobe
->flags
& UPROBE_SKIP_SSTEP
))
1524 * cannot singlestep; cannot skip instruction;
1525 * re-execute the instruction.
1527 instruction_pointer_set(regs
, bp_vaddr
);
1534 * Perform required fix-ups and disable singlestep.
1535 * Allow pending signals to take effect.
1537 static void handle_singlestep(struct uprobe_task
*utask
, struct pt_regs
*regs
)
1539 struct uprobe
*uprobe
;
1541 uprobe
= utask
->active_uprobe
;
1542 if (utask
->state
== UTASK_SSTEP_ACK
)
1543 arch_uprobe_post_xol(&uprobe
->arch
, regs
);
1544 else if (utask
->state
== UTASK_SSTEP_TRAPPED
)
1545 arch_uprobe_abort_xol(&uprobe
->arch
, regs
);
1550 utask
->active_uprobe
= NULL
;
1551 utask
->state
= UTASK_RUNNING
;
1552 user_disable_single_step(current
);
1553 xol_free_insn_slot(current
);
1555 spin_lock_irq(¤t
->sighand
->siglock
);
1556 recalc_sigpending(); /* see uprobe_deny_signal() */
1557 spin_unlock_irq(¤t
->sighand
->siglock
);
1561 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on
1562 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1563 * allows the thread to return from interrupt.
1565 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1566 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1569 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1570 * uprobe_notify_resume().
1572 void uprobe_notify_resume(struct pt_regs
*regs
)
1574 struct uprobe_task
*utask
;
1576 utask
= current
->utask
;
1577 if (!utask
|| utask
->state
== UTASK_BP_HIT
)
1580 handle_singlestep(utask
, regs
);
1584 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1585 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1587 int uprobe_pre_sstep_notifier(struct pt_regs
*regs
)
1589 struct uprobe_task
*utask
;
1591 if (!current
->mm
|| !atomic_read(¤t
->mm
->uprobes_state
.count
))
1592 /* task is currently not uprobed */
1595 utask
= current
->utask
;
1597 utask
->state
= UTASK_BP_HIT
;
1599 set_thread_flag(TIF_UPROBE
);
1605 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1606 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1608 int uprobe_post_sstep_notifier(struct pt_regs
*regs
)
1610 struct uprobe_task
*utask
= current
->utask
;
1612 if (!current
->mm
|| !utask
|| !utask
->active_uprobe
)
1613 /* task is currently not uprobed */
1616 utask
->state
= UTASK_SSTEP_ACK
;
1617 set_thread_flag(TIF_UPROBE
);
1621 static struct notifier_block uprobe_exception_nb
= {
1622 .notifier_call
= arch_uprobe_exception_notify
,
1623 .priority
= INT_MAX
-1, /* notified after kprobes, kgdb */
1626 static int __init
init_uprobes(void)
1630 for (i
= 0; i
< UPROBES_HASH_SZ
; i
++) {
1631 mutex_init(&uprobes_mutex
[i
]);
1632 mutex_init(&uprobes_mmap_mutex
[i
]);
1635 return register_die_notifier(&uprobe_exception_nb
);
1637 module_init(init_uprobes
);
1639 static void __exit
exit_uprobes(void)
1642 module_exit(exit_uprobes
);