2 * User-space Probes (UProbes)
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright (C) IBM Corporation, 2008-2012
22 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
25 #include <linux/kernel.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h> /* read_mapping_page */
28 #include <linux/slab.h>
29 #include <linux/sched.h>
30 #include <linux/export.h>
31 #include <linux/rmap.h> /* anon_vma_prepare */
32 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
33 #include <linux/swap.h> /* try_to_free_swap */
34 #include <linux/ptrace.h> /* user_enable_single_step */
35 #include <linux/kdebug.h> /* notifier mechanism */
36 #include "../../mm/internal.h" /* munlock_vma_page */
37 #include <linux/percpu-rwsem.h>
38 #include <linux/task_work.h>
40 #include <linux/uprobes.h>
42 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
43 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
45 static struct rb_root uprobes_tree = RB_ROOT;
47 * allows us to skip the uprobe_mmap if there are no uprobe events active
48 * at this time. Probably a fine grained per inode count is better?
50 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
52 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
54 #define UPROBES_HASH_SZ 13
55 /* serialize uprobe->pending_list */
56 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
57 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
59 static struct percpu_rw_semaphore dup_mmap_sem;
61 /* Have a copy of original instruction */
62 #define UPROBE_COPY_INSN 0
63 /* Can skip singlestep */
64 #define UPROBE_SKIP_SSTEP 1
67 struct rb_node rb_node; /* node in the rb tree */
69 struct rw_semaphore register_rwsem;
70 struct rw_semaphore consumer_rwsem;
71 struct list_head pending_list;
72 struct uprobe_consumer *consumers;
73 struct inode *inode; /* Also hold a ref to inode */
76 struct arch_uprobe arch;
79 struct return_instance {
80 struct uprobe *uprobe;
82 unsigned long orig_ret_vaddr; /* original return address */
83 bool chained; /* true, if instance is nested */
85 struct return_instance *next; /* keep as stack */
89 * On a breakpoint hit, thread contests for a slot. It frees the
90 * slot after singlestep. Currently a fixed number of slots are
94 wait_queue_head_t wq; /* if all slots are busy */
95 atomic_t slot_count; /* number of in-use slots */
96 unsigned long *bitmap; /* 0 = free slot */
100 * We keep the vma's vm_start rather than a pointer to the vma
101 * itself. The probed process or a naughty kernel module could make
102 * the vma go away, and we must handle that reasonably gracefully.
104 unsigned long vaddr; /* Page(s) of instruction slots */
108 * valid_vma: Verify if the specified vma is an executable vma
109 * Relax restrictions while unregistering: vm_flags might have
110 * changed after breakpoint was inserted.
111 * - is_register: indicates if we are in register context.
112 * - Return 1 if the specified virtual address is in an
115 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
117 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
122 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
125 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
127 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
130 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
132 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
136 * __replace_page - replace page in vma by new page.
137 * based on replace_page in mm/ksm.c
139 * @vma: vma that holds the pte pointing to page
140 * @addr: address the old @page is mapped at
141 * @page: the cowed page we are replacing by kpage
142 * @kpage: the modified page we replace page by
144 * Returns 0 on success, -EFAULT on failure.
146 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
147 struct page *page, struct page *kpage)
149 struct mm_struct *mm = vma->vm_mm;
153 /* For mmu_notifiers */
154 const unsigned long mmun_start = addr;
155 const unsigned long mmun_end = addr + PAGE_SIZE;
157 /* For try_to_free_swap() and munlock_vma_page() below */
160 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
162 ptep = page_check_address(page, mm, addr, &ptl, 0);
167 page_add_new_anon_rmap(kpage, vma, addr);
169 if (!PageAnon(page)) {
170 dec_mm_counter(mm, MM_FILEPAGES);
171 inc_mm_counter(mm, MM_ANONPAGES);
174 flush_cache_page(vma, addr, pte_pfn(*ptep));
175 ptep_clear_flush(vma, addr, ptep);
176 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
178 page_remove_rmap(page);
179 if (!page_mapped(page))
180 try_to_free_swap(page);
181 pte_unmap_unlock(ptep, ptl);
183 if (vma->vm_flags & VM_LOCKED)
184 munlock_vma_page(page);
189 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
195 * is_swbp_insn - check if instruction is breakpoint instruction.
196 * @insn: instruction to be checked.
197 * Default implementation of is_swbp_insn
198 * Returns true if @insn is a breakpoint instruction.
200 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
202 return *insn == UPROBE_SWBP_INSN;
206 * is_trap_insn - check if instruction is breakpoint instruction.
207 * @insn: instruction to be checked.
208 * Default implementation of is_trap_insn
209 * Returns true if @insn is a breakpoint instruction.
211 * This function is needed for the case where an architecture has multiple
212 * trap instructions (like powerpc).
214 bool __weak is_trap_insn(uprobe_opcode_t *insn)
216 return is_swbp_insn(insn);
219 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
221 void *kaddr = kmap_atomic(page);
222 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
223 kunmap_atomic(kaddr);
226 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
228 void *kaddr = kmap_atomic(page);
229 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
230 kunmap_atomic(kaddr);
233 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
235 uprobe_opcode_t old_opcode;
239 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
240 * We do not check if it is any other 'trap variant' which could
241 * be conditional trap instruction such as the one powerpc supports.
243 * The logic is that we do not care if the underlying instruction
244 * is a trap variant; uprobes always wins over any other (gdb)
247 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
248 is_swbp = is_swbp_insn(&old_opcode);
250 if (is_swbp_insn(new_opcode)) {
251 if (is_swbp) /* register: already installed? */
254 if (!is_swbp) /* unregister: was it changed by us? */
263 * Expect the breakpoint instruction to be the smallest size instruction for
264 * the architecture. If an arch has variable length instruction and the
265 * breakpoint instruction is not of the smallest length instruction
266 * supported by that architecture then we need to modify is_trap_at_addr and
267 * uprobe_write_opcode accordingly. This would never be a problem for archs
268 * that have fixed length instructions.
272 * uprobe_write_opcode - write the opcode at a given virtual address.
273 * @mm: the probed process address space.
274 * @vaddr: the virtual address to store the opcode.
275 * @opcode: opcode to be written at @vaddr.
277 * Called with mm->mmap_sem held (for read and with a reference to
280 * For mm @mm, write the opcode at @vaddr.
281 * Return 0 (success) or a negative errno.
283 int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr,
284 uprobe_opcode_t opcode)
286 struct page *old_page, *new_page;
287 struct vm_area_struct *vma;
291 /* Read the page with vaddr into memory */
292 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
296 ret = verify_opcode(old_page, vaddr, &opcode);
301 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
305 __SetPageUptodate(new_page);
307 copy_highpage(new_page, old_page);
308 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
310 ret = anon_vma_prepare(vma);
314 ret = __replace_page(vma, vaddr, old_page, new_page);
317 page_cache_release(new_page);
321 if (unlikely(ret == -EAGAIN))
327 * set_swbp - store breakpoint at a given address.
328 * @auprobe: arch specific probepoint information.
329 * @mm: the probed process address space.
330 * @vaddr: the virtual address to insert the opcode.
332 * For mm @mm, store the breakpoint instruction at @vaddr.
333 * Return 0 (success) or a negative errno.
335 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
337 return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
341 * set_orig_insn - Restore the original instruction.
342 * @mm: the probed process address space.
343 * @auprobe: arch specific probepoint information.
344 * @vaddr: the virtual address to insert the opcode.
346 * For mm @mm, restore the original opcode (opcode) at @vaddr.
347 * Return 0 (success) or a negative errno.
350 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
352 return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn);
355 static int match_uprobe(struct uprobe *l, struct uprobe *r)
357 if (l->inode < r->inode)
360 if (l->inode > r->inode)
363 if (l->offset < r->offset)
366 if (l->offset > r->offset)
372 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
374 struct uprobe u = { .inode = inode, .offset = offset };
375 struct rb_node *n = uprobes_tree.rb_node;
376 struct uprobe *uprobe;
380 uprobe = rb_entry(n, struct uprobe, rb_node);
381 match = match_uprobe(&u, uprobe);
383 atomic_inc(&uprobe->ref);
396 * Find a uprobe corresponding to a given inode:offset
397 * Acquires uprobes_treelock
399 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
401 struct uprobe *uprobe;
403 spin_lock(&uprobes_treelock);
404 uprobe = __find_uprobe(inode, offset);
405 spin_unlock(&uprobes_treelock);
410 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
412 struct rb_node **p = &uprobes_tree.rb_node;
413 struct rb_node *parent = NULL;
419 u = rb_entry(parent, struct uprobe, rb_node);
420 match = match_uprobe(uprobe, u);
427 p = &parent->rb_left;
429 p = &parent->rb_right;
434 rb_link_node(&uprobe->rb_node, parent, p);
435 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
436 /* get access + creation ref */
437 atomic_set(&uprobe->ref, 2);
443 * Acquire uprobes_treelock.
444 * Matching uprobe already exists in rbtree;
445 * increment (access refcount) and return the matching uprobe.
447 * No matching uprobe; insert the uprobe in rb_tree;
448 * get a double refcount (access + creation) and return NULL.
450 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
454 spin_lock(&uprobes_treelock);
455 u = __insert_uprobe(uprobe);
456 spin_unlock(&uprobes_treelock);
461 static void put_uprobe(struct uprobe *uprobe)
463 if (atomic_dec_and_test(&uprobe->ref))
467 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
469 struct uprobe *uprobe, *cur_uprobe;
471 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
475 uprobe->inode = igrab(inode);
476 uprobe->offset = offset;
477 init_rwsem(&uprobe->register_rwsem);
478 init_rwsem(&uprobe->consumer_rwsem);
479 /* For now assume that the instruction need not be single-stepped */
480 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
482 /* add to uprobes_tree, sorted on inode:offset */
483 cur_uprobe = insert_uprobe(uprobe);
485 /* a uprobe exists for this inode:offset combination */
495 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
497 down_write(&uprobe->consumer_rwsem);
498 uc->next = uprobe->consumers;
499 uprobe->consumers = uc;
500 up_write(&uprobe->consumer_rwsem);
504 * For uprobe @uprobe, delete the consumer @uc.
505 * Return true if the @uc is deleted successfully
508 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
510 struct uprobe_consumer **con;
513 down_write(&uprobe->consumer_rwsem);
514 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
521 up_write(&uprobe->consumer_rwsem);
526 static int __copy_insn(struct address_space *mapping, struct file *filp,
527 void *insn, int nbytes, loff_t offset)
531 if (!mapping->a_ops->readpage)
534 * Ensure that the page that has the original instruction is
535 * populated and in page-cache.
537 page = read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT, filp);
539 return PTR_ERR(page);
541 copy_from_page(page, offset, insn, nbytes);
542 page_cache_release(page);
547 static int copy_insn(struct uprobe *uprobe, struct file *filp)
549 struct address_space *mapping = uprobe->inode->i_mapping;
550 loff_t offs = uprobe->offset;
551 void *insn = &uprobe->arch.insn;
552 int size = sizeof(uprobe->arch.insn);
555 /* Copy only available bytes, -EIO if nothing was read */
557 if (offs >= i_size_read(uprobe->inode))
560 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
561 err = __copy_insn(mapping, filp, insn, len, offs);
573 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
574 struct mm_struct *mm, unsigned long vaddr)
578 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
581 /* TODO: move this into _register, until then we abuse this sem. */
582 down_write(&uprobe->consumer_rwsem);
583 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
586 ret = copy_insn(uprobe, file);
591 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
594 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
598 /* uprobe_write_opcode() assumes we don't cross page boundary */
599 BUG_ON((uprobe->offset & ~PAGE_MASK) +
600 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
602 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
603 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
606 up_write(&uprobe->consumer_rwsem);
611 static inline bool consumer_filter(struct uprobe_consumer *uc,
612 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
614 return !uc->filter || uc->filter(uc, ctx, mm);
617 static bool filter_chain(struct uprobe *uprobe,
618 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
620 struct uprobe_consumer *uc;
623 down_read(&uprobe->consumer_rwsem);
624 for (uc = uprobe->consumers; uc; uc = uc->next) {
625 ret = consumer_filter(uc, ctx, mm);
629 up_read(&uprobe->consumer_rwsem);
635 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
636 struct vm_area_struct *vma, unsigned long vaddr)
641 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
646 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
647 * the task can hit this breakpoint right after __replace_page().
649 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
651 set_bit(MMF_HAS_UPROBES, &mm->flags);
653 ret = set_swbp(&uprobe->arch, mm, vaddr);
655 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
656 else if (first_uprobe)
657 clear_bit(MMF_HAS_UPROBES, &mm->flags);
663 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
665 set_bit(MMF_RECALC_UPROBES, &mm->flags);
666 return set_orig_insn(&uprobe->arch, mm, vaddr);
669 static inline bool uprobe_is_active(struct uprobe *uprobe)
671 return !RB_EMPTY_NODE(&uprobe->rb_node);
674 * There could be threads that have already hit the breakpoint. They
675 * will recheck the current insn and restart if find_uprobe() fails.
676 * See find_active_uprobe().
678 static void delete_uprobe(struct uprobe *uprobe)
680 if (WARN_ON(!uprobe_is_active(uprobe)))
683 spin_lock(&uprobes_treelock);
684 rb_erase(&uprobe->rb_node, &uprobes_tree);
685 spin_unlock(&uprobes_treelock);
686 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
692 struct map_info *next;
693 struct mm_struct *mm;
697 static inline struct map_info *free_map_info(struct map_info *info)
699 struct map_info *next = info->next;
704 static struct map_info *
705 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
707 unsigned long pgoff = offset >> PAGE_SHIFT;
708 struct vm_area_struct *vma;
709 struct map_info *curr = NULL;
710 struct map_info *prev = NULL;
711 struct map_info *info;
715 mutex_lock(&mapping->i_mmap_mutex);
716 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
717 if (!valid_vma(vma, is_register))
720 if (!prev && !more) {
722 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
723 * reclaim. This is optimistic, no harm done if it fails.
725 prev = kmalloc(sizeof(struct map_info),
726 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
735 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
743 info->mm = vma->vm_mm;
744 info->vaddr = offset_to_vaddr(vma, offset);
746 mutex_unlock(&mapping->i_mmap_mutex);
758 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
760 curr = ERR_PTR(-ENOMEM);
770 prev = free_map_info(prev);
775 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
777 bool is_register = !!new;
778 struct map_info *info;
781 percpu_down_write(&dup_mmap_sem);
782 info = build_map_info(uprobe->inode->i_mapping,
783 uprobe->offset, is_register);
790 struct mm_struct *mm = info->mm;
791 struct vm_area_struct *vma;
793 if (err && is_register)
796 down_write(&mm->mmap_sem);
797 vma = find_vma(mm, info->vaddr);
798 if (!vma || !valid_vma(vma, is_register) ||
799 file_inode(vma->vm_file) != uprobe->inode)
802 if (vma->vm_start > info->vaddr ||
803 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
807 /* consult only the "caller", new consumer. */
808 if (consumer_filter(new,
809 UPROBE_FILTER_REGISTER, mm))
810 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
811 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
812 if (!filter_chain(uprobe,
813 UPROBE_FILTER_UNREGISTER, mm))
814 err |= remove_breakpoint(uprobe, mm, info->vaddr);
818 up_write(&mm->mmap_sem);
821 info = free_map_info(info);
824 percpu_up_write(&dup_mmap_sem);
828 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
830 consumer_add(uprobe, uc);
831 return register_for_each_vma(uprobe, uc);
834 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
838 if (!consumer_del(uprobe, uc)) /* WARN? */
841 err = register_for_each_vma(uprobe, NULL);
842 /* TODO : cant unregister? schedule a worker thread */
843 if (!uprobe->consumers && !err)
844 delete_uprobe(uprobe);
848 * uprobe_register - register a probe
849 * @inode: the file in which the probe has to be placed.
850 * @offset: offset from the start of the file.
851 * @uc: information on howto handle the probe..
853 * Apart from the access refcount, uprobe_register() takes a creation
854 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
855 * inserted into the rbtree (i.e first consumer for a @inode:@offset
856 * tuple). Creation refcount stops uprobe_unregister from freeing the
857 * @uprobe even before the register operation is complete. Creation
858 * refcount is released when the last @uc for the @uprobe
861 * Return errno if it cannot successully install probes
862 * else return 0 (success)
864 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
866 struct uprobe *uprobe;
869 /* Uprobe must have at least one set consumer */
870 if (!uc->handler && !uc->ret_handler)
873 /* Racy, just to catch the obvious mistakes */
874 if (offset > i_size_read(inode))
878 uprobe = alloc_uprobe(inode, offset);
882 * We can race with uprobe_unregister()->delete_uprobe().
883 * Check uprobe_is_active() and retry if it is false.
885 down_write(&uprobe->register_rwsem);
887 if (likely(uprobe_is_active(uprobe))) {
888 ret = __uprobe_register(uprobe, uc);
890 __uprobe_unregister(uprobe, uc);
892 up_write(&uprobe->register_rwsem);
895 if (unlikely(ret == -EAGAIN))
899 EXPORT_SYMBOL_GPL(uprobe_register);
902 * uprobe_apply - unregister a already registered probe.
903 * @inode: the file in which the probe has to be removed.
904 * @offset: offset from the start of the file.
905 * @uc: consumer which wants to add more or remove some breakpoints
906 * @add: add or remove the breakpoints
908 int uprobe_apply(struct inode *inode, loff_t offset,
909 struct uprobe_consumer *uc, bool add)
911 struct uprobe *uprobe;
912 struct uprobe_consumer *con;
915 uprobe = find_uprobe(inode, offset);
919 down_write(&uprobe->register_rwsem);
920 for (con = uprobe->consumers; con && con != uc ; con = con->next)
923 ret = register_for_each_vma(uprobe, add ? uc : NULL);
924 up_write(&uprobe->register_rwsem);
931 * uprobe_unregister - unregister a already registered probe.
932 * @inode: the file in which the probe has to be removed.
933 * @offset: offset from the start of the file.
934 * @uc: identify which probe if multiple probes are colocated.
936 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
938 struct uprobe *uprobe;
940 uprobe = find_uprobe(inode, offset);
944 down_write(&uprobe->register_rwsem);
945 __uprobe_unregister(uprobe, uc);
946 up_write(&uprobe->register_rwsem);
949 EXPORT_SYMBOL_GPL(uprobe_unregister);
951 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
953 struct vm_area_struct *vma;
956 down_read(&mm->mmap_sem);
957 for (vma = mm->mmap; vma; vma = vma->vm_next) {
961 if (!valid_vma(vma, false) ||
962 file_inode(vma->vm_file) != uprobe->inode)
965 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
966 if (uprobe->offset < offset ||
967 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
970 vaddr = offset_to_vaddr(vma, uprobe->offset);
971 err |= remove_breakpoint(uprobe, mm, vaddr);
973 up_read(&mm->mmap_sem);
978 static struct rb_node *
979 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
981 struct rb_node *n = uprobes_tree.rb_node;
984 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
986 if (inode < u->inode) {
988 } else if (inode > u->inode) {
993 else if (min > u->offset)
1004 * For a given range in vma, build a list of probes that need to be inserted.
1006 static void build_probe_list(struct inode *inode,
1007 struct vm_area_struct *vma,
1008 unsigned long start, unsigned long end,
1009 struct list_head *head)
1012 struct rb_node *n, *t;
1015 INIT_LIST_HEAD(head);
1016 min = vaddr_to_offset(vma, start);
1017 max = min + (end - start) - 1;
1019 spin_lock(&uprobes_treelock);
1020 n = find_node_in_range(inode, min, max);
1022 for (t = n; t; t = rb_prev(t)) {
1023 u = rb_entry(t, struct uprobe, rb_node);
1024 if (u->inode != inode || u->offset < min)
1026 list_add(&u->pending_list, head);
1027 atomic_inc(&u->ref);
1029 for (t = n; (t = rb_next(t)); ) {
1030 u = rb_entry(t, struct uprobe, rb_node);
1031 if (u->inode != inode || u->offset > max)
1033 list_add(&u->pending_list, head);
1034 atomic_inc(&u->ref);
1037 spin_unlock(&uprobes_treelock);
1041 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1043 * Currently we ignore all errors and always return 0, the callers
1044 * can't handle the failure anyway.
1046 int uprobe_mmap(struct vm_area_struct *vma)
1048 struct list_head tmp_list;
1049 struct uprobe *uprobe, *u;
1050 struct inode *inode;
1052 if (no_uprobe_events() || !valid_vma(vma, true))
1055 inode = file_inode(vma->vm_file);
1059 mutex_lock(uprobes_mmap_hash(inode));
1060 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1062 * We can race with uprobe_unregister(), this uprobe can be already
1063 * removed. But in this case filter_chain() must return false, all
1064 * consumers have gone away.
1066 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1067 if (!fatal_signal_pending(current) &&
1068 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1069 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1070 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1074 mutex_unlock(uprobes_mmap_hash(inode));
1080 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1083 struct inode *inode;
1086 inode = file_inode(vma->vm_file);
1088 min = vaddr_to_offset(vma, start);
1089 max = min + (end - start) - 1;
1091 spin_lock(&uprobes_treelock);
1092 n = find_node_in_range(inode, min, max);
1093 spin_unlock(&uprobes_treelock);
1099 * Called in context of a munmap of a vma.
1101 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1103 if (no_uprobe_events() || !valid_vma(vma, false))
1106 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1109 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1110 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1113 if (vma_has_uprobes(vma, start, end))
1114 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1117 /* Slot allocation for XOL */
1118 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1120 int ret = -EALREADY;
1122 down_write(&mm->mmap_sem);
1123 if (mm->uprobes_state.xol_area)
1127 /* Try to map as high as possible, this is only a hint. */
1128 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1130 if (area->vaddr & ~PAGE_MASK) {
1136 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1137 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1141 smp_wmb(); /* pairs with get_xol_area() */
1142 mm->uprobes_state.xol_area = area;
1144 up_write(&mm->mmap_sem);
1149 static struct xol_area *__create_xol_area(unsigned long vaddr)
1151 struct mm_struct *mm = current->mm;
1152 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1153 struct xol_area *area;
1155 area = kmalloc(sizeof(*area), GFP_KERNEL);
1156 if (unlikely(!area))
1159 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1163 area->page = alloc_page(GFP_HIGHUSER);
1167 area->vaddr = vaddr;
1168 init_waitqueue_head(&area->wq);
1169 /* Reserve the 1st slot for get_trampoline_vaddr() */
1170 set_bit(0, area->bitmap);
1171 atomic_set(&area->slot_count, 1);
1172 copy_to_page(area->page, 0, &insn, UPROBE_SWBP_INSN_SIZE);
1174 if (!xol_add_vma(mm, area))
1177 __free_page(area->page);
1179 kfree(area->bitmap);
1187 * get_xol_area - Allocate process's xol_area if necessary.
1188 * This area will be used for storing instructions for execution out of line.
1190 * Returns the allocated area or NULL.
1192 static struct xol_area *get_xol_area(void)
1194 struct mm_struct *mm = current->mm;
1195 struct xol_area *area;
1197 if (!mm->uprobes_state.xol_area)
1198 __create_xol_area(0);
1200 area = mm->uprobes_state.xol_area;
1201 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1206 * uprobe_clear_state - Free the area allocated for slots.
1208 void uprobe_clear_state(struct mm_struct *mm)
1210 struct xol_area *area = mm->uprobes_state.xol_area;
1215 put_page(area->page);
1216 kfree(area->bitmap);
1220 void uprobe_start_dup_mmap(void)
1222 percpu_down_read(&dup_mmap_sem);
1225 void uprobe_end_dup_mmap(void)
1227 percpu_up_read(&dup_mmap_sem);
1230 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1232 newmm->uprobes_state.xol_area = NULL;
1234 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1235 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1236 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1237 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1242 * - search for a free slot.
1244 static unsigned long xol_take_insn_slot(struct xol_area *area)
1246 unsigned long slot_addr;
1250 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1251 if (slot_nr < UINSNS_PER_PAGE) {
1252 if (!test_and_set_bit(slot_nr, area->bitmap))
1255 slot_nr = UINSNS_PER_PAGE;
1258 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1259 } while (slot_nr >= UINSNS_PER_PAGE);
1261 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1262 atomic_inc(&area->slot_count);
1268 * xol_get_insn_slot - allocate a slot for xol.
1269 * Returns the allocated slot address or 0.
1271 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1273 struct xol_area *area;
1274 unsigned long xol_vaddr;
1276 area = get_xol_area();
1280 xol_vaddr = xol_take_insn_slot(area);
1281 if (unlikely(!xol_vaddr))
1284 /* Initialize the slot */
1285 copy_to_page(area->page, xol_vaddr,
1286 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1288 * We probably need flush_icache_user_range() but it needs vma.
1289 * This should work on supported architectures too.
1291 flush_dcache_page(area->page);
1297 * xol_free_insn_slot - If slot was earlier allocated by
1298 * @xol_get_insn_slot(), make the slot available for
1299 * subsequent requests.
1301 static void xol_free_insn_slot(struct task_struct *tsk)
1303 struct xol_area *area;
1304 unsigned long vma_end;
1305 unsigned long slot_addr;
1307 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1310 slot_addr = tsk->utask->xol_vaddr;
1311 if (unlikely(!slot_addr))
1314 area = tsk->mm->uprobes_state.xol_area;
1315 vma_end = area->vaddr + PAGE_SIZE;
1316 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1317 unsigned long offset;
1320 offset = slot_addr - area->vaddr;
1321 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1322 if (slot_nr >= UINSNS_PER_PAGE)
1325 clear_bit(slot_nr, area->bitmap);
1326 atomic_dec(&area->slot_count);
1327 if (waitqueue_active(&area->wq))
1330 tsk->utask->xol_vaddr = 0;
1335 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1336 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1338 * Return the address of the breakpoint instruction.
1340 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1342 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1346 * Called with no locks held.
1347 * Called in context of a exiting or a exec-ing thread.
1349 void uprobe_free_utask(struct task_struct *t)
1351 struct uprobe_task *utask = t->utask;
1352 struct return_instance *ri, *tmp;
1357 if (utask->active_uprobe)
1358 put_uprobe(utask->active_uprobe);
1360 ri = utask->return_instances;
1365 put_uprobe(tmp->uprobe);
1369 xol_free_insn_slot(t);
1375 * Allocate a uprobe_task object for the task if if necessary.
1376 * Called when the thread hits a breakpoint.
1379 * - pointer to new uprobe_task on success
1382 static struct uprobe_task *get_utask(void)
1384 if (!current->utask)
1385 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1386 return current->utask;
1389 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1391 struct uprobe_task *n_utask;
1392 struct return_instance **p, *o, *n;
1394 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1399 p = &n_utask->return_instances;
1400 for (o = o_utask->return_instances; o; o = o->next) {
1401 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1406 atomic_inc(&n->uprobe->ref);
1417 static void uprobe_warn(struct task_struct *t, const char *msg)
1419 pr_warn("uprobe: %s:%d failed to %s\n",
1420 current->comm, current->pid, msg);
1423 static void dup_xol_work(struct callback_head *work)
1425 if (current->flags & PF_EXITING)
1428 if (!__create_xol_area(current->utask->dup_xol_addr))
1429 uprobe_warn(current, "dup xol area");
1433 * Called in context of a new clone/fork from copy_process.
1435 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1437 struct uprobe_task *utask = current->utask;
1438 struct mm_struct *mm = current->mm;
1439 struct xol_area *area;
1443 if (!utask || !utask->return_instances)
1446 if (mm == t->mm && !(flags & CLONE_VFORK))
1449 if (dup_utask(t, utask))
1450 return uprobe_warn(t, "dup ret instances");
1452 /* The task can fork() after dup_xol_work() fails */
1453 area = mm->uprobes_state.xol_area;
1455 return uprobe_warn(t, "dup xol area");
1460 t->utask->dup_xol_addr = area->vaddr;
1461 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1462 task_work_add(t, &t->utask->dup_xol_work, true);
1466 * Current area->vaddr notion assume the trampoline address is always
1467 * equal area->vaddr.
1469 * Returns -1 in case the xol_area is not allocated.
1471 static unsigned long get_trampoline_vaddr(void)
1473 struct xol_area *area;
1474 unsigned long trampoline_vaddr = -1;
1476 area = current->mm->uprobes_state.xol_area;
1477 smp_read_barrier_depends();
1479 trampoline_vaddr = area->vaddr;
1481 return trampoline_vaddr;
1484 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1486 struct return_instance *ri;
1487 struct uprobe_task *utask;
1488 unsigned long orig_ret_vaddr, trampoline_vaddr;
1489 bool chained = false;
1491 if (!get_xol_area())
1494 utask = get_utask();
1498 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1499 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1500 " nestedness limit pid/tgid=%d/%d\n",
1501 current->pid, current->tgid);
1505 ri = kzalloc(sizeof(struct return_instance), GFP_KERNEL);
1509 trampoline_vaddr = get_trampoline_vaddr();
1510 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1511 if (orig_ret_vaddr == -1)
1515 * We don't want to keep trampoline address in stack, rather keep the
1516 * original return address of first caller thru all the consequent
1517 * instances. This also makes breakpoint unwrapping easier.
1519 if (orig_ret_vaddr == trampoline_vaddr) {
1520 if (!utask->return_instances) {
1522 * This situation is not possible. Likely we have an
1523 * attack from user-space.
1525 pr_warn("uprobe: unable to set uretprobe pid/tgid=%d/%d\n",
1526 current->pid, current->tgid);
1531 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1534 atomic_inc(&uprobe->ref);
1535 ri->uprobe = uprobe;
1536 ri->func = instruction_pointer(regs);
1537 ri->orig_ret_vaddr = orig_ret_vaddr;
1538 ri->chained = chained;
1542 /* add instance to the stack */
1543 ri->next = utask->return_instances;
1544 utask->return_instances = ri;
1552 /* Prepare to single-step probed instruction out of line. */
1554 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1556 struct uprobe_task *utask;
1557 unsigned long xol_vaddr;
1560 utask = get_utask();
1564 xol_vaddr = xol_get_insn_slot(uprobe);
1568 utask->xol_vaddr = xol_vaddr;
1569 utask->vaddr = bp_vaddr;
1571 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1572 if (unlikely(err)) {
1573 xol_free_insn_slot(current);
1577 utask->active_uprobe = uprobe;
1578 utask->state = UTASK_SSTEP;
1583 * If we are singlestepping, then ensure this thread is not connected to
1584 * non-fatal signals until completion of singlestep. When xol insn itself
1585 * triggers the signal, restart the original insn even if the task is
1586 * already SIGKILL'ed (since coredump should report the correct ip). This
1587 * is even more important if the task has a handler for SIGSEGV/etc, The
1588 * _same_ instruction should be repeated again after return from the signal
1589 * handler, and SSTEP can never finish in this case.
1591 bool uprobe_deny_signal(void)
1593 struct task_struct *t = current;
1594 struct uprobe_task *utask = t->utask;
1596 if (likely(!utask || !utask->active_uprobe))
1599 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1601 if (signal_pending(t)) {
1602 spin_lock_irq(&t->sighand->siglock);
1603 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1604 spin_unlock_irq(&t->sighand->siglock);
1606 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1607 utask->state = UTASK_SSTEP_TRAPPED;
1608 set_tsk_thread_flag(t, TIF_UPROBE);
1609 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1617 * Avoid singlestepping the original instruction if the original instruction
1618 * is a NOP or can be emulated.
1620 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1622 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1623 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1625 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1630 static void mmf_recalc_uprobes(struct mm_struct *mm)
1632 struct vm_area_struct *vma;
1634 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1635 if (!valid_vma(vma, false))
1638 * This is not strictly accurate, we can race with
1639 * uprobe_unregister() and see the already removed
1640 * uprobe if delete_uprobe() was not yet called.
1641 * Or this uprobe can be filtered out.
1643 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1647 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1650 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1653 uprobe_opcode_t opcode;
1656 pagefault_disable();
1657 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1661 if (likely(result == 0))
1664 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1668 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1671 /* This needs to return true for any variant of the trap insn */
1672 return is_trap_insn(&opcode);
1675 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1677 struct mm_struct *mm = current->mm;
1678 struct uprobe *uprobe = NULL;
1679 struct vm_area_struct *vma;
1681 down_read(&mm->mmap_sem);
1682 vma = find_vma(mm, bp_vaddr);
1683 if (vma && vma->vm_start <= bp_vaddr) {
1684 if (valid_vma(vma, false)) {
1685 struct inode *inode = file_inode(vma->vm_file);
1686 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1688 uprobe = find_uprobe(inode, offset);
1692 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1697 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1698 mmf_recalc_uprobes(mm);
1699 up_read(&mm->mmap_sem);
1704 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1706 struct uprobe_consumer *uc;
1707 int remove = UPROBE_HANDLER_REMOVE;
1708 bool need_prep = false; /* prepare return uprobe, when needed */
1710 down_read(&uprobe->register_rwsem);
1711 for (uc = uprobe->consumers; uc; uc = uc->next) {
1715 rc = uc->handler(uc, regs);
1716 WARN(rc & ~UPROBE_HANDLER_MASK,
1717 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1720 if (uc->ret_handler)
1726 if (need_prep && !remove)
1727 prepare_uretprobe(uprobe, regs); /* put bp at return */
1729 if (remove && uprobe->consumers) {
1730 WARN_ON(!uprobe_is_active(uprobe));
1731 unapply_uprobe(uprobe, current->mm);
1733 up_read(&uprobe->register_rwsem);
1737 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1739 struct uprobe *uprobe = ri->uprobe;
1740 struct uprobe_consumer *uc;
1742 down_read(&uprobe->register_rwsem);
1743 for (uc = uprobe->consumers; uc; uc = uc->next) {
1744 if (uc->ret_handler)
1745 uc->ret_handler(uc, ri->func, regs);
1747 up_read(&uprobe->register_rwsem);
1750 static bool handle_trampoline(struct pt_regs *regs)
1752 struct uprobe_task *utask;
1753 struct return_instance *ri, *tmp;
1756 utask = current->utask;
1760 ri = utask->return_instances;
1765 * TODO: we should throw out return_instance's invalidated by
1766 * longjmp(), currently we assume that the probed function always
1769 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1772 handle_uretprobe_chain(ri, regs);
1774 chained = ri->chained;
1775 put_uprobe(ri->uprobe);
1787 utask->return_instances = ri;
1793 * Run handler and ask thread to singlestep.
1794 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1796 static void handle_swbp(struct pt_regs *regs)
1798 struct uprobe *uprobe;
1799 unsigned long bp_vaddr;
1800 int uninitialized_var(is_swbp);
1802 bp_vaddr = uprobe_get_swbp_addr(regs);
1803 if (bp_vaddr == get_trampoline_vaddr()) {
1804 if (handle_trampoline(regs))
1807 pr_warn("uprobe: unable to handle uretprobe pid/tgid=%d/%d\n",
1808 current->pid, current->tgid);
1811 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1814 /* No matching uprobe; signal SIGTRAP. */
1815 send_sig(SIGTRAP, current, 0);
1818 * Either we raced with uprobe_unregister() or we can't
1819 * access this memory. The latter is only possible if
1820 * another thread plays with our ->mm. In both cases
1821 * we can simply restart. If this vma was unmapped we
1822 * can pretend this insn was not executed yet and get
1823 * the (correct) SIGSEGV after restart.
1825 instruction_pointer_set(regs, bp_vaddr);
1830 /* change it in advance for ->handler() and restart */
1831 instruction_pointer_set(regs, bp_vaddr);
1834 * TODO: move copy_insn/etc into _register and remove this hack.
1835 * After we hit the bp, _unregister + _register can install the
1836 * new and not-yet-analyzed uprobe at the same address, restart.
1838 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1839 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1842 handler_chain(uprobe, regs);
1843 if (can_skip_sstep(uprobe, regs))
1846 if (!pre_ssout(uprobe, regs, bp_vaddr))
1849 /* can_skip_sstep() succeeded, or restart if can't singlestep */
1855 * Perform required fix-ups and disable singlestep.
1856 * Allow pending signals to take effect.
1858 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1860 struct uprobe *uprobe;
1862 uprobe = utask->active_uprobe;
1863 if (utask->state == UTASK_SSTEP_ACK)
1864 arch_uprobe_post_xol(&uprobe->arch, regs);
1865 else if (utask->state == UTASK_SSTEP_TRAPPED)
1866 arch_uprobe_abort_xol(&uprobe->arch, regs);
1871 utask->active_uprobe = NULL;
1872 utask->state = UTASK_RUNNING;
1873 xol_free_insn_slot(current);
1875 spin_lock_irq(¤t->sighand->siglock);
1876 recalc_sigpending(); /* see uprobe_deny_signal() */
1877 spin_unlock_irq(¤t->sighand->siglock);
1881 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1882 * allows the thread to return from interrupt. After that handle_swbp()
1883 * sets utask->active_uprobe.
1885 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1886 * and allows the thread to return from interrupt.
1888 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1889 * uprobe_notify_resume().
1891 void uprobe_notify_resume(struct pt_regs *regs)
1893 struct uprobe_task *utask;
1895 clear_thread_flag(TIF_UPROBE);
1897 utask = current->utask;
1898 if (utask && utask->active_uprobe)
1899 handle_singlestep(utask, regs);
1905 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1906 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1908 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1913 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
1914 (!current->utask || !current->utask->return_instances))
1917 set_thread_flag(TIF_UPROBE);
1922 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1923 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1925 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1927 struct uprobe_task *utask = current->utask;
1929 if (!current->mm || !utask || !utask->active_uprobe)
1930 /* task is currently not uprobed */
1933 utask->state = UTASK_SSTEP_ACK;
1934 set_thread_flag(TIF_UPROBE);
1938 static struct notifier_block uprobe_exception_nb = {
1939 .notifier_call = arch_uprobe_exception_notify,
1940 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1943 static int __init init_uprobes(void)
1947 for (i = 0; i < UPROBES_HASH_SZ; i++)
1948 mutex_init(&uprobes_mmap_mutex[i]);
1950 if (percpu_init_rwsem(&dup_mmap_sem))
1953 return register_die_notifier(&uprobe_exception_nb);
1955 __initcall(init_uprobes);