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 *old_page, struct page *new_page)
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(new_page, 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(old_page, mm, addr, &ptl, 0);
176 mem_cgroup_cancel_charge(new_page, memcg, false);
181 page_add_new_anon_rmap(new_page, vma, addr, false);
182 mem_cgroup_commit_charge(new_page, memcg, false, false);
183 lru_cache_add_active_or_unevictable(new_page, vma);
185 if (!PageAnon(old_page)) {
186 dec_mm_counter(mm, mm_counter_file(old_page));
187 inc_mm_counter(mm, MM_ANONPAGES);
190 flush_cache_page(vma, addr, pte_pfn(*ptep));
191 ptep_clear_flush_notify(vma, addr, ptep);
192 set_pte_at_notify(mm, addr, ptep, mk_pte(new_page, vma->vm_page_prot));
194 page_remove_rmap(old_page, false);
195 if (!page_mapped(old_page))
196 try_to_free_swap(old_page);
197 pte_unmap_unlock(ptep, ptl);
199 if (vma->vm_flags & VM_LOCKED)
200 munlock_vma_page(old_page);
205 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
206 unlock_page(old_page);
211 * is_swbp_insn - check if instruction is breakpoint instruction.
212 * @insn: instruction to be checked.
213 * Default implementation of is_swbp_insn
214 * Returns true if @insn is a breakpoint instruction.
216 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
218 return *insn == UPROBE_SWBP_INSN;
222 * is_trap_insn - check if instruction is breakpoint instruction.
223 * @insn: instruction to be checked.
224 * Default implementation of is_trap_insn
225 * Returns true if @insn is a breakpoint instruction.
227 * This function is needed for the case where an architecture has multiple
228 * trap instructions (like powerpc).
230 bool __weak is_trap_insn(uprobe_opcode_t *insn)
232 return is_swbp_insn(insn);
235 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
237 void *kaddr = kmap_atomic(page);
238 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
239 kunmap_atomic(kaddr);
242 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
244 void *kaddr = kmap_atomic(page);
245 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
246 kunmap_atomic(kaddr);
249 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
251 uprobe_opcode_t old_opcode;
255 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
256 * We do not check if it is any other 'trap variant' which could
257 * be conditional trap instruction such as the one powerpc supports.
259 * The logic is that we do not care if the underlying instruction
260 * is a trap variant; uprobes always wins over any other (gdb)
263 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
264 is_swbp = is_swbp_insn(&old_opcode);
266 if (is_swbp_insn(new_opcode)) {
267 if (is_swbp) /* register: already installed? */
270 if (!is_swbp) /* unregister: was it changed by us? */
279 * Expect the breakpoint instruction to be the smallest size instruction for
280 * the architecture. If an arch has variable length instruction and the
281 * breakpoint instruction is not of the smallest length instruction
282 * supported by that architecture then we need to modify is_trap_at_addr and
283 * uprobe_write_opcode accordingly. This would never be a problem for archs
284 * that have fixed length instructions.
286 * uprobe_write_opcode - write the opcode at a given virtual address.
287 * @mm: the probed process address space.
288 * @vaddr: the virtual address to store the opcode.
289 * @opcode: opcode to be written at @vaddr.
291 * Called with mm->mmap_sem held for write.
292 * Return 0 (success) or a negative errno.
294 int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr,
295 uprobe_opcode_t opcode)
297 struct page *old_page, *new_page;
298 struct vm_area_struct *vma;
302 /* Read the page with vaddr into memory */
303 ret = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &old_page,
308 ret = verify_opcode(old_page, vaddr, &opcode);
312 ret = anon_vma_prepare(vma);
317 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
321 __SetPageUptodate(new_page);
322 copy_highpage(new_page, old_page);
323 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
325 ret = __replace_page(vma, vaddr, old_page, new_page);
330 if (unlikely(ret == -EAGAIN))
336 * set_swbp - store breakpoint at a given address.
337 * @auprobe: arch specific probepoint information.
338 * @mm: the probed process address space.
339 * @vaddr: the virtual address to insert the opcode.
341 * For mm @mm, store the breakpoint instruction at @vaddr.
342 * Return 0 (success) or a negative errno.
344 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
346 return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
350 * set_orig_insn - Restore the original instruction.
351 * @mm: the probed process address space.
352 * @auprobe: arch specific probepoint information.
353 * @vaddr: the virtual address to insert the opcode.
355 * For mm @mm, restore the original opcode (opcode) at @vaddr.
356 * Return 0 (success) or a negative errno.
359 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
361 return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn);
364 static struct uprobe *get_uprobe(struct uprobe *uprobe)
366 atomic_inc(&uprobe->ref);
370 static void put_uprobe(struct uprobe *uprobe)
372 if (atomic_dec_and_test(&uprobe->ref))
376 static int match_uprobe(struct uprobe *l, struct uprobe *r)
378 if (l->inode < r->inode)
381 if (l->inode > r->inode)
384 if (l->offset < r->offset)
387 if (l->offset > r->offset)
393 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
395 struct uprobe u = { .inode = inode, .offset = offset };
396 struct rb_node *n = uprobes_tree.rb_node;
397 struct uprobe *uprobe;
401 uprobe = rb_entry(n, struct uprobe, rb_node);
402 match = match_uprobe(&u, uprobe);
404 return get_uprobe(uprobe);
415 * Find a uprobe corresponding to a given inode:offset
416 * Acquires uprobes_treelock
418 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
420 struct uprobe *uprobe;
422 spin_lock(&uprobes_treelock);
423 uprobe = __find_uprobe(inode, offset);
424 spin_unlock(&uprobes_treelock);
429 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
431 struct rb_node **p = &uprobes_tree.rb_node;
432 struct rb_node *parent = NULL;
438 u = rb_entry(parent, struct uprobe, rb_node);
439 match = match_uprobe(uprobe, u);
441 return get_uprobe(u);
444 p = &parent->rb_left;
446 p = &parent->rb_right;
451 rb_link_node(&uprobe->rb_node, parent, p);
452 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
453 /* get access + creation ref */
454 atomic_set(&uprobe->ref, 2);
460 * Acquire uprobes_treelock.
461 * Matching uprobe already exists in rbtree;
462 * increment (access refcount) and return the matching uprobe.
464 * No matching uprobe; insert the uprobe in rb_tree;
465 * get a double refcount (access + creation) and return NULL.
467 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
471 spin_lock(&uprobes_treelock);
472 u = __insert_uprobe(uprobe);
473 spin_unlock(&uprobes_treelock);
478 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
480 struct uprobe *uprobe, *cur_uprobe;
482 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
486 uprobe->inode = igrab(inode);
487 uprobe->offset = offset;
488 init_rwsem(&uprobe->register_rwsem);
489 init_rwsem(&uprobe->consumer_rwsem);
491 /* add to uprobes_tree, sorted on inode:offset */
492 cur_uprobe = insert_uprobe(uprobe);
493 /* a uprobe exists for this inode:offset combination */
503 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
505 down_write(&uprobe->consumer_rwsem);
506 uc->next = uprobe->consumers;
507 uprobe->consumers = uc;
508 up_write(&uprobe->consumer_rwsem);
512 * For uprobe @uprobe, delete the consumer @uc.
513 * Return true if the @uc is deleted successfully
516 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
518 struct uprobe_consumer **con;
521 down_write(&uprobe->consumer_rwsem);
522 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
529 up_write(&uprobe->consumer_rwsem);
534 static int __copy_insn(struct address_space *mapping, struct file *filp,
535 void *insn, int nbytes, loff_t offset)
539 * Ensure that the page that has the original instruction is populated
540 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
541 * see uprobe_register().
543 if (mapping->a_ops->readpage)
544 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
546 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
548 return PTR_ERR(page);
550 copy_from_page(page, offset, insn, nbytes);
556 static int copy_insn(struct uprobe *uprobe, struct file *filp)
558 struct address_space *mapping = uprobe->inode->i_mapping;
559 loff_t offs = uprobe->offset;
560 void *insn = &uprobe->arch.insn;
561 int size = sizeof(uprobe->arch.insn);
564 /* Copy only available bytes, -EIO if nothing was read */
566 if (offs >= i_size_read(uprobe->inode))
569 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
570 err = __copy_insn(mapping, filp, insn, len, offs);
582 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
583 struct mm_struct *mm, unsigned long vaddr)
587 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
590 /* TODO: move this into _register, until then we abuse this sem. */
591 down_write(&uprobe->consumer_rwsem);
592 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
595 ret = copy_insn(uprobe, file);
600 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
603 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
607 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
608 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
611 up_write(&uprobe->consumer_rwsem);
616 static inline bool consumer_filter(struct uprobe_consumer *uc,
617 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
619 return !uc->filter || uc->filter(uc, ctx, mm);
622 static bool filter_chain(struct uprobe *uprobe,
623 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
625 struct uprobe_consumer *uc;
628 down_read(&uprobe->consumer_rwsem);
629 for (uc = uprobe->consumers; uc; uc = uc->next) {
630 ret = consumer_filter(uc, ctx, mm);
634 up_read(&uprobe->consumer_rwsem);
640 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
641 struct vm_area_struct *vma, unsigned long vaddr)
646 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
651 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
652 * the task can hit this breakpoint right after __replace_page().
654 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
656 set_bit(MMF_HAS_UPROBES, &mm->flags);
658 ret = set_swbp(&uprobe->arch, mm, vaddr);
660 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
661 else if (first_uprobe)
662 clear_bit(MMF_HAS_UPROBES, &mm->flags);
668 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
670 set_bit(MMF_RECALC_UPROBES, &mm->flags);
671 return set_orig_insn(&uprobe->arch, mm, vaddr);
674 static inline bool uprobe_is_active(struct uprobe *uprobe)
676 return !RB_EMPTY_NODE(&uprobe->rb_node);
679 * There could be threads that have already hit the breakpoint. They
680 * will recheck the current insn and restart if find_uprobe() fails.
681 * See find_active_uprobe().
683 static void delete_uprobe(struct uprobe *uprobe)
685 if (WARN_ON(!uprobe_is_active(uprobe)))
688 spin_lock(&uprobes_treelock);
689 rb_erase(&uprobe->rb_node, &uprobes_tree);
690 spin_unlock(&uprobes_treelock);
691 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
697 struct map_info *next;
698 struct mm_struct *mm;
702 static inline struct map_info *free_map_info(struct map_info *info)
704 struct map_info *next = info->next;
709 static struct map_info *
710 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
712 unsigned long pgoff = offset >> PAGE_SHIFT;
713 struct vm_area_struct *vma;
714 struct map_info *curr = NULL;
715 struct map_info *prev = NULL;
716 struct map_info *info;
720 i_mmap_lock_read(mapping);
721 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
722 if (!valid_vma(vma, is_register))
725 if (!prev && !more) {
727 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
728 * reclaim. This is optimistic, no harm done if it fails.
730 prev = kmalloc(sizeof(struct map_info),
731 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
740 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
748 info->mm = vma->vm_mm;
749 info->vaddr = offset_to_vaddr(vma, offset);
751 i_mmap_unlock_read(mapping);
763 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
765 curr = ERR_PTR(-ENOMEM);
775 prev = free_map_info(prev);
780 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
782 bool is_register = !!new;
783 struct map_info *info;
786 percpu_down_write(&dup_mmap_sem);
787 info = build_map_info(uprobe->inode->i_mapping,
788 uprobe->offset, is_register);
795 struct mm_struct *mm = info->mm;
796 struct vm_area_struct *vma;
798 if (err && is_register)
801 down_write(&mm->mmap_sem);
802 vma = find_vma(mm, info->vaddr);
803 if (!vma || !valid_vma(vma, is_register) ||
804 file_inode(vma->vm_file) != uprobe->inode)
807 if (vma->vm_start > info->vaddr ||
808 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
812 /* consult only the "caller", new consumer. */
813 if (consumer_filter(new,
814 UPROBE_FILTER_REGISTER, mm))
815 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
816 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
817 if (!filter_chain(uprobe,
818 UPROBE_FILTER_UNREGISTER, mm))
819 err |= remove_breakpoint(uprobe, mm, info->vaddr);
823 up_write(&mm->mmap_sem);
826 info = free_map_info(info);
829 percpu_up_write(&dup_mmap_sem);
833 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
835 consumer_add(uprobe, uc);
836 return register_for_each_vma(uprobe, uc);
839 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
843 if (WARN_ON(!consumer_del(uprobe, uc)))
846 err = register_for_each_vma(uprobe, NULL);
847 /* TODO : cant unregister? schedule a worker thread */
848 if (!uprobe->consumers && !err)
849 delete_uprobe(uprobe);
853 * uprobe_register - register a probe
854 * @inode: the file in which the probe has to be placed.
855 * @offset: offset from the start of the file.
856 * @uc: information on howto handle the probe..
858 * Apart from the access refcount, uprobe_register() takes a creation
859 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
860 * inserted into the rbtree (i.e first consumer for a @inode:@offset
861 * tuple). Creation refcount stops uprobe_unregister from freeing the
862 * @uprobe even before the register operation is complete. Creation
863 * refcount is released when the last @uc for the @uprobe
866 * Return errno if it cannot successully install probes
867 * else return 0 (success)
869 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
871 struct uprobe *uprobe;
874 /* Uprobe must have at least one set consumer */
875 if (!uc->handler && !uc->ret_handler)
878 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
879 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
881 /* Racy, just to catch the obvious mistakes */
882 if (offset > i_size_read(inode))
886 * This ensures that copy_from_page() and copy_to_page()
887 * can't cross page boundary.
889 if (!IS_ALIGNED(offset, UPROBE_SWBP_INSN_SIZE))
893 uprobe = alloc_uprobe(inode, offset);
897 * We can race with uprobe_unregister()->delete_uprobe().
898 * Check uprobe_is_active() and retry if it is false.
900 down_write(&uprobe->register_rwsem);
902 if (likely(uprobe_is_active(uprobe))) {
903 ret = __uprobe_register(uprobe, uc);
905 __uprobe_unregister(uprobe, uc);
907 up_write(&uprobe->register_rwsem);
910 if (unlikely(ret == -EAGAIN))
914 EXPORT_SYMBOL_GPL(uprobe_register);
917 * uprobe_apply - unregister a already registered probe.
918 * @inode: the file in which the probe has to be removed.
919 * @offset: offset from the start of the file.
920 * @uc: consumer which wants to add more or remove some breakpoints
921 * @add: add or remove the breakpoints
923 int uprobe_apply(struct inode *inode, loff_t offset,
924 struct uprobe_consumer *uc, bool add)
926 struct uprobe *uprobe;
927 struct uprobe_consumer *con;
930 uprobe = find_uprobe(inode, offset);
931 if (WARN_ON(!uprobe))
934 down_write(&uprobe->register_rwsem);
935 for (con = uprobe->consumers; con && con != uc ; con = con->next)
938 ret = register_for_each_vma(uprobe, add ? uc : NULL);
939 up_write(&uprobe->register_rwsem);
946 * uprobe_unregister - unregister a already registered probe.
947 * @inode: the file in which the probe has to be removed.
948 * @offset: offset from the start of the file.
949 * @uc: identify which probe if multiple probes are colocated.
951 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
953 struct uprobe *uprobe;
955 uprobe = find_uprobe(inode, offset);
956 if (WARN_ON(!uprobe))
959 down_write(&uprobe->register_rwsem);
960 __uprobe_unregister(uprobe, uc);
961 up_write(&uprobe->register_rwsem);
964 EXPORT_SYMBOL_GPL(uprobe_unregister);
966 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
968 struct vm_area_struct *vma;
971 down_read(&mm->mmap_sem);
972 for (vma = mm->mmap; vma; vma = vma->vm_next) {
976 if (!valid_vma(vma, false) ||
977 file_inode(vma->vm_file) != uprobe->inode)
980 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
981 if (uprobe->offset < offset ||
982 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
985 vaddr = offset_to_vaddr(vma, uprobe->offset);
986 err |= remove_breakpoint(uprobe, mm, vaddr);
988 up_read(&mm->mmap_sem);
993 static struct rb_node *
994 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
996 struct rb_node *n = uprobes_tree.rb_node;
999 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1001 if (inode < u->inode) {
1003 } else if (inode > u->inode) {
1006 if (max < u->offset)
1008 else if (min > u->offset)
1019 * For a given range in vma, build a list of probes that need to be inserted.
1021 static void build_probe_list(struct inode *inode,
1022 struct vm_area_struct *vma,
1023 unsigned long start, unsigned long end,
1024 struct list_head *head)
1027 struct rb_node *n, *t;
1030 INIT_LIST_HEAD(head);
1031 min = vaddr_to_offset(vma, start);
1032 max = min + (end - start) - 1;
1034 spin_lock(&uprobes_treelock);
1035 n = find_node_in_range(inode, min, max);
1037 for (t = n; t; t = rb_prev(t)) {
1038 u = rb_entry(t, struct uprobe, rb_node);
1039 if (u->inode != inode || u->offset < min)
1041 list_add(&u->pending_list, head);
1044 for (t = n; (t = rb_next(t)); ) {
1045 u = rb_entry(t, struct uprobe, rb_node);
1046 if (u->inode != inode || u->offset > max)
1048 list_add(&u->pending_list, head);
1052 spin_unlock(&uprobes_treelock);
1056 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1058 * Currently we ignore all errors and always return 0, the callers
1059 * can't handle the failure anyway.
1061 int uprobe_mmap(struct vm_area_struct *vma)
1063 struct list_head tmp_list;
1064 struct uprobe *uprobe, *u;
1065 struct inode *inode;
1067 if (no_uprobe_events() || !valid_vma(vma, true))
1070 inode = file_inode(vma->vm_file);
1074 mutex_lock(uprobes_mmap_hash(inode));
1075 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1077 * We can race with uprobe_unregister(), this uprobe can be already
1078 * removed. But in this case filter_chain() must return false, all
1079 * consumers have gone away.
1081 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1082 if (!fatal_signal_pending(current) &&
1083 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1084 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1085 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1089 mutex_unlock(uprobes_mmap_hash(inode));
1095 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1098 struct inode *inode;
1101 inode = file_inode(vma->vm_file);
1103 min = vaddr_to_offset(vma, start);
1104 max = min + (end - start) - 1;
1106 spin_lock(&uprobes_treelock);
1107 n = find_node_in_range(inode, min, max);
1108 spin_unlock(&uprobes_treelock);
1114 * Called in context of a munmap of a vma.
1116 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1118 if (no_uprobe_events() || !valid_vma(vma, false))
1121 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1124 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1125 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1128 if (vma_has_uprobes(vma, start, end))
1129 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1132 /* Slot allocation for XOL */
1133 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1135 struct vm_area_struct *vma;
1138 if (down_write_killable(&mm->mmap_sem))
1141 if (mm->uprobes_state.xol_area) {
1147 /* Try to map as high as possible, this is only a hint. */
1148 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1150 if (area->vaddr & ~PAGE_MASK) {
1156 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1157 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1158 &area->xol_mapping);
1165 smp_wmb(); /* pairs with get_xol_area() */
1166 mm->uprobes_state.xol_area = area;
1168 up_write(&mm->mmap_sem);
1173 static struct xol_area *__create_xol_area(unsigned long vaddr)
1175 struct mm_struct *mm = current->mm;
1176 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1177 struct xol_area *area;
1179 area = kmalloc(sizeof(*area), GFP_KERNEL);
1180 if (unlikely(!area))
1183 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1187 area->xol_mapping.name = "[uprobes]";
1188 area->xol_mapping.fault = NULL;
1189 area->xol_mapping.pages = area->pages;
1190 area->pages[0] = alloc_page(GFP_HIGHUSER);
1191 if (!area->pages[0])
1193 area->pages[1] = NULL;
1195 area->vaddr = vaddr;
1196 init_waitqueue_head(&area->wq);
1197 /* Reserve the 1st slot for get_trampoline_vaddr() */
1198 set_bit(0, area->bitmap);
1199 atomic_set(&area->slot_count, 1);
1200 copy_to_page(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1202 if (!xol_add_vma(mm, area))
1205 __free_page(area->pages[0]);
1207 kfree(area->bitmap);
1215 * get_xol_area - Allocate process's xol_area if necessary.
1216 * This area will be used for storing instructions for execution out of line.
1218 * Returns the allocated area or NULL.
1220 static struct xol_area *get_xol_area(void)
1222 struct mm_struct *mm = current->mm;
1223 struct xol_area *area;
1225 if (!mm->uprobes_state.xol_area)
1226 __create_xol_area(0);
1228 area = mm->uprobes_state.xol_area;
1229 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1234 * uprobe_clear_state - Free the area allocated for slots.
1236 void uprobe_clear_state(struct mm_struct *mm)
1238 struct xol_area *area = mm->uprobes_state.xol_area;
1243 put_page(area->pages[0]);
1244 kfree(area->bitmap);
1248 void uprobe_start_dup_mmap(void)
1250 percpu_down_read(&dup_mmap_sem);
1253 void uprobe_end_dup_mmap(void)
1255 percpu_up_read(&dup_mmap_sem);
1258 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1260 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1261 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1262 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1263 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1268 * - search for a free slot.
1270 static unsigned long xol_take_insn_slot(struct xol_area *area)
1272 unsigned long slot_addr;
1276 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1277 if (slot_nr < UINSNS_PER_PAGE) {
1278 if (!test_and_set_bit(slot_nr, area->bitmap))
1281 slot_nr = UINSNS_PER_PAGE;
1284 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1285 } while (slot_nr >= UINSNS_PER_PAGE);
1287 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1288 atomic_inc(&area->slot_count);
1294 * xol_get_insn_slot - allocate a slot for xol.
1295 * Returns the allocated slot address or 0.
1297 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1299 struct xol_area *area;
1300 unsigned long xol_vaddr;
1302 area = get_xol_area();
1306 xol_vaddr = xol_take_insn_slot(area);
1307 if (unlikely(!xol_vaddr))
1310 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1311 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1317 * xol_free_insn_slot - If slot was earlier allocated by
1318 * @xol_get_insn_slot(), make the slot available for
1319 * subsequent requests.
1321 static void xol_free_insn_slot(struct task_struct *tsk)
1323 struct xol_area *area;
1324 unsigned long vma_end;
1325 unsigned long slot_addr;
1327 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1330 slot_addr = tsk->utask->xol_vaddr;
1331 if (unlikely(!slot_addr))
1334 area = tsk->mm->uprobes_state.xol_area;
1335 vma_end = area->vaddr + PAGE_SIZE;
1336 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1337 unsigned long offset;
1340 offset = slot_addr - area->vaddr;
1341 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1342 if (slot_nr >= UINSNS_PER_PAGE)
1345 clear_bit(slot_nr, area->bitmap);
1346 atomic_dec(&area->slot_count);
1347 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1348 if (waitqueue_active(&area->wq))
1351 tsk->utask->xol_vaddr = 0;
1355 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1356 void *src, unsigned long len)
1358 /* Initialize the slot */
1359 copy_to_page(page, vaddr, src, len);
1362 * We probably need flush_icache_user_range() but it needs vma.
1363 * This should work on most of architectures by default. If
1364 * architecture needs to do something different it can define
1365 * its own version of the function.
1367 flush_dcache_page(page);
1371 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1372 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1374 * Return the address of the breakpoint instruction.
1376 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1378 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1381 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1383 struct uprobe_task *utask = current->utask;
1385 if (unlikely(utask && utask->active_uprobe))
1386 return utask->vaddr;
1388 return instruction_pointer(regs);
1391 static struct return_instance *free_ret_instance(struct return_instance *ri)
1393 struct return_instance *next = ri->next;
1394 put_uprobe(ri->uprobe);
1400 * Called with no locks held.
1401 * Called in context of a exiting or a exec-ing thread.
1403 void uprobe_free_utask(struct task_struct *t)
1405 struct uprobe_task *utask = t->utask;
1406 struct return_instance *ri;
1411 if (utask->active_uprobe)
1412 put_uprobe(utask->active_uprobe);
1414 ri = utask->return_instances;
1416 ri = free_ret_instance(ri);
1418 xol_free_insn_slot(t);
1424 * Allocate a uprobe_task object for the task if if necessary.
1425 * Called when the thread hits a breakpoint.
1428 * - pointer to new uprobe_task on success
1431 static struct uprobe_task *get_utask(void)
1433 if (!current->utask)
1434 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1435 return current->utask;
1438 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1440 struct uprobe_task *n_utask;
1441 struct return_instance **p, *o, *n;
1443 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1448 p = &n_utask->return_instances;
1449 for (o = o_utask->return_instances; o; o = o->next) {
1450 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1455 get_uprobe(n->uprobe);
1466 static void uprobe_warn(struct task_struct *t, const char *msg)
1468 pr_warn("uprobe: %s:%d failed to %s\n",
1469 current->comm, current->pid, msg);
1472 static void dup_xol_work(struct callback_head *work)
1474 if (current->flags & PF_EXITING)
1477 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1478 !fatal_signal_pending(current))
1479 uprobe_warn(current, "dup xol area");
1483 * Called in context of a new clone/fork from copy_process.
1485 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1487 struct uprobe_task *utask = current->utask;
1488 struct mm_struct *mm = current->mm;
1489 struct xol_area *area;
1493 if (!utask || !utask->return_instances)
1496 if (mm == t->mm && !(flags & CLONE_VFORK))
1499 if (dup_utask(t, utask))
1500 return uprobe_warn(t, "dup ret instances");
1502 /* The task can fork() after dup_xol_work() fails */
1503 area = mm->uprobes_state.xol_area;
1505 return uprobe_warn(t, "dup xol area");
1510 t->utask->dup_xol_addr = area->vaddr;
1511 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1512 task_work_add(t, &t->utask->dup_xol_work, true);
1516 * Current area->vaddr notion assume the trampoline address is always
1517 * equal area->vaddr.
1519 * Returns -1 in case the xol_area is not allocated.
1521 static unsigned long get_trampoline_vaddr(void)
1523 struct xol_area *area;
1524 unsigned long trampoline_vaddr = -1;
1526 area = current->mm->uprobes_state.xol_area;
1527 smp_read_barrier_depends();
1529 trampoline_vaddr = area->vaddr;
1531 return trampoline_vaddr;
1534 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1535 struct pt_regs *regs)
1537 struct return_instance *ri = utask->return_instances;
1538 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1540 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1541 ri = free_ret_instance(ri);
1544 utask->return_instances = ri;
1547 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1549 struct return_instance *ri;
1550 struct uprobe_task *utask;
1551 unsigned long orig_ret_vaddr, trampoline_vaddr;
1554 if (!get_xol_area())
1557 utask = get_utask();
1561 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1562 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1563 " nestedness limit pid/tgid=%d/%d\n",
1564 current->pid, current->tgid);
1568 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1572 trampoline_vaddr = get_trampoline_vaddr();
1573 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1574 if (orig_ret_vaddr == -1)
1577 /* drop the entries invalidated by longjmp() */
1578 chained = (orig_ret_vaddr == trampoline_vaddr);
1579 cleanup_return_instances(utask, chained, regs);
1582 * We don't want to keep trampoline address in stack, rather keep the
1583 * original return address of first caller thru all the consequent
1584 * instances. This also makes breakpoint unwrapping easier.
1587 if (!utask->return_instances) {
1589 * This situation is not possible. Likely we have an
1590 * attack from user-space.
1592 uprobe_warn(current, "handle tail call");
1595 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1598 ri->uprobe = get_uprobe(uprobe);
1599 ri->func = instruction_pointer(regs);
1600 ri->stack = user_stack_pointer(regs);
1601 ri->orig_ret_vaddr = orig_ret_vaddr;
1602 ri->chained = chained;
1605 ri->next = utask->return_instances;
1606 utask->return_instances = ri;
1613 /* Prepare to single-step probed instruction out of line. */
1615 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1617 struct uprobe_task *utask;
1618 unsigned long xol_vaddr;
1621 utask = get_utask();
1625 xol_vaddr = xol_get_insn_slot(uprobe);
1629 utask->xol_vaddr = xol_vaddr;
1630 utask->vaddr = bp_vaddr;
1632 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1633 if (unlikely(err)) {
1634 xol_free_insn_slot(current);
1638 utask->active_uprobe = uprobe;
1639 utask->state = UTASK_SSTEP;
1644 * If we are singlestepping, then ensure this thread is not connected to
1645 * non-fatal signals until completion of singlestep. When xol insn itself
1646 * triggers the signal, restart the original insn even if the task is
1647 * already SIGKILL'ed (since coredump should report the correct ip). This
1648 * is even more important if the task has a handler for SIGSEGV/etc, The
1649 * _same_ instruction should be repeated again after return from the signal
1650 * handler, and SSTEP can never finish in this case.
1652 bool uprobe_deny_signal(void)
1654 struct task_struct *t = current;
1655 struct uprobe_task *utask = t->utask;
1657 if (likely(!utask || !utask->active_uprobe))
1660 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1662 if (signal_pending(t)) {
1663 spin_lock_irq(&t->sighand->siglock);
1664 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1665 spin_unlock_irq(&t->sighand->siglock);
1667 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1668 utask->state = UTASK_SSTEP_TRAPPED;
1669 set_tsk_thread_flag(t, TIF_UPROBE);
1676 static void mmf_recalc_uprobes(struct mm_struct *mm)
1678 struct vm_area_struct *vma;
1680 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1681 if (!valid_vma(vma, false))
1684 * This is not strictly accurate, we can race with
1685 * uprobe_unregister() and see the already removed
1686 * uprobe if delete_uprobe() was not yet called.
1687 * Or this uprobe can be filtered out.
1689 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1693 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1696 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1699 uprobe_opcode_t opcode;
1702 if (WARN_ON_ONCE(!IS_ALIGNED(vaddr, UPROBE_SWBP_INSN_SIZE)))
1705 pagefault_disable();
1706 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
1709 if (likely(result == 0))
1713 * The NULL 'tsk' here ensures that any faults that occur here
1714 * will not be accounted to the task. 'mm' *is* current->mm,
1715 * but we treat this as a 'remote' access since it is
1716 * essentially a kernel access to the memory.
1718 result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
1723 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1726 /* This needs to return true for any variant of the trap insn */
1727 return is_trap_insn(&opcode);
1730 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1732 struct mm_struct *mm = current->mm;
1733 struct uprobe *uprobe = NULL;
1734 struct vm_area_struct *vma;
1736 down_read(&mm->mmap_sem);
1737 vma = find_vma(mm, bp_vaddr);
1738 if (vma && vma->vm_start <= bp_vaddr) {
1739 if (valid_vma(vma, false)) {
1740 struct inode *inode = file_inode(vma->vm_file);
1741 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1743 uprobe = find_uprobe(inode, offset);
1747 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1752 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1753 mmf_recalc_uprobes(mm);
1754 up_read(&mm->mmap_sem);
1759 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1761 struct uprobe_consumer *uc;
1762 int remove = UPROBE_HANDLER_REMOVE;
1763 bool need_prep = false; /* prepare return uprobe, when needed */
1765 down_read(&uprobe->register_rwsem);
1766 for (uc = uprobe->consumers; uc; uc = uc->next) {
1770 rc = uc->handler(uc, regs);
1771 WARN(rc & ~UPROBE_HANDLER_MASK,
1772 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1775 if (uc->ret_handler)
1781 if (need_prep && !remove)
1782 prepare_uretprobe(uprobe, regs); /* put bp at return */
1784 if (remove && uprobe->consumers) {
1785 WARN_ON(!uprobe_is_active(uprobe));
1786 unapply_uprobe(uprobe, current->mm);
1788 up_read(&uprobe->register_rwsem);
1792 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1794 struct uprobe *uprobe = ri->uprobe;
1795 struct uprobe_consumer *uc;
1797 down_read(&uprobe->register_rwsem);
1798 for (uc = uprobe->consumers; uc; uc = uc->next) {
1799 if (uc->ret_handler)
1800 uc->ret_handler(uc, ri->func, regs);
1802 up_read(&uprobe->register_rwsem);
1805 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
1810 chained = ri->chained;
1811 ri = ri->next; /* can't be NULL if chained */
1817 static void handle_trampoline(struct pt_regs *regs)
1819 struct uprobe_task *utask;
1820 struct return_instance *ri, *next;
1823 utask = current->utask;
1827 ri = utask->return_instances;
1833 * We should throw out the frames invalidated by longjmp().
1834 * If this chain is valid, then the next one should be alive
1835 * or NULL; the latter case means that nobody but ri->func
1836 * could hit this trampoline on return. TODO: sigaltstack().
1838 next = find_next_ret_chain(ri);
1839 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
1841 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1844 handle_uretprobe_chain(ri, regs);
1845 ri = free_ret_instance(ri);
1847 } while (ri != next);
1850 utask->return_instances = ri;
1854 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
1855 force_sig(SIGILL, current);
1859 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
1864 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
1865 struct pt_regs *regs)
1871 * Run handler and ask thread to singlestep.
1872 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1874 static void handle_swbp(struct pt_regs *regs)
1876 struct uprobe *uprobe;
1877 unsigned long bp_vaddr;
1878 int uninitialized_var(is_swbp);
1880 bp_vaddr = uprobe_get_swbp_addr(regs);
1881 if (bp_vaddr == get_trampoline_vaddr())
1882 return handle_trampoline(regs);
1884 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1887 /* No matching uprobe; signal SIGTRAP. */
1888 force_sig(SIGTRAP, current);
1891 * Either we raced with uprobe_unregister() or we can't
1892 * access this memory. The latter is only possible if
1893 * another thread plays with our ->mm. In both cases
1894 * we can simply restart. If this vma was unmapped we
1895 * can pretend this insn was not executed yet and get
1896 * the (correct) SIGSEGV after restart.
1898 instruction_pointer_set(regs, bp_vaddr);
1903 /* change it in advance for ->handler() and restart */
1904 instruction_pointer_set(regs, bp_vaddr);
1907 * TODO: move copy_insn/etc into _register and remove this hack.
1908 * After we hit the bp, _unregister + _register can install the
1909 * new and not-yet-analyzed uprobe at the same address, restart.
1911 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1915 * Pairs with the smp_wmb() in prepare_uprobe().
1917 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
1918 * we must also see the stores to &uprobe->arch performed by the
1919 * prepare_uprobe() call.
1923 /* Tracing handlers use ->utask to communicate with fetch methods */
1927 if (arch_uprobe_ignore(&uprobe->arch, regs))
1930 handler_chain(uprobe, regs);
1932 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1935 if (!pre_ssout(uprobe, regs, bp_vaddr))
1938 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1944 * Perform required fix-ups and disable singlestep.
1945 * Allow pending signals to take effect.
1947 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1949 struct uprobe *uprobe;
1952 uprobe = utask->active_uprobe;
1953 if (utask->state == UTASK_SSTEP_ACK)
1954 err = arch_uprobe_post_xol(&uprobe->arch, regs);
1955 else if (utask->state == UTASK_SSTEP_TRAPPED)
1956 arch_uprobe_abort_xol(&uprobe->arch, regs);
1961 utask->active_uprobe = NULL;
1962 utask->state = UTASK_RUNNING;
1963 xol_free_insn_slot(current);
1965 spin_lock_irq(¤t->sighand->siglock);
1966 recalc_sigpending(); /* see uprobe_deny_signal() */
1967 spin_unlock_irq(¤t->sighand->siglock);
1969 if (unlikely(err)) {
1970 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
1971 force_sig(SIGILL, current);
1976 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1977 * allows the thread to return from interrupt. After that handle_swbp()
1978 * sets utask->active_uprobe.
1980 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1981 * and allows the thread to return from interrupt.
1983 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1984 * uprobe_notify_resume().
1986 void uprobe_notify_resume(struct pt_regs *regs)
1988 struct uprobe_task *utask;
1990 clear_thread_flag(TIF_UPROBE);
1992 utask = current->utask;
1993 if (utask && utask->active_uprobe)
1994 handle_singlestep(utask, regs);
2000 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2001 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2003 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2008 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2009 (!current->utask || !current->utask->return_instances))
2012 set_thread_flag(TIF_UPROBE);
2017 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2018 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2020 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2022 struct uprobe_task *utask = current->utask;
2024 if (!current->mm || !utask || !utask->active_uprobe)
2025 /* task is currently not uprobed */
2028 utask->state = UTASK_SSTEP_ACK;
2029 set_thread_flag(TIF_UPROBE);
2033 static struct notifier_block uprobe_exception_nb = {
2034 .notifier_call = arch_uprobe_exception_notify,
2035 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2038 static int __init init_uprobes(void)
2042 for (i = 0; i < UPROBES_HASH_SZ; i++)
2043 mutex_init(&uprobes_mmap_mutex[i]);
2045 if (percpu_init_rwsem(&dup_mmap_sem))
2048 return register_die_notifier(&uprobe_exception_nb);
2050 __initcall(init_uprobes);