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>
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 */
105 * We keep the vma's vm_start rather than a pointer to the vma
106 * itself. The probed process or a naughty kernel module could make
107 * the vma go away, and we must handle that reasonably gracefully.
109 unsigned long vaddr; /* Page(s) of instruction slots */
113 * valid_vma: Verify if the specified vma is an executable vma
114 * Relax restrictions while unregistering: vm_flags might have
115 * changed after breakpoint was inserted.
116 * - is_register: indicates if we are in register context.
117 * - Return 1 if the specified virtual address is in an
120 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
122 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
127 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
130 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
132 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
135 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
137 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
141 * __replace_page - replace page in vma by new page.
142 * based on replace_page in mm/ksm.c
144 * @vma: vma that holds the pte pointing to page
145 * @addr: address the old @page is mapped at
146 * @page: the cowed page we are replacing by kpage
147 * @kpage: the modified page we replace page by
149 * Returns 0 on success, -EFAULT on failure.
151 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
152 struct page *page, struct page *kpage)
154 struct mm_struct *mm = vma->vm_mm;
158 /* For mmu_notifiers */
159 const unsigned long mmun_start = addr;
160 const unsigned long mmun_end = addr + PAGE_SIZE;
161 struct mem_cgroup *memcg;
163 err = mem_cgroup_try_charge(kpage, vma->vm_mm, GFP_KERNEL, &memcg);
167 /* For try_to_free_swap() and munlock_vma_page() below */
170 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
172 ptep = page_check_address(page, mm, addr, &ptl, 0);
177 page_add_new_anon_rmap(kpage, vma, addr);
178 mem_cgroup_commit_charge(kpage, memcg, false);
179 lru_cache_add_active_or_unevictable(kpage, vma);
181 if (!PageAnon(page)) {
182 dec_mm_counter(mm, MM_FILEPAGES);
183 inc_mm_counter(mm, MM_ANONPAGES);
186 flush_cache_page(vma, addr, pte_pfn(*ptep));
187 ptep_clear_flush_notify(vma, addr, ptep);
188 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
190 page_remove_rmap(page);
191 if (!page_mapped(page))
192 try_to_free_swap(page);
193 pte_unmap_unlock(ptep, ptl);
195 if (vma->vm_flags & VM_LOCKED)
196 munlock_vma_page(page);
201 mem_cgroup_cancel_charge(kpage, memcg);
202 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
208 * is_swbp_insn - check if instruction is breakpoint instruction.
209 * @insn: instruction to be checked.
210 * Default implementation of is_swbp_insn
211 * Returns true if @insn is a breakpoint instruction.
213 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
215 return *insn == UPROBE_SWBP_INSN;
219 * is_trap_insn - check if instruction is breakpoint instruction.
220 * @insn: instruction to be checked.
221 * Default implementation of is_trap_insn
222 * Returns true if @insn is a breakpoint instruction.
224 * This function is needed for the case where an architecture has multiple
225 * trap instructions (like powerpc).
227 bool __weak is_trap_insn(uprobe_opcode_t *insn)
229 return is_swbp_insn(insn);
232 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
234 void *kaddr = kmap_atomic(page);
235 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
236 kunmap_atomic(kaddr);
239 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
241 void *kaddr = kmap_atomic(page);
242 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
243 kunmap_atomic(kaddr);
246 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
248 uprobe_opcode_t old_opcode;
252 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
253 * We do not check if it is any other 'trap variant' which could
254 * be conditional trap instruction such as the one powerpc supports.
256 * The logic is that we do not care if the underlying instruction
257 * is a trap variant; uprobes always wins over any other (gdb)
260 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
261 is_swbp = is_swbp_insn(&old_opcode);
263 if (is_swbp_insn(new_opcode)) {
264 if (is_swbp) /* register: already installed? */
267 if (!is_swbp) /* unregister: was it changed by us? */
276 * Expect the breakpoint instruction to be the smallest size instruction for
277 * the architecture. If an arch has variable length instruction and the
278 * breakpoint instruction is not of the smallest length instruction
279 * supported by that architecture then we need to modify is_trap_at_addr and
280 * uprobe_write_opcode accordingly. This would never be a problem for archs
281 * that have fixed length instructions.
283 * uprobe_write_opcode - write the opcode at a given virtual address.
284 * @mm: the probed process address space.
285 * @vaddr: the virtual address to store the opcode.
286 * @opcode: opcode to be written at @vaddr.
288 * Called with mm->mmap_sem held for write.
289 * Return 0 (success) or a negative errno.
291 int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr,
292 uprobe_opcode_t opcode)
294 struct page *old_page, *new_page;
295 struct vm_area_struct *vma;
299 /* Read the page with vaddr into memory */
300 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
304 ret = verify_opcode(old_page, vaddr, &opcode);
308 ret = anon_vma_prepare(vma);
313 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
317 __SetPageUptodate(new_page);
318 copy_highpage(new_page, old_page);
319 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
321 ret = __replace_page(vma, vaddr, old_page, new_page);
322 page_cache_release(new_page);
326 if (unlikely(ret == -EAGAIN))
332 * set_swbp - store breakpoint at a given address.
333 * @auprobe: arch specific probepoint information.
334 * @mm: the probed process address space.
335 * @vaddr: the virtual address to insert the opcode.
337 * For mm @mm, store the breakpoint instruction at @vaddr.
338 * Return 0 (success) or a negative errno.
340 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
342 return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
346 * set_orig_insn - Restore the original instruction.
347 * @mm: the probed process address space.
348 * @auprobe: arch specific probepoint information.
349 * @vaddr: the virtual address to insert the opcode.
351 * For mm @mm, restore the original opcode (opcode) at @vaddr.
352 * Return 0 (success) or a negative errno.
355 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
357 return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn);
360 static struct uprobe *get_uprobe(struct uprobe *uprobe)
362 atomic_inc(&uprobe->ref);
366 static void put_uprobe(struct uprobe *uprobe)
368 if (atomic_dec_and_test(&uprobe->ref))
372 static int match_uprobe(struct uprobe *l, struct uprobe *r)
374 if (l->inode < r->inode)
377 if (l->inode > r->inode)
380 if (l->offset < r->offset)
383 if (l->offset > r->offset)
389 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
391 struct uprobe u = { .inode = inode, .offset = offset };
392 struct rb_node *n = uprobes_tree.rb_node;
393 struct uprobe *uprobe;
397 uprobe = rb_entry(n, struct uprobe, rb_node);
398 match = match_uprobe(&u, uprobe);
400 return get_uprobe(uprobe);
411 * Find a uprobe corresponding to a given inode:offset
412 * Acquires uprobes_treelock
414 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
416 struct uprobe *uprobe;
418 spin_lock(&uprobes_treelock);
419 uprobe = __find_uprobe(inode, offset);
420 spin_unlock(&uprobes_treelock);
425 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
427 struct rb_node **p = &uprobes_tree.rb_node;
428 struct rb_node *parent = NULL;
434 u = rb_entry(parent, struct uprobe, rb_node);
435 match = match_uprobe(uprobe, u);
437 return get_uprobe(u);
440 p = &parent->rb_left;
442 p = &parent->rb_right;
447 rb_link_node(&uprobe->rb_node, parent, p);
448 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
449 /* get access + creation ref */
450 atomic_set(&uprobe->ref, 2);
456 * Acquire uprobes_treelock.
457 * Matching uprobe already exists in rbtree;
458 * increment (access refcount) and return the matching uprobe.
460 * No matching uprobe; insert the uprobe in rb_tree;
461 * get a double refcount (access + creation) and return NULL.
463 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
467 spin_lock(&uprobes_treelock);
468 u = __insert_uprobe(uprobe);
469 spin_unlock(&uprobes_treelock);
474 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
476 struct uprobe *uprobe, *cur_uprobe;
478 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
482 uprobe->inode = igrab(inode);
483 uprobe->offset = offset;
484 init_rwsem(&uprobe->register_rwsem);
485 init_rwsem(&uprobe->consumer_rwsem);
487 /* add to uprobes_tree, sorted on inode:offset */
488 cur_uprobe = insert_uprobe(uprobe);
489 /* a uprobe exists for this inode:offset combination */
499 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
501 down_write(&uprobe->consumer_rwsem);
502 uc->next = uprobe->consumers;
503 uprobe->consumers = uc;
504 up_write(&uprobe->consumer_rwsem);
508 * For uprobe @uprobe, delete the consumer @uc.
509 * Return true if the @uc is deleted successfully
512 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
514 struct uprobe_consumer **con;
517 down_write(&uprobe->consumer_rwsem);
518 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
525 up_write(&uprobe->consumer_rwsem);
530 static int __copy_insn(struct address_space *mapping, struct file *filp,
531 void *insn, int nbytes, loff_t offset)
535 * Ensure that the page that has the original instruction is populated
536 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
537 * see uprobe_register().
539 if (mapping->a_ops->readpage)
540 page = read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT, filp);
542 page = shmem_read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT);
544 return PTR_ERR(page);
546 copy_from_page(page, offset, insn, nbytes);
547 page_cache_release(page);
552 static int copy_insn(struct uprobe *uprobe, struct file *filp)
554 struct address_space *mapping = uprobe->inode->i_mapping;
555 loff_t offs = uprobe->offset;
556 void *insn = &uprobe->arch.insn;
557 int size = sizeof(uprobe->arch.insn);
560 /* Copy only available bytes, -EIO if nothing was read */
562 if (offs >= i_size_read(uprobe->inode))
565 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
566 err = __copy_insn(mapping, filp, insn, len, offs);
578 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
579 struct mm_struct *mm, unsigned long vaddr)
583 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
586 /* TODO: move this into _register, until then we abuse this sem. */
587 down_write(&uprobe->consumer_rwsem);
588 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
591 ret = copy_insn(uprobe, file);
596 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
599 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
603 /* uprobe_write_opcode() assumes we don't cross page boundary */
604 BUG_ON((uprobe->offset & ~PAGE_MASK) +
605 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
607 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
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 uprobe = alloc_uprobe(inode, offset);
890 * We can race with uprobe_unregister()->delete_uprobe().
891 * Check uprobe_is_active() and retry if it is false.
893 down_write(&uprobe->register_rwsem);
895 if (likely(uprobe_is_active(uprobe))) {
896 ret = __uprobe_register(uprobe, uc);
898 __uprobe_unregister(uprobe, uc);
900 up_write(&uprobe->register_rwsem);
903 if (unlikely(ret == -EAGAIN))
907 EXPORT_SYMBOL_GPL(uprobe_register);
910 * uprobe_apply - unregister a already registered probe.
911 * @inode: the file in which the probe has to be removed.
912 * @offset: offset from the start of the file.
913 * @uc: consumer which wants to add more or remove some breakpoints
914 * @add: add or remove the breakpoints
916 int uprobe_apply(struct inode *inode, loff_t offset,
917 struct uprobe_consumer *uc, bool add)
919 struct uprobe *uprobe;
920 struct uprobe_consumer *con;
923 uprobe = find_uprobe(inode, offset);
924 if (WARN_ON(!uprobe))
927 down_write(&uprobe->register_rwsem);
928 for (con = uprobe->consumers; con && con != uc ; con = con->next)
931 ret = register_for_each_vma(uprobe, add ? uc : NULL);
932 up_write(&uprobe->register_rwsem);
939 * uprobe_unregister - unregister a already registered probe.
940 * @inode: the file in which the probe has to be removed.
941 * @offset: offset from the start of the file.
942 * @uc: identify which probe if multiple probes are colocated.
944 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
946 struct uprobe *uprobe;
948 uprobe = find_uprobe(inode, offset);
949 if (WARN_ON(!uprobe))
952 down_write(&uprobe->register_rwsem);
953 __uprobe_unregister(uprobe, uc);
954 up_write(&uprobe->register_rwsem);
957 EXPORT_SYMBOL_GPL(uprobe_unregister);
959 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
961 struct vm_area_struct *vma;
964 down_read(&mm->mmap_sem);
965 for (vma = mm->mmap; vma; vma = vma->vm_next) {
969 if (!valid_vma(vma, false) ||
970 file_inode(vma->vm_file) != uprobe->inode)
973 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
974 if (uprobe->offset < offset ||
975 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
978 vaddr = offset_to_vaddr(vma, uprobe->offset);
979 err |= remove_breakpoint(uprobe, mm, vaddr);
981 up_read(&mm->mmap_sem);
986 static struct rb_node *
987 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
989 struct rb_node *n = uprobes_tree.rb_node;
992 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
994 if (inode < u->inode) {
996 } else if (inode > u->inode) {
1001 else if (min > u->offset)
1012 * For a given range in vma, build a list of probes that need to be inserted.
1014 static void build_probe_list(struct inode *inode,
1015 struct vm_area_struct *vma,
1016 unsigned long start, unsigned long end,
1017 struct list_head *head)
1020 struct rb_node *n, *t;
1023 INIT_LIST_HEAD(head);
1024 min = vaddr_to_offset(vma, start);
1025 max = min + (end - start) - 1;
1027 spin_lock(&uprobes_treelock);
1028 n = find_node_in_range(inode, min, max);
1030 for (t = n; t; t = rb_prev(t)) {
1031 u = rb_entry(t, struct uprobe, rb_node);
1032 if (u->inode != inode || u->offset < min)
1034 list_add(&u->pending_list, head);
1037 for (t = n; (t = rb_next(t)); ) {
1038 u = rb_entry(t, struct uprobe, rb_node);
1039 if (u->inode != inode || u->offset > max)
1041 list_add(&u->pending_list, head);
1045 spin_unlock(&uprobes_treelock);
1049 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1051 * Currently we ignore all errors and always return 0, the callers
1052 * can't handle the failure anyway.
1054 int uprobe_mmap(struct vm_area_struct *vma)
1056 struct list_head tmp_list;
1057 struct uprobe *uprobe, *u;
1058 struct inode *inode;
1060 if (no_uprobe_events() || !valid_vma(vma, true))
1063 inode = file_inode(vma->vm_file);
1067 mutex_lock(uprobes_mmap_hash(inode));
1068 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1070 * We can race with uprobe_unregister(), this uprobe can be already
1071 * removed. But in this case filter_chain() must return false, all
1072 * consumers have gone away.
1074 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1075 if (!fatal_signal_pending(current) &&
1076 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1077 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1078 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1082 mutex_unlock(uprobes_mmap_hash(inode));
1088 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1091 struct inode *inode;
1094 inode = file_inode(vma->vm_file);
1096 min = vaddr_to_offset(vma, start);
1097 max = min + (end - start) - 1;
1099 spin_lock(&uprobes_treelock);
1100 n = find_node_in_range(inode, min, max);
1101 spin_unlock(&uprobes_treelock);
1107 * Called in context of a munmap of a vma.
1109 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1111 if (no_uprobe_events() || !valid_vma(vma, false))
1114 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1117 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1118 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1121 if (vma_has_uprobes(vma, start, end))
1122 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1125 /* Slot allocation for XOL */
1126 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1128 int ret = -EALREADY;
1130 down_write(&mm->mmap_sem);
1131 if (mm->uprobes_state.xol_area)
1135 /* Try to map as high as possible, this is only a hint. */
1136 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1138 if (area->vaddr & ~PAGE_MASK) {
1144 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1145 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1149 smp_wmb(); /* pairs with get_xol_area() */
1150 mm->uprobes_state.xol_area = area;
1152 up_write(&mm->mmap_sem);
1157 static struct xol_area *__create_xol_area(unsigned long vaddr)
1159 struct mm_struct *mm = current->mm;
1160 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1161 struct xol_area *area;
1163 area = kmalloc(sizeof(*area), GFP_KERNEL);
1164 if (unlikely(!area))
1167 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1171 area->page = alloc_page(GFP_HIGHUSER);
1175 area->vaddr = vaddr;
1176 init_waitqueue_head(&area->wq);
1177 /* Reserve the 1st slot for get_trampoline_vaddr() */
1178 set_bit(0, area->bitmap);
1179 atomic_set(&area->slot_count, 1);
1180 copy_to_page(area->page, 0, &insn, UPROBE_SWBP_INSN_SIZE);
1182 if (!xol_add_vma(mm, area))
1185 __free_page(area->page);
1187 kfree(area->bitmap);
1195 * get_xol_area - Allocate process's xol_area if necessary.
1196 * This area will be used for storing instructions for execution out of line.
1198 * Returns the allocated area or NULL.
1200 static struct xol_area *get_xol_area(void)
1202 struct mm_struct *mm = current->mm;
1203 struct xol_area *area;
1205 if (!mm->uprobes_state.xol_area)
1206 __create_xol_area(0);
1208 area = mm->uprobes_state.xol_area;
1209 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1214 * uprobe_clear_state - Free the area allocated for slots.
1216 void uprobe_clear_state(struct mm_struct *mm)
1218 struct xol_area *area = mm->uprobes_state.xol_area;
1223 put_page(area->page);
1224 kfree(area->bitmap);
1228 void uprobe_start_dup_mmap(void)
1230 percpu_down_read(&dup_mmap_sem);
1233 void uprobe_end_dup_mmap(void)
1235 percpu_up_read(&dup_mmap_sem);
1238 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1240 newmm->uprobes_state.xol_area = NULL;
1242 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1243 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1244 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1245 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1250 * - search for a free slot.
1252 static unsigned long xol_take_insn_slot(struct xol_area *area)
1254 unsigned long slot_addr;
1258 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1259 if (slot_nr < UINSNS_PER_PAGE) {
1260 if (!test_and_set_bit(slot_nr, area->bitmap))
1263 slot_nr = UINSNS_PER_PAGE;
1266 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1267 } while (slot_nr >= UINSNS_PER_PAGE);
1269 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1270 atomic_inc(&area->slot_count);
1276 * xol_get_insn_slot - allocate a slot for xol.
1277 * Returns the allocated slot address or 0.
1279 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1281 struct xol_area *area;
1282 unsigned long xol_vaddr;
1284 area = get_xol_area();
1288 xol_vaddr = xol_take_insn_slot(area);
1289 if (unlikely(!xol_vaddr))
1292 arch_uprobe_copy_ixol(area->page, xol_vaddr,
1293 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1299 * xol_free_insn_slot - If slot was earlier allocated by
1300 * @xol_get_insn_slot(), make the slot available for
1301 * subsequent requests.
1303 static void xol_free_insn_slot(struct task_struct *tsk)
1305 struct xol_area *area;
1306 unsigned long vma_end;
1307 unsigned long slot_addr;
1309 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1312 slot_addr = tsk->utask->xol_vaddr;
1313 if (unlikely(!slot_addr))
1316 area = tsk->mm->uprobes_state.xol_area;
1317 vma_end = area->vaddr + PAGE_SIZE;
1318 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1319 unsigned long offset;
1322 offset = slot_addr - area->vaddr;
1323 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1324 if (slot_nr >= UINSNS_PER_PAGE)
1327 clear_bit(slot_nr, area->bitmap);
1328 atomic_dec(&area->slot_count);
1329 if (waitqueue_active(&area->wq))
1332 tsk->utask->xol_vaddr = 0;
1336 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1337 void *src, unsigned long len)
1339 /* Initialize the slot */
1340 copy_to_page(page, vaddr, src, len);
1343 * We probably need flush_icache_user_range() but it needs vma.
1344 * This should work on most of architectures by default. If
1345 * architecture needs to do something different it can define
1346 * its own version of the function.
1348 flush_dcache_page(page);
1352 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1353 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1355 * Return the address of the breakpoint instruction.
1357 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1359 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1362 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1364 struct uprobe_task *utask = current->utask;
1366 if (unlikely(utask && utask->active_uprobe))
1367 return utask->vaddr;
1369 return instruction_pointer(regs);
1372 static struct return_instance *free_ret_instance(struct return_instance *ri)
1374 struct return_instance *next = ri->next;
1375 put_uprobe(ri->uprobe);
1381 * Called with no locks held.
1382 * Called in context of a exiting or a exec-ing thread.
1384 void uprobe_free_utask(struct task_struct *t)
1386 struct uprobe_task *utask = t->utask;
1387 struct return_instance *ri;
1392 if (utask->active_uprobe)
1393 put_uprobe(utask->active_uprobe);
1395 ri = utask->return_instances;
1397 ri = free_ret_instance(ri);
1399 xol_free_insn_slot(t);
1405 * Allocate a uprobe_task object for the task if if necessary.
1406 * Called when the thread hits a breakpoint.
1409 * - pointer to new uprobe_task on success
1412 static struct uprobe_task *get_utask(void)
1414 if (!current->utask)
1415 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1416 return current->utask;
1419 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1421 struct uprobe_task *n_utask;
1422 struct return_instance **p, *o, *n;
1424 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1429 p = &n_utask->return_instances;
1430 for (o = o_utask->return_instances; o; o = o->next) {
1431 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1436 get_uprobe(n->uprobe);
1447 static void uprobe_warn(struct task_struct *t, const char *msg)
1449 pr_warn("uprobe: %s:%d failed to %s\n",
1450 current->comm, current->pid, msg);
1453 static void dup_xol_work(struct callback_head *work)
1455 if (current->flags & PF_EXITING)
1458 if (!__create_xol_area(current->utask->dup_xol_addr))
1459 uprobe_warn(current, "dup xol area");
1463 * Called in context of a new clone/fork from copy_process.
1465 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1467 struct uprobe_task *utask = current->utask;
1468 struct mm_struct *mm = current->mm;
1469 struct xol_area *area;
1473 if (!utask || !utask->return_instances)
1476 if (mm == t->mm && !(flags & CLONE_VFORK))
1479 if (dup_utask(t, utask))
1480 return uprobe_warn(t, "dup ret instances");
1482 /* The task can fork() after dup_xol_work() fails */
1483 area = mm->uprobes_state.xol_area;
1485 return uprobe_warn(t, "dup xol area");
1490 t->utask->dup_xol_addr = area->vaddr;
1491 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1492 task_work_add(t, &t->utask->dup_xol_work, true);
1496 * Current area->vaddr notion assume the trampoline address is always
1497 * equal area->vaddr.
1499 * Returns -1 in case the xol_area is not allocated.
1501 static unsigned long get_trampoline_vaddr(void)
1503 struct xol_area *area;
1504 unsigned long trampoline_vaddr = -1;
1506 area = current->mm->uprobes_state.xol_area;
1507 smp_read_barrier_depends();
1509 trampoline_vaddr = area->vaddr;
1511 return trampoline_vaddr;
1514 static void cleanup_return_instances(struct uprobe_task *utask, struct pt_regs *regs)
1516 struct return_instance *ri = utask->return_instances;
1517 while (ri && !arch_uretprobe_is_alive(ri, regs)) {
1518 ri = free_ret_instance(ri);
1521 utask->return_instances = ri;
1524 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1526 struct return_instance *ri;
1527 struct uprobe_task *utask;
1528 unsigned long orig_ret_vaddr, trampoline_vaddr;
1529 bool chained = false;
1531 if (!get_xol_area())
1534 utask = get_utask();
1538 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1539 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1540 " nestedness limit pid/tgid=%d/%d\n",
1541 current->pid, current->tgid);
1545 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1549 trampoline_vaddr = get_trampoline_vaddr();
1550 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1551 if (orig_ret_vaddr == -1)
1554 /* drop the entries invalidated by longjmp() */
1555 cleanup_return_instances(utask, regs);
1558 * We don't want to keep trampoline address in stack, rather keep the
1559 * original return address of first caller thru all the consequent
1560 * instances. This also makes breakpoint unwrapping easier.
1562 if (orig_ret_vaddr == trampoline_vaddr) {
1563 if (!utask->return_instances) {
1565 * This situation is not possible. Likely we have an
1566 * attack from user-space.
1568 uprobe_warn(current, "handle tail call");
1573 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1576 ri->uprobe = get_uprobe(uprobe);
1577 ri->func = instruction_pointer(regs);
1578 ri->stack = user_stack_pointer(regs);
1579 ri->orig_ret_vaddr = orig_ret_vaddr;
1580 ri->chained = chained;
1583 ri->next = utask->return_instances;
1584 utask->return_instances = ri;
1591 /* Prepare to single-step probed instruction out of line. */
1593 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1595 struct uprobe_task *utask;
1596 unsigned long xol_vaddr;
1599 utask = get_utask();
1603 xol_vaddr = xol_get_insn_slot(uprobe);
1607 utask->xol_vaddr = xol_vaddr;
1608 utask->vaddr = bp_vaddr;
1610 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1611 if (unlikely(err)) {
1612 xol_free_insn_slot(current);
1616 utask->active_uprobe = uprobe;
1617 utask->state = UTASK_SSTEP;
1622 * If we are singlestepping, then ensure this thread is not connected to
1623 * non-fatal signals until completion of singlestep. When xol insn itself
1624 * triggers the signal, restart the original insn even if the task is
1625 * already SIGKILL'ed (since coredump should report the correct ip). This
1626 * is even more important if the task has a handler for SIGSEGV/etc, The
1627 * _same_ instruction should be repeated again after return from the signal
1628 * handler, and SSTEP can never finish in this case.
1630 bool uprobe_deny_signal(void)
1632 struct task_struct *t = current;
1633 struct uprobe_task *utask = t->utask;
1635 if (likely(!utask || !utask->active_uprobe))
1638 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1640 if (signal_pending(t)) {
1641 spin_lock_irq(&t->sighand->siglock);
1642 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1643 spin_unlock_irq(&t->sighand->siglock);
1645 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1646 utask->state = UTASK_SSTEP_TRAPPED;
1647 set_tsk_thread_flag(t, TIF_UPROBE);
1654 static void mmf_recalc_uprobes(struct mm_struct *mm)
1656 struct vm_area_struct *vma;
1658 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1659 if (!valid_vma(vma, false))
1662 * This is not strictly accurate, we can race with
1663 * uprobe_unregister() and see the already removed
1664 * uprobe if delete_uprobe() was not yet called.
1665 * Or this uprobe can be filtered out.
1667 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1671 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1674 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1677 uprobe_opcode_t opcode;
1680 pagefault_disable();
1681 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1685 if (likely(result == 0))
1688 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1692 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1695 /* This needs to return true for any variant of the trap insn */
1696 return is_trap_insn(&opcode);
1699 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1701 struct mm_struct *mm = current->mm;
1702 struct uprobe *uprobe = NULL;
1703 struct vm_area_struct *vma;
1705 down_read(&mm->mmap_sem);
1706 vma = find_vma(mm, bp_vaddr);
1707 if (vma && vma->vm_start <= bp_vaddr) {
1708 if (valid_vma(vma, false)) {
1709 struct inode *inode = file_inode(vma->vm_file);
1710 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1712 uprobe = find_uprobe(inode, offset);
1716 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1721 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1722 mmf_recalc_uprobes(mm);
1723 up_read(&mm->mmap_sem);
1728 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1730 struct uprobe_consumer *uc;
1731 int remove = UPROBE_HANDLER_REMOVE;
1732 bool need_prep = false; /* prepare return uprobe, when needed */
1734 down_read(&uprobe->register_rwsem);
1735 for (uc = uprobe->consumers; uc; uc = uc->next) {
1739 rc = uc->handler(uc, regs);
1740 WARN(rc & ~UPROBE_HANDLER_MASK,
1741 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1744 if (uc->ret_handler)
1750 if (need_prep && !remove)
1751 prepare_uretprobe(uprobe, regs); /* put bp at return */
1753 if (remove && uprobe->consumers) {
1754 WARN_ON(!uprobe_is_active(uprobe));
1755 unapply_uprobe(uprobe, current->mm);
1757 up_read(&uprobe->register_rwsem);
1761 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1763 struct uprobe *uprobe = ri->uprobe;
1764 struct uprobe_consumer *uc;
1766 down_read(&uprobe->register_rwsem);
1767 for (uc = uprobe->consumers; uc; uc = uc->next) {
1768 if (uc->ret_handler)
1769 uc->ret_handler(uc, ri->func, regs);
1771 up_read(&uprobe->register_rwsem);
1774 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
1779 chained = ri->chained;
1780 ri = ri->next; /* can't be NULL if chained */
1786 static void handle_trampoline(struct pt_regs *regs)
1788 struct uprobe_task *utask;
1789 struct return_instance *ri, *next;
1792 utask = current->utask;
1796 ri = utask->return_instances;
1802 * We should throw out the frames invalidated by longjmp().
1803 * If this chain is valid, then the next one should be alive
1804 * or NULL; the latter case means that nobody but ri->func
1805 * could hit this trampoline on return. TODO: sigaltstack().
1807 next = find_next_ret_chain(ri);
1808 valid = !next || arch_uretprobe_is_alive(next, regs);
1810 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1813 handle_uretprobe_chain(ri, regs);
1814 ri = free_ret_instance(ri);
1816 } while (ri != next);
1819 utask->return_instances = ri;
1823 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
1824 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1828 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
1833 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, struct pt_regs *regs)
1839 * Run handler and ask thread to singlestep.
1840 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1842 static void handle_swbp(struct pt_regs *regs)
1844 struct uprobe *uprobe;
1845 unsigned long bp_vaddr;
1846 int uninitialized_var(is_swbp);
1848 bp_vaddr = uprobe_get_swbp_addr(regs);
1849 if (bp_vaddr == get_trampoline_vaddr())
1850 return handle_trampoline(regs);
1852 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1855 /* No matching uprobe; signal SIGTRAP. */
1856 send_sig(SIGTRAP, current, 0);
1859 * Either we raced with uprobe_unregister() or we can't
1860 * access this memory. The latter is only possible if
1861 * another thread plays with our ->mm. In both cases
1862 * we can simply restart. If this vma was unmapped we
1863 * can pretend this insn was not executed yet and get
1864 * the (correct) SIGSEGV after restart.
1866 instruction_pointer_set(regs, bp_vaddr);
1871 /* change it in advance for ->handler() and restart */
1872 instruction_pointer_set(regs, bp_vaddr);
1875 * TODO: move copy_insn/etc into _register and remove this hack.
1876 * After we hit the bp, _unregister + _register can install the
1877 * new and not-yet-analyzed uprobe at the same address, restart.
1879 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1880 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1883 /* Tracing handlers use ->utask to communicate with fetch methods */
1887 if (arch_uprobe_ignore(&uprobe->arch, regs))
1890 handler_chain(uprobe, regs);
1892 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1895 if (!pre_ssout(uprobe, regs, bp_vaddr))
1898 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1904 * Perform required fix-ups and disable singlestep.
1905 * Allow pending signals to take effect.
1907 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1909 struct uprobe *uprobe;
1912 uprobe = utask->active_uprobe;
1913 if (utask->state == UTASK_SSTEP_ACK)
1914 err = arch_uprobe_post_xol(&uprobe->arch, regs);
1915 else if (utask->state == UTASK_SSTEP_TRAPPED)
1916 arch_uprobe_abort_xol(&uprobe->arch, regs);
1921 utask->active_uprobe = NULL;
1922 utask->state = UTASK_RUNNING;
1923 xol_free_insn_slot(current);
1925 spin_lock_irq(¤t->sighand->siglock);
1926 recalc_sigpending(); /* see uprobe_deny_signal() */
1927 spin_unlock_irq(¤t->sighand->siglock);
1929 if (unlikely(err)) {
1930 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
1931 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1936 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1937 * allows the thread to return from interrupt. After that handle_swbp()
1938 * sets utask->active_uprobe.
1940 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1941 * and allows the thread to return from interrupt.
1943 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1944 * uprobe_notify_resume().
1946 void uprobe_notify_resume(struct pt_regs *regs)
1948 struct uprobe_task *utask;
1950 clear_thread_flag(TIF_UPROBE);
1952 utask = current->utask;
1953 if (utask && utask->active_uprobe)
1954 handle_singlestep(utask, regs);
1960 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1961 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1963 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1968 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
1969 (!current->utask || !current->utask->return_instances))
1972 set_thread_flag(TIF_UPROBE);
1977 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1978 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1980 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1982 struct uprobe_task *utask = current->utask;
1984 if (!current->mm || !utask || !utask->active_uprobe)
1985 /* task is currently not uprobed */
1988 utask->state = UTASK_SSTEP_ACK;
1989 set_thread_flag(TIF_UPROBE);
1993 static struct notifier_block uprobe_exception_nb = {
1994 .notifier_call = arch_uprobe_exception_notify,
1995 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1998 static int __init init_uprobes(void)
2002 for (i = 0; i < UPROBES_HASH_SZ; i++)
2003 mutex_init(&uprobes_mmap_mutex[i]);
2005 if (percpu_init_rwsem(&dup_mmap_sem))
2008 return register_die_notifier(&uprobe_exception_nb);
2010 __initcall(init_uprobes);