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;
89 struct return_instance {
90 struct uprobe *uprobe;
92 unsigned long orig_ret_vaddr; /* original return address */
93 bool chained; /* true, if instance is nested */
95 struct return_instance *next; /* keep as stack */
99 * Execute out of line area: anonymous executable mapping installed
100 * by the probed task to execute the copy of the original instruction
101 * mangled by set_swbp().
103 * On a breakpoint hit, thread contests for a slot. It frees the
104 * slot after singlestep. Currently a fixed number of slots are
108 wait_queue_head_t wq; /* if all slots are busy */
109 atomic_t slot_count; /* number of in-use slots */
110 unsigned long *bitmap; /* 0 = free slot */
114 * We keep the vma's vm_start rather than a pointer to the vma
115 * itself. The probed process or a naughty kernel module could make
116 * the vma go away, and we must handle that reasonably gracefully.
118 unsigned long vaddr; /* Page(s) of instruction slots */
122 * valid_vma: Verify if the specified vma is an executable vma
123 * Relax restrictions while unregistering: vm_flags might have
124 * changed after breakpoint was inserted.
125 * - is_register: indicates if we are in register context.
126 * - Return 1 if the specified virtual address is in an
129 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
131 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
136 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
139 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
141 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
144 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
146 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
150 * __replace_page - replace page in vma by new page.
151 * based on replace_page in mm/ksm.c
153 * @vma: vma that holds the pte pointing to page
154 * @addr: address the old @page is mapped at
155 * @page: the cowed page we are replacing by kpage
156 * @kpage: the modified page we replace page by
158 * Returns 0 on success, -EFAULT on failure.
160 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
161 struct page *page, struct page *kpage)
163 struct mm_struct *mm = vma->vm_mm;
167 /* For mmu_notifiers */
168 const unsigned long mmun_start = addr;
169 const unsigned long mmun_end = addr + PAGE_SIZE;
170 struct mem_cgroup *memcg;
172 err = mem_cgroup_try_charge(kpage, vma->vm_mm, GFP_KERNEL, &memcg);
176 /* For try_to_free_swap() and munlock_vma_page() below */
179 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
181 ptep = page_check_address(page, mm, addr, &ptl, 0);
186 page_add_new_anon_rmap(kpage, vma, addr);
187 mem_cgroup_commit_charge(kpage, memcg, false);
188 lru_cache_add_active_or_unevictable(kpage, vma);
190 if (!PageAnon(page)) {
191 dec_mm_counter(mm, MM_FILEPAGES);
192 inc_mm_counter(mm, MM_ANONPAGES);
195 flush_cache_page(vma, addr, pte_pfn(*ptep));
196 ptep_clear_flush_notify(vma, addr, ptep);
197 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
199 page_remove_rmap(page);
200 if (!page_mapped(page))
201 try_to_free_swap(page);
202 pte_unmap_unlock(ptep, ptl);
204 if (vma->vm_flags & VM_LOCKED)
205 munlock_vma_page(page);
210 mem_cgroup_cancel_charge(kpage, memcg);
211 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
217 * is_swbp_insn - check if instruction is breakpoint instruction.
218 * @insn: instruction to be checked.
219 * Default implementation of is_swbp_insn
220 * Returns true if @insn is a breakpoint instruction.
222 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
224 return *insn == UPROBE_SWBP_INSN;
228 * is_trap_insn - check if instruction is breakpoint instruction.
229 * @insn: instruction to be checked.
230 * Default implementation of is_trap_insn
231 * Returns true if @insn is a breakpoint instruction.
233 * This function is needed for the case where an architecture has multiple
234 * trap instructions (like powerpc).
236 bool __weak is_trap_insn(uprobe_opcode_t *insn)
238 return is_swbp_insn(insn);
241 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
243 void *kaddr = kmap_atomic(page);
244 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
245 kunmap_atomic(kaddr);
248 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
250 void *kaddr = kmap_atomic(page);
251 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
252 kunmap_atomic(kaddr);
255 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
257 uprobe_opcode_t old_opcode;
261 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
262 * We do not check if it is any other 'trap variant' which could
263 * be conditional trap instruction such as the one powerpc supports.
265 * The logic is that we do not care if the underlying instruction
266 * is a trap variant; uprobes always wins over any other (gdb)
269 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
270 is_swbp = is_swbp_insn(&old_opcode);
272 if (is_swbp_insn(new_opcode)) {
273 if (is_swbp) /* register: already installed? */
276 if (!is_swbp) /* unregister: was it changed by us? */
285 * Expect the breakpoint instruction to be the smallest size instruction for
286 * the architecture. If an arch has variable length instruction and the
287 * breakpoint instruction is not of the smallest length instruction
288 * supported by that architecture then we need to modify is_trap_at_addr and
289 * uprobe_write_opcode accordingly. This would never be a problem for archs
290 * that have fixed length instructions.
292 * uprobe_write_opcode - write the opcode at a given virtual address.
293 * @mm: the probed process address space.
294 * @vaddr: the virtual address to store the opcode.
295 * @opcode: opcode to be written at @vaddr.
297 * Called with mm->mmap_sem held for write.
298 * Return 0 (success) or a negative errno.
300 int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr,
301 uprobe_opcode_t opcode)
303 struct page *old_page, *new_page;
304 struct vm_area_struct *vma;
308 /* Read the page with vaddr into memory */
309 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
313 ret = verify_opcode(old_page, vaddr, &opcode);
317 ret = anon_vma_prepare(vma);
322 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
326 __SetPageUptodate(new_page);
327 copy_highpage(new_page, old_page);
328 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
330 ret = __replace_page(vma, vaddr, old_page, new_page);
331 page_cache_release(new_page);
335 if (unlikely(ret == -EAGAIN))
341 * set_swbp - store breakpoint at a given address.
342 * @auprobe: arch specific probepoint information.
343 * @mm: the probed process address space.
344 * @vaddr: the virtual address to insert the opcode.
346 * For mm @mm, store the breakpoint instruction at @vaddr.
347 * Return 0 (success) or a negative errno.
349 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
351 return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
355 * set_orig_insn - Restore the original instruction.
356 * @mm: the probed process address space.
357 * @auprobe: arch specific probepoint information.
358 * @vaddr: the virtual address to insert the opcode.
360 * For mm @mm, restore the original opcode (opcode) at @vaddr.
361 * Return 0 (success) or a negative errno.
364 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
366 return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn);
369 static struct uprobe *get_uprobe(struct uprobe *uprobe)
371 atomic_inc(&uprobe->ref);
375 static void put_uprobe(struct uprobe *uprobe)
377 if (atomic_dec_and_test(&uprobe->ref))
381 static int match_uprobe(struct uprobe *l, struct uprobe *r)
383 if (l->inode < r->inode)
386 if (l->inode > r->inode)
389 if (l->offset < r->offset)
392 if (l->offset > r->offset)
398 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
400 struct uprobe u = { .inode = inode, .offset = offset };
401 struct rb_node *n = uprobes_tree.rb_node;
402 struct uprobe *uprobe;
406 uprobe = rb_entry(n, struct uprobe, rb_node);
407 match = match_uprobe(&u, uprobe);
409 return get_uprobe(uprobe);
420 * Find a uprobe corresponding to a given inode:offset
421 * Acquires uprobes_treelock
423 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
425 struct uprobe *uprobe;
427 spin_lock(&uprobes_treelock);
428 uprobe = __find_uprobe(inode, offset);
429 spin_unlock(&uprobes_treelock);
434 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
436 struct rb_node **p = &uprobes_tree.rb_node;
437 struct rb_node *parent = NULL;
443 u = rb_entry(parent, struct uprobe, rb_node);
444 match = match_uprobe(uprobe, u);
446 return get_uprobe(u);
449 p = &parent->rb_left;
451 p = &parent->rb_right;
456 rb_link_node(&uprobe->rb_node, parent, p);
457 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
458 /* get access + creation ref */
459 atomic_set(&uprobe->ref, 2);
465 * Acquire uprobes_treelock.
466 * Matching uprobe already exists in rbtree;
467 * increment (access refcount) and return the matching uprobe.
469 * No matching uprobe; insert the uprobe in rb_tree;
470 * get a double refcount (access + creation) and return NULL.
472 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
476 spin_lock(&uprobes_treelock);
477 u = __insert_uprobe(uprobe);
478 spin_unlock(&uprobes_treelock);
483 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
485 struct uprobe *uprobe, *cur_uprobe;
487 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
491 uprobe->inode = igrab(inode);
492 uprobe->offset = offset;
493 init_rwsem(&uprobe->register_rwsem);
494 init_rwsem(&uprobe->consumer_rwsem);
496 /* add to uprobes_tree, sorted on inode:offset */
497 cur_uprobe = insert_uprobe(uprobe);
498 /* a uprobe exists for this inode:offset combination */
508 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
510 down_write(&uprobe->consumer_rwsem);
511 uc->next = uprobe->consumers;
512 uprobe->consumers = uc;
513 up_write(&uprobe->consumer_rwsem);
517 * For uprobe @uprobe, delete the consumer @uc.
518 * Return true if the @uc is deleted successfully
521 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
523 struct uprobe_consumer **con;
526 down_write(&uprobe->consumer_rwsem);
527 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
534 up_write(&uprobe->consumer_rwsem);
539 static int __copy_insn(struct address_space *mapping, struct file *filp,
540 void *insn, int nbytes, loff_t offset)
544 * Ensure that the page that has the original instruction is populated
545 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
546 * see uprobe_register().
548 if (mapping->a_ops->readpage)
549 page = read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT, filp);
551 page = shmem_read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT);
553 return PTR_ERR(page);
555 copy_from_page(page, offset, insn, nbytes);
556 page_cache_release(page);
561 static int copy_insn(struct uprobe *uprobe, struct file *filp)
563 struct address_space *mapping = uprobe->inode->i_mapping;
564 loff_t offs = uprobe->offset;
565 void *insn = &uprobe->arch.insn;
566 int size = sizeof(uprobe->arch.insn);
569 /* Copy only available bytes, -EIO if nothing was read */
571 if (offs >= i_size_read(uprobe->inode))
574 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
575 err = __copy_insn(mapping, filp, insn, len, offs);
587 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
588 struct mm_struct *mm, unsigned long vaddr)
592 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
595 /* TODO: move this into _register, until then we abuse this sem. */
596 down_write(&uprobe->consumer_rwsem);
597 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
600 ret = copy_insn(uprobe, file);
605 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
608 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
612 /* uprobe_write_opcode() assumes we don't cross page boundary */
613 BUG_ON((uprobe->offset & ~PAGE_MASK) +
614 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
616 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
617 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
620 up_write(&uprobe->consumer_rwsem);
625 static inline bool consumer_filter(struct uprobe_consumer *uc,
626 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
628 return !uc->filter || uc->filter(uc, ctx, mm);
631 static bool filter_chain(struct uprobe *uprobe,
632 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
634 struct uprobe_consumer *uc;
637 down_read(&uprobe->consumer_rwsem);
638 for (uc = uprobe->consumers; uc; uc = uc->next) {
639 ret = consumer_filter(uc, ctx, mm);
643 up_read(&uprobe->consumer_rwsem);
649 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
650 struct vm_area_struct *vma, unsigned long vaddr)
655 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
660 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
661 * the task can hit this breakpoint right after __replace_page().
663 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
665 set_bit(MMF_HAS_UPROBES, &mm->flags);
667 ret = set_swbp(&uprobe->arch, mm, vaddr);
669 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
670 else if (first_uprobe)
671 clear_bit(MMF_HAS_UPROBES, &mm->flags);
677 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
679 set_bit(MMF_RECALC_UPROBES, &mm->flags);
680 return set_orig_insn(&uprobe->arch, mm, vaddr);
683 static inline bool uprobe_is_active(struct uprobe *uprobe)
685 return !RB_EMPTY_NODE(&uprobe->rb_node);
688 * There could be threads that have already hit the breakpoint. They
689 * will recheck the current insn and restart if find_uprobe() fails.
690 * See find_active_uprobe().
692 static void delete_uprobe(struct uprobe *uprobe)
694 if (WARN_ON(!uprobe_is_active(uprobe)))
697 spin_lock(&uprobes_treelock);
698 rb_erase(&uprobe->rb_node, &uprobes_tree);
699 spin_unlock(&uprobes_treelock);
700 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
706 struct map_info *next;
707 struct mm_struct *mm;
711 static inline struct map_info *free_map_info(struct map_info *info)
713 struct map_info *next = info->next;
718 static struct map_info *
719 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
721 unsigned long pgoff = offset >> PAGE_SHIFT;
722 struct vm_area_struct *vma;
723 struct map_info *curr = NULL;
724 struct map_info *prev = NULL;
725 struct map_info *info;
729 i_mmap_lock_read(mapping);
730 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
731 if (!valid_vma(vma, is_register))
734 if (!prev && !more) {
736 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
737 * reclaim. This is optimistic, no harm done if it fails.
739 prev = kmalloc(sizeof(struct map_info),
740 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
749 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
757 info->mm = vma->vm_mm;
758 info->vaddr = offset_to_vaddr(vma, offset);
760 i_mmap_unlock_read(mapping);
772 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
774 curr = ERR_PTR(-ENOMEM);
784 prev = free_map_info(prev);
789 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
791 bool is_register = !!new;
792 struct map_info *info;
795 percpu_down_write(&dup_mmap_sem);
796 info = build_map_info(uprobe->inode->i_mapping,
797 uprobe->offset, is_register);
804 struct mm_struct *mm = info->mm;
805 struct vm_area_struct *vma;
807 if (err && is_register)
810 down_write(&mm->mmap_sem);
811 vma = find_vma(mm, info->vaddr);
812 if (!vma || !valid_vma(vma, is_register) ||
813 file_inode(vma->vm_file) != uprobe->inode)
816 if (vma->vm_start > info->vaddr ||
817 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
821 /* consult only the "caller", new consumer. */
822 if (consumer_filter(new,
823 UPROBE_FILTER_REGISTER, mm))
824 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
825 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
826 if (!filter_chain(uprobe,
827 UPROBE_FILTER_UNREGISTER, mm))
828 err |= remove_breakpoint(uprobe, mm, info->vaddr);
832 up_write(&mm->mmap_sem);
835 info = free_map_info(info);
838 percpu_up_write(&dup_mmap_sem);
842 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
844 consumer_add(uprobe, uc);
845 return register_for_each_vma(uprobe, uc);
848 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
852 if (WARN_ON(!consumer_del(uprobe, uc)))
855 err = register_for_each_vma(uprobe, NULL);
856 /* TODO : cant unregister? schedule a worker thread */
857 if (!uprobe->consumers && !err)
858 delete_uprobe(uprobe);
862 * uprobe_register - register a probe
863 * @inode: the file in which the probe has to be placed.
864 * @offset: offset from the start of the file.
865 * @uc: information on howto handle the probe..
867 * Apart from the access refcount, uprobe_register() takes a creation
868 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
869 * inserted into the rbtree (i.e first consumer for a @inode:@offset
870 * tuple). Creation refcount stops uprobe_unregister from freeing the
871 * @uprobe even before the register operation is complete. Creation
872 * refcount is released when the last @uc for the @uprobe
875 * Return errno if it cannot successully install probes
876 * else return 0 (success)
878 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
880 struct uprobe *uprobe;
883 /* Uprobe must have at least one set consumer */
884 if (!uc->handler && !uc->ret_handler)
887 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
888 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
890 /* Racy, just to catch the obvious mistakes */
891 if (offset > i_size_read(inode))
895 uprobe = alloc_uprobe(inode, offset);
899 * We can race with uprobe_unregister()->delete_uprobe().
900 * Check uprobe_is_active() and retry if it is false.
902 down_write(&uprobe->register_rwsem);
904 if (likely(uprobe_is_active(uprobe))) {
905 ret = __uprobe_register(uprobe, uc);
907 __uprobe_unregister(uprobe, uc);
909 up_write(&uprobe->register_rwsem);
912 if (unlikely(ret == -EAGAIN))
916 EXPORT_SYMBOL_GPL(uprobe_register);
919 * uprobe_apply - unregister a already registered probe.
920 * @inode: the file in which the probe has to be removed.
921 * @offset: offset from the start of the file.
922 * @uc: consumer which wants to add more or remove some breakpoints
923 * @add: add or remove the breakpoints
925 int uprobe_apply(struct inode *inode, loff_t offset,
926 struct uprobe_consumer *uc, bool add)
928 struct uprobe *uprobe;
929 struct uprobe_consumer *con;
932 uprobe = find_uprobe(inode, offset);
933 if (WARN_ON(!uprobe))
936 down_write(&uprobe->register_rwsem);
937 for (con = uprobe->consumers; con && con != uc ; con = con->next)
940 ret = register_for_each_vma(uprobe, add ? uc : NULL);
941 up_write(&uprobe->register_rwsem);
948 * uprobe_unregister - unregister a already registered probe.
949 * @inode: the file in which the probe has to be removed.
950 * @offset: offset from the start of the file.
951 * @uc: identify which probe if multiple probes are colocated.
953 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
955 struct uprobe *uprobe;
957 uprobe = find_uprobe(inode, offset);
958 if (WARN_ON(!uprobe))
961 down_write(&uprobe->register_rwsem);
962 __uprobe_unregister(uprobe, uc);
963 up_write(&uprobe->register_rwsem);
966 EXPORT_SYMBOL_GPL(uprobe_unregister);
968 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
970 struct vm_area_struct *vma;
973 down_read(&mm->mmap_sem);
974 for (vma = mm->mmap; vma; vma = vma->vm_next) {
978 if (!valid_vma(vma, false) ||
979 file_inode(vma->vm_file) != uprobe->inode)
982 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
983 if (uprobe->offset < offset ||
984 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
987 vaddr = offset_to_vaddr(vma, uprobe->offset);
988 err |= remove_breakpoint(uprobe, mm, vaddr);
990 up_read(&mm->mmap_sem);
995 static struct rb_node *
996 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
998 struct rb_node *n = uprobes_tree.rb_node;
1001 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1003 if (inode < u->inode) {
1005 } else if (inode > u->inode) {
1008 if (max < u->offset)
1010 else if (min > u->offset)
1021 * For a given range in vma, build a list of probes that need to be inserted.
1023 static void build_probe_list(struct inode *inode,
1024 struct vm_area_struct *vma,
1025 unsigned long start, unsigned long end,
1026 struct list_head *head)
1029 struct rb_node *n, *t;
1032 INIT_LIST_HEAD(head);
1033 min = vaddr_to_offset(vma, start);
1034 max = min + (end - start) - 1;
1036 spin_lock(&uprobes_treelock);
1037 n = find_node_in_range(inode, min, max);
1039 for (t = n; t; t = rb_prev(t)) {
1040 u = rb_entry(t, struct uprobe, rb_node);
1041 if (u->inode != inode || u->offset < min)
1043 list_add(&u->pending_list, head);
1046 for (t = n; (t = rb_next(t)); ) {
1047 u = rb_entry(t, struct uprobe, rb_node);
1048 if (u->inode != inode || u->offset > max)
1050 list_add(&u->pending_list, head);
1054 spin_unlock(&uprobes_treelock);
1058 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1060 * Currently we ignore all errors and always return 0, the callers
1061 * can't handle the failure anyway.
1063 int uprobe_mmap(struct vm_area_struct *vma)
1065 struct list_head tmp_list;
1066 struct uprobe *uprobe, *u;
1067 struct inode *inode;
1069 if (no_uprobe_events() || !valid_vma(vma, true))
1072 inode = file_inode(vma->vm_file);
1076 mutex_lock(uprobes_mmap_hash(inode));
1077 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1079 * We can race with uprobe_unregister(), this uprobe can be already
1080 * removed. But in this case filter_chain() must return false, all
1081 * consumers have gone away.
1083 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1084 if (!fatal_signal_pending(current) &&
1085 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1086 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1087 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1091 mutex_unlock(uprobes_mmap_hash(inode));
1097 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1100 struct inode *inode;
1103 inode = file_inode(vma->vm_file);
1105 min = vaddr_to_offset(vma, start);
1106 max = min + (end - start) - 1;
1108 spin_lock(&uprobes_treelock);
1109 n = find_node_in_range(inode, min, max);
1110 spin_unlock(&uprobes_treelock);
1116 * Called in context of a munmap of a vma.
1118 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1120 if (no_uprobe_events() || !valid_vma(vma, false))
1123 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1126 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1127 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1130 if (vma_has_uprobes(vma, start, end))
1131 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1134 /* Slot allocation for XOL */
1135 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1137 int ret = -EALREADY;
1139 down_write(&mm->mmap_sem);
1140 if (mm->uprobes_state.xol_area)
1144 /* Try to map as high as possible, this is only a hint. */
1145 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1147 if (area->vaddr & ~PAGE_MASK) {
1153 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1154 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1158 smp_wmb(); /* pairs with get_xol_area() */
1159 mm->uprobes_state.xol_area = area;
1161 up_write(&mm->mmap_sem);
1166 static struct xol_area *__create_xol_area(unsigned long vaddr)
1168 struct mm_struct *mm = current->mm;
1169 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1170 struct xol_area *area;
1172 area = kmalloc(sizeof(*area), GFP_KERNEL);
1173 if (unlikely(!area))
1176 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1180 area->page = alloc_page(GFP_HIGHUSER);
1184 area->vaddr = vaddr;
1185 init_waitqueue_head(&area->wq);
1186 /* Reserve the 1st slot for get_trampoline_vaddr() */
1187 set_bit(0, area->bitmap);
1188 atomic_set(&area->slot_count, 1);
1189 copy_to_page(area->page, 0, &insn, UPROBE_SWBP_INSN_SIZE);
1191 if (!xol_add_vma(mm, area))
1194 __free_page(area->page);
1196 kfree(area->bitmap);
1204 * get_xol_area - Allocate process's xol_area if necessary.
1205 * This area will be used for storing instructions for execution out of line.
1207 * Returns the allocated area or NULL.
1209 static struct xol_area *get_xol_area(void)
1211 struct mm_struct *mm = current->mm;
1212 struct xol_area *area;
1214 if (!mm->uprobes_state.xol_area)
1215 __create_xol_area(0);
1217 area = mm->uprobes_state.xol_area;
1218 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1223 * uprobe_clear_state - Free the area allocated for slots.
1225 void uprobe_clear_state(struct mm_struct *mm)
1227 struct xol_area *area = mm->uprobes_state.xol_area;
1232 put_page(area->page);
1233 kfree(area->bitmap);
1237 void uprobe_start_dup_mmap(void)
1239 percpu_down_read(&dup_mmap_sem);
1242 void uprobe_end_dup_mmap(void)
1244 percpu_up_read(&dup_mmap_sem);
1247 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1249 newmm->uprobes_state.xol_area = NULL;
1251 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1252 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1253 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1254 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1259 * - search for a free slot.
1261 static unsigned long xol_take_insn_slot(struct xol_area *area)
1263 unsigned long slot_addr;
1267 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1268 if (slot_nr < UINSNS_PER_PAGE) {
1269 if (!test_and_set_bit(slot_nr, area->bitmap))
1272 slot_nr = UINSNS_PER_PAGE;
1275 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1276 } while (slot_nr >= UINSNS_PER_PAGE);
1278 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1279 atomic_inc(&area->slot_count);
1285 * xol_get_insn_slot - allocate a slot for xol.
1286 * Returns the allocated slot address or 0.
1288 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1290 struct xol_area *area;
1291 unsigned long xol_vaddr;
1293 area = get_xol_area();
1297 xol_vaddr = xol_take_insn_slot(area);
1298 if (unlikely(!xol_vaddr))
1301 arch_uprobe_copy_ixol(area->page, xol_vaddr,
1302 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1308 * xol_free_insn_slot - If slot was earlier allocated by
1309 * @xol_get_insn_slot(), make the slot available for
1310 * subsequent requests.
1312 static void xol_free_insn_slot(struct task_struct *tsk)
1314 struct xol_area *area;
1315 unsigned long vma_end;
1316 unsigned long slot_addr;
1318 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1321 slot_addr = tsk->utask->xol_vaddr;
1322 if (unlikely(!slot_addr))
1325 area = tsk->mm->uprobes_state.xol_area;
1326 vma_end = area->vaddr + PAGE_SIZE;
1327 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1328 unsigned long offset;
1331 offset = slot_addr - area->vaddr;
1332 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1333 if (slot_nr >= UINSNS_PER_PAGE)
1336 clear_bit(slot_nr, area->bitmap);
1337 atomic_dec(&area->slot_count);
1338 if (waitqueue_active(&area->wq))
1341 tsk->utask->xol_vaddr = 0;
1345 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1346 void *src, unsigned long len)
1348 /* Initialize the slot */
1349 copy_to_page(page, vaddr, src, len);
1352 * We probably need flush_icache_user_range() but it needs vma.
1353 * This should work on most of architectures by default. If
1354 * architecture needs to do something different it can define
1355 * its own version of the function.
1357 flush_dcache_page(page);
1361 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1362 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1364 * Return the address of the breakpoint instruction.
1366 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1368 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1371 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1373 struct uprobe_task *utask = current->utask;
1375 if (unlikely(utask && utask->active_uprobe))
1376 return utask->vaddr;
1378 return instruction_pointer(regs);
1381 static struct return_instance *free_ret_instance(struct return_instance *ri)
1383 struct return_instance *next = ri->next;
1384 put_uprobe(ri->uprobe);
1390 * Called with no locks held.
1391 * Called in context of a exiting or a exec-ing thread.
1393 void uprobe_free_utask(struct task_struct *t)
1395 struct uprobe_task *utask = t->utask;
1396 struct return_instance *ri;
1401 if (utask->active_uprobe)
1402 put_uprobe(utask->active_uprobe);
1404 ri = utask->return_instances;
1406 ri = free_ret_instance(ri);
1408 xol_free_insn_slot(t);
1414 * Allocate a uprobe_task object for the task if if necessary.
1415 * Called when the thread hits a breakpoint.
1418 * - pointer to new uprobe_task on success
1421 static struct uprobe_task *get_utask(void)
1423 if (!current->utask)
1424 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1425 return current->utask;
1428 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1430 struct uprobe_task *n_utask;
1431 struct return_instance **p, *o, *n;
1433 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1438 p = &n_utask->return_instances;
1439 for (o = o_utask->return_instances; o; o = o->next) {
1440 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1445 get_uprobe(n->uprobe);
1456 static void uprobe_warn(struct task_struct *t, const char *msg)
1458 pr_warn("uprobe: %s:%d failed to %s\n",
1459 current->comm, current->pid, msg);
1462 static void dup_xol_work(struct callback_head *work)
1464 if (current->flags & PF_EXITING)
1467 if (!__create_xol_area(current->utask->dup_xol_addr))
1468 uprobe_warn(current, "dup xol area");
1472 * Called in context of a new clone/fork from copy_process.
1474 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1476 struct uprobe_task *utask = current->utask;
1477 struct mm_struct *mm = current->mm;
1478 struct xol_area *area;
1482 if (!utask || !utask->return_instances)
1485 if (mm == t->mm && !(flags & CLONE_VFORK))
1488 if (dup_utask(t, utask))
1489 return uprobe_warn(t, "dup ret instances");
1491 /* The task can fork() after dup_xol_work() fails */
1492 area = mm->uprobes_state.xol_area;
1494 return uprobe_warn(t, "dup xol area");
1499 t->utask->dup_xol_addr = area->vaddr;
1500 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1501 task_work_add(t, &t->utask->dup_xol_work, true);
1505 * Current area->vaddr notion assume the trampoline address is always
1506 * equal area->vaddr.
1508 * Returns -1 in case the xol_area is not allocated.
1510 static unsigned long get_trampoline_vaddr(void)
1512 struct xol_area *area;
1513 unsigned long trampoline_vaddr = -1;
1515 area = current->mm->uprobes_state.xol_area;
1516 smp_read_barrier_depends();
1518 trampoline_vaddr = area->vaddr;
1520 return trampoline_vaddr;
1523 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1525 struct return_instance *ri;
1526 struct uprobe_task *utask;
1527 unsigned long orig_ret_vaddr, trampoline_vaddr;
1528 bool chained = false;
1530 if (!get_xol_area())
1533 utask = get_utask();
1537 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1538 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1539 " nestedness limit pid/tgid=%d/%d\n",
1540 current->pid, current->tgid);
1544 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1548 trampoline_vaddr = get_trampoline_vaddr();
1549 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1550 if (orig_ret_vaddr == -1)
1554 * We don't want to keep trampoline address in stack, rather keep the
1555 * original return address of first caller thru all the consequent
1556 * instances. This also makes breakpoint unwrapping easier.
1558 if (orig_ret_vaddr == trampoline_vaddr) {
1559 if (!utask->return_instances) {
1561 * This situation is not possible. Likely we have an
1562 * attack from user-space.
1564 uprobe_warn(current, "handle tail call");
1569 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1572 ri->uprobe = get_uprobe(uprobe);
1573 ri->func = instruction_pointer(regs);
1574 ri->orig_ret_vaddr = orig_ret_vaddr;
1575 ri->chained = chained;
1578 ri->next = utask->return_instances;
1579 utask->return_instances = ri;
1586 /* Prepare to single-step probed instruction out of line. */
1588 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1590 struct uprobe_task *utask;
1591 unsigned long xol_vaddr;
1594 utask = get_utask();
1598 xol_vaddr = xol_get_insn_slot(uprobe);
1602 utask->xol_vaddr = xol_vaddr;
1603 utask->vaddr = bp_vaddr;
1605 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1606 if (unlikely(err)) {
1607 xol_free_insn_slot(current);
1611 utask->active_uprobe = uprobe;
1612 utask->state = UTASK_SSTEP;
1617 * If we are singlestepping, then ensure this thread is not connected to
1618 * non-fatal signals until completion of singlestep. When xol insn itself
1619 * triggers the signal, restart the original insn even if the task is
1620 * already SIGKILL'ed (since coredump should report the correct ip). This
1621 * is even more important if the task has a handler for SIGSEGV/etc, The
1622 * _same_ instruction should be repeated again after return from the signal
1623 * handler, and SSTEP can never finish in this case.
1625 bool uprobe_deny_signal(void)
1627 struct task_struct *t = current;
1628 struct uprobe_task *utask = t->utask;
1630 if (likely(!utask || !utask->active_uprobe))
1633 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1635 if (signal_pending(t)) {
1636 spin_lock_irq(&t->sighand->siglock);
1637 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1638 spin_unlock_irq(&t->sighand->siglock);
1640 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1641 utask->state = UTASK_SSTEP_TRAPPED;
1642 set_tsk_thread_flag(t, TIF_UPROBE);
1649 static void mmf_recalc_uprobes(struct mm_struct *mm)
1651 struct vm_area_struct *vma;
1653 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1654 if (!valid_vma(vma, false))
1657 * This is not strictly accurate, we can race with
1658 * uprobe_unregister() and see the already removed
1659 * uprobe if delete_uprobe() was not yet called.
1660 * Or this uprobe can be filtered out.
1662 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1666 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1669 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1672 uprobe_opcode_t opcode;
1675 pagefault_disable();
1676 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1680 if (likely(result == 0))
1683 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1687 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1690 /* This needs to return true for any variant of the trap insn */
1691 return is_trap_insn(&opcode);
1694 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1696 struct mm_struct *mm = current->mm;
1697 struct uprobe *uprobe = NULL;
1698 struct vm_area_struct *vma;
1700 down_read(&mm->mmap_sem);
1701 vma = find_vma(mm, bp_vaddr);
1702 if (vma && vma->vm_start <= bp_vaddr) {
1703 if (valid_vma(vma, false)) {
1704 struct inode *inode = file_inode(vma->vm_file);
1705 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1707 uprobe = find_uprobe(inode, offset);
1711 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1716 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1717 mmf_recalc_uprobes(mm);
1718 up_read(&mm->mmap_sem);
1723 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1725 struct uprobe_consumer *uc;
1726 int remove = UPROBE_HANDLER_REMOVE;
1727 bool need_prep = false; /* prepare return uprobe, when needed */
1729 down_read(&uprobe->register_rwsem);
1730 for (uc = uprobe->consumers; uc; uc = uc->next) {
1734 rc = uc->handler(uc, regs);
1735 WARN(rc & ~UPROBE_HANDLER_MASK,
1736 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1739 if (uc->ret_handler)
1745 if (need_prep && !remove)
1746 prepare_uretprobe(uprobe, regs); /* put bp at return */
1748 if (remove && uprobe->consumers) {
1749 WARN_ON(!uprobe_is_active(uprobe));
1750 unapply_uprobe(uprobe, current->mm);
1752 up_read(&uprobe->register_rwsem);
1756 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1758 struct uprobe *uprobe = ri->uprobe;
1759 struct uprobe_consumer *uc;
1761 down_read(&uprobe->register_rwsem);
1762 for (uc = uprobe->consumers; uc; uc = uc->next) {
1763 if (uc->ret_handler)
1764 uc->ret_handler(uc, ri->func, regs);
1766 up_read(&uprobe->register_rwsem);
1769 static void handle_trampoline(struct pt_regs *regs)
1771 struct uprobe_task *utask;
1772 struct return_instance *ri;
1775 utask = current->utask;
1779 ri = utask->return_instances;
1784 * TODO: we should throw out return_instance's invalidated by
1785 * longjmp(), currently we assume that the probed function always
1788 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1791 handle_uretprobe_chain(ri, regs);
1793 chained = ri->chained;
1794 ri = free_ret_instance(ri);
1802 utask->return_instances = ri;
1806 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
1807 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1811 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
1817 * Run handler and ask thread to singlestep.
1818 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1820 static void handle_swbp(struct pt_regs *regs)
1822 struct uprobe *uprobe;
1823 unsigned long bp_vaddr;
1824 int uninitialized_var(is_swbp);
1826 bp_vaddr = uprobe_get_swbp_addr(regs);
1827 if (bp_vaddr == get_trampoline_vaddr())
1828 return handle_trampoline(regs);
1830 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1833 /* No matching uprobe; signal SIGTRAP. */
1834 send_sig(SIGTRAP, current, 0);
1837 * Either we raced with uprobe_unregister() or we can't
1838 * access this memory. The latter is only possible if
1839 * another thread plays with our ->mm. In both cases
1840 * we can simply restart. If this vma was unmapped we
1841 * can pretend this insn was not executed yet and get
1842 * the (correct) SIGSEGV after restart.
1844 instruction_pointer_set(regs, bp_vaddr);
1849 /* change it in advance for ->handler() and restart */
1850 instruction_pointer_set(regs, bp_vaddr);
1853 * TODO: move copy_insn/etc into _register and remove this hack.
1854 * After we hit the bp, _unregister + _register can install the
1855 * new and not-yet-analyzed uprobe at the same address, restart.
1857 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1858 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1861 /* Tracing handlers use ->utask to communicate with fetch methods */
1865 if (arch_uprobe_ignore(&uprobe->arch, regs))
1868 handler_chain(uprobe, regs);
1870 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1873 if (!pre_ssout(uprobe, regs, bp_vaddr))
1876 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1882 * Perform required fix-ups and disable singlestep.
1883 * Allow pending signals to take effect.
1885 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1887 struct uprobe *uprobe;
1890 uprobe = utask->active_uprobe;
1891 if (utask->state == UTASK_SSTEP_ACK)
1892 err = arch_uprobe_post_xol(&uprobe->arch, regs);
1893 else if (utask->state == UTASK_SSTEP_TRAPPED)
1894 arch_uprobe_abort_xol(&uprobe->arch, regs);
1899 utask->active_uprobe = NULL;
1900 utask->state = UTASK_RUNNING;
1901 xol_free_insn_slot(current);
1903 spin_lock_irq(¤t->sighand->siglock);
1904 recalc_sigpending(); /* see uprobe_deny_signal() */
1905 spin_unlock_irq(¤t->sighand->siglock);
1907 if (unlikely(err)) {
1908 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
1909 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1914 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1915 * allows the thread to return from interrupt. After that handle_swbp()
1916 * sets utask->active_uprobe.
1918 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1919 * and allows the thread to return from interrupt.
1921 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1922 * uprobe_notify_resume().
1924 void uprobe_notify_resume(struct pt_regs *regs)
1926 struct uprobe_task *utask;
1928 clear_thread_flag(TIF_UPROBE);
1930 utask = current->utask;
1931 if (utask && utask->active_uprobe)
1932 handle_singlestep(utask, regs);
1938 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1939 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1941 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1946 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
1947 (!current->utask || !current->utask->return_instances))
1950 set_thread_flag(TIF_UPROBE);
1955 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1956 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1958 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1960 struct uprobe_task *utask = current->utask;
1962 if (!current->mm || !utask || !utask->active_uprobe)
1963 /* task is currently not uprobed */
1966 utask->state = UTASK_SSTEP_ACK;
1967 set_thread_flag(TIF_UPROBE);
1971 static struct notifier_block uprobe_exception_nb = {
1972 .notifier_call = arch_uprobe_exception_notify,
1973 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1976 static int __init init_uprobes(void)
1980 for (i = 0; i < UPROBES_HASH_SZ; i++)
1981 mutex_init(&uprobes_mmap_mutex[i]);
1983 if (percpu_init_rwsem(&dup_mmap_sem))
1986 return register_die_notifier(&uprobe_exception_nb);
1988 __initcall(init_uprobes);