1 // SPDX-License-Identifier: GPL-2.0+
3 * User-space Probes (UProbes)
5 * Copyright (C) IBM Corporation, 2008-2012
9 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra
12 #include <linux/kernel.h>
13 #include <linux/highmem.h>
14 #include <linux/pagemap.h> /* read_mapping_page */
15 #include <linux/slab.h>
16 #include <linux/sched.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/coredump.h>
19 #include <linux/export.h>
20 #include <linux/rmap.h> /* anon_vma_prepare */
21 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
22 #include <linux/swap.h> /* try_to_free_swap */
23 #include <linux/ptrace.h> /* user_enable_single_step */
24 #include <linux/kdebug.h> /* notifier mechanism */
25 #include "../../mm/internal.h" /* munlock_vma_page */
26 #include <linux/percpu-rwsem.h>
27 #include <linux/task_work.h>
28 #include <linux/shmem_fs.h>
29 #include <linux/khugepaged.h>
31 #include <linux/uprobes.h>
33 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
34 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
36 static struct rb_root uprobes_tree = RB_ROOT;
38 * allows us to skip the uprobe_mmap if there are no uprobe events active
39 * at this time. Probably a fine grained per inode count is better?
41 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
43 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
45 #define UPROBES_HASH_SZ 13
46 /* serialize uprobe->pending_list */
47 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
48 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
50 DEFINE_STATIC_PERCPU_RWSEM(dup_mmap_sem);
52 /* Have a copy of original instruction */
53 #define UPROBE_COPY_INSN 0
56 struct rb_node rb_node; /* node in the rb tree */
58 struct rw_semaphore register_rwsem;
59 struct rw_semaphore consumer_rwsem;
60 struct list_head pending_list;
61 struct uprobe_consumer *consumers;
62 struct inode *inode; /* Also hold a ref to inode */
64 loff_t ref_ctr_offset;
68 * The generic code assumes that it has two members of unknown type
69 * owned by the arch-specific code:
71 * insn - copy_insn() saves the original instruction here for
72 * arch_uprobe_analyze_insn().
74 * ixol - potentially modified instruction to execute out of
75 * line, copied to xol_area by xol_get_insn_slot().
77 struct arch_uprobe arch;
80 struct delayed_uprobe {
81 struct list_head list;
82 struct uprobe *uprobe;
86 static DEFINE_MUTEX(delayed_uprobe_lock);
87 static LIST_HEAD(delayed_uprobe_list);
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 * @old_page: the page we are replacing by new_page
148 * @new_page: the modified page we replace page by
150 * If @new_page is NULL, only unmap @old_page.
152 * Returns 0 on success, negative error code otherwise.
154 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
155 struct page *old_page, struct page *new_page)
157 struct mm_struct *mm = vma->vm_mm;
158 struct page_vma_mapped_walk pvmw = {
159 .page = compound_head(old_page),
164 struct mmu_notifier_range range;
166 mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, addr,
170 err = mem_cgroup_charge(new_page, vma->vm_mm, GFP_KERNEL);
175 /* For try_to_free_swap() and munlock_vma_page() below */
178 mmu_notifier_invalidate_range_start(&range);
180 if (!page_vma_mapped_walk(&pvmw))
182 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
186 page_add_new_anon_rmap(new_page, vma, addr, false);
187 lru_cache_add_inactive_or_unevictable(new_page, vma);
189 /* no new page, just dec_mm_counter for old_page */
190 dec_mm_counter(mm, MM_ANONPAGES);
192 if (!PageAnon(old_page)) {
193 dec_mm_counter(mm, mm_counter_file(old_page));
194 inc_mm_counter(mm, MM_ANONPAGES);
197 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
198 ptep_clear_flush_notify(vma, addr, pvmw.pte);
200 set_pte_at_notify(mm, addr, pvmw.pte,
201 mk_pte(new_page, vma->vm_page_prot));
203 page_remove_rmap(old_page, false);
204 if (!page_mapped(old_page))
205 try_to_free_swap(old_page);
206 page_vma_mapped_walk_done(&pvmw);
208 if ((vma->vm_flags & VM_LOCKED) && !PageCompound(old_page))
209 munlock_vma_page(old_page);
214 mmu_notifier_invalidate_range_end(&range);
215 unlock_page(old_page);
220 * is_swbp_insn - check if instruction is breakpoint instruction.
221 * @insn: instruction to be checked.
222 * Default implementation of is_swbp_insn
223 * Returns true if @insn is a breakpoint instruction.
225 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
227 return *insn == UPROBE_SWBP_INSN;
231 * is_trap_insn - check if instruction is breakpoint instruction.
232 * @insn: instruction to be checked.
233 * Default implementation of is_trap_insn
234 * Returns true if @insn is a breakpoint instruction.
236 * This function is needed for the case where an architecture has multiple
237 * trap instructions (like powerpc).
239 bool __weak is_trap_insn(uprobe_opcode_t *insn)
241 return is_swbp_insn(insn);
244 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
246 void *kaddr = kmap_atomic(page);
247 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
248 kunmap_atomic(kaddr);
251 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
253 void *kaddr = kmap_atomic(page);
254 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
255 kunmap_atomic(kaddr);
258 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
260 uprobe_opcode_t old_opcode;
264 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
265 * We do not check if it is any other 'trap variant' which could
266 * be conditional trap instruction such as the one powerpc supports.
268 * The logic is that we do not care if the underlying instruction
269 * is a trap variant; uprobes always wins over any other (gdb)
272 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
273 is_swbp = is_swbp_insn(&old_opcode);
275 if (is_swbp_insn(new_opcode)) {
276 if (is_swbp) /* register: already installed? */
279 if (!is_swbp) /* unregister: was it changed by us? */
286 static struct delayed_uprobe *
287 delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm)
289 struct delayed_uprobe *du;
291 list_for_each_entry(du, &delayed_uprobe_list, list)
292 if (du->uprobe == uprobe && du->mm == mm)
297 static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
299 struct delayed_uprobe *du;
301 if (delayed_uprobe_check(uprobe, mm))
304 du = kzalloc(sizeof(*du), GFP_KERNEL);
310 list_add(&du->list, &delayed_uprobe_list);
314 static void delayed_uprobe_delete(struct delayed_uprobe *du)
322 static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm)
324 struct list_head *pos, *q;
325 struct delayed_uprobe *du;
330 list_for_each_safe(pos, q, &delayed_uprobe_list) {
331 du = list_entry(pos, struct delayed_uprobe, list);
333 if (uprobe && du->uprobe != uprobe)
335 if (mm && du->mm != mm)
338 delayed_uprobe_delete(du);
342 static bool valid_ref_ctr_vma(struct uprobe *uprobe,
343 struct vm_area_struct *vma)
345 unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
347 return uprobe->ref_ctr_offset &&
349 file_inode(vma->vm_file) == uprobe->inode &&
350 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
351 vma->vm_start <= vaddr &&
355 static struct vm_area_struct *
356 find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm)
358 struct vm_area_struct *tmp;
360 for (tmp = mm->mmap; tmp; tmp = tmp->vm_next)
361 if (valid_ref_ctr_vma(uprobe, tmp))
368 __update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d)
372 struct vm_area_struct *vma;
379 ret = get_user_pages_remote(mm, vaddr, 1,
380 FOLL_WRITE, &page, &vma, NULL);
381 if (unlikely(ret <= 0)) {
383 * We are asking for 1 page. If get_user_pages_remote() fails,
384 * it may return 0, in that case we have to return error.
386 return ret == 0 ? -EBUSY : ret;
389 kaddr = kmap_atomic(page);
390 ptr = kaddr + (vaddr & ~PAGE_MASK);
392 if (unlikely(*ptr + d < 0)) {
393 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
394 "curr val: %d, delta: %d\n", vaddr, *ptr, d);
402 kunmap_atomic(kaddr);
407 static void update_ref_ctr_warn(struct uprobe *uprobe,
408 struct mm_struct *mm, short d)
410 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
411 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
412 d > 0 ? "increment" : "decrement", uprobe->inode->i_ino,
413 (unsigned long long) uprobe->offset,
414 (unsigned long long) uprobe->ref_ctr_offset, mm);
417 static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
420 struct vm_area_struct *rc_vma;
421 unsigned long rc_vaddr;
424 rc_vma = find_ref_ctr_vma(uprobe, mm);
427 rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset);
428 ret = __update_ref_ctr(mm, rc_vaddr, d);
430 update_ref_ctr_warn(uprobe, mm, d);
436 mutex_lock(&delayed_uprobe_lock);
438 ret = delayed_uprobe_add(uprobe, mm);
440 delayed_uprobe_remove(uprobe, mm);
441 mutex_unlock(&delayed_uprobe_lock);
448 * Expect the breakpoint instruction to be the smallest size instruction for
449 * the architecture. If an arch has variable length instruction and the
450 * breakpoint instruction is not of the smallest length instruction
451 * supported by that architecture then we need to modify is_trap_at_addr and
452 * uprobe_write_opcode accordingly. This would never be a problem for archs
453 * that have fixed length instructions.
455 * uprobe_write_opcode - write the opcode at a given virtual address.
456 * @auprobe: arch specific probepoint information.
457 * @mm: the probed process address space.
458 * @vaddr: the virtual address to store the opcode.
459 * @opcode: opcode to be written at @vaddr.
461 * Called with mm->mmap_lock held for write.
462 * Return 0 (success) or a negative errno.
464 int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
465 unsigned long vaddr, uprobe_opcode_t opcode)
467 struct uprobe *uprobe;
468 struct page *old_page, *new_page;
469 struct vm_area_struct *vma;
470 int ret, is_register, ref_ctr_updated = 0;
471 bool orig_page_huge = false;
472 unsigned int gup_flags = FOLL_FORCE;
474 is_register = is_swbp_insn(&opcode);
475 uprobe = container_of(auprobe, struct uprobe, arch);
479 gup_flags |= FOLL_SPLIT_PMD;
480 /* Read the page with vaddr into memory */
481 ret = get_user_pages_remote(mm, vaddr, 1, gup_flags,
482 &old_page, &vma, NULL);
486 ret = verify_opcode(old_page, vaddr, &opcode);
490 if (WARN(!is_register && PageCompound(old_page),
491 "uprobe unregister should never work on compound page\n")) {
496 /* We are going to replace instruction, update ref_ctr. */
497 if (!ref_ctr_updated && uprobe->ref_ctr_offset) {
498 ret = update_ref_ctr(uprobe, mm, is_register ? 1 : -1);
506 if (!is_register && !PageAnon(old_page))
509 ret = anon_vma_prepare(vma);
514 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
518 __SetPageUptodate(new_page);
519 copy_highpage(new_page, old_page);
520 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
523 struct page *orig_page;
526 VM_BUG_ON_PAGE(!PageAnon(old_page), old_page);
528 index = vaddr_to_offset(vma, vaddr & PAGE_MASK) >> PAGE_SHIFT;
529 orig_page = find_get_page(vma->vm_file->f_inode->i_mapping,
533 if (PageUptodate(orig_page) &&
534 pages_identical(new_page, orig_page)) {
535 /* let go new_page */
539 if (PageCompound(orig_page))
540 orig_page_huge = true;
546 ret = __replace_page(vma, vaddr, old_page, new_page);
552 if (unlikely(ret == -EAGAIN))
555 /* Revert back reference counter if instruction update failed. */
556 if (ret && is_register && ref_ctr_updated)
557 update_ref_ctr(uprobe, mm, -1);
559 /* try collapse pmd for compound page */
560 if (!ret && orig_page_huge)
561 collapse_pte_mapped_thp(mm, vaddr);
567 * set_swbp - store breakpoint at a given address.
568 * @auprobe: arch specific probepoint information.
569 * @mm: the probed process address space.
570 * @vaddr: the virtual address to insert the opcode.
572 * For mm @mm, store the breakpoint instruction at @vaddr.
573 * Return 0 (success) or a negative errno.
575 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
577 return uprobe_write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
581 * set_orig_insn - Restore the original instruction.
582 * @mm: the probed process address space.
583 * @auprobe: arch specific probepoint information.
584 * @vaddr: the virtual address to insert the opcode.
586 * For mm @mm, restore the original opcode (opcode) at @vaddr.
587 * Return 0 (success) or a negative errno.
590 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
592 return uprobe_write_opcode(auprobe, mm, vaddr,
593 *(uprobe_opcode_t *)&auprobe->insn);
596 static struct uprobe *get_uprobe(struct uprobe *uprobe)
598 refcount_inc(&uprobe->ref);
602 static void put_uprobe(struct uprobe *uprobe)
604 if (refcount_dec_and_test(&uprobe->ref)) {
606 * If application munmap(exec_vma) before uprobe_unregister()
607 * gets called, we don't get a chance to remove uprobe from
608 * delayed_uprobe_list from remove_breakpoint(). Do it here.
610 mutex_lock(&delayed_uprobe_lock);
611 delayed_uprobe_remove(uprobe, NULL);
612 mutex_unlock(&delayed_uprobe_lock);
617 static __always_inline
618 int uprobe_cmp(const struct inode *l_inode, const loff_t l_offset,
619 const struct uprobe *r)
621 if (l_inode < r->inode)
624 if (l_inode > r->inode)
627 if (l_offset < r->offset)
630 if (l_offset > r->offset)
636 #define __node_2_uprobe(node) \
637 rb_entry((node), struct uprobe, rb_node)
639 struct __uprobe_key {
644 static inline int __uprobe_cmp_key(const void *key, const struct rb_node *b)
646 const struct __uprobe_key *a = key;
647 return uprobe_cmp(a->inode, a->offset, __node_2_uprobe(b));
650 static inline int __uprobe_cmp(struct rb_node *a, const struct rb_node *b)
652 struct uprobe *u = __node_2_uprobe(a);
653 return uprobe_cmp(u->inode, u->offset, __node_2_uprobe(b));
656 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
658 struct __uprobe_key key = {
662 struct rb_node *node = rb_find(&key, &uprobes_tree, __uprobe_cmp_key);
665 return get_uprobe(__node_2_uprobe(node));
671 * Find a uprobe corresponding to a given inode:offset
672 * Acquires uprobes_treelock
674 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
676 struct uprobe *uprobe;
678 spin_lock(&uprobes_treelock);
679 uprobe = __find_uprobe(inode, offset);
680 spin_unlock(&uprobes_treelock);
685 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
687 struct rb_node *node;
689 node = rb_find_add(&uprobe->rb_node, &uprobes_tree, __uprobe_cmp);
691 return get_uprobe(__node_2_uprobe(node));
693 /* get access + creation ref */
694 refcount_set(&uprobe->ref, 2);
699 * Acquire uprobes_treelock.
700 * Matching uprobe already exists in rbtree;
701 * increment (access refcount) and return the matching uprobe.
703 * No matching uprobe; insert the uprobe in rb_tree;
704 * get a double refcount (access + creation) and return NULL.
706 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
710 spin_lock(&uprobes_treelock);
711 u = __insert_uprobe(uprobe);
712 spin_unlock(&uprobes_treelock);
718 ref_ctr_mismatch_warn(struct uprobe *cur_uprobe, struct uprobe *uprobe)
720 pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx "
721 "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n",
722 uprobe->inode->i_ino, (unsigned long long) uprobe->offset,
723 (unsigned long long) cur_uprobe->ref_ctr_offset,
724 (unsigned long long) uprobe->ref_ctr_offset);
727 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
728 loff_t ref_ctr_offset)
730 struct uprobe *uprobe, *cur_uprobe;
732 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
736 uprobe->inode = inode;
737 uprobe->offset = offset;
738 uprobe->ref_ctr_offset = ref_ctr_offset;
739 init_rwsem(&uprobe->register_rwsem);
740 init_rwsem(&uprobe->consumer_rwsem);
742 /* add to uprobes_tree, sorted on inode:offset */
743 cur_uprobe = insert_uprobe(uprobe);
744 /* a uprobe exists for this inode:offset combination */
746 if (cur_uprobe->ref_ctr_offset != uprobe->ref_ctr_offset) {
747 ref_ctr_mismatch_warn(cur_uprobe, uprobe);
748 put_uprobe(cur_uprobe);
750 return ERR_PTR(-EINVAL);
759 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
761 down_write(&uprobe->consumer_rwsem);
762 uc->next = uprobe->consumers;
763 uprobe->consumers = uc;
764 up_write(&uprobe->consumer_rwsem);
768 * For uprobe @uprobe, delete the consumer @uc.
769 * Return true if the @uc is deleted successfully
772 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
774 struct uprobe_consumer **con;
777 down_write(&uprobe->consumer_rwsem);
778 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
785 up_write(&uprobe->consumer_rwsem);
790 static int __copy_insn(struct address_space *mapping, struct file *filp,
791 void *insn, int nbytes, loff_t offset)
795 * Ensure that the page that has the original instruction is populated
796 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
797 * see uprobe_register().
799 if (mapping->a_ops->readpage)
800 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
802 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
804 return PTR_ERR(page);
806 copy_from_page(page, offset, insn, nbytes);
812 static int copy_insn(struct uprobe *uprobe, struct file *filp)
814 struct address_space *mapping = uprobe->inode->i_mapping;
815 loff_t offs = uprobe->offset;
816 void *insn = &uprobe->arch.insn;
817 int size = sizeof(uprobe->arch.insn);
820 /* Copy only available bytes, -EIO if nothing was read */
822 if (offs >= i_size_read(uprobe->inode))
825 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
826 err = __copy_insn(mapping, filp, insn, len, offs);
838 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
839 struct mm_struct *mm, unsigned long vaddr)
843 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
846 /* TODO: move this into _register, until then we abuse this sem. */
847 down_write(&uprobe->consumer_rwsem);
848 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
851 ret = copy_insn(uprobe, file);
856 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
859 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
863 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
864 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
867 up_write(&uprobe->consumer_rwsem);
872 static inline bool consumer_filter(struct uprobe_consumer *uc,
873 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
875 return !uc->filter || uc->filter(uc, ctx, mm);
878 static bool filter_chain(struct uprobe *uprobe,
879 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
881 struct uprobe_consumer *uc;
884 down_read(&uprobe->consumer_rwsem);
885 for (uc = uprobe->consumers; uc; uc = uc->next) {
886 ret = consumer_filter(uc, ctx, mm);
890 up_read(&uprobe->consumer_rwsem);
896 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
897 struct vm_area_struct *vma, unsigned long vaddr)
902 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
907 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
908 * the task can hit this breakpoint right after __replace_page().
910 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
912 set_bit(MMF_HAS_UPROBES, &mm->flags);
914 ret = set_swbp(&uprobe->arch, mm, vaddr);
916 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
917 else if (first_uprobe)
918 clear_bit(MMF_HAS_UPROBES, &mm->flags);
924 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
926 set_bit(MMF_RECALC_UPROBES, &mm->flags);
927 return set_orig_insn(&uprobe->arch, mm, vaddr);
930 static inline bool uprobe_is_active(struct uprobe *uprobe)
932 return !RB_EMPTY_NODE(&uprobe->rb_node);
935 * There could be threads that have already hit the breakpoint. They
936 * will recheck the current insn and restart if find_uprobe() fails.
937 * See find_active_uprobe().
939 static void delete_uprobe(struct uprobe *uprobe)
941 if (WARN_ON(!uprobe_is_active(uprobe)))
944 spin_lock(&uprobes_treelock);
945 rb_erase(&uprobe->rb_node, &uprobes_tree);
946 spin_unlock(&uprobes_treelock);
947 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
952 struct map_info *next;
953 struct mm_struct *mm;
957 static inline struct map_info *free_map_info(struct map_info *info)
959 struct map_info *next = info->next;
964 static struct map_info *
965 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
967 unsigned long pgoff = offset >> PAGE_SHIFT;
968 struct vm_area_struct *vma;
969 struct map_info *curr = NULL;
970 struct map_info *prev = NULL;
971 struct map_info *info;
975 i_mmap_lock_read(mapping);
976 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
977 if (!valid_vma(vma, is_register))
980 if (!prev && !more) {
982 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
983 * reclaim. This is optimistic, no harm done if it fails.
985 prev = kmalloc(sizeof(struct map_info),
986 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
995 if (!mmget_not_zero(vma->vm_mm))
1003 info->mm = vma->vm_mm;
1004 info->vaddr = offset_to_vaddr(vma, offset);
1006 i_mmap_unlock_read(mapping);
1018 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
1020 curr = ERR_PTR(-ENOMEM);
1030 prev = free_map_info(prev);
1035 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
1037 bool is_register = !!new;
1038 struct map_info *info;
1041 percpu_down_write(&dup_mmap_sem);
1042 info = build_map_info(uprobe->inode->i_mapping,
1043 uprobe->offset, is_register);
1045 err = PTR_ERR(info);
1050 struct mm_struct *mm = info->mm;
1051 struct vm_area_struct *vma;
1053 if (err && is_register)
1056 mmap_write_lock(mm);
1057 vma = find_vma(mm, info->vaddr);
1058 if (!vma || !valid_vma(vma, is_register) ||
1059 file_inode(vma->vm_file) != uprobe->inode)
1062 if (vma->vm_start > info->vaddr ||
1063 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
1067 /* consult only the "caller", new consumer. */
1068 if (consumer_filter(new,
1069 UPROBE_FILTER_REGISTER, mm))
1070 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
1071 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
1072 if (!filter_chain(uprobe,
1073 UPROBE_FILTER_UNREGISTER, mm))
1074 err |= remove_breakpoint(uprobe, mm, info->vaddr);
1078 mmap_write_unlock(mm);
1081 info = free_map_info(info);
1084 percpu_up_write(&dup_mmap_sem);
1089 __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
1093 if (WARN_ON(!consumer_del(uprobe, uc)))
1096 err = register_for_each_vma(uprobe, NULL);
1097 /* TODO : cant unregister? schedule a worker thread */
1098 if (!uprobe->consumers && !err)
1099 delete_uprobe(uprobe);
1103 * uprobe_unregister - unregister an already registered probe.
1104 * @inode: the file in which the probe has to be removed.
1105 * @offset: offset from the start of the file.
1106 * @uc: identify which probe if multiple probes are colocated.
1108 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
1110 struct uprobe *uprobe;
1112 uprobe = find_uprobe(inode, offset);
1113 if (WARN_ON(!uprobe))
1116 down_write(&uprobe->register_rwsem);
1117 __uprobe_unregister(uprobe, uc);
1118 up_write(&uprobe->register_rwsem);
1121 EXPORT_SYMBOL_GPL(uprobe_unregister);
1124 * __uprobe_register - register a probe
1125 * @inode: the file in which the probe has to be placed.
1126 * @offset: offset from the start of the file.
1127 * @uc: information on howto handle the probe..
1129 * Apart from the access refcount, __uprobe_register() takes a creation
1130 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
1131 * inserted into the rbtree (i.e first consumer for a @inode:@offset
1132 * tuple). Creation refcount stops uprobe_unregister from freeing the
1133 * @uprobe even before the register operation is complete. Creation
1134 * refcount is released when the last @uc for the @uprobe
1135 * unregisters. Caller of __uprobe_register() is required to keep @inode
1136 * (and the containing mount) referenced.
1138 * Return errno if it cannot successully install probes
1139 * else return 0 (success)
1141 static int __uprobe_register(struct inode *inode, loff_t offset,
1142 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1144 struct uprobe *uprobe;
1147 /* Uprobe must have at least one set consumer */
1148 if (!uc->handler && !uc->ret_handler)
1151 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
1152 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
1154 /* Racy, just to catch the obvious mistakes */
1155 if (offset > i_size_read(inode))
1159 * This ensures that copy_from_page(), copy_to_page() and
1160 * __update_ref_ctr() can't cross page boundary.
1162 if (!IS_ALIGNED(offset, UPROBE_SWBP_INSN_SIZE))
1164 if (!IS_ALIGNED(ref_ctr_offset, sizeof(short)))
1168 uprobe = alloc_uprobe(inode, offset, ref_ctr_offset);
1172 return PTR_ERR(uprobe);
1175 * We can race with uprobe_unregister()->delete_uprobe().
1176 * Check uprobe_is_active() and retry if it is false.
1178 down_write(&uprobe->register_rwsem);
1180 if (likely(uprobe_is_active(uprobe))) {
1181 consumer_add(uprobe, uc);
1182 ret = register_for_each_vma(uprobe, uc);
1184 __uprobe_unregister(uprobe, uc);
1186 up_write(&uprobe->register_rwsem);
1189 if (unlikely(ret == -EAGAIN))
1194 int uprobe_register(struct inode *inode, loff_t offset,
1195 struct uprobe_consumer *uc)
1197 return __uprobe_register(inode, offset, 0, uc);
1199 EXPORT_SYMBOL_GPL(uprobe_register);
1201 int uprobe_register_refctr(struct inode *inode, loff_t offset,
1202 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1204 return __uprobe_register(inode, offset, ref_ctr_offset, uc);
1206 EXPORT_SYMBOL_GPL(uprobe_register_refctr);
1209 * uprobe_apply - unregister an already registered probe.
1210 * @inode: the file in which the probe has to be removed.
1211 * @offset: offset from the start of the file.
1212 * @uc: consumer which wants to add more or remove some breakpoints
1213 * @add: add or remove the breakpoints
1215 int uprobe_apply(struct inode *inode, loff_t offset,
1216 struct uprobe_consumer *uc, bool add)
1218 struct uprobe *uprobe;
1219 struct uprobe_consumer *con;
1222 uprobe = find_uprobe(inode, offset);
1223 if (WARN_ON(!uprobe))
1226 down_write(&uprobe->register_rwsem);
1227 for (con = uprobe->consumers; con && con != uc ; con = con->next)
1230 ret = register_for_each_vma(uprobe, add ? uc : NULL);
1231 up_write(&uprobe->register_rwsem);
1237 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
1239 struct vm_area_struct *vma;
1243 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1244 unsigned long vaddr;
1247 if (!valid_vma(vma, false) ||
1248 file_inode(vma->vm_file) != uprobe->inode)
1251 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
1252 if (uprobe->offset < offset ||
1253 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
1256 vaddr = offset_to_vaddr(vma, uprobe->offset);
1257 err |= remove_breakpoint(uprobe, mm, vaddr);
1259 mmap_read_unlock(mm);
1264 static struct rb_node *
1265 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1267 struct rb_node *n = uprobes_tree.rb_node;
1270 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1272 if (inode < u->inode) {
1274 } else if (inode > u->inode) {
1277 if (max < u->offset)
1279 else if (min > u->offset)
1290 * For a given range in vma, build a list of probes that need to be inserted.
1292 static void build_probe_list(struct inode *inode,
1293 struct vm_area_struct *vma,
1294 unsigned long start, unsigned long end,
1295 struct list_head *head)
1298 struct rb_node *n, *t;
1301 INIT_LIST_HEAD(head);
1302 min = vaddr_to_offset(vma, start);
1303 max = min + (end - start) - 1;
1305 spin_lock(&uprobes_treelock);
1306 n = find_node_in_range(inode, min, max);
1308 for (t = n; t; t = rb_prev(t)) {
1309 u = rb_entry(t, struct uprobe, rb_node);
1310 if (u->inode != inode || u->offset < min)
1312 list_add(&u->pending_list, head);
1315 for (t = n; (t = rb_next(t)); ) {
1316 u = rb_entry(t, struct uprobe, rb_node);
1317 if (u->inode != inode || u->offset > max)
1319 list_add(&u->pending_list, head);
1323 spin_unlock(&uprobes_treelock);
1326 /* @vma contains reference counter, not the probed instruction. */
1327 static int delayed_ref_ctr_inc(struct vm_area_struct *vma)
1329 struct list_head *pos, *q;
1330 struct delayed_uprobe *du;
1331 unsigned long vaddr;
1332 int ret = 0, err = 0;
1334 mutex_lock(&delayed_uprobe_lock);
1335 list_for_each_safe(pos, q, &delayed_uprobe_list) {
1336 du = list_entry(pos, struct delayed_uprobe, list);
1338 if (du->mm != vma->vm_mm ||
1339 !valid_ref_ctr_vma(du->uprobe, vma))
1342 vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
1343 ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
1345 update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1);
1349 delayed_uprobe_delete(du);
1351 mutex_unlock(&delayed_uprobe_lock);
1356 * Called from mmap_region/vma_adjust with mm->mmap_lock acquired.
1358 * Currently we ignore all errors and always return 0, the callers
1359 * can't handle the failure anyway.
1361 int uprobe_mmap(struct vm_area_struct *vma)
1363 struct list_head tmp_list;
1364 struct uprobe *uprobe, *u;
1365 struct inode *inode;
1367 if (no_uprobe_events())
1371 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
1372 test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
1373 delayed_ref_ctr_inc(vma);
1375 if (!valid_vma(vma, true))
1378 inode = file_inode(vma->vm_file);
1382 mutex_lock(uprobes_mmap_hash(inode));
1383 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1385 * We can race with uprobe_unregister(), this uprobe can be already
1386 * removed. But in this case filter_chain() must return false, all
1387 * consumers have gone away.
1389 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1390 if (!fatal_signal_pending(current) &&
1391 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1392 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1393 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1397 mutex_unlock(uprobes_mmap_hash(inode));
1403 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1406 struct inode *inode;
1409 inode = file_inode(vma->vm_file);
1411 min = vaddr_to_offset(vma, start);
1412 max = min + (end - start) - 1;
1414 spin_lock(&uprobes_treelock);
1415 n = find_node_in_range(inode, min, max);
1416 spin_unlock(&uprobes_treelock);
1422 * Called in context of a munmap of a vma.
1424 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1426 if (no_uprobe_events() || !valid_vma(vma, false))
1429 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1432 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1433 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1436 if (vma_has_uprobes(vma, start, end))
1437 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1440 /* Slot allocation for XOL */
1441 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1443 struct vm_area_struct *vma;
1446 if (mmap_write_lock_killable(mm))
1449 if (mm->uprobes_state.xol_area) {
1455 /* Try to map as high as possible, this is only a hint. */
1456 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1458 if (IS_ERR_VALUE(area->vaddr)) {
1464 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1465 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1466 &area->xol_mapping);
1473 /* pairs with get_xol_area() */
1474 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1476 mmap_write_unlock(mm);
1481 static struct xol_area *__create_xol_area(unsigned long vaddr)
1483 struct mm_struct *mm = current->mm;
1484 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1485 struct xol_area *area;
1487 area = kmalloc(sizeof(*area), GFP_KERNEL);
1488 if (unlikely(!area))
1491 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1496 area->xol_mapping.name = "[uprobes]";
1497 area->xol_mapping.fault = NULL;
1498 area->xol_mapping.pages = area->pages;
1499 area->pages[0] = alloc_page(GFP_HIGHUSER);
1500 if (!area->pages[0])
1502 area->pages[1] = NULL;
1504 area->vaddr = vaddr;
1505 init_waitqueue_head(&area->wq);
1506 /* Reserve the 1st slot for get_trampoline_vaddr() */
1507 set_bit(0, area->bitmap);
1508 atomic_set(&area->slot_count, 1);
1509 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1511 if (!xol_add_vma(mm, area))
1514 __free_page(area->pages[0]);
1516 kfree(area->bitmap);
1524 * get_xol_area - Allocate process's xol_area if necessary.
1525 * This area will be used for storing instructions for execution out of line.
1527 * Returns the allocated area or NULL.
1529 static struct xol_area *get_xol_area(void)
1531 struct mm_struct *mm = current->mm;
1532 struct xol_area *area;
1534 if (!mm->uprobes_state.xol_area)
1535 __create_xol_area(0);
1537 /* Pairs with xol_add_vma() smp_store_release() */
1538 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1543 * uprobe_clear_state - Free the area allocated for slots.
1545 void uprobe_clear_state(struct mm_struct *mm)
1547 struct xol_area *area = mm->uprobes_state.xol_area;
1549 mutex_lock(&delayed_uprobe_lock);
1550 delayed_uprobe_remove(NULL, mm);
1551 mutex_unlock(&delayed_uprobe_lock);
1556 put_page(area->pages[0]);
1557 kfree(area->bitmap);
1561 void uprobe_start_dup_mmap(void)
1563 percpu_down_read(&dup_mmap_sem);
1566 void uprobe_end_dup_mmap(void)
1568 percpu_up_read(&dup_mmap_sem);
1571 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1573 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1574 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1575 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1576 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1581 * - search for a free slot.
1583 static unsigned long xol_take_insn_slot(struct xol_area *area)
1585 unsigned long slot_addr;
1589 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1590 if (slot_nr < UINSNS_PER_PAGE) {
1591 if (!test_and_set_bit(slot_nr, area->bitmap))
1594 slot_nr = UINSNS_PER_PAGE;
1597 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1598 } while (slot_nr >= UINSNS_PER_PAGE);
1600 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1601 atomic_inc(&area->slot_count);
1607 * xol_get_insn_slot - allocate a slot for xol.
1608 * Returns the allocated slot address or 0.
1610 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1612 struct xol_area *area;
1613 unsigned long xol_vaddr;
1615 area = get_xol_area();
1619 xol_vaddr = xol_take_insn_slot(area);
1620 if (unlikely(!xol_vaddr))
1623 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1624 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1630 * xol_free_insn_slot - If slot was earlier allocated by
1631 * @xol_get_insn_slot(), make the slot available for
1632 * subsequent requests.
1634 static void xol_free_insn_slot(struct task_struct *tsk)
1636 struct xol_area *area;
1637 unsigned long vma_end;
1638 unsigned long slot_addr;
1640 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1643 slot_addr = tsk->utask->xol_vaddr;
1644 if (unlikely(!slot_addr))
1647 area = tsk->mm->uprobes_state.xol_area;
1648 vma_end = area->vaddr + PAGE_SIZE;
1649 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1650 unsigned long offset;
1653 offset = slot_addr - area->vaddr;
1654 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1655 if (slot_nr >= UINSNS_PER_PAGE)
1658 clear_bit(slot_nr, area->bitmap);
1659 atomic_dec(&area->slot_count);
1660 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1661 if (waitqueue_active(&area->wq))
1664 tsk->utask->xol_vaddr = 0;
1668 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1669 void *src, unsigned long len)
1671 /* Initialize the slot */
1672 copy_to_page(page, vaddr, src, len);
1675 * We probably need flush_icache_user_page() but it needs vma.
1676 * This should work on most of architectures by default. If
1677 * architecture needs to do something different it can define
1678 * its own version of the function.
1680 flush_dcache_page(page);
1684 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1685 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1687 * Return the address of the breakpoint instruction.
1689 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1691 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1694 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1696 struct uprobe_task *utask = current->utask;
1698 if (unlikely(utask && utask->active_uprobe))
1699 return utask->vaddr;
1701 return instruction_pointer(regs);
1704 static struct return_instance *free_ret_instance(struct return_instance *ri)
1706 struct return_instance *next = ri->next;
1707 put_uprobe(ri->uprobe);
1713 * Called with no locks held.
1714 * Called in context of an exiting or an exec-ing thread.
1716 void uprobe_free_utask(struct task_struct *t)
1718 struct uprobe_task *utask = t->utask;
1719 struct return_instance *ri;
1724 if (utask->active_uprobe)
1725 put_uprobe(utask->active_uprobe);
1727 ri = utask->return_instances;
1729 ri = free_ret_instance(ri);
1731 xol_free_insn_slot(t);
1737 * Allocate a uprobe_task object for the task if necessary.
1738 * Called when the thread hits a breakpoint.
1741 * - pointer to new uprobe_task on success
1744 static struct uprobe_task *get_utask(void)
1746 if (!current->utask)
1747 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1748 return current->utask;
1751 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1753 struct uprobe_task *n_utask;
1754 struct return_instance **p, *o, *n;
1756 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1761 p = &n_utask->return_instances;
1762 for (o = o_utask->return_instances; o; o = o->next) {
1763 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1768 get_uprobe(n->uprobe);
1779 static void uprobe_warn(struct task_struct *t, const char *msg)
1781 pr_warn("uprobe: %s:%d failed to %s\n",
1782 current->comm, current->pid, msg);
1785 static void dup_xol_work(struct callback_head *work)
1787 if (current->flags & PF_EXITING)
1790 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1791 !fatal_signal_pending(current))
1792 uprobe_warn(current, "dup xol area");
1796 * Called in context of a new clone/fork from copy_process.
1798 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1800 struct uprobe_task *utask = current->utask;
1801 struct mm_struct *mm = current->mm;
1802 struct xol_area *area;
1806 if (!utask || !utask->return_instances)
1809 if (mm == t->mm && !(flags & CLONE_VFORK))
1812 if (dup_utask(t, utask))
1813 return uprobe_warn(t, "dup ret instances");
1815 /* The task can fork() after dup_xol_work() fails */
1816 area = mm->uprobes_state.xol_area;
1818 return uprobe_warn(t, "dup xol area");
1823 t->utask->dup_xol_addr = area->vaddr;
1824 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1825 task_work_add(t, &t->utask->dup_xol_work, TWA_RESUME);
1829 * Current area->vaddr notion assume the trampoline address is always
1830 * equal area->vaddr.
1832 * Returns -1 in case the xol_area is not allocated.
1834 static unsigned long get_trampoline_vaddr(void)
1836 struct xol_area *area;
1837 unsigned long trampoline_vaddr = -1;
1839 /* Pairs with xol_add_vma() smp_store_release() */
1840 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1842 trampoline_vaddr = area->vaddr;
1844 return trampoline_vaddr;
1847 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1848 struct pt_regs *regs)
1850 struct return_instance *ri = utask->return_instances;
1851 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1853 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1854 ri = free_ret_instance(ri);
1857 utask->return_instances = ri;
1860 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1862 struct return_instance *ri;
1863 struct uprobe_task *utask;
1864 unsigned long orig_ret_vaddr, trampoline_vaddr;
1867 if (!get_xol_area())
1870 utask = get_utask();
1874 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1875 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1876 " nestedness limit pid/tgid=%d/%d\n",
1877 current->pid, current->tgid);
1881 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1885 trampoline_vaddr = get_trampoline_vaddr();
1886 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1887 if (orig_ret_vaddr == -1)
1890 /* drop the entries invalidated by longjmp() */
1891 chained = (orig_ret_vaddr == trampoline_vaddr);
1892 cleanup_return_instances(utask, chained, regs);
1895 * We don't want to keep trampoline address in stack, rather keep the
1896 * original return address of first caller thru all the consequent
1897 * instances. This also makes breakpoint unwrapping easier.
1900 if (!utask->return_instances) {
1902 * This situation is not possible. Likely we have an
1903 * attack from user-space.
1905 uprobe_warn(current, "handle tail call");
1908 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1911 ri->uprobe = get_uprobe(uprobe);
1912 ri->func = instruction_pointer(regs);
1913 ri->stack = user_stack_pointer(regs);
1914 ri->orig_ret_vaddr = orig_ret_vaddr;
1915 ri->chained = chained;
1918 ri->next = utask->return_instances;
1919 utask->return_instances = ri;
1926 /* Prepare to single-step probed instruction out of line. */
1928 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1930 struct uprobe_task *utask;
1931 unsigned long xol_vaddr;
1934 utask = get_utask();
1938 xol_vaddr = xol_get_insn_slot(uprobe);
1942 utask->xol_vaddr = xol_vaddr;
1943 utask->vaddr = bp_vaddr;
1945 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1946 if (unlikely(err)) {
1947 xol_free_insn_slot(current);
1951 utask->active_uprobe = uprobe;
1952 utask->state = UTASK_SSTEP;
1957 * If we are singlestepping, then ensure this thread is not connected to
1958 * non-fatal signals until completion of singlestep. When xol insn itself
1959 * triggers the signal, restart the original insn even if the task is
1960 * already SIGKILL'ed (since coredump should report the correct ip). This
1961 * is even more important if the task has a handler for SIGSEGV/etc, The
1962 * _same_ instruction should be repeated again after return from the signal
1963 * handler, and SSTEP can never finish in this case.
1965 bool uprobe_deny_signal(void)
1967 struct task_struct *t = current;
1968 struct uprobe_task *utask = t->utask;
1970 if (likely(!utask || !utask->active_uprobe))
1973 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1975 if (task_sigpending(t)) {
1976 spin_lock_irq(&t->sighand->siglock);
1977 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1978 spin_unlock_irq(&t->sighand->siglock);
1980 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1981 utask->state = UTASK_SSTEP_TRAPPED;
1982 set_tsk_thread_flag(t, TIF_UPROBE);
1989 static void mmf_recalc_uprobes(struct mm_struct *mm)
1991 struct vm_area_struct *vma;
1993 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1994 if (!valid_vma(vma, false))
1997 * This is not strictly accurate, we can race with
1998 * uprobe_unregister() and see the already removed
1999 * uprobe if delete_uprobe() was not yet called.
2000 * Or this uprobe can be filtered out.
2002 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
2006 clear_bit(MMF_HAS_UPROBES, &mm->flags);
2009 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
2012 uprobe_opcode_t opcode;
2015 if (WARN_ON_ONCE(!IS_ALIGNED(vaddr, UPROBE_SWBP_INSN_SIZE)))
2018 pagefault_disable();
2019 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
2022 if (likely(result == 0))
2026 * The NULL 'tsk' here ensures that any faults that occur here
2027 * will not be accounted to the task. 'mm' *is* current->mm,
2028 * but we treat this as a 'remote' access since it is
2029 * essentially a kernel access to the memory.
2031 result = get_user_pages_remote(mm, vaddr, 1, FOLL_FORCE, &page,
2036 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
2039 /* This needs to return true for any variant of the trap insn */
2040 return is_trap_insn(&opcode);
2043 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
2045 struct mm_struct *mm = current->mm;
2046 struct uprobe *uprobe = NULL;
2047 struct vm_area_struct *vma;
2050 vma = vma_lookup(mm, bp_vaddr);
2052 if (valid_vma(vma, false)) {
2053 struct inode *inode = file_inode(vma->vm_file);
2054 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
2056 uprobe = find_uprobe(inode, offset);
2060 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
2065 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
2066 mmf_recalc_uprobes(mm);
2067 mmap_read_unlock(mm);
2072 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
2074 struct uprobe_consumer *uc;
2075 int remove = UPROBE_HANDLER_REMOVE;
2076 bool need_prep = false; /* prepare return uprobe, when needed */
2078 down_read(&uprobe->register_rwsem);
2079 for (uc = uprobe->consumers; uc; uc = uc->next) {
2083 rc = uc->handler(uc, regs);
2084 WARN(rc & ~UPROBE_HANDLER_MASK,
2085 "bad rc=0x%x from %ps()\n", rc, uc->handler);
2088 if (uc->ret_handler)
2094 if (need_prep && !remove)
2095 prepare_uretprobe(uprobe, regs); /* put bp at return */
2097 if (remove && uprobe->consumers) {
2098 WARN_ON(!uprobe_is_active(uprobe));
2099 unapply_uprobe(uprobe, current->mm);
2101 up_read(&uprobe->register_rwsem);
2105 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
2107 struct uprobe *uprobe = ri->uprobe;
2108 struct uprobe_consumer *uc;
2110 down_read(&uprobe->register_rwsem);
2111 for (uc = uprobe->consumers; uc; uc = uc->next) {
2112 if (uc->ret_handler)
2113 uc->ret_handler(uc, ri->func, regs);
2115 up_read(&uprobe->register_rwsem);
2118 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
2123 chained = ri->chained;
2124 ri = ri->next; /* can't be NULL if chained */
2130 static void handle_trampoline(struct pt_regs *regs)
2132 struct uprobe_task *utask;
2133 struct return_instance *ri, *next;
2136 utask = current->utask;
2140 ri = utask->return_instances;
2146 * We should throw out the frames invalidated by longjmp().
2147 * If this chain is valid, then the next one should be alive
2148 * or NULL; the latter case means that nobody but ri->func
2149 * could hit this trampoline on return. TODO: sigaltstack().
2151 next = find_next_ret_chain(ri);
2152 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
2154 instruction_pointer_set(regs, ri->orig_ret_vaddr);
2157 handle_uretprobe_chain(ri, regs);
2158 ri = free_ret_instance(ri);
2160 } while (ri != next);
2163 utask->return_instances = ri;
2167 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
2172 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
2177 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
2178 struct pt_regs *regs)
2184 * Run handler and ask thread to singlestep.
2185 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2187 static void handle_swbp(struct pt_regs *regs)
2189 struct uprobe *uprobe;
2190 unsigned long bp_vaddr;
2193 bp_vaddr = uprobe_get_swbp_addr(regs);
2194 if (bp_vaddr == get_trampoline_vaddr())
2195 return handle_trampoline(regs);
2197 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
2200 /* No matching uprobe; signal SIGTRAP. */
2204 * Either we raced with uprobe_unregister() or we can't
2205 * access this memory. The latter is only possible if
2206 * another thread plays with our ->mm. In both cases
2207 * we can simply restart. If this vma was unmapped we
2208 * can pretend this insn was not executed yet and get
2209 * the (correct) SIGSEGV after restart.
2211 instruction_pointer_set(regs, bp_vaddr);
2216 /* change it in advance for ->handler() and restart */
2217 instruction_pointer_set(regs, bp_vaddr);
2220 * TODO: move copy_insn/etc into _register and remove this hack.
2221 * After we hit the bp, _unregister + _register can install the
2222 * new and not-yet-analyzed uprobe at the same address, restart.
2224 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
2228 * Pairs with the smp_wmb() in prepare_uprobe().
2230 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2231 * we must also see the stores to &uprobe->arch performed by the
2232 * prepare_uprobe() call.
2236 /* Tracing handlers use ->utask to communicate with fetch methods */
2240 if (arch_uprobe_ignore(&uprobe->arch, regs))
2243 handler_chain(uprobe, regs);
2245 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
2248 if (!pre_ssout(uprobe, regs, bp_vaddr))
2251 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2257 * Perform required fix-ups and disable singlestep.
2258 * Allow pending signals to take effect.
2260 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
2262 struct uprobe *uprobe;
2265 uprobe = utask->active_uprobe;
2266 if (utask->state == UTASK_SSTEP_ACK)
2267 err = arch_uprobe_post_xol(&uprobe->arch, regs);
2268 else if (utask->state == UTASK_SSTEP_TRAPPED)
2269 arch_uprobe_abort_xol(&uprobe->arch, regs);
2274 utask->active_uprobe = NULL;
2275 utask->state = UTASK_RUNNING;
2276 xol_free_insn_slot(current);
2278 spin_lock_irq(¤t->sighand->siglock);
2279 recalc_sigpending(); /* see uprobe_deny_signal() */
2280 spin_unlock_irq(¤t->sighand->siglock);
2282 if (unlikely(err)) {
2283 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
2289 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2290 * allows the thread to return from interrupt. After that handle_swbp()
2291 * sets utask->active_uprobe.
2293 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2294 * and allows the thread to return from interrupt.
2296 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2297 * uprobe_notify_resume().
2299 void uprobe_notify_resume(struct pt_regs *regs)
2301 struct uprobe_task *utask;
2303 clear_thread_flag(TIF_UPROBE);
2305 utask = current->utask;
2306 if (utask && utask->active_uprobe)
2307 handle_singlestep(utask, regs);
2313 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2314 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2316 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2321 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2322 (!current->utask || !current->utask->return_instances))
2325 set_thread_flag(TIF_UPROBE);
2330 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2331 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2333 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2335 struct uprobe_task *utask = current->utask;
2337 if (!current->mm || !utask || !utask->active_uprobe)
2338 /* task is currently not uprobed */
2341 utask->state = UTASK_SSTEP_ACK;
2342 set_thread_flag(TIF_UPROBE);
2346 static struct notifier_block uprobe_exception_nb = {
2347 .notifier_call = arch_uprobe_exception_notify,
2348 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2351 void __init uprobes_init(void)
2355 for (i = 0; i < UPROBES_HASH_SZ; i++)
2356 mutex_init(&uprobes_mmap_mutex[i]);
2358 BUG_ON(register_die_notifier(&uprobe_exception_nb));