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>
30 #include <linux/uprobes.h>
32 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
33 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
35 static struct rb_root uprobes_tree = RB_ROOT;
37 * allows us to skip the uprobe_mmap if there are no uprobe events active
38 * at this time. Probably a fine grained per inode count is better?
40 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
42 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
44 #define UPROBES_HASH_SZ 13
45 /* serialize uprobe->pending_list */
46 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
47 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
49 static struct percpu_rw_semaphore dup_mmap_sem;
51 /* Have a copy of original instruction */
52 #define UPROBE_COPY_INSN 0
55 struct rb_node rb_node; /* node in the rb tree */
57 struct rw_semaphore register_rwsem;
58 struct rw_semaphore consumer_rwsem;
59 struct list_head pending_list;
60 struct uprobe_consumer *consumers;
61 struct inode *inode; /* Also hold a ref to inode */
63 loff_t ref_ctr_offset;
67 * The generic code assumes that it has two members of unknown type
68 * owned by the arch-specific code:
70 * insn - copy_insn() saves the original instruction here for
71 * arch_uprobe_analyze_insn().
73 * ixol - potentially modified instruction to execute out of
74 * line, copied to xol_area by xol_get_insn_slot().
76 struct arch_uprobe arch;
79 struct delayed_uprobe {
80 struct list_head list;
81 struct uprobe *uprobe;
85 static DEFINE_MUTEX(delayed_uprobe_lock);
86 static LIST_HEAD(delayed_uprobe_list);
89 * Execute out of line area: anonymous executable mapping installed
90 * by the probed task to execute the copy of the original instruction
91 * mangled by set_swbp().
93 * On a breakpoint hit, thread contests for a slot. It frees the
94 * slot after singlestep. Currently a fixed number of slots are
98 wait_queue_head_t wq; /* if all slots are busy */
99 atomic_t slot_count; /* number of in-use slots */
100 unsigned long *bitmap; /* 0 = free slot */
102 struct vm_special_mapping xol_mapping;
103 struct page *pages[2];
105 * We keep the vma's vm_start rather than a pointer to the vma
106 * itself. The probed process or a naughty kernel module could make
107 * the vma go away, and we must handle that reasonably gracefully.
109 unsigned long vaddr; /* Page(s) of instruction slots */
113 * valid_vma: Verify if the specified vma is an executable vma
114 * Relax restrictions while unregistering: vm_flags might have
115 * changed after breakpoint was inserted.
116 * - is_register: indicates if we are in register context.
117 * - Return 1 if the specified virtual address is in an
120 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
122 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
127 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
130 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
132 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
135 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
137 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
141 * __replace_page - replace page in vma by new page.
142 * based on replace_page in mm/ksm.c
144 * @vma: vma that holds the pte pointing to page
145 * @addr: address the old @page is mapped at
146 * @page: the cowed page we are replacing by kpage
147 * @kpage: the modified page we replace page by
149 * Returns 0 on success, -EFAULT on failure.
151 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
152 struct page *old_page, struct page *new_page)
154 struct mm_struct *mm = vma->vm_mm;
155 struct page_vma_mapped_walk pvmw = {
161 struct mmu_notifier_range range;
162 struct mem_cgroup *memcg;
164 mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, addr,
167 VM_BUG_ON_PAGE(PageTransHuge(old_page), old_page);
169 err = mem_cgroup_try_charge(new_page, vma->vm_mm, GFP_KERNEL, &memcg,
174 /* For try_to_free_swap() and munlock_vma_page() below */
177 mmu_notifier_invalidate_range_start(&range);
179 if (!page_vma_mapped_walk(&pvmw)) {
180 mem_cgroup_cancel_charge(new_page, memcg, false);
183 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
186 page_add_new_anon_rmap(new_page, vma, addr, false);
187 mem_cgroup_commit_charge(new_page, memcg, false, false);
188 lru_cache_add_active_or_unevictable(new_page, vma);
190 if (!PageAnon(old_page)) {
191 dec_mm_counter(mm, mm_counter_file(old_page));
192 inc_mm_counter(mm, MM_ANONPAGES);
195 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
196 ptep_clear_flush_notify(vma, addr, pvmw.pte);
197 set_pte_at_notify(mm, addr, pvmw.pte,
198 mk_pte(new_page, vma->vm_page_prot));
200 page_remove_rmap(old_page, false);
201 if (!page_mapped(old_page))
202 try_to_free_swap(old_page);
203 page_vma_mapped_walk_done(&pvmw);
205 if (vma->vm_flags & VM_LOCKED)
206 munlock_vma_page(old_page);
211 mmu_notifier_invalidate_range_end(&range);
212 unlock_page(old_page);
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? */
283 static struct delayed_uprobe *
284 delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm)
286 struct delayed_uprobe *du;
288 list_for_each_entry(du, &delayed_uprobe_list, list)
289 if (du->uprobe == uprobe && du->mm == mm)
294 static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
296 struct delayed_uprobe *du;
298 if (delayed_uprobe_check(uprobe, mm))
301 du = kzalloc(sizeof(*du), GFP_KERNEL);
307 list_add(&du->list, &delayed_uprobe_list);
311 static void delayed_uprobe_delete(struct delayed_uprobe *du)
319 static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm)
321 struct list_head *pos, *q;
322 struct delayed_uprobe *du;
327 list_for_each_safe(pos, q, &delayed_uprobe_list) {
328 du = list_entry(pos, struct delayed_uprobe, list);
330 if (uprobe && du->uprobe != uprobe)
332 if (mm && du->mm != mm)
335 delayed_uprobe_delete(du);
339 static bool valid_ref_ctr_vma(struct uprobe *uprobe,
340 struct vm_area_struct *vma)
342 unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
344 return uprobe->ref_ctr_offset &&
346 file_inode(vma->vm_file) == uprobe->inode &&
347 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
348 vma->vm_start <= vaddr &&
352 static struct vm_area_struct *
353 find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm)
355 struct vm_area_struct *tmp;
357 for (tmp = mm->mmap; tmp; tmp = tmp->vm_next)
358 if (valid_ref_ctr_vma(uprobe, tmp))
365 __update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d)
369 struct vm_area_struct *vma;
376 ret = get_user_pages_remote(NULL, mm, vaddr, 1,
377 FOLL_WRITE, &page, &vma, NULL);
378 if (unlikely(ret <= 0)) {
380 * We are asking for 1 page. If get_user_pages_remote() fails,
381 * it may return 0, in that case we have to return error.
383 return ret == 0 ? -EBUSY : ret;
386 kaddr = kmap_atomic(page);
387 ptr = kaddr + (vaddr & ~PAGE_MASK);
389 if (unlikely(*ptr + d < 0)) {
390 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
391 "curr val: %d, delta: %d\n", vaddr, *ptr, d);
399 kunmap_atomic(kaddr);
404 static void update_ref_ctr_warn(struct uprobe *uprobe,
405 struct mm_struct *mm, short d)
407 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
408 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
409 d > 0 ? "increment" : "decrement", uprobe->inode->i_ino,
410 (unsigned long long) uprobe->offset,
411 (unsigned long long) uprobe->ref_ctr_offset, mm);
414 static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
417 struct vm_area_struct *rc_vma;
418 unsigned long rc_vaddr;
421 rc_vma = find_ref_ctr_vma(uprobe, mm);
424 rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset);
425 ret = __update_ref_ctr(mm, rc_vaddr, d);
427 update_ref_ctr_warn(uprobe, mm, d);
433 mutex_lock(&delayed_uprobe_lock);
435 ret = delayed_uprobe_add(uprobe, mm);
437 delayed_uprobe_remove(uprobe, mm);
438 mutex_unlock(&delayed_uprobe_lock);
445 * Expect the breakpoint instruction to be the smallest size instruction for
446 * the architecture. If an arch has variable length instruction and the
447 * breakpoint instruction is not of the smallest length instruction
448 * supported by that architecture then we need to modify is_trap_at_addr and
449 * uprobe_write_opcode accordingly. This would never be a problem for archs
450 * that have fixed length instructions.
452 * uprobe_write_opcode - write the opcode at a given virtual address.
453 * @mm: the probed process address space.
454 * @vaddr: the virtual address to store the opcode.
455 * @opcode: opcode to be written at @vaddr.
457 * Called with mm->mmap_sem held for write.
458 * Return 0 (success) or a negative errno.
460 int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
461 unsigned long vaddr, uprobe_opcode_t opcode)
463 struct uprobe *uprobe;
464 struct page *old_page, *new_page;
465 struct vm_area_struct *vma;
466 int ret, is_register, ref_ctr_updated = 0;
468 is_register = is_swbp_insn(&opcode);
469 uprobe = container_of(auprobe, struct uprobe, arch);
472 /* Read the page with vaddr into memory */
473 ret = get_user_pages_remote(NULL, mm, vaddr, 1,
474 FOLL_FORCE | FOLL_SPLIT, &old_page, &vma, NULL);
478 ret = verify_opcode(old_page, vaddr, &opcode);
482 /* We are going to replace instruction, update ref_ctr. */
483 if (!ref_ctr_updated && uprobe->ref_ctr_offset) {
484 ret = update_ref_ctr(uprobe, mm, is_register ? 1 : -1);
491 ret = anon_vma_prepare(vma);
496 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
500 __SetPageUptodate(new_page);
501 copy_highpage(new_page, old_page);
502 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
504 ret = __replace_page(vma, vaddr, old_page, new_page);
509 if (unlikely(ret == -EAGAIN))
512 /* Revert back reference counter if instruction update failed. */
513 if (ret && is_register && ref_ctr_updated)
514 update_ref_ctr(uprobe, mm, -1);
520 * set_swbp - store breakpoint at a given address.
521 * @auprobe: arch specific probepoint information.
522 * @mm: the probed process address space.
523 * @vaddr: the virtual address to insert the opcode.
525 * For mm @mm, store the breakpoint instruction at @vaddr.
526 * Return 0 (success) or a negative errno.
528 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
530 return uprobe_write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
534 * set_orig_insn - Restore the original instruction.
535 * @mm: the probed process address space.
536 * @auprobe: arch specific probepoint information.
537 * @vaddr: the virtual address to insert the opcode.
539 * For mm @mm, restore the original opcode (opcode) at @vaddr.
540 * Return 0 (success) or a negative errno.
543 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
545 return uprobe_write_opcode(auprobe, mm, vaddr,
546 *(uprobe_opcode_t *)&auprobe->insn);
549 static struct uprobe *get_uprobe(struct uprobe *uprobe)
551 refcount_inc(&uprobe->ref);
555 static void put_uprobe(struct uprobe *uprobe)
557 if (refcount_dec_and_test(&uprobe->ref)) {
559 * If application munmap(exec_vma) before uprobe_unregister()
560 * gets called, we don't get a chance to remove uprobe from
561 * delayed_uprobe_list from remove_breakpoint(). Do it here.
563 mutex_lock(&delayed_uprobe_lock);
564 delayed_uprobe_remove(uprobe, NULL);
565 mutex_unlock(&delayed_uprobe_lock);
570 static int match_uprobe(struct uprobe *l, struct uprobe *r)
572 if (l->inode < r->inode)
575 if (l->inode > r->inode)
578 if (l->offset < r->offset)
581 if (l->offset > r->offset)
587 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
589 struct uprobe u = { .inode = inode, .offset = offset };
590 struct rb_node *n = uprobes_tree.rb_node;
591 struct uprobe *uprobe;
595 uprobe = rb_entry(n, struct uprobe, rb_node);
596 match = match_uprobe(&u, uprobe);
598 return get_uprobe(uprobe);
609 * Find a uprobe corresponding to a given inode:offset
610 * Acquires uprobes_treelock
612 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
614 struct uprobe *uprobe;
616 spin_lock(&uprobes_treelock);
617 uprobe = __find_uprobe(inode, offset);
618 spin_unlock(&uprobes_treelock);
623 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
625 struct rb_node **p = &uprobes_tree.rb_node;
626 struct rb_node *parent = NULL;
632 u = rb_entry(parent, struct uprobe, rb_node);
633 match = match_uprobe(uprobe, u);
635 return get_uprobe(u);
638 p = &parent->rb_left;
640 p = &parent->rb_right;
645 rb_link_node(&uprobe->rb_node, parent, p);
646 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
647 /* get access + creation ref */
648 refcount_set(&uprobe->ref, 2);
654 * Acquire uprobes_treelock.
655 * Matching uprobe already exists in rbtree;
656 * increment (access refcount) and return the matching uprobe.
658 * No matching uprobe; insert the uprobe in rb_tree;
659 * get a double refcount (access + creation) and return NULL.
661 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
665 spin_lock(&uprobes_treelock);
666 u = __insert_uprobe(uprobe);
667 spin_unlock(&uprobes_treelock);
673 ref_ctr_mismatch_warn(struct uprobe *cur_uprobe, struct uprobe *uprobe)
675 pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx "
676 "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n",
677 uprobe->inode->i_ino, (unsigned long long) uprobe->offset,
678 (unsigned long long) cur_uprobe->ref_ctr_offset,
679 (unsigned long long) uprobe->ref_ctr_offset);
682 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
683 loff_t ref_ctr_offset)
685 struct uprobe *uprobe, *cur_uprobe;
687 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
691 uprobe->inode = inode;
692 uprobe->offset = offset;
693 uprobe->ref_ctr_offset = ref_ctr_offset;
694 init_rwsem(&uprobe->register_rwsem);
695 init_rwsem(&uprobe->consumer_rwsem);
697 /* add to uprobes_tree, sorted on inode:offset */
698 cur_uprobe = insert_uprobe(uprobe);
699 /* a uprobe exists for this inode:offset combination */
701 if (cur_uprobe->ref_ctr_offset != uprobe->ref_ctr_offset) {
702 ref_ctr_mismatch_warn(cur_uprobe, uprobe);
703 put_uprobe(cur_uprobe);
705 return ERR_PTR(-EINVAL);
714 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
716 down_write(&uprobe->consumer_rwsem);
717 uc->next = uprobe->consumers;
718 uprobe->consumers = uc;
719 up_write(&uprobe->consumer_rwsem);
723 * For uprobe @uprobe, delete the consumer @uc.
724 * Return true if the @uc is deleted successfully
727 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
729 struct uprobe_consumer **con;
732 down_write(&uprobe->consumer_rwsem);
733 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
740 up_write(&uprobe->consumer_rwsem);
745 static int __copy_insn(struct address_space *mapping, struct file *filp,
746 void *insn, int nbytes, loff_t offset)
750 * Ensure that the page that has the original instruction is populated
751 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
752 * see uprobe_register().
754 if (mapping->a_ops->readpage)
755 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
757 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
759 return PTR_ERR(page);
761 copy_from_page(page, offset, insn, nbytes);
767 static int copy_insn(struct uprobe *uprobe, struct file *filp)
769 struct address_space *mapping = uprobe->inode->i_mapping;
770 loff_t offs = uprobe->offset;
771 void *insn = &uprobe->arch.insn;
772 int size = sizeof(uprobe->arch.insn);
775 /* Copy only available bytes, -EIO if nothing was read */
777 if (offs >= i_size_read(uprobe->inode))
780 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
781 err = __copy_insn(mapping, filp, insn, len, offs);
793 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
794 struct mm_struct *mm, unsigned long vaddr)
798 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
801 /* TODO: move this into _register, until then we abuse this sem. */
802 down_write(&uprobe->consumer_rwsem);
803 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
806 ret = copy_insn(uprobe, file);
811 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
814 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
818 /* uprobe_write_opcode() assumes we don't cross page boundary */
819 BUG_ON((uprobe->offset & ~PAGE_MASK) +
820 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
822 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
823 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
826 up_write(&uprobe->consumer_rwsem);
831 static inline bool consumer_filter(struct uprobe_consumer *uc,
832 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
834 return !uc->filter || uc->filter(uc, ctx, mm);
837 static bool filter_chain(struct uprobe *uprobe,
838 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
840 struct uprobe_consumer *uc;
843 down_read(&uprobe->consumer_rwsem);
844 for (uc = uprobe->consumers; uc; uc = uc->next) {
845 ret = consumer_filter(uc, ctx, mm);
849 up_read(&uprobe->consumer_rwsem);
855 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
856 struct vm_area_struct *vma, unsigned long vaddr)
861 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
866 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
867 * the task can hit this breakpoint right after __replace_page().
869 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
871 set_bit(MMF_HAS_UPROBES, &mm->flags);
873 ret = set_swbp(&uprobe->arch, mm, vaddr);
875 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
876 else if (first_uprobe)
877 clear_bit(MMF_HAS_UPROBES, &mm->flags);
883 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
885 set_bit(MMF_RECALC_UPROBES, &mm->flags);
886 return set_orig_insn(&uprobe->arch, mm, vaddr);
889 static inline bool uprobe_is_active(struct uprobe *uprobe)
891 return !RB_EMPTY_NODE(&uprobe->rb_node);
894 * There could be threads that have already hit the breakpoint. They
895 * will recheck the current insn and restart if find_uprobe() fails.
896 * See find_active_uprobe().
898 static void delete_uprobe(struct uprobe *uprobe)
900 if (WARN_ON(!uprobe_is_active(uprobe)))
903 spin_lock(&uprobes_treelock);
904 rb_erase(&uprobe->rb_node, &uprobes_tree);
905 spin_unlock(&uprobes_treelock);
906 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
911 struct map_info *next;
912 struct mm_struct *mm;
916 static inline struct map_info *free_map_info(struct map_info *info)
918 struct map_info *next = info->next;
923 static struct map_info *
924 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
926 unsigned long pgoff = offset >> PAGE_SHIFT;
927 struct vm_area_struct *vma;
928 struct map_info *curr = NULL;
929 struct map_info *prev = NULL;
930 struct map_info *info;
934 i_mmap_lock_read(mapping);
935 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
936 if (!valid_vma(vma, is_register))
939 if (!prev && !more) {
941 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
942 * reclaim. This is optimistic, no harm done if it fails.
944 prev = kmalloc(sizeof(struct map_info),
945 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
954 if (!mmget_not_zero(vma->vm_mm))
962 info->mm = vma->vm_mm;
963 info->vaddr = offset_to_vaddr(vma, offset);
965 i_mmap_unlock_read(mapping);
977 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
979 curr = ERR_PTR(-ENOMEM);
989 prev = free_map_info(prev);
994 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
996 bool is_register = !!new;
997 struct map_info *info;
1000 percpu_down_write(&dup_mmap_sem);
1001 info = build_map_info(uprobe->inode->i_mapping,
1002 uprobe->offset, is_register);
1004 err = PTR_ERR(info);
1009 struct mm_struct *mm = info->mm;
1010 struct vm_area_struct *vma;
1012 if (err && is_register)
1015 down_write(&mm->mmap_sem);
1016 vma = find_vma(mm, info->vaddr);
1017 if (!vma || !valid_vma(vma, is_register) ||
1018 file_inode(vma->vm_file) != uprobe->inode)
1021 if (vma->vm_start > info->vaddr ||
1022 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
1026 /* consult only the "caller", new consumer. */
1027 if (consumer_filter(new,
1028 UPROBE_FILTER_REGISTER, mm))
1029 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
1030 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
1031 if (!filter_chain(uprobe,
1032 UPROBE_FILTER_UNREGISTER, mm))
1033 err |= remove_breakpoint(uprobe, mm, info->vaddr);
1037 up_write(&mm->mmap_sem);
1040 info = free_map_info(info);
1043 percpu_up_write(&dup_mmap_sem);
1048 __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
1052 if (WARN_ON(!consumer_del(uprobe, uc)))
1055 err = register_for_each_vma(uprobe, NULL);
1056 /* TODO : cant unregister? schedule a worker thread */
1057 if (!uprobe->consumers && !err)
1058 delete_uprobe(uprobe);
1062 * uprobe_unregister - unregister an already registered probe.
1063 * @inode: the file in which the probe has to be removed.
1064 * @offset: offset from the start of the file.
1065 * @uc: identify which probe if multiple probes are colocated.
1067 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
1069 struct uprobe *uprobe;
1071 uprobe = find_uprobe(inode, offset);
1072 if (WARN_ON(!uprobe))
1075 down_write(&uprobe->register_rwsem);
1076 __uprobe_unregister(uprobe, uc);
1077 up_write(&uprobe->register_rwsem);
1080 EXPORT_SYMBOL_GPL(uprobe_unregister);
1083 * __uprobe_register - register a probe
1084 * @inode: the file in which the probe has to be placed.
1085 * @offset: offset from the start of the file.
1086 * @uc: information on howto handle the probe..
1088 * Apart from the access refcount, __uprobe_register() takes a creation
1089 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
1090 * inserted into the rbtree (i.e first consumer for a @inode:@offset
1091 * tuple). Creation refcount stops uprobe_unregister from freeing the
1092 * @uprobe even before the register operation is complete. Creation
1093 * refcount is released when the last @uc for the @uprobe
1094 * unregisters. Caller of __uprobe_register() is required to keep @inode
1095 * (and the containing mount) referenced.
1097 * Return errno if it cannot successully install probes
1098 * else return 0 (success)
1100 static int __uprobe_register(struct inode *inode, loff_t offset,
1101 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1103 struct uprobe *uprobe;
1106 /* Uprobe must have at least one set consumer */
1107 if (!uc->handler && !uc->ret_handler)
1110 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
1111 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
1113 /* Racy, just to catch the obvious mistakes */
1114 if (offset > i_size_read(inode))
1118 uprobe = alloc_uprobe(inode, offset, ref_ctr_offset);
1122 return PTR_ERR(uprobe);
1125 * We can race with uprobe_unregister()->delete_uprobe().
1126 * Check uprobe_is_active() and retry if it is false.
1128 down_write(&uprobe->register_rwsem);
1130 if (likely(uprobe_is_active(uprobe))) {
1131 consumer_add(uprobe, uc);
1132 ret = register_for_each_vma(uprobe, uc);
1134 __uprobe_unregister(uprobe, uc);
1136 up_write(&uprobe->register_rwsem);
1139 if (unlikely(ret == -EAGAIN))
1144 int uprobe_register(struct inode *inode, loff_t offset,
1145 struct uprobe_consumer *uc)
1147 return __uprobe_register(inode, offset, 0, uc);
1149 EXPORT_SYMBOL_GPL(uprobe_register);
1151 int uprobe_register_refctr(struct inode *inode, loff_t offset,
1152 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1154 return __uprobe_register(inode, offset, ref_ctr_offset, uc);
1156 EXPORT_SYMBOL_GPL(uprobe_register_refctr);
1159 * uprobe_apply - unregister an already registered probe.
1160 * @inode: the file in which the probe has to be removed.
1161 * @offset: offset from the start of the file.
1162 * @uc: consumer which wants to add more or remove some breakpoints
1163 * @add: add or remove the breakpoints
1165 int uprobe_apply(struct inode *inode, loff_t offset,
1166 struct uprobe_consumer *uc, bool add)
1168 struct uprobe *uprobe;
1169 struct uprobe_consumer *con;
1172 uprobe = find_uprobe(inode, offset);
1173 if (WARN_ON(!uprobe))
1176 down_write(&uprobe->register_rwsem);
1177 for (con = uprobe->consumers; con && con != uc ; con = con->next)
1180 ret = register_for_each_vma(uprobe, add ? uc : NULL);
1181 up_write(&uprobe->register_rwsem);
1187 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
1189 struct vm_area_struct *vma;
1192 down_read(&mm->mmap_sem);
1193 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1194 unsigned long vaddr;
1197 if (!valid_vma(vma, false) ||
1198 file_inode(vma->vm_file) != uprobe->inode)
1201 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
1202 if (uprobe->offset < offset ||
1203 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
1206 vaddr = offset_to_vaddr(vma, uprobe->offset);
1207 err |= remove_breakpoint(uprobe, mm, vaddr);
1209 up_read(&mm->mmap_sem);
1214 static struct rb_node *
1215 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1217 struct rb_node *n = uprobes_tree.rb_node;
1220 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1222 if (inode < u->inode) {
1224 } else if (inode > u->inode) {
1227 if (max < u->offset)
1229 else if (min > u->offset)
1240 * For a given range in vma, build a list of probes that need to be inserted.
1242 static void build_probe_list(struct inode *inode,
1243 struct vm_area_struct *vma,
1244 unsigned long start, unsigned long end,
1245 struct list_head *head)
1248 struct rb_node *n, *t;
1251 INIT_LIST_HEAD(head);
1252 min = vaddr_to_offset(vma, start);
1253 max = min + (end - start) - 1;
1255 spin_lock(&uprobes_treelock);
1256 n = find_node_in_range(inode, min, max);
1258 for (t = n; t; t = rb_prev(t)) {
1259 u = rb_entry(t, struct uprobe, rb_node);
1260 if (u->inode != inode || u->offset < min)
1262 list_add(&u->pending_list, head);
1265 for (t = n; (t = rb_next(t)); ) {
1266 u = rb_entry(t, struct uprobe, rb_node);
1267 if (u->inode != inode || u->offset > max)
1269 list_add(&u->pending_list, head);
1273 spin_unlock(&uprobes_treelock);
1276 /* @vma contains reference counter, not the probed instruction. */
1277 static int delayed_ref_ctr_inc(struct vm_area_struct *vma)
1279 struct list_head *pos, *q;
1280 struct delayed_uprobe *du;
1281 unsigned long vaddr;
1282 int ret = 0, err = 0;
1284 mutex_lock(&delayed_uprobe_lock);
1285 list_for_each_safe(pos, q, &delayed_uprobe_list) {
1286 du = list_entry(pos, struct delayed_uprobe, list);
1288 if (du->mm != vma->vm_mm ||
1289 !valid_ref_ctr_vma(du->uprobe, vma))
1292 vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
1293 ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
1295 update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1);
1299 delayed_uprobe_delete(du);
1301 mutex_unlock(&delayed_uprobe_lock);
1306 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1308 * Currently we ignore all errors and always return 0, the callers
1309 * can't handle the failure anyway.
1311 int uprobe_mmap(struct vm_area_struct *vma)
1313 struct list_head tmp_list;
1314 struct uprobe *uprobe, *u;
1315 struct inode *inode;
1317 if (no_uprobe_events())
1321 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
1322 test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
1323 delayed_ref_ctr_inc(vma);
1325 if (!valid_vma(vma, true))
1328 inode = file_inode(vma->vm_file);
1332 mutex_lock(uprobes_mmap_hash(inode));
1333 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1335 * We can race with uprobe_unregister(), this uprobe can be already
1336 * removed. But in this case filter_chain() must return false, all
1337 * consumers have gone away.
1339 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1340 if (!fatal_signal_pending(current) &&
1341 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1342 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1343 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1347 mutex_unlock(uprobes_mmap_hash(inode));
1353 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1356 struct inode *inode;
1359 inode = file_inode(vma->vm_file);
1361 min = vaddr_to_offset(vma, start);
1362 max = min + (end - start) - 1;
1364 spin_lock(&uprobes_treelock);
1365 n = find_node_in_range(inode, min, max);
1366 spin_unlock(&uprobes_treelock);
1372 * Called in context of a munmap of a vma.
1374 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1376 if (no_uprobe_events() || !valid_vma(vma, false))
1379 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1382 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1383 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1386 if (vma_has_uprobes(vma, start, end))
1387 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1390 /* Slot allocation for XOL */
1391 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1393 struct vm_area_struct *vma;
1396 if (down_write_killable(&mm->mmap_sem))
1399 if (mm->uprobes_state.xol_area) {
1405 /* Try to map as high as possible, this is only a hint. */
1406 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1408 if (area->vaddr & ~PAGE_MASK) {
1414 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1415 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1416 &area->xol_mapping);
1423 /* pairs with get_xol_area() */
1424 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1426 up_write(&mm->mmap_sem);
1431 static struct xol_area *__create_xol_area(unsigned long vaddr)
1433 struct mm_struct *mm = current->mm;
1434 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1435 struct xol_area *area;
1437 area = kmalloc(sizeof(*area), GFP_KERNEL);
1438 if (unlikely(!area))
1441 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1446 area->xol_mapping.name = "[uprobes]";
1447 area->xol_mapping.fault = NULL;
1448 area->xol_mapping.pages = area->pages;
1449 area->pages[0] = alloc_page(GFP_HIGHUSER);
1450 if (!area->pages[0])
1452 area->pages[1] = NULL;
1454 area->vaddr = vaddr;
1455 init_waitqueue_head(&area->wq);
1456 /* Reserve the 1st slot for get_trampoline_vaddr() */
1457 set_bit(0, area->bitmap);
1458 atomic_set(&area->slot_count, 1);
1459 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1461 if (!xol_add_vma(mm, area))
1464 __free_page(area->pages[0]);
1466 kfree(area->bitmap);
1474 * get_xol_area - Allocate process's xol_area if necessary.
1475 * This area will be used for storing instructions for execution out of line.
1477 * Returns the allocated area or NULL.
1479 static struct xol_area *get_xol_area(void)
1481 struct mm_struct *mm = current->mm;
1482 struct xol_area *area;
1484 if (!mm->uprobes_state.xol_area)
1485 __create_xol_area(0);
1487 /* Pairs with xol_add_vma() smp_store_release() */
1488 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1493 * uprobe_clear_state - Free the area allocated for slots.
1495 void uprobe_clear_state(struct mm_struct *mm)
1497 struct xol_area *area = mm->uprobes_state.xol_area;
1499 mutex_lock(&delayed_uprobe_lock);
1500 delayed_uprobe_remove(NULL, mm);
1501 mutex_unlock(&delayed_uprobe_lock);
1506 put_page(area->pages[0]);
1507 kfree(area->bitmap);
1511 void uprobe_start_dup_mmap(void)
1513 percpu_down_read(&dup_mmap_sem);
1516 void uprobe_end_dup_mmap(void)
1518 percpu_up_read(&dup_mmap_sem);
1521 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1523 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1524 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1525 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1526 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1531 * - search for a free slot.
1533 static unsigned long xol_take_insn_slot(struct xol_area *area)
1535 unsigned long slot_addr;
1539 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1540 if (slot_nr < UINSNS_PER_PAGE) {
1541 if (!test_and_set_bit(slot_nr, area->bitmap))
1544 slot_nr = UINSNS_PER_PAGE;
1547 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1548 } while (slot_nr >= UINSNS_PER_PAGE);
1550 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1551 atomic_inc(&area->slot_count);
1557 * xol_get_insn_slot - allocate a slot for xol.
1558 * Returns the allocated slot address or 0.
1560 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1562 struct xol_area *area;
1563 unsigned long xol_vaddr;
1565 area = get_xol_area();
1569 xol_vaddr = xol_take_insn_slot(area);
1570 if (unlikely(!xol_vaddr))
1573 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1574 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1580 * xol_free_insn_slot - If slot was earlier allocated by
1581 * @xol_get_insn_slot(), make the slot available for
1582 * subsequent requests.
1584 static void xol_free_insn_slot(struct task_struct *tsk)
1586 struct xol_area *area;
1587 unsigned long vma_end;
1588 unsigned long slot_addr;
1590 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1593 slot_addr = tsk->utask->xol_vaddr;
1594 if (unlikely(!slot_addr))
1597 area = tsk->mm->uprobes_state.xol_area;
1598 vma_end = area->vaddr + PAGE_SIZE;
1599 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1600 unsigned long offset;
1603 offset = slot_addr - area->vaddr;
1604 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1605 if (slot_nr >= UINSNS_PER_PAGE)
1608 clear_bit(slot_nr, area->bitmap);
1609 atomic_dec(&area->slot_count);
1610 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1611 if (waitqueue_active(&area->wq))
1614 tsk->utask->xol_vaddr = 0;
1618 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1619 void *src, unsigned long len)
1621 /* Initialize the slot */
1622 copy_to_page(page, vaddr, src, len);
1625 * We probably need flush_icache_user_range() but it needs vma.
1626 * This should work on most of architectures by default. If
1627 * architecture needs to do something different it can define
1628 * its own version of the function.
1630 flush_dcache_page(page);
1634 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1635 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1637 * Return the address of the breakpoint instruction.
1639 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1641 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1644 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1646 struct uprobe_task *utask = current->utask;
1648 if (unlikely(utask && utask->active_uprobe))
1649 return utask->vaddr;
1651 return instruction_pointer(regs);
1654 static struct return_instance *free_ret_instance(struct return_instance *ri)
1656 struct return_instance *next = ri->next;
1657 put_uprobe(ri->uprobe);
1663 * Called with no locks held.
1664 * Called in context of an exiting or an exec-ing thread.
1666 void uprobe_free_utask(struct task_struct *t)
1668 struct uprobe_task *utask = t->utask;
1669 struct return_instance *ri;
1674 if (utask->active_uprobe)
1675 put_uprobe(utask->active_uprobe);
1677 ri = utask->return_instances;
1679 ri = free_ret_instance(ri);
1681 xol_free_insn_slot(t);
1687 * Allocate a uprobe_task object for the task if if necessary.
1688 * Called when the thread hits a breakpoint.
1691 * - pointer to new uprobe_task on success
1694 static struct uprobe_task *get_utask(void)
1696 if (!current->utask)
1697 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1698 return current->utask;
1701 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1703 struct uprobe_task *n_utask;
1704 struct return_instance **p, *o, *n;
1706 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1711 p = &n_utask->return_instances;
1712 for (o = o_utask->return_instances; o; o = o->next) {
1713 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1718 get_uprobe(n->uprobe);
1729 static void uprobe_warn(struct task_struct *t, const char *msg)
1731 pr_warn("uprobe: %s:%d failed to %s\n",
1732 current->comm, current->pid, msg);
1735 static void dup_xol_work(struct callback_head *work)
1737 if (current->flags & PF_EXITING)
1740 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1741 !fatal_signal_pending(current))
1742 uprobe_warn(current, "dup xol area");
1746 * Called in context of a new clone/fork from copy_process.
1748 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1750 struct uprobe_task *utask = current->utask;
1751 struct mm_struct *mm = current->mm;
1752 struct xol_area *area;
1756 if (!utask || !utask->return_instances)
1759 if (mm == t->mm && !(flags & CLONE_VFORK))
1762 if (dup_utask(t, utask))
1763 return uprobe_warn(t, "dup ret instances");
1765 /* The task can fork() after dup_xol_work() fails */
1766 area = mm->uprobes_state.xol_area;
1768 return uprobe_warn(t, "dup xol area");
1773 t->utask->dup_xol_addr = area->vaddr;
1774 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1775 task_work_add(t, &t->utask->dup_xol_work, true);
1779 * Current area->vaddr notion assume the trampoline address is always
1780 * equal area->vaddr.
1782 * Returns -1 in case the xol_area is not allocated.
1784 static unsigned long get_trampoline_vaddr(void)
1786 struct xol_area *area;
1787 unsigned long trampoline_vaddr = -1;
1789 /* Pairs with xol_add_vma() smp_store_release() */
1790 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1792 trampoline_vaddr = area->vaddr;
1794 return trampoline_vaddr;
1797 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1798 struct pt_regs *regs)
1800 struct return_instance *ri = utask->return_instances;
1801 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1803 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1804 ri = free_ret_instance(ri);
1807 utask->return_instances = ri;
1810 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1812 struct return_instance *ri;
1813 struct uprobe_task *utask;
1814 unsigned long orig_ret_vaddr, trampoline_vaddr;
1817 if (!get_xol_area())
1820 utask = get_utask();
1824 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1825 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1826 " nestedness limit pid/tgid=%d/%d\n",
1827 current->pid, current->tgid);
1831 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1835 trampoline_vaddr = get_trampoline_vaddr();
1836 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1837 if (orig_ret_vaddr == -1)
1840 /* drop the entries invalidated by longjmp() */
1841 chained = (orig_ret_vaddr == trampoline_vaddr);
1842 cleanup_return_instances(utask, chained, regs);
1845 * We don't want to keep trampoline address in stack, rather keep the
1846 * original return address of first caller thru all the consequent
1847 * instances. This also makes breakpoint unwrapping easier.
1850 if (!utask->return_instances) {
1852 * This situation is not possible. Likely we have an
1853 * attack from user-space.
1855 uprobe_warn(current, "handle tail call");
1858 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1861 ri->uprobe = get_uprobe(uprobe);
1862 ri->func = instruction_pointer(regs);
1863 ri->stack = user_stack_pointer(regs);
1864 ri->orig_ret_vaddr = orig_ret_vaddr;
1865 ri->chained = chained;
1868 ri->next = utask->return_instances;
1869 utask->return_instances = ri;
1876 /* Prepare to single-step probed instruction out of line. */
1878 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1880 struct uprobe_task *utask;
1881 unsigned long xol_vaddr;
1884 utask = get_utask();
1888 xol_vaddr = xol_get_insn_slot(uprobe);
1892 utask->xol_vaddr = xol_vaddr;
1893 utask->vaddr = bp_vaddr;
1895 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1896 if (unlikely(err)) {
1897 xol_free_insn_slot(current);
1901 utask->active_uprobe = uprobe;
1902 utask->state = UTASK_SSTEP;
1907 * If we are singlestepping, then ensure this thread is not connected to
1908 * non-fatal signals until completion of singlestep. When xol insn itself
1909 * triggers the signal, restart the original insn even if the task is
1910 * already SIGKILL'ed (since coredump should report the correct ip). This
1911 * is even more important if the task has a handler for SIGSEGV/etc, The
1912 * _same_ instruction should be repeated again after return from the signal
1913 * handler, and SSTEP can never finish in this case.
1915 bool uprobe_deny_signal(void)
1917 struct task_struct *t = current;
1918 struct uprobe_task *utask = t->utask;
1920 if (likely(!utask || !utask->active_uprobe))
1923 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1925 if (signal_pending(t)) {
1926 spin_lock_irq(&t->sighand->siglock);
1927 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1928 spin_unlock_irq(&t->sighand->siglock);
1930 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1931 utask->state = UTASK_SSTEP_TRAPPED;
1932 set_tsk_thread_flag(t, TIF_UPROBE);
1939 static void mmf_recalc_uprobes(struct mm_struct *mm)
1941 struct vm_area_struct *vma;
1943 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1944 if (!valid_vma(vma, false))
1947 * This is not strictly accurate, we can race with
1948 * uprobe_unregister() and see the already removed
1949 * uprobe if delete_uprobe() was not yet called.
1950 * Or this uprobe can be filtered out.
1952 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1956 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1959 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1962 uprobe_opcode_t opcode;
1965 pagefault_disable();
1966 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
1969 if (likely(result == 0))
1973 * The NULL 'tsk' here ensures that any faults that occur here
1974 * will not be accounted to the task. 'mm' *is* current->mm,
1975 * but we treat this as a 'remote' access since it is
1976 * essentially a kernel access to the memory.
1978 result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
1983 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1986 /* This needs to return true for any variant of the trap insn */
1987 return is_trap_insn(&opcode);
1990 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1992 struct mm_struct *mm = current->mm;
1993 struct uprobe *uprobe = NULL;
1994 struct vm_area_struct *vma;
1996 down_read(&mm->mmap_sem);
1997 vma = find_vma(mm, bp_vaddr);
1998 if (vma && vma->vm_start <= bp_vaddr) {
1999 if (valid_vma(vma, false)) {
2000 struct inode *inode = file_inode(vma->vm_file);
2001 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
2003 uprobe = find_uprobe(inode, offset);
2007 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
2012 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
2013 mmf_recalc_uprobes(mm);
2014 up_read(&mm->mmap_sem);
2019 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
2021 struct uprobe_consumer *uc;
2022 int remove = UPROBE_HANDLER_REMOVE;
2023 bool need_prep = false; /* prepare return uprobe, when needed */
2025 down_read(&uprobe->register_rwsem);
2026 for (uc = uprobe->consumers; uc; uc = uc->next) {
2030 rc = uc->handler(uc, regs);
2031 WARN(rc & ~UPROBE_HANDLER_MASK,
2032 "bad rc=0x%x from %ps()\n", rc, uc->handler);
2035 if (uc->ret_handler)
2041 if (need_prep && !remove)
2042 prepare_uretprobe(uprobe, regs); /* put bp at return */
2044 if (remove && uprobe->consumers) {
2045 WARN_ON(!uprobe_is_active(uprobe));
2046 unapply_uprobe(uprobe, current->mm);
2048 up_read(&uprobe->register_rwsem);
2052 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
2054 struct uprobe *uprobe = ri->uprobe;
2055 struct uprobe_consumer *uc;
2057 down_read(&uprobe->register_rwsem);
2058 for (uc = uprobe->consumers; uc; uc = uc->next) {
2059 if (uc->ret_handler)
2060 uc->ret_handler(uc, ri->func, regs);
2062 up_read(&uprobe->register_rwsem);
2065 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
2070 chained = ri->chained;
2071 ri = ri->next; /* can't be NULL if chained */
2077 static void handle_trampoline(struct pt_regs *regs)
2079 struct uprobe_task *utask;
2080 struct return_instance *ri, *next;
2083 utask = current->utask;
2087 ri = utask->return_instances;
2093 * We should throw out the frames invalidated by longjmp().
2094 * If this chain is valid, then the next one should be alive
2095 * or NULL; the latter case means that nobody but ri->func
2096 * could hit this trampoline on return. TODO: sigaltstack().
2098 next = find_next_ret_chain(ri);
2099 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
2101 instruction_pointer_set(regs, ri->orig_ret_vaddr);
2104 handle_uretprobe_chain(ri, regs);
2105 ri = free_ret_instance(ri);
2107 } while (ri != next);
2110 utask->return_instances = ri;
2114 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
2115 force_sig(SIGILL, current);
2119 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
2124 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
2125 struct pt_regs *regs)
2131 * Run handler and ask thread to singlestep.
2132 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2134 static void handle_swbp(struct pt_regs *regs)
2136 struct uprobe *uprobe;
2137 unsigned long bp_vaddr;
2138 int uninitialized_var(is_swbp);
2140 bp_vaddr = uprobe_get_swbp_addr(regs);
2141 if (bp_vaddr == get_trampoline_vaddr())
2142 return handle_trampoline(regs);
2144 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
2147 /* No matching uprobe; signal SIGTRAP. */
2148 send_sig(SIGTRAP, current, 0);
2151 * Either we raced with uprobe_unregister() or we can't
2152 * access this memory. The latter is only possible if
2153 * another thread plays with our ->mm. In both cases
2154 * we can simply restart. If this vma was unmapped we
2155 * can pretend this insn was not executed yet and get
2156 * the (correct) SIGSEGV after restart.
2158 instruction_pointer_set(regs, bp_vaddr);
2163 /* change it in advance for ->handler() and restart */
2164 instruction_pointer_set(regs, bp_vaddr);
2167 * TODO: move copy_insn/etc into _register and remove this hack.
2168 * After we hit the bp, _unregister + _register can install the
2169 * new and not-yet-analyzed uprobe at the same address, restart.
2171 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
2175 * Pairs with the smp_wmb() in prepare_uprobe().
2177 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2178 * we must also see the stores to &uprobe->arch performed by the
2179 * prepare_uprobe() call.
2183 /* Tracing handlers use ->utask to communicate with fetch methods */
2187 if (arch_uprobe_ignore(&uprobe->arch, regs))
2190 handler_chain(uprobe, regs);
2192 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
2195 if (!pre_ssout(uprobe, regs, bp_vaddr))
2198 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2204 * Perform required fix-ups and disable singlestep.
2205 * Allow pending signals to take effect.
2207 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
2209 struct uprobe *uprobe;
2212 uprobe = utask->active_uprobe;
2213 if (utask->state == UTASK_SSTEP_ACK)
2214 err = arch_uprobe_post_xol(&uprobe->arch, regs);
2215 else if (utask->state == UTASK_SSTEP_TRAPPED)
2216 arch_uprobe_abort_xol(&uprobe->arch, regs);
2221 utask->active_uprobe = NULL;
2222 utask->state = UTASK_RUNNING;
2223 xol_free_insn_slot(current);
2225 spin_lock_irq(¤t->sighand->siglock);
2226 recalc_sigpending(); /* see uprobe_deny_signal() */
2227 spin_unlock_irq(¤t->sighand->siglock);
2229 if (unlikely(err)) {
2230 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
2231 force_sig(SIGILL, current);
2236 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2237 * allows the thread to return from interrupt. After that handle_swbp()
2238 * sets utask->active_uprobe.
2240 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2241 * and allows the thread to return from interrupt.
2243 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2244 * uprobe_notify_resume().
2246 void uprobe_notify_resume(struct pt_regs *regs)
2248 struct uprobe_task *utask;
2250 clear_thread_flag(TIF_UPROBE);
2252 utask = current->utask;
2253 if (utask && utask->active_uprobe)
2254 handle_singlestep(utask, regs);
2260 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2261 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2263 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2268 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2269 (!current->utask || !current->utask->return_instances))
2272 set_thread_flag(TIF_UPROBE);
2277 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2278 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2280 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2282 struct uprobe_task *utask = current->utask;
2284 if (!current->mm || !utask || !utask->active_uprobe)
2285 /* task is currently not uprobed */
2288 utask->state = UTASK_SSTEP_ACK;
2289 set_thread_flag(TIF_UPROBE);
2293 static struct notifier_block uprobe_exception_nb = {
2294 .notifier_call = arch_uprobe_exception_notify,
2295 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2298 void __init uprobes_init(void)
2302 for (i = 0; i < UPROBES_HASH_SZ; i++)
2303 mutex_init(&uprobes_mmap_mutex[i]);
2305 BUG_ON(percpu_init_rwsem(&dup_mmap_sem));
2307 BUG_ON(register_die_notifier(&uprobe_exception_nb));