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,
176 /* For try_to_free_swap() and munlock_vma_page() below */
179 mmu_notifier_invalidate_range_start(&range);
181 if (!page_vma_mapped_walk(&pvmw))
183 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
187 page_add_new_anon_rmap(new_page, vma, addr, false);
188 lru_cache_add_active_or_unevictable(new_page, vma);
190 /* no new page, just dec_mm_counter for old_page */
191 dec_mm_counter(mm, MM_ANONPAGES);
193 if (!PageAnon(old_page)) {
194 dec_mm_counter(mm, mm_counter_file(old_page));
195 inc_mm_counter(mm, MM_ANONPAGES);
198 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
199 ptep_clear_flush_notify(vma, addr, pvmw.pte);
201 set_pte_at_notify(mm, addr, pvmw.pte,
202 mk_pte(new_page, vma->vm_page_prot));
204 page_remove_rmap(old_page, false);
205 if (!page_mapped(old_page))
206 try_to_free_swap(old_page);
207 page_vma_mapped_walk_done(&pvmw);
209 if (vma->vm_flags & VM_LOCKED)
210 munlock_vma_page(old_page);
215 mmu_notifier_invalidate_range_end(&range);
216 unlock_page(old_page);
221 * is_swbp_insn - check if instruction is breakpoint instruction.
222 * @insn: instruction to be checked.
223 * Default implementation of is_swbp_insn
224 * Returns true if @insn is a breakpoint instruction.
226 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
228 return *insn == UPROBE_SWBP_INSN;
232 * is_trap_insn - check if instruction is breakpoint instruction.
233 * @insn: instruction to be checked.
234 * Default implementation of is_trap_insn
235 * Returns true if @insn is a breakpoint instruction.
237 * This function is needed for the case where an architecture has multiple
238 * trap instructions (like powerpc).
240 bool __weak is_trap_insn(uprobe_opcode_t *insn)
242 return is_swbp_insn(insn);
245 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
247 void *kaddr = kmap_atomic(page);
248 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
249 kunmap_atomic(kaddr);
252 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
254 void *kaddr = kmap_atomic(page);
255 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
256 kunmap_atomic(kaddr);
259 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
261 uprobe_opcode_t old_opcode;
265 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
266 * We do not check if it is any other 'trap variant' which could
267 * be conditional trap instruction such as the one powerpc supports.
269 * The logic is that we do not care if the underlying instruction
270 * is a trap variant; uprobes always wins over any other (gdb)
273 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
274 is_swbp = is_swbp_insn(&old_opcode);
276 if (is_swbp_insn(new_opcode)) {
277 if (is_swbp) /* register: already installed? */
280 if (!is_swbp) /* unregister: was it changed by us? */
287 static struct delayed_uprobe *
288 delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm)
290 struct delayed_uprobe *du;
292 list_for_each_entry(du, &delayed_uprobe_list, list)
293 if (du->uprobe == uprobe && du->mm == mm)
298 static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
300 struct delayed_uprobe *du;
302 if (delayed_uprobe_check(uprobe, mm))
305 du = kzalloc(sizeof(*du), GFP_KERNEL);
311 list_add(&du->list, &delayed_uprobe_list);
315 static void delayed_uprobe_delete(struct delayed_uprobe *du)
323 static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm)
325 struct list_head *pos, *q;
326 struct delayed_uprobe *du;
331 list_for_each_safe(pos, q, &delayed_uprobe_list) {
332 du = list_entry(pos, struct delayed_uprobe, list);
334 if (uprobe && du->uprobe != uprobe)
336 if (mm && du->mm != mm)
339 delayed_uprobe_delete(du);
343 static bool valid_ref_ctr_vma(struct uprobe *uprobe,
344 struct vm_area_struct *vma)
346 unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
348 return uprobe->ref_ctr_offset &&
350 file_inode(vma->vm_file) == uprobe->inode &&
351 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
352 vma->vm_start <= vaddr &&
356 static struct vm_area_struct *
357 find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm)
359 struct vm_area_struct *tmp;
361 for (tmp = mm->mmap; tmp; tmp = tmp->vm_next)
362 if (valid_ref_ctr_vma(uprobe, tmp))
369 __update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d)
373 struct vm_area_struct *vma;
380 ret = get_user_pages_remote(NULL, mm, vaddr, 1,
381 FOLL_WRITE, &page, &vma, NULL);
382 if (unlikely(ret <= 0)) {
384 * We are asking for 1 page. If get_user_pages_remote() fails,
385 * it may return 0, in that case we have to return error.
387 return ret == 0 ? -EBUSY : ret;
390 kaddr = kmap_atomic(page);
391 ptr = kaddr + (vaddr & ~PAGE_MASK);
393 if (unlikely(*ptr + d < 0)) {
394 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
395 "curr val: %d, delta: %d\n", vaddr, *ptr, d);
403 kunmap_atomic(kaddr);
408 static void update_ref_ctr_warn(struct uprobe *uprobe,
409 struct mm_struct *mm, short d)
411 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
412 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
413 d > 0 ? "increment" : "decrement", uprobe->inode->i_ino,
414 (unsigned long long) uprobe->offset,
415 (unsigned long long) uprobe->ref_ctr_offset, mm);
418 static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
421 struct vm_area_struct *rc_vma;
422 unsigned long rc_vaddr;
425 rc_vma = find_ref_ctr_vma(uprobe, mm);
428 rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset);
429 ret = __update_ref_ctr(mm, rc_vaddr, d);
431 update_ref_ctr_warn(uprobe, mm, d);
437 mutex_lock(&delayed_uprobe_lock);
439 ret = delayed_uprobe_add(uprobe, mm);
441 delayed_uprobe_remove(uprobe, mm);
442 mutex_unlock(&delayed_uprobe_lock);
449 * Expect the breakpoint instruction to be the smallest size instruction for
450 * the architecture. If an arch has variable length instruction and the
451 * breakpoint instruction is not of the smallest length instruction
452 * supported by that architecture then we need to modify is_trap_at_addr and
453 * uprobe_write_opcode accordingly. This would never be a problem for archs
454 * that have fixed length instructions.
456 * uprobe_write_opcode - write the opcode at a given virtual address.
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_sem 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(NULL, 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 int match_uprobe(struct uprobe *l, struct uprobe *r)
619 if (l->inode < r->inode)
622 if (l->inode > r->inode)
625 if (l->offset < r->offset)
628 if (l->offset > r->offset)
634 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
636 struct uprobe u = { .inode = inode, .offset = offset };
637 struct rb_node *n = uprobes_tree.rb_node;
638 struct uprobe *uprobe;
642 uprobe = rb_entry(n, struct uprobe, rb_node);
643 match = match_uprobe(&u, uprobe);
645 return get_uprobe(uprobe);
656 * Find a uprobe corresponding to a given inode:offset
657 * Acquires uprobes_treelock
659 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
661 struct uprobe *uprobe;
663 spin_lock(&uprobes_treelock);
664 uprobe = __find_uprobe(inode, offset);
665 spin_unlock(&uprobes_treelock);
670 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
672 struct rb_node **p = &uprobes_tree.rb_node;
673 struct rb_node *parent = NULL;
679 u = rb_entry(parent, struct uprobe, rb_node);
680 match = match_uprobe(uprobe, u);
682 return get_uprobe(u);
685 p = &parent->rb_left;
687 p = &parent->rb_right;
692 rb_link_node(&uprobe->rb_node, parent, p);
693 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
694 /* get access + creation ref */
695 refcount_set(&uprobe->ref, 2);
701 * Acquire uprobes_treelock.
702 * Matching uprobe already exists in rbtree;
703 * increment (access refcount) and return the matching uprobe.
705 * No matching uprobe; insert the uprobe in rb_tree;
706 * get a double refcount (access + creation) and return NULL.
708 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
712 spin_lock(&uprobes_treelock);
713 u = __insert_uprobe(uprobe);
714 spin_unlock(&uprobes_treelock);
720 ref_ctr_mismatch_warn(struct uprobe *cur_uprobe, struct uprobe *uprobe)
722 pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx "
723 "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n",
724 uprobe->inode->i_ino, (unsigned long long) uprobe->offset,
725 (unsigned long long) cur_uprobe->ref_ctr_offset,
726 (unsigned long long) uprobe->ref_ctr_offset);
729 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
730 loff_t ref_ctr_offset)
732 struct uprobe *uprobe, *cur_uprobe;
734 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
738 uprobe->inode = inode;
739 uprobe->offset = offset;
740 uprobe->ref_ctr_offset = ref_ctr_offset;
741 init_rwsem(&uprobe->register_rwsem);
742 init_rwsem(&uprobe->consumer_rwsem);
744 /* add to uprobes_tree, sorted on inode:offset */
745 cur_uprobe = insert_uprobe(uprobe);
746 /* a uprobe exists for this inode:offset combination */
748 if (cur_uprobe->ref_ctr_offset != uprobe->ref_ctr_offset) {
749 ref_ctr_mismatch_warn(cur_uprobe, uprobe);
750 put_uprobe(cur_uprobe);
752 return ERR_PTR(-EINVAL);
761 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
763 down_write(&uprobe->consumer_rwsem);
764 uc->next = uprobe->consumers;
765 uprobe->consumers = uc;
766 up_write(&uprobe->consumer_rwsem);
770 * For uprobe @uprobe, delete the consumer @uc.
771 * Return true if the @uc is deleted successfully
774 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
776 struct uprobe_consumer **con;
779 down_write(&uprobe->consumer_rwsem);
780 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
787 up_write(&uprobe->consumer_rwsem);
792 static int __copy_insn(struct address_space *mapping, struct file *filp,
793 void *insn, int nbytes, loff_t offset)
797 * Ensure that the page that has the original instruction is populated
798 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
799 * see uprobe_register().
801 if (mapping->a_ops->readpage)
802 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
804 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
806 return PTR_ERR(page);
808 copy_from_page(page, offset, insn, nbytes);
814 static int copy_insn(struct uprobe *uprobe, struct file *filp)
816 struct address_space *mapping = uprobe->inode->i_mapping;
817 loff_t offs = uprobe->offset;
818 void *insn = &uprobe->arch.insn;
819 int size = sizeof(uprobe->arch.insn);
822 /* Copy only available bytes, -EIO if nothing was read */
824 if (offs >= i_size_read(uprobe->inode))
827 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
828 err = __copy_insn(mapping, filp, insn, len, offs);
840 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
841 struct mm_struct *mm, unsigned long vaddr)
845 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
848 /* TODO: move this into _register, until then we abuse this sem. */
849 down_write(&uprobe->consumer_rwsem);
850 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
853 ret = copy_insn(uprobe, file);
858 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
861 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
865 /* uprobe_write_opcode() assumes we don't cross page boundary */
866 BUG_ON((uprobe->offset & ~PAGE_MASK) +
867 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
869 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
870 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
873 up_write(&uprobe->consumer_rwsem);
878 static inline bool consumer_filter(struct uprobe_consumer *uc,
879 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
881 return !uc->filter || uc->filter(uc, ctx, mm);
884 static bool filter_chain(struct uprobe *uprobe,
885 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
887 struct uprobe_consumer *uc;
890 down_read(&uprobe->consumer_rwsem);
891 for (uc = uprobe->consumers; uc; uc = uc->next) {
892 ret = consumer_filter(uc, ctx, mm);
896 up_read(&uprobe->consumer_rwsem);
902 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
903 struct vm_area_struct *vma, unsigned long vaddr)
908 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
913 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
914 * the task can hit this breakpoint right after __replace_page().
916 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
918 set_bit(MMF_HAS_UPROBES, &mm->flags);
920 ret = set_swbp(&uprobe->arch, mm, vaddr);
922 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
923 else if (first_uprobe)
924 clear_bit(MMF_HAS_UPROBES, &mm->flags);
930 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
932 set_bit(MMF_RECALC_UPROBES, &mm->flags);
933 return set_orig_insn(&uprobe->arch, mm, vaddr);
936 static inline bool uprobe_is_active(struct uprobe *uprobe)
938 return !RB_EMPTY_NODE(&uprobe->rb_node);
941 * There could be threads that have already hit the breakpoint. They
942 * will recheck the current insn and restart if find_uprobe() fails.
943 * See find_active_uprobe().
945 static void delete_uprobe(struct uprobe *uprobe)
947 if (WARN_ON(!uprobe_is_active(uprobe)))
950 spin_lock(&uprobes_treelock);
951 rb_erase(&uprobe->rb_node, &uprobes_tree);
952 spin_unlock(&uprobes_treelock);
953 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
958 struct map_info *next;
959 struct mm_struct *mm;
963 static inline struct map_info *free_map_info(struct map_info *info)
965 struct map_info *next = info->next;
970 static struct map_info *
971 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
973 unsigned long pgoff = offset >> PAGE_SHIFT;
974 struct vm_area_struct *vma;
975 struct map_info *curr = NULL;
976 struct map_info *prev = NULL;
977 struct map_info *info;
981 i_mmap_lock_read(mapping);
982 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
983 if (!valid_vma(vma, is_register))
986 if (!prev && !more) {
988 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
989 * reclaim. This is optimistic, no harm done if it fails.
991 prev = kmalloc(sizeof(struct map_info),
992 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
1001 if (!mmget_not_zero(vma->vm_mm))
1009 info->mm = vma->vm_mm;
1010 info->vaddr = offset_to_vaddr(vma, offset);
1012 i_mmap_unlock_read(mapping);
1024 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
1026 curr = ERR_PTR(-ENOMEM);
1036 prev = free_map_info(prev);
1041 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
1043 bool is_register = !!new;
1044 struct map_info *info;
1047 percpu_down_write(&dup_mmap_sem);
1048 info = build_map_info(uprobe->inode->i_mapping,
1049 uprobe->offset, is_register);
1051 err = PTR_ERR(info);
1056 struct mm_struct *mm = info->mm;
1057 struct vm_area_struct *vma;
1059 if (err && is_register)
1062 down_write(&mm->mmap_sem);
1063 vma = find_vma(mm, info->vaddr);
1064 if (!vma || !valid_vma(vma, is_register) ||
1065 file_inode(vma->vm_file) != uprobe->inode)
1068 if (vma->vm_start > info->vaddr ||
1069 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
1073 /* consult only the "caller", new consumer. */
1074 if (consumer_filter(new,
1075 UPROBE_FILTER_REGISTER, mm))
1076 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
1077 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
1078 if (!filter_chain(uprobe,
1079 UPROBE_FILTER_UNREGISTER, mm))
1080 err |= remove_breakpoint(uprobe, mm, info->vaddr);
1084 up_write(&mm->mmap_sem);
1087 info = free_map_info(info);
1090 percpu_up_write(&dup_mmap_sem);
1095 __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
1099 if (WARN_ON(!consumer_del(uprobe, uc)))
1102 err = register_for_each_vma(uprobe, NULL);
1103 /* TODO : cant unregister? schedule a worker thread */
1104 if (!uprobe->consumers && !err)
1105 delete_uprobe(uprobe);
1109 * uprobe_unregister - unregister an already registered probe.
1110 * @inode: the file in which the probe has to be removed.
1111 * @offset: offset from the start of the file.
1112 * @uc: identify which probe if multiple probes are colocated.
1114 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
1116 struct uprobe *uprobe;
1118 uprobe = find_uprobe(inode, offset);
1119 if (WARN_ON(!uprobe))
1122 down_write(&uprobe->register_rwsem);
1123 __uprobe_unregister(uprobe, uc);
1124 up_write(&uprobe->register_rwsem);
1127 EXPORT_SYMBOL_GPL(uprobe_unregister);
1130 * __uprobe_register - register a probe
1131 * @inode: the file in which the probe has to be placed.
1132 * @offset: offset from the start of the file.
1133 * @uc: information on howto handle the probe..
1135 * Apart from the access refcount, __uprobe_register() takes a creation
1136 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
1137 * inserted into the rbtree (i.e first consumer for a @inode:@offset
1138 * tuple). Creation refcount stops uprobe_unregister from freeing the
1139 * @uprobe even before the register operation is complete. Creation
1140 * refcount is released when the last @uc for the @uprobe
1141 * unregisters. Caller of __uprobe_register() is required to keep @inode
1142 * (and the containing mount) referenced.
1144 * Return errno if it cannot successully install probes
1145 * else return 0 (success)
1147 static int __uprobe_register(struct inode *inode, loff_t offset,
1148 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1150 struct uprobe *uprobe;
1153 /* Uprobe must have at least one set consumer */
1154 if (!uc->handler && !uc->ret_handler)
1157 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
1158 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
1160 /* Racy, just to catch the obvious mistakes */
1161 if (offset > i_size_read(inode))
1165 uprobe = alloc_uprobe(inode, offset, ref_ctr_offset);
1169 return PTR_ERR(uprobe);
1172 * We can race with uprobe_unregister()->delete_uprobe().
1173 * Check uprobe_is_active() and retry if it is false.
1175 down_write(&uprobe->register_rwsem);
1177 if (likely(uprobe_is_active(uprobe))) {
1178 consumer_add(uprobe, uc);
1179 ret = register_for_each_vma(uprobe, uc);
1181 __uprobe_unregister(uprobe, uc);
1183 up_write(&uprobe->register_rwsem);
1186 if (unlikely(ret == -EAGAIN))
1191 int uprobe_register(struct inode *inode, loff_t offset,
1192 struct uprobe_consumer *uc)
1194 return __uprobe_register(inode, offset, 0, uc);
1196 EXPORT_SYMBOL_GPL(uprobe_register);
1198 int uprobe_register_refctr(struct inode *inode, loff_t offset,
1199 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1201 return __uprobe_register(inode, offset, ref_ctr_offset, uc);
1203 EXPORT_SYMBOL_GPL(uprobe_register_refctr);
1206 * uprobe_apply - unregister an already registered probe.
1207 * @inode: the file in which the probe has to be removed.
1208 * @offset: offset from the start of the file.
1209 * @uc: consumer which wants to add more or remove some breakpoints
1210 * @add: add or remove the breakpoints
1212 int uprobe_apply(struct inode *inode, loff_t offset,
1213 struct uprobe_consumer *uc, bool add)
1215 struct uprobe *uprobe;
1216 struct uprobe_consumer *con;
1219 uprobe = find_uprobe(inode, offset);
1220 if (WARN_ON(!uprobe))
1223 down_write(&uprobe->register_rwsem);
1224 for (con = uprobe->consumers; con && con != uc ; con = con->next)
1227 ret = register_for_each_vma(uprobe, add ? uc : NULL);
1228 up_write(&uprobe->register_rwsem);
1234 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
1236 struct vm_area_struct *vma;
1239 down_read(&mm->mmap_sem);
1240 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1241 unsigned long vaddr;
1244 if (!valid_vma(vma, false) ||
1245 file_inode(vma->vm_file) != uprobe->inode)
1248 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
1249 if (uprobe->offset < offset ||
1250 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
1253 vaddr = offset_to_vaddr(vma, uprobe->offset);
1254 err |= remove_breakpoint(uprobe, mm, vaddr);
1256 up_read(&mm->mmap_sem);
1261 static struct rb_node *
1262 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1264 struct rb_node *n = uprobes_tree.rb_node;
1267 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1269 if (inode < u->inode) {
1271 } else if (inode > u->inode) {
1274 if (max < u->offset)
1276 else if (min > u->offset)
1287 * For a given range in vma, build a list of probes that need to be inserted.
1289 static void build_probe_list(struct inode *inode,
1290 struct vm_area_struct *vma,
1291 unsigned long start, unsigned long end,
1292 struct list_head *head)
1295 struct rb_node *n, *t;
1298 INIT_LIST_HEAD(head);
1299 min = vaddr_to_offset(vma, start);
1300 max = min + (end - start) - 1;
1302 spin_lock(&uprobes_treelock);
1303 n = find_node_in_range(inode, min, max);
1305 for (t = n; t; t = rb_prev(t)) {
1306 u = rb_entry(t, struct uprobe, rb_node);
1307 if (u->inode != inode || u->offset < min)
1309 list_add(&u->pending_list, head);
1312 for (t = n; (t = rb_next(t)); ) {
1313 u = rb_entry(t, struct uprobe, rb_node);
1314 if (u->inode != inode || u->offset > max)
1316 list_add(&u->pending_list, head);
1320 spin_unlock(&uprobes_treelock);
1323 /* @vma contains reference counter, not the probed instruction. */
1324 static int delayed_ref_ctr_inc(struct vm_area_struct *vma)
1326 struct list_head *pos, *q;
1327 struct delayed_uprobe *du;
1328 unsigned long vaddr;
1329 int ret = 0, err = 0;
1331 mutex_lock(&delayed_uprobe_lock);
1332 list_for_each_safe(pos, q, &delayed_uprobe_list) {
1333 du = list_entry(pos, struct delayed_uprobe, list);
1335 if (du->mm != vma->vm_mm ||
1336 !valid_ref_ctr_vma(du->uprobe, vma))
1339 vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
1340 ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
1342 update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1);
1346 delayed_uprobe_delete(du);
1348 mutex_unlock(&delayed_uprobe_lock);
1353 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1355 * Currently we ignore all errors and always return 0, the callers
1356 * can't handle the failure anyway.
1358 int uprobe_mmap(struct vm_area_struct *vma)
1360 struct list_head tmp_list;
1361 struct uprobe *uprobe, *u;
1362 struct inode *inode;
1364 if (no_uprobe_events())
1368 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
1369 test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
1370 delayed_ref_ctr_inc(vma);
1372 if (!valid_vma(vma, true))
1375 inode = file_inode(vma->vm_file);
1379 mutex_lock(uprobes_mmap_hash(inode));
1380 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1382 * We can race with uprobe_unregister(), this uprobe can be already
1383 * removed. But in this case filter_chain() must return false, all
1384 * consumers have gone away.
1386 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1387 if (!fatal_signal_pending(current) &&
1388 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1389 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1390 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1394 mutex_unlock(uprobes_mmap_hash(inode));
1400 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1403 struct inode *inode;
1406 inode = file_inode(vma->vm_file);
1408 min = vaddr_to_offset(vma, start);
1409 max = min + (end - start) - 1;
1411 spin_lock(&uprobes_treelock);
1412 n = find_node_in_range(inode, min, max);
1413 spin_unlock(&uprobes_treelock);
1419 * Called in context of a munmap of a vma.
1421 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1423 if (no_uprobe_events() || !valid_vma(vma, false))
1426 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1429 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1430 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1433 if (vma_has_uprobes(vma, start, end))
1434 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1437 /* Slot allocation for XOL */
1438 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1440 struct vm_area_struct *vma;
1443 if (down_write_killable(&mm->mmap_sem))
1446 if (mm->uprobes_state.xol_area) {
1452 /* Try to map as high as possible, this is only a hint. */
1453 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1455 if (IS_ERR_VALUE(area->vaddr)) {
1461 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1462 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1463 &area->xol_mapping);
1470 /* pairs with get_xol_area() */
1471 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1473 up_write(&mm->mmap_sem);
1478 static struct xol_area *__create_xol_area(unsigned long vaddr)
1480 struct mm_struct *mm = current->mm;
1481 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1482 struct xol_area *area;
1484 area = kmalloc(sizeof(*area), GFP_KERNEL);
1485 if (unlikely(!area))
1488 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1493 area->xol_mapping.name = "[uprobes]";
1494 area->xol_mapping.fault = NULL;
1495 area->xol_mapping.pages = area->pages;
1496 area->pages[0] = alloc_page(GFP_HIGHUSER);
1497 if (!area->pages[0])
1499 area->pages[1] = NULL;
1501 area->vaddr = vaddr;
1502 init_waitqueue_head(&area->wq);
1503 /* Reserve the 1st slot for get_trampoline_vaddr() */
1504 set_bit(0, area->bitmap);
1505 atomic_set(&area->slot_count, 1);
1506 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1508 if (!xol_add_vma(mm, area))
1511 __free_page(area->pages[0]);
1513 kfree(area->bitmap);
1521 * get_xol_area - Allocate process's xol_area if necessary.
1522 * This area will be used for storing instructions for execution out of line.
1524 * Returns the allocated area or NULL.
1526 static struct xol_area *get_xol_area(void)
1528 struct mm_struct *mm = current->mm;
1529 struct xol_area *area;
1531 if (!mm->uprobes_state.xol_area)
1532 __create_xol_area(0);
1534 /* Pairs with xol_add_vma() smp_store_release() */
1535 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1540 * uprobe_clear_state - Free the area allocated for slots.
1542 void uprobe_clear_state(struct mm_struct *mm)
1544 struct xol_area *area = mm->uprobes_state.xol_area;
1546 mutex_lock(&delayed_uprobe_lock);
1547 delayed_uprobe_remove(NULL, mm);
1548 mutex_unlock(&delayed_uprobe_lock);
1553 put_page(area->pages[0]);
1554 kfree(area->bitmap);
1558 void uprobe_start_dup_mmap(void)
1560 percpu_down_read(&dup_mmap_sem);
1563 void uprobe_end_dup_mmap(void)
1565 percpu_up_read(&dup_mmap_sem);
1568 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1570 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1571 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1572 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1573 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1578 * - search for a free slot.
1580 static unsigned long xol_take_insn_slot(struct xol_area *area)
1582 unsigned long slot_addr;
1586 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1587 if (slot_nr < UINSNS_PER_PAGE) {
1588 if (!test_and_set_bit(slot_nr, area->bitmap))
1591 slot_nr = UINSNS_PER_PAGE;
1594 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1595 } while (slot_nr >= UINSNS_PER_PAGE);
1597 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1598 atomic_inc(&area->slot_count);
1604 * xol_get_insn_slot - allocate a slot for xol.
1605 * Returns the allocated slot address or 0.
1607 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1609 struct xol_area *area;
1610 unsigned long xol_vaddr;
1612 area = get_xol_area();
1616 xol_vaddr = xol_take_insn_slot(area);
1617 if (unlikely(!xol_vaddr))
1620 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1621 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1627 * xol_free_insn_slot - If slot was earlier allocated by
1628 * @xol_get_insn_slot(), make the slot available for
1629 * subsequent requests.
1631 static void xol_free_insn_slot(struct task_struct *tsk)
1633 struct xol_area *area;
1634 unsigned long vma_end;
1635 unsigned long slot_addr;
1637 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1640 slot_addr = tsk->utask->xol_vaddr;
1641 if (unlikely(!slot_addr))
1644 area = tsk->mm->uprobes_state.xol_area;
1645 vma_end = area->vaddr + PAGE_SIZE;
1646 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1647 unsigned long offset;
1650 offset = slot_addr - area->vaddr;
1651 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1652 if (slot_nr >= UINSNS_PER_PAGE)
1655 clear_bit(slot_nr, area->bitmap);
1656 atomic_dec(&area->slot_count);
1657 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1658 if (waitqueue_active(&area->wq))
1661 tsk->utask->xol_vaddr = 0;
1665 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1666 void *src, unsigned long len)
1668 /* Initialize the slot */
1669 copy_to_page(page, vaddr, src, len);
1672 * We probably need flush_icache_user_range() but it needs vma.
1673 * This should work on most of architectures by default. If
1674 * architecture needs to do something different it can define
1675 * its own version of the function.
1677 flush_dcache_page(page);
1681 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1682 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1684 * Return the address of the breakpoint instruction.
1686 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1688 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1691 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1693 struct uprobe_task *utask = current->utask;
1695 if (unlikely(utask && utask->active_uprobe))
1696 return utask->vaddr;
1698 return instruction_pointer(regs);
1701 static struct return_instance *free_ret_instance(struct return_instance *ri)
1703 struct return_instance *next = ri->next;
1704 put_uprobe(ri->uprobe);
1710 * Called with no locks held.
1711 * Called in context of an exiting or an exec-ing thread.
1713 void uprobe_free_utask(struct task_struct *t)
1715 struct uprobe_task *utask = t->utask;
1716 struct return_instance *ri;
1721 if (utask->active_uprobe)
1722 put_uprobe(utask->active_uprobe);
1724 ri = utask->return_instances;
1726 ri = free_ret_instance(ri);
1728 xol_free_insn_slot(t);
1734 * Allocate a uprobe_task object for the task if if necessary.
1735 * Called when the thread hits a breakpoint.
1738 * - pointer to new uprobe_task on success
1741 static struct uprobe_task *get_utask(void)
1743 if (!current->utask)
1744 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1745 return current->utask;
1748 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1750 struct uprobe_task *n_utask;
1751 struct return_instance **p, *o, *n;
1753 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1758 p = &n_utask->return_instances;
1759 for (o = o_utask->return_instances; o; o = o->next) {
1760 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1765 get_uprobe(n->uprobe);
1776 static void uprobe_warn(struct task_struct *t, const char *msg)
1778 pr_warn("uprobe: %s:%d failed to %s\n",
1779 current->comm, current->pid, msg);
1782 static void dup_xol_work(struct callback_head *work)
1784 if (current->flags & PF_EXITING)
1787 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1788 !fatal_signal_pending(current))
1789 uprobe_warn(current, "dup xol area");
1793 * Called in context of a new clone/fork from copy_process.
1795 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1797 struct uprobe_task *utask = current->utask;
1798 struct mm_struct *mm = current->mm;
1799 struct xol_area *area;
1803 if (!utask || !utask->return_instances)
1806 if (mm == t->mm && !(flags & CLONE_VFORK))
1809 if (dup_utask(t, utask))
1810 return uprobe_warn(t, "dup ret instances");
1812 /* The task can fork() after dup_xol_work() fails */
1813 area = mm->uprobes_state.xol_area;
1815 return uprobe_warn(t, "dup xol area");
1820 t->utask->dup_xol_addr = area->vaddr;
1821 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1822 task_work_add(t, &t->utask->dup_xol_work, true);
1826 * Current area->vaddr notion assume the trampoline address is always
1827 * equal area->vaddr.
1829 * Returns -1 in case the xol_area is not allocated.
1831 static unsigned long get_trampoline_vaddr(void)
1833 struct xol_area *area;
1834 unsigned long trampoline_vaddr = -1;
1836 /* Pairs with xol_add_vma() smp_store_release() */
1837 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1839 trampoline_vaddr = area->vaddr;
1841 return trampoline_vaddr;
1844 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1845 struct pt_regs *regs)
1847 struct return_instance *ri = utask->return_instances;
1848 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1850 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1851 ri = free_ret_instance(ri);
1854 utask->return_instances = ri;
1857 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1859 struct return_instance *ri;
1860 struct uprobe_task *utask;
1861 unsigned long orig_ret_vaddr, trampoline_vaddr;
1864 if (!get_xol_area())
1867 utask = get_utask();
1871 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1872 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1873 " nestedness limit pid/tgid=%d/%d\n",
1874 current->pid, current->tgid);
1878 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1882 trampoline_vaddr = get_trampoline_vaddr();
1883 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1884 if (orig_ret_vaddr == -1)
1887 /* drop the entries invalidated by longjmp() */
1888 chained = (orig_ret_vaddr == trampoline_vaddr);
1889 cleanup_return_instances(utask, chained, regs);
1892 * We don't want to keep trampoline address in stack, rather keep the
1893 * original return address of first caller thru all the consequent
1894 * instances. This also makes breakpoint unwrapping easier.
1897 if (!utask->return_instances) {
1899 * This situation is not possible. Likely we have an
1900 * attack from user-space.
1902 uprobe_warn(current, "handle tail call");
1905 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1908 ri->uprobe = get_uprobe(uprobe);
1909 ri->func = instruction_pointer(regs);
1910 ri->stack = user_stack_pointer(regs);
1911 ri->orig_ret_vaddr = orig_ret_vaddr;
1912 ri->chained = chained;
1915 ri->next = utask->return_instances;
1916 utask->return_instances = ri;
1923 /* Prepare to single-step probed instruction out of line. */
1925 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1927 struct uprobe_task *utask;
1928 unsigned long xol_vaddr;
1931 utask = get_utask();
1935 xol_vaddr = xol_get_insn_slot(uprobe);
1939 utask->xol_vaddr = xol_vaddr;
1940 utask->vaddr = bp_vaddr;
1942 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1943 if (unlikely(err)) {
1944 xol_free_insn_slot(current);
1948 utask->active_uprobe = uprobe;
1949 utask->state = UTASK_SSTEP;
1954 * If we are singlestepping, then ensure this thread is not connected to
1955 * non-fatal signals until completion of singlestep. When xol insn itself
1956 * triggers the signal, restart the original insn even if the task is
1957 * already SIGKILL'ed (since coredump should report the correct ip). This
1958 * is even more important if the task has a handler for SIGSEGV/etc, The
1959 * _same_ instruction should be repeated again after return from the signal
1960 * handler, and SSTEP can never finish in this case.
1962 bool uprobe_deny_signal(void)
1964 struct task_struct *t = current;
1965 struct uprobe_task *utask = t->utask;
1967 if (likely(!utask || !utask->active_uprobe))
1970 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1972 if (signal_pending(t)) {
1973 spin_lock_irq(&t->sighand->siglock);
1974 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1975 spin_unlock_irq(&t->sighand->siglock);
1977 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1978 utask->state = UTASK_SSTEP_TRAPPED;
1979 set_tsk_thread_flag(t, TIF_UPROBE);
1986 static void mmf_recalc_uprobes(struct mm_struct *mm)
1988 struct vm_area_struct *vma;
1990 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1991 if (!valid_vma(vma, false))
1994 * This is not strictly accurate, we can race with
1995 * uprobe_unregister() and see the already removed
1996 * uprobe if delete_uprobe() was not yet called.
1997 * Or this uprobe can be filtered out.
1999 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
2003 clear_bit(MMF_HAS_UPROBES, &mm->flags);
2006 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
2009 uprobe_opcode_t opcode;
2012 pagefault_disable();
2013 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
2016 if (likely(result == 0))
2020 * The NULL 'tsk' here ensures that any faults that occur here
2021 * will not be accounted to the task. 'mm' *is* current->mm,
2022 * but we treat this as a 'remote' access since it is
2023 * essentially a kernel access to the memory.
2025 result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
2030 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
2033 /* This needs to return true for any variant of the trap insn */
2034 return is_trap_insn(&opcode);
2037 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
2039 struct mm_struct *mm = current->mm;
2040 struct uprobe *uprobe = NULL;
2041 struct vm_area_struct *vma;
2043 down_read(&mm->mmap_sem);
2044 vma = find_vma(mm, bp_vaddr);
2045 if (vma && vma->vm_start <= bp_vaddr) {
2046 if (valid_vma(vma, false)) {
2047 struct inode *inode = file_inode(vma->vm_file);
2048 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
2050 uprobe = find_uprobe(inode, offset);
2054 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
2059 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
2060 mmf_recalc_uprobes(mm);
2061 up_read(&mm->mmap_sem);
2066 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
2068 struct uprobe_consumer *uc;
2069 int remove = UPROBE_HANDLER_REMOVE;
2070 bool need_prep = false; /* prepare return uprobe, when needed */
2072 down_read(&uprobe->register_rwsem);
2073 for (uc = uprobe->consumers; uc; uc = uc->next) {
2077 rc = uc->handler(uc, regs);
2078 WARN(rc & ~UPROBE_HANDLER_MASK,
2079 "bad rc=0x%x from %ps()\n", rc, uc->handler);
2082 if (uc->ret_handler)
2088 if (need_prep && !remove)
2089 prepare_uretprobe(uprobe, regs); /* put bp at return */
2091 if (remove && uprobe->consumers) {
2092 WARN_ON(!uprobe_is_active(uprobe));
2093 unapply_uprobe(uprobe, current->mm);
2095 up_read(&uprobe->register_rwsem);
2099 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
2101 struct uprobe *uprobe = ri->uprobe;
2102 struct uprobe_consumer *uc;
2104 down_read(&uprobe->register_rwsem);
2105 for (uc = uprobe->consumers; uc; uc = uc->next) {
2106 if (uc->ret_handler)
2107 uc->ret_handler(uc, ri->func, regs);
2109 up_read(&uprobe->register_rwsem);
2112 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
2117 chained = ri->chained;
2118 ri = ri->next; /* can't be NULL if chained */
2124 static void handle_trampoline(struct pt_regs *regs)
2126 struct uprobe_task *utask;
2127 struct return_instance *ri, *next;
2130 utask = current->utask;
2134 ri = utask->return_instances;
2140 * We should throw out the frames invalidated by longjmp().
2141 * If this chain is valid, then the next one should be alive
2142 * or NULL; the latter case means that nobody but ri->func
2143 * could hit this trampoline on return. TODO: sigaltstack().
2145 next = find_next_ret_chain(ri);
2146 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
2148 instruction_pointer_set(regs, ri->orig_ret_vaddr);
2151 handle_uretprobe_chain(ri, regs);
2152 ri = free_ret_instance(ri);
2154 } while (ri != next);
2157 utask->return_instances = ri;
2161 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
2166 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
2171 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
2172 struct pt_regs *regs)
2178 * Run handler and ask thread to singlestep.
2179 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2181 static void handle_swbp(struct pt_regs *regs)
2183 struct uprobe *uprobe;
2184 unsigned long bp_vaddr;
2185 int uninitialized_var(is_swbp);
2187 bp_vaddr = uprobe_get_swbp_addr(regs);
2188 if (bp_vaddr == get_trampoline_vaddr())
2189 return handle_trampoline(regs);
2191 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
2194 /* No matching uprobe; signal SIGTRAP. */
2195 send_sig(SIGTRAP, current, 0);
2198 * Either we raced with uprobe_unregister() or we can't
2199 * access this memory. The latter is only possible if
2200 * another thread plays with our ->mm. In both cases
2201 * we can simply restart. If this vma was unmapped we
2202 * can pretend this insn was not executed yet and get
2203 * the (correct) SIGSEGV after restart.
2205 instruction_pointer_set(regs, bp_vaddr);
2210 /* change it in advance for ->handler() and restart */
2211 instruction_pointer_set(regs, bp_vaddr);
2214 * TODO: move copy_insn/etc into _register and remove this hack.
2215 * After we hit the bp, _unregister + _register can install the
2216 * new and not-yet-analyzed uprobe at the same address, restart.
2218 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
2222 * Pairs with the smp_wmb() in prepare_uprobe().
2224 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2225 * we must also see the stores to &uprobe->arch performed by the
2226 * prepare_uprobe() call.
2230 /* Tracing handlers use ->utask to communicate with fetch methods */
2234 if (arch_uprobe_ignore(&uprobe->arch, regs))
2237 handler_chain(uprobe, regs);
2239 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
2242 if (!pre_ssout(uprobe, regs, bp_vaddr))
2245 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2251 * Perform required fix-ups and disable singlestep.
2252 * Allow pending signals to take effect.
2254 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
2256 struct uprobe *uprobe;
2259 uprobe = utask->active_uprobe;
2260 if (utask->state == UTASK_SSTEP_ACK)
2261 err = arch_uprobe_post_xol(&uprobe->arch, regs);
2262 else if (utask->state == UTASK_SSTEP_TRAPPED)
2263 arch_uprobe_abort_xol(&uprobe->arch, regs);
2268 utask->active_uprobe = NULL;
2269 utask->state = UTASK_RUNNING;
2270 xol_free_insn_slot(current);
2272 spin_lock_irq(¤t->sighand->siglock);
2273 recalc_sigpending(); /* see uprobe_deny_signal() */
2274 spin_unlock_irq(¤t->sighand->siglock);
2276 if (unlikely(err)) {
2277 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
2283 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2284 * allows the thread to return from interrupt. After that handle_swbp()
2285 * sets utask->active_uprobe.
2287 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2288 * and allows the thread to return from interrupt.
2290 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2291 * uprobe_notify_resume().
2293 void uprobe_notify_resume(struct pt_regs *regs)
2295 struct uprobe_task *utask;
2297 clear_thread_flag(TIF_UPROBE);
2299 utask = current->utask;
2300 if (utask && utask->active_uprobe)
2301 handle_singlestep(utask, regs);
2307 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2308 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2310 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2315 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2316 (!current->utask || !current->utask->return_instances))
2319 set_thread_flag(TIF_UPROBE);
2324 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2325 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2327 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2329 struct uprobe_task *utask = current->utask;
2331 if (!current->mm || !utask || !utask->active_uprobe)
2332 /* task is currently not uprobed */
2335 utask->state = UTASK_SSTEP_ACK;
2336 set_thread_flag(TIF_UPROBE);
2340 static struct notifier_block uprobe_exception_nb = {
2341 .notifier_call = arch_uprobe_exception_notify,
2342 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2345 void __init uprobes_init(void)
2349 for (i = 0; i < UPROBES_HASH_SZ; i++)
2350 mutex_init(&uprobes_mmap_mutex[i]);
2352 BUG_ON(register_die_notifier(&uprobe_exception_nb));