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, mm, addr, addr + PAGE_SIZE);
166 VM_BUG_ON_PAGE(PageTransHuge(old_page), old_page);
168 err = mem_cgroup_try_charge(new_page, vma->vm_mm, GFP_KERNEL, &memcg,
173 /* For try_to_free_swap() and munlock_vma_page() below */
176 mmu_notifier_invalidate_range_start(&range);
178 if (!page_vma_mapped_walk(&pvmw)) {
179 mem_cgroup_cancel_charge(new_page, memcg, false);
182 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
185 page_add_new_anon_rmap(new_page, vma, addr, false);
186 mem_cgroup_commit_charge(new_page, memcg, false, false);
187 lru_cache_add_active_or_unevictable(new_page, vma);
189 if (!PageAnon(old_page)) {
190 dec_mm_counter(mm, mm_counter_file(old_page));
191 inc_mm_counter(mm, MM_ANONPAGES);
194 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
195 ptep_clear_flush_notify(vma, addr, pvmw.pte);
196 set_pte_at_notify(mm, addr, pvmw.pte,
197 mk_pte(new_page, vma->vm_page_prot));
199 page_remove_rmap(old_page, false);
200 if (!page_mapped(old_page))
201 try_to_free_swap(old_page);
202 page_vma_mapped_walk_done(&pvmw);
204 if (vma->vm_flags & VM_LOCKED)
205 munlock_vma_page(old_page);
210 mmu_notifier_invalidate_range_end(&range);
211 unlock_page(old_page);
216 * is_swbp_insn - check if instruction is breakpoint instruction.
217 * @insn: instruction to be checked.
218 * Default implementation of is_swbp_insn
219 * Returns true if @insn is a breakpoint instruction.
221 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
223 return *insn == UPROBE_SWBP_INSN;
227 * is_trap_insn - check if instruction is breakpoint instruction.
228 * @insn: instruction to be checked.
229 * Default implementation of is_trap_insn
230 * Returns true if @insn is a breakpoint instruction.
232 * This function is needed for the case where an architecture has multiple
233 * trap instructions (like powerpc).
235 bool __weak is_trap_insn(uprobe_opcode_t *insn)
237 return is_swbp_insn(insn);
240 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
242 void *kaddr = kmap_atomic(page);
243 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
244 kunmap_atomic(kaddr);
247 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
249 void *kaddr = kmap_atomic(page);
250 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
251 kunmap_atomic(kaddr);
254 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
256 uprobe_opcode_t old_opcode;
260 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
261 * We do not check if it is any other 'trap variant' which could
262 * be conditional trap instruction such as the one powerpc supports.
264 * The logic is that we do not care if the underlying instruction
265 * is a trap variant; uprobes always wins over any other (gdb)
268 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
269 is_swbp = is_swbp_insn(&old_opcode);
271 if (is_swbp_insn(new_opcode)) {
272 if (is_swbp) /* register: already installed? */
275 if (!is_swbp) /* unregister: was it changed by us? */
282 static struct delayed_uprobe *
283 delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm)
285 struct delayed_uprobe *du;
287 list_for_each_entry(du, &delayed_uprobe_list, list)
288 if (du->uprobe == uprobe && du->mm == mm)
293 static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
295 struct delayed_uprobe *du;
297 if (delayed_uprobe_check(uprobe, mm))
300 du = kzalloc(sizeof(*du), GFP_KERNEL);
306 list_add(&du->list, &delayed_uprobe_list);
310 static void delayed_uprobe_delete(struct delayed_uprobe *du)
318 static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm)
320 struct list_head *pos, *q;
321 struct delayed_uprobe *du;
326 list_for_each_safe(pos, q, &delayed_uprobe_list) {
327 du = list_entry(pos, struct delayed_uprobe, list);
329 if (uprobe && du->uprobe != uprobe)
331 if (mm && du->mm != mm)
334 delayed_uprobe_delete(du);
338 static bool valid_ref_ctr_vma(struct uprobe *uprobe,
339 struct vm_area_struct *vma)
341 unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
343 return uprobe->ref_ctr_offset &&
345 file_inode(vma->vm_file) == uprobe->inode &&
346 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
347 vma->vm_start <= vaddr &&
351 static struct vm_area_struct *
352 find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm)
354 struct vm_area_struct *tmp;
356 for (tmp = mm->mmap; tmp; tmp = tmp->vm_next)
357 if (valid_ref_ctr_vma(uprobe, tmp))
364 __update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d)
368 struct vm_area_struct *vma;
375 ret = get_user_pages_remote(NULL, mm, vaddr, 1,
376 FOLL_WRITE, &page, &vma, NULL);
377 if (unlikely(ret <= 0)) {
379 * We are asking for 1 page. If get_user_pages_remote() fails,
380 * it may return 0, in that case we have to return error.
382 return ret == 0 ? -EBUSY : ret;
385 kaddr = kmap_atomic(page);
386 ptr = kaddr + (vaddr & ~PAGE_MASK);
388 if (unlikely(*ptr + d < 0)) {
389 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
390 "curr val: %d, delta: %d\n", vaddr, *ptr, d);
398 kunmap_atomic(kaddr);
403 static void update_ref_ctr_warn(struct uprobe *uprobe,
404 struct mm_struct *mm, short d)
406 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
407 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
408 d > 0 ? "increment" : "decrement", uprobe->inode->i_ino,
409 (unsigned long long) uprobe->offset,
410 (unsigned long long) uprobe->ref_ctr_offset, mm);
413 static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
416 struct vm_area_struct *rc_vma;
417 unsigned long rc_vaddr;
420 rc_vma = find_ref_ctr_vma(uprobe, mm);
423 rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset);
424 ret = __update_ref_ctr(mm, rc_vaddr, d);
426 update_ref_ctr_warn(uprobe, mm, d);
432 mutex_lock(&delayed_uprobe_lock);
434 ret = delayed_uprobe_add(uprobe, mm);
436 delayed_uprobe_remove(uprobe, mm);
437 mutex_unlock(&delayed_uprobe_lock);
444 * Expect the breakpoint instruction to be the smallest size instruction for
445 * the architecture. If an arch has variable length instruction and the
446 * breakpoint instruction is not of the smallest length instruction
447 * supported by that architecture then we need to modify is_trap_at_addr and
448 * uprobe_write_opcode accordingly. This would never be a problem for archs
449 * that have fixed length instructions.
451 * uprobe_write_opcode - write the opcode at a given virtual address.
452 * @mm: the probed process address space.
453 * @vaddr: the virtual address to store the opcode.
454 * @opcode: opcode to be written at @vaddr.
456 * Called with mm->mmap_sem held for write.
457 * Return 0 (success) or a negative errno.
459 int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
460 unsigned long vaddr, uprobe_opcode_t opcode)
462 struct uprobe *uprobe;
463 struct page *old_page, *new_page;
464 struct vm_area_struct *vma;
465 int ret, is_register, ref_ctr_updated = 0;
467 is_register = is_swbp_insn(&opcode);
468 uprobe = container_of(auprobe, struct uprobe, arch);
471 /* Read the page with vaddr into memory */
472 ret = get_user_pages_remote(NULL, mm, vaddr, 1,
473 FOLL_FORCE | FOLL_SPLIT, &old_page, &vma, NULL);
477 ret = verify_opcode(old_page, vaddr, &opcode);
481 /* We are going to replace instruction, update ref_ctr. */
482 if (!ref_ctr_updated && uprobe->ref_ctr_offset) {
483 ret = update_ref_ctr(uprobe, mm, is_register ? 1 : -1);
490 ret = anon_vma_prepare(vma);
495 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
499 __SetPageUptodate(new_page);
500 copy_highpage(new_page, old_page);
501 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
503 ret = __replace_page(vma, vaddr, old_page, new_page);
508 if (unlikely(ret == -EAGAIN))
511 /* Revert back reference counter if instruction update failed. */
512 if (ret && is_register && ref_ctr_updated)
513 update_ref_ctr(uprobe, mm, -1);
519 * set_swbp - store breakpoint at a given address.
520 * @auprobe: arch specific probepoint information.
521 * @mm: the probed process address space.
522 * @vaddr: the virtual address to insert the opcode.
524 * For mm @mm, store the breakpoint instruction at @vaddr.
525 * Return 0 (success) or a negative errno.
527 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
529 return uprobe_write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
533 * set_orig_insn - Restore the original instruction.
534 * @mm: the probed process address space.
535 * @auprobe: arch specific probepoint information.
536 * @vaddr: the virtual address to insert the opcode.
538 * For mm @mm, restore the original opcode (opcode) at @vaddr.
539 * Return 0 (success) or a negative errno.
542 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
544 return uprobe_write_opcode(auprobe, mm, vaddr,
545 *(uprobe_opcode_t *)&auprobe->insn);
548 static struct uprobe *get_uprobe(struct uprobe *uprobe)
550 atomic_inc(&uprobe->ref);
554 static void put_uprobe(struct uprobe *uprobe)
556 if (atomic_dec_and_test(&uprobe->ref)) {
558 * If application munmap(exec_vma) before uprobe_unregister()
559 * gets called, we don't get a chance to remove uprobe from
560 * delayed_uprobe_list from remove_breakpoint(). Do it here.
562 mutex_lock(&delayed_uprobe_lock);
563 delayed_uprobe_remove(uprobe, NULL);
564 mutex_unlock(&delayed_uprobe_lock);
569 static int match_uprobe(struct uprobe *l, struct uprobe *r)
571 if (l->inode < r->inode)
574 if (l->inode > r->inode)
577 if (l->offset < r->offset)
580 if (l->offset > r->offset)
586 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
588 struct uprobe u = { .inode = inode, .offset = offset };
589 struct rb_node *n = uprobes_tree.rb_node;
590 struct uprobe *uprobe;
594 uprobe = rb_entry(n, struct uprobe, rb_node);
595 match = match_uprobe(&u, uprobe);
597 return get_uprobe(uprobe);
608 * Find a uprobe corresponding to a given inode:offset
609 * Acquires uprobes_treelock
611 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
613 struct uprobe *uprobe;
615 spin_lock(&uprobes_treelock);
616 uprobe = __find_uprobe(inode, offset);
617 spin_unlock(&uprobes_treelock);
622 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
624 struct rb_node **p = &uprobes_tree.rb_node;
625 struct rb_node *parent = NULL;
631 u = rb_entry(parent, struct uprobe, rb_node);
632 match = match_uprobe(uprobe, u);
634 return get_uprobe(u);
637 p = &parent->rb_left;
639 p = &parent->rb_right;
644 rb_link_node(&uprobe->rb_node, parent, p);
645 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
646 /* get access + creation ref */
647 atomic_set(&uprobe->ref, 2);
653 * Acquire uprobes_treelock.
654 * Matching uprobe already exists in rbtree;
655 * increment (access refcount) and return the matching uprobe.
657 * No matching uprobe; insert the uprobe in rb_tree;
658 * get a double refcount (access + creation) and return NULL.
660 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
664 spin_lock(&uprobes_treelock);
665 u = __insert_uprobe(uprobe);
666 spin_unlock(&uprobes_treelock);
672 ref_ctr_mismatch_warn(struct uprobe *cur_uprobe, struct uprobe *uprobe)
674 pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx "
675 "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n",
676 uprobe->inode->i_ino, (unsigned long long) uprobe->offset,
677 (unsigned long long) cur_uprobe->ref_ctr_offset,
678 (unsigned long long) uprobe->ref_ctr_offset);
681 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
682 loff_t ref_ctr_offset)
684 struct uprobe *uprobe, *cur_uprobe;
686 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
690 uprobe->inode = inode;
691 uprobe->offset = offset;
692 uprobe->ref_ctr_offset = ref_ctr_offset;
693 init_rwsem(&uprobe->register_rwsem);
694 init_rwsem(&uprobe->consumer_rwsem);
696 /* add to uprobes_tree, sorted on inode:offset */
697 cur_uprobe = insert_uprobe(uprobe);
698 /* a uprobe exists for this inode:offset combination */
700 if (cur_uprobe->ref_ctr_offset != uprobe->ref_ctr_offset) {
701 ref_ctr_mismatch_warn(cur_uprobe, uprobe);
702 put_uprobe(cur_uprobe);
704 return ERR_PTR(-EINVAL);
713 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
715 down_write(&uprobe->consumer_rwsem);
716 uc->next = uprobe->consumers;
717 uprobe->consumers = uc;
718 up_write(&uprobe->consumer_rwsem);
722 * For uprobe @uprobe, delete the consumer @uc.
723 * Return true if the @uc is deleted successfully
726 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
728 struct uprobe_consumer **con;
731 down_write(&uprobe->consumer_rwsem);
732 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
739 up_write(&uprobe->consumer_rwsem);
744 static int __copy_insn(struct address_space *mapping, struct file *filp,
745 void *insn, int nbytes, loff_t offset)
749 * Ensure that the page that has the original instruction is populated
750 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
751 * see uprobe_register().
753 if (mapping->a_ops->readpage)
754 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
756 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
758 return PTR_ERR(page);
760 copy_from_page(page, offset, insn, nbytes);
766 static int copy_insn(struct uprobe *uprobe, struct file *filp)
768 struct address_space *mapping = uprobe->inode->i_mapping;
769 loff_t offs = uprobe->offset;
770 void *insn = &uprobe->arch.insn;
771 int size = sizeof(uprobe->arch.insn);
774 /* Copy only available bytes, -EIO if nothing was read */
776 if (offs >= i_size_read(uprobe->inode))
779 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
780 err = __copy_insn(mapping, filp, insn, len, offs);
792 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
793 struct mm_struct *mm, unsigned long vaddr)
797 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
800 /* TODO: move this into _register, until then we abuse this sem. */
801 down_write(&uprobe->consumer_rwsem);
802 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
805 ret = copy_insn(uprobe, file);
810 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
813 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
817 /* uprobe_write_opcode() assumes we don't cross page boundary */
818 BUG_ON((uprobe->offset & ~PAGE_MASK) +
819 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
821 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
822 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
825 up_write(&uprobe->consumer_rwsem);
830 static inline bool consumer_filter(struct uprobe_consumer *uc,
831 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
833 return !uc->filter || uc->filter(uc, ctx, mm);
836 static bool filter_chain(struct uprobe *uprobe,
837 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
839 struct uprobe_consumer *uc;
842 down_read(&uprobe->consumer_rwsem);
843 for (uc = uprobe->consumers; uc; uc = uc->next) {
844 ret = consumer_filter(uc, ctx, mm);
848 up_read(&uprobe->consumer_rwsem);
854 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
855 struct vm_area_struct *vma, unsigned long vaddr)
860 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
865 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
866 * the task can hit this breakpoint right after __replace_page().
868 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
870 set_bit(MMF_HAS_UPROBES, &mm->flags);
872 ret = set_swbp(&uprobe->arch, mm, vaddr);
874 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
875 else if (first_uprobe)
876 clear_bit(MMF_HAS_UPROBES, &mm->flags);
882 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
884 set_bit(MMF_RECALC_UPROBES, &mm->flags);
885 return set_orig_insn(&uprobe->arch, mm, vaddr);
888 static inline bool uprobe_is_active(struct uprobe *uprobe)
890 return !RB_EMPTY_NODE(&uprobe->rb_node);
893 * There could be threads that have already hit the breakpoint. They
894 * will recheck the current insn and restart if find_uprobe() fails.
895 * See find_active_uprobe().
897 static void delete_uprobe(struct uprobe *uprobe)
899 if (WARN_ON(!uprobe_is_active(uprobe)))
902 spin_lock(&uprobes_treelock);
903 rb_erase(&uprobe->rb_node, &uprobes_tree);
904 spin_unlock(&uprobes_treelock);
905 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
910 struct map_info *next;
911 struct mm_struct *mm;
915 static inline struct map_info *free_map_info(struct map_info *info)
917 struct map_info *next = info->next;
922 static struct map_info *
923 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
925 unsigned long pgoff = offset >> PAGE_SHIFT;
926 struct vm_area_struct *vma;
927 struct map_info *curr = NULL;
928 struct map_info *prev = NULL;
929 struct map_info *info;
933 i_mmap_lock_read(mapping);
934 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
935 if (!valid_vma(vma, is_register))
938 if (!prev && !more) {
940 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
941 * reclaim. This is optimistic, no harm done if it fails.
943 prev = kmalloc(sizeof(struct map_info),
944 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
953 if (!mmget_not_zero(vma->vm_mm))
961 info->mm = vma->vm_mm;
962 info->vaddr = offset_to_vaddr(vma, offset);
964 i_mmap_unlock_read(mapping);
976 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
978 curr = ERR_PTR(-ENOMEM);
988 prev = free_map_info(prev);
993 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
995 bool is_register = !!new;
996 struct map_info *info;
999 percpu_down_write(&dup_mmap_sem);
1000 info = build_map_info(uprobe->inode->i_mapping,
1001 uprobe->offset, is_register);
1003 err = PTR_ERR(info);
1008 struct mm_struct *mm = info->mm;
1009 struct vm_area_struct *vma;
1011 if (err && is_register)
1014 down_write(&mm->mmap_sem);
1015 vma = find_vma(mm, info->vaddr);
1016 if (!vma || !valid_vma(vma, is_register) ||
1017 file_inode(vma->vm_file) != uprobe->inode)
1020 if (vma->vm_start > info->vaddr ||
1021 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
1025 /* consult only the "caller", new consumer. */
1026 if (consumer_filter(new,
1027 UPROBE_FILTER_REGISTER, mm))
1028 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
1029 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
1030 if (!filter_chain(uprobe,
1031 UPROBE_FILTER_UNREGISTER, mm))
1032 err |= remove_breakpoint(uprobe, mm, info->vaddr);
1036 up_write(&mm->mmap_sem);
1039 info = free_map_info(info);
1042 percpu_up_write(&dup_mmap_sem);
1047 __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
1051 if (WARN_ON(!consumer_del(uprobe, uc)))
1054 err = register_for_each_vma(uprobe, NULL);
1055 /* TODO : cant unregister? schedule a worker thread */
1056 if (!uprobe->consumers && !err)
1057 delete_uprobe(uprobe);
1061 * uprobe_unregister - unregister an already registered probe.
1062 * @inode: the file in which the probe has to be removed.
1063 * @offset: offset from the start of the file.
1064 * @uc: identify which probe if multiple probes are colocated.
1066 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
1068 struct uprobe *uprobe;
1070 uprobe = find_uprobe(inode, offset);
1071 if (WARN_ON(!uprobe))
1074 down_write(&uprobe->register_rwsem);
1075 __uprobe_unregister(uprobe, uc);
1076 up_write(&uprobe->register_rwsem);
1079 EXPORT_SYMBOL_GPL(uprobe_unregister);
1082 * __uprobe_register - register a probe
1083 * @inode: the file in which the probe has to be placed.
1084 * @offset: offset from the start of the file.
1085 * @uc: information on howto handle the probe..
1087 * Apart from the access refcount, __uprobe_register() takes a creation
1088 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
1089 * inserted into the rbtree (i.e first consumer for a @inode:@offset
1090 * tuple). Creation refcount stops uprobe_unregister from freeing the
1091 * @uprobe even before the register operation is complete. Creation
1092 * refcount is released when the last @uc for the @uprobe
1093 * unregisters. Caller of __uprobe_register() is required to keep @inode
1094 * (and the containing mount) referenced.
1096 * Return errno if it cannot successully install probes
1097 * else return 0 (success)
1099 static int __uprobe_register(struct inode *inode, loff_t offset,
1100 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1102 struct uprobe *uprobe;
1105 /* Uprobe must have at least one set consumer */
1106 if (!uc->handler && !uc->ret_handler)
1109 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
1110 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
1112 /* Racy, just to catch the obvious mistakes */
1113 if (offset > i_size_read(inode))
1117 uprobe = alloc_uprobe(inode, offset, ref_ctr_offset);
1121 return PTR_ERR(uprobe);
1124 * We can race with uprobe_unregister()->delete_uprobe().
1125 * Check uprobe_is_active() and retry if it is false.
1127 down_write(&uprobe->register_rwsem);
1129 if (likely(uprobe_is_active(uprobe))) {
1130 consumer_add(uprobe, uc);
1131 ret = register_for_each_vma(uprobe, uc);
1133 __uprobe_unregister(uprobe, uc);
1135 up_write(&uprobe->register_rwsem);
1138 if (unlikely(ret == -EAGAIN))
1143 int uprobe_register(struct inode *inode, loff_t offset,
1144 struct uprobe_consumer *uc)
1146 return __uprobe_register(inode, offset, 0, uc);
1148 EXPORT_SYMBOL_GPL(uprobe_register);
1150 int uprobe_register_refctr(struct inode *inode, loff_t offset,
1151 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1153 return __uprobe_register(inode, offset, ref_ctr_offset, uc);
1155 EXPORT_SYMBOL_GPL(uprobe_register_refctr);
1158 * uprobe_apply - unregister an already registered probe.
1159 * @inode: the file in which the probe has to be removed.
1160 * @offset: offset from the start of the file.
1161 * @uc: consumer which wants to add more or remove some breakpoints
1162 * @add: add or remove the breakpoints
1164 int uprobe_apply(struct inode *inode, loff_t offset,
1165 struct uprobe_consumer *uc, bool add)
1167 struct uprobe *uprobe;
1168 struct uprobe_consumer *con;
1171 uprobe = find_uprobe(inode, offset);
1172 if (WARN_ON(!uprobe))
1175 down_write(&uprobe->register_rwsem);
1176 for (con = uprobe->consumers; con && con != uc ; con = con->next)
1179 ret = register_for_each_vma(uprobe, add ? uc : NULL);
1180 up_write(&uprobe->register_rwsem);
1186 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
1188 struct vm_area_struct *vma;
1191 down_read(&mm->mmap_sem);
1192 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1193 unsigned long vaddr;
1196 if (!valid_vma(vma, false) ||
1197 file_inode(vma->vm_file) != uprobe->inode)
1200 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
1201 if (uprobe->offset < offset ||
1202 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
1205 vaddr = offset_to_vaddr(vma, uprobe->offset);
1206 err |= remove_breakpoint(uprobe, mm, vaddr);
1208 up_read(&mm->mmap_sem);
1213 static struct rb_node *
1214 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1216 struct rb_node *n = uprobes_tree.rb_node;
1219 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1221 if (inode < u->inode) {
1223 } else if (inode > u->inode) {
1226 if (max < u->offset)
1228 else if (min > u->offset)
1239 * For a given range in vma, build a list of probes that need to be inserted.
1241 static void build_probe_list(struct inode *inode,
1242 struct vm_area_struct *vma,
1243 unsigned long start, unsigned long end,
1244 struct list_head *head)
1247 struct rb_node *n, *t;
1250 INIT_LIST_HEAD(head);
1251 min = vaddr_to_offset(vma, start);
1252 max = min + (end - start) - 1;
1254 spin_lock(&uprobes_treelock);
1255 n = find_node_in_range(inode, min, max);
1257 for (t = n; t; t = rb_prev(t)) {
1258 u = rb_entry(t, struct uprobe, rb_node);
1259 if (u->inode != inode || u->offset < min)
1261 list_add(&u->pending_list, head);
1264 for (t = n; (t = rb_next(t)); ) {
1265 u = rb_entry(t, struct uprobe, rb_node);
1266 if (u->inode != inode || u->offset > max)
1268 list_add(&u->pending_list, head);
1272 spin_unlock(&uprobes_treelock);
1275 /* @vma contains reference counter, not the probed instruction. */
1276 static int delayed_ref_ctr_inc(struct vm_area_struct *vma)
1278 struct list_head *pos, *q;
1279 struct delayed_uprobe *du;
1280 unsigned long vaddr;
1281 int ret = 0, err = 0;
1283 mutex_lock(&delayed_uprobe_lock);
1284 list_for_each_safe(pos, q, &delayed_uprobe_list) {
1285 du = list_entry(pos, struct delayed_uprobe, list);
1287 if (du->mm != vma->vm_mm ||
1288 !valid_ref_ctr_vma(du->uprobe, vma))
1291 vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
1292 ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
1294 update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1);
1298 delayed_uprobe_delete(du);
1300 mutex_unlock(&delayed_uprobe_lock);
1305 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1307 * Currently we ignore all errors and always return 0, the callers
1308 * can't handle the failure anyway.
1310 int uprobe_mmap(struct vm_area_struct *vma)
1312 struct list_head tmp_list;
1313 struct uprobe *uprobe, *u;
1314 struct inode *inode;
1316 if (no_uprobe_events())
1320 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
1321 test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
1322 delayed_ref_ctr_inc(vma);
1324 if (!valid_vma(vma, true))
1327 inode = file_inode(vma->vm_file);
1331 mutex_lock(uprobes_mmap_hash(inode));
1332 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1334 * We can race with uprobe_unregister(), this uprobe can be already
1335 * removed. But in this case filter_chain() must return false, all
1336 * consumers have gone away.
1338 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1339 if (!fatal_signal_pending(current) &&
1340 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1341 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1342 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1346 mutex_unlock(uprobes_mmap_hash(inode));
1352 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1355 struct inode *inode;
1358 inode = file_inode(vma->vm_file);
1360 min = vaddr_to_offset(vma, start);
1361 max = min + (end - start) - 1;
1363 spin_lock(&uprobes_treelock);
1364 n = find_node_in_range(inode, min, max);
1365 spin_unlock(&uprobes_treelock);
1371 * Called in context of a munmap of a vma.
1373 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1375 if (no_uprobe_events() || !valid_vma(vma, false))
1378 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1381 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1382 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1385 if (vma_has_uprobes(vma, start, end))
1386 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1389 /* Slot allocation for XOL */
1390 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1392 struct vm_area_struct *vma;
1395 if (down_write_killable(&mm->mmap_sem))
1398 if (mm->uprobes_state.xol_area) {
1404 /* Try to map as high as possible, this is only a hint. */
1405 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1407 if (area->vaddr & ~PAGE_MASK) {
1413 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1414 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1415 &area->xol_mapping);
1422 /* pairs with get_xol_area() */
1423 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1425 up_write(&mm->mmap_sem);
1430 static struct xol_area *__create_xol_area(unsigned long vaddr)
1432 struct mm_struct *mm = current->mm;
1433 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1434 struct xol_area *area;
1436 area = kmalloc(sizeof(*area), GFP_KERNEL);
1437 if (unlikely(!area))
1440 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1445 area->xol_mapping.name = "[uprobes]";
1446 area->xol_mapping.fault = NULL;
1447 area->xol_mapping.pages = area->pages;
1448 area->pages[0] = alloc_page(GFP_HIGHUSER);
1449 if (!area->pages[0])
1451 area->pages[1] = NULL;
1453 area->vaddr = vaddr;
1454 init_waitqueue_head(&area->wq);
1455 /* Reserve the 1st slot for get_trampoline_vaddr() */
1456 set_bit(0, area->bitmap);
1457 atomic_set(&area->slot_count, 1);
1458 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1460 if (!xol_add_vma(mm, area))
1463 __free_page(area->pages[0]);
1465 kfree(area->bitmap);
1473 * get_xol_area - Allocate process's xol_area if necessary.
1474 * This area will be used for storing instructions for execution out of line.
1476 * Returns the allocated area or NULL.
1478 static struct xol_area *get_xol_area(void)
1480 struct mm_struct *mm = current->mm;
1481 struct xol_area *area;
1483 if (!mm->uprobes_state.xol_area)
1484 __create_xol_area(0);
1486 /* Pairs with xol_add_vma() smp_store_release() */
1487 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1492 * uprobe_clear_state - Free the area allocated for slots.
1494 void uprobe_clear_state(struct mm_struct *mm)
1496 struct xol_area *area = mm->uprobes_state.xol_area;
1498 mutex_lock(&delayed_uprobe_lock);
1499 delayed_uprobe_remove(NULL, mm);
1500 mutex_unlock(&delayed_uprobe_lock);
1505 put_page(area->pages[0]);
1506 kfree(area->bitmap);
1510 void uprobe_start_dup_mmap(void)
1512 percpu_down_read(&dup_mmap_sem);
1515 void uprobe_end_dup_mmap(void)
1517 percpu_up_read(&dup_mmap_sem);
1520 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1522 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1523 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1524 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1525 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1530 * - search for a free slot.
1532 static unsigned long xol_take_insn_slot(struct xol_area *area)
1534 unsigned long slot_addr;
1538 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1539 if (slot_nr < UINSNS_PER_PAGE) {
1540 if (!test_and_set_bit(slot_nr, area->bitmap))
1543 slot_nr = UINSNS_PER_PAGE;
1546 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1547 } while (slot_nr >= UINSNS_PER_PAGE);
1549 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1550 atomic_inc(&area->slot_count);
1556 * xol_get_insn_slot - allocate a slot for xol.
1557 * Returns the allocated slot address or 0.
1559 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1561 struct xol_area *area;
1562 unsigned long xol_vaddr;
1564 area = get_xol_area();
1568 xol_vaddr = xol_take_insn_slot(area);
1569 if (unlikely(!xol_vaddr))
1572 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1573 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1579 * xol_free_insn_slot - If slot was earlier allocated by
1580 * @xol_get_insn_slot(), make the slot available for
1581 * subsequent requests.
1583 static void xol_free_insn_slot(struct task_struct *tsk)
1585 struct xol_area *area;
1586 unsigned long vma_end;
1587 unsigned long slot_addr;
1589 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1592 slot_addr = tsk->utask->xol_vaddr;
1593 if (unlikely(!slot_addr))
1596 area = tsk->mm->uprobes_state.xol_area;
1597 vma_end = area->vaddr + PAGE_SIZE;
1598 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1599 unsigned long offset;
1602 offset = slot_addr - area->vaddr;
1603 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1604 if (slot_nr >= UINSNS_PER_PAGE)
1607 clear_bit(slot_nr, area->bitmap);
1608 atomic_dec(&area->slot_count);
1609 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1610 if (waitqueue_active(&area->wq))
1613 tsk->utask->xol_vaddr = 0;
1617 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1618 void *src, unsigned long len)
1620 /* Initialize the slot */
1621 copy_to_page(page, vaddr, src, len);
1624 * We probably need flush_icache_user_range() but it needs vma.
1625 * This should work on most of architectures by default. If
1626 * architecture needs to do something different it can define
1627 * its own version of the function.
1629 flush_dcache_page(page);
1633 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1634 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1636 * Return the address of the breakpoint instruction.
1638 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1640 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1643 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1645 struct uprobe_task *utask = current->utask;
1647 if (unlikely(utask && utask->active_uprobe))
1648 return utask->vaddr;
1650 return instruction_pointer(regs);
1653 static struct return_instance *free_ret_instance(struct return_instance *ri)
1655 struct return_instance *next = ri->next;
1656 put_uprobe(ri->uprobe);
1662 * Called with no locks held.
1663 * Called in context of an exiting or an exec-ing thread.
1665 void uprobe_free_utask(struct task_struct *t)
1667 struct uprobe_task *utask = t->utask;
1668 struct return_instance *ri;
1673 if (utask->active_uprobe)
1674 put_uprobe(utask->active_uprobe);
1676 ri = utask->return_instances;
1678 ri = free_ret_instance(ri);
1680 xol_free_insn_slot(t);
1686 * Allocate a uprobe_task object for the task if if necessary.
1687 * Called when the thread hits a breakpoint.
1690 * - pointer to new uprobe_task on success
1693 static struct uprobe_task *get_utask(void)
1695 if (!current->utask)
1696 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1697 return current->utask;
1700 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1702 struct uprobe_task *n_utask;
1703 struct return_instance **p, *o, *n;
1705 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1710 p = &n_utask->return_instances;
1711 for (o = o_utask->return_instances; o; o = o->next) {
1712 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1717 get_uprobe(n->uprobe);
1728 static void uprobe_warn(struct task_struct *t, const char *msg)
1730 pr_warn("uprobe: %s:%d failed to %s\n",
1731 current->comm, current->pid, msg);
1734 static void dup_xol_work(struct callback_head *work)
1736 if (current->flags & PF_EXITING)
1739 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1740 !fatal_signal_pending(current))
1741 uprobe_warn(current, "dup xol area");
1745 * Called in context of a new clone/fork from copy_process.
1747 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1749 struct uprobe_task *utask = current->utask;
1750 struct mm_struct *mm = current->mm;
1751 struct xol_area *area;
1755 if (!utask || !utask->return_instances)
1758 if (mm == t->mm && !(flags & CLONE_VFORK))
1761 if (dup_utask(t, utask))
1762 return uprobe_warn(t, "dup ret instances");
1764 /* The task can fork() after dup_xol_work() fails */
1765 area = mm->uprobes_state.xol_area;
1767 return uprobe_warn(t, "dup xol area");
1772 t->utask->dup_xol_addr = area->vaddr;
1773 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1774 task_work_add(t, &t->utask->dup_xol_work, true);
1778 * Current area->vaddr notion assume the trampoline address is always
1779 * equal area->vaddr.
1781 * Returns -1 in case the xol_area is not allocated.
1783 static unsigned long get_trampoline_vaddr(void)
1785 struct xol_area *area;
1786 unsigned long trampoline_vaddr = -1;
1788 /* Pairs with xol_add_vma() smp_store_release() */
1789 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1791 trampoline_vaddr = area->vaddr;
1793 return trampoline_vaddr;
1796 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1797 struct pt_regs *regs)
1799 struct return_instance *ri = utask->return_instances;
1800 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1802 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1803 ri = free_ret_instance(ri);
1806 utask->return_instances = ri;
1809 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1811 struct return_instance *ri;
1812 struct uprobe_task *utask;
1813 unsigned long orig_ret_vaddr, trampoline_vaddr;
1816 if (!get_xol_area())
1819 utask = get_utask();
1823 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1824 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1825 " nestedness limit pid/tgid=%d/%d\n",
1826 current->pid, current->tgid);
1830 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1834 trampoline_vaddr = get_trampoline_vaddr();
1835 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1836 if (orig_ret_vaddr == -1)
1839 /* drop the entries invalidated by longjmp() */
1840 chained = (orig_ret_vaddr == trampoline_vaddr);
1841 cleanup_return_instances(utask, chained, regs);
1844 * We don't want to keep trampoline address in stack, rather keep the
1845 * original return address of first caller thru all the consequent
1846 * instances. This also makes breakpoint unwrapping easier.
1849 if (!utask->return_instances) {
1851 * This situation is not possible. Likely we have an
1852 * attack from user-space.
1854 uprobe_warn(current, "handle tail call");
1857 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1860 ri->uprobe = get_uprobe(uprobe);
1861 ri->func = instruction_pointer(regs);
1862 ri->stack = user_stack_pointer(regs);
1863 ri->orig_ret_vaddr = orig_ret_vaddr;
1864 ri->chained = chained;
1867 ri->next = utask->return_instances;
1868 utask->return_instances = ri;
1875 /* Prepare to single-step probed instruction out of line. */
1877 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1879 struct uprobe_task *utask;
1880 unsigned long xol_vaddr;
1883 utask = get_utask();
1887 xol_vaddr = xol_get_insn_slot(uprobe);
1891 utask->xol_vaddr = xol_vaddr;
1892 utask->vaddr = bp_vaddr;
1894 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1895 if (unlikely(err)) {
1896 xol_free_insn_slot(current);
1900 utask->active_uprobe = uprobe;
1901 utask->state = UTASK_SSTEP;
1906 * If we are singlestepping, then ensure this thread is not connected to
1907 * non-fatal signals until completion of singlestep. When xol insn itself
1908 * triggers the signal, restart the original insn even if the task is
1909 * already SIGKILL'ed (since coredump should report the correct ip). This
1910 * is even more important if the task has a handler for SIGSEGV/etc, The
1911 * _same_ instruction should be repeated again after return from the signal
1912 * handler, and SSTEP can never finish in this case.
1914 bool uprobe_deny_signal(void)
1916 struct task_struct *t = current;
1917 struct uprobe_task *utask = t->utask;
1919 if (likely(!utask || !utask->active_uprobe))
1922 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1924 if (signal_pending(t)) {
1925 spin_lock_irq(&t->sighand->siglock);
1926 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1927 spin_unlock_irq(&t->sighand->siglock);
1929 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1930 utask->state = UTASK_SSTEP_TRAPPED;
1931 set_tsk_thread_flag(t, TIF_UPROBE);
1938 static void mmf_recalc_uprobes(struct mm_struct *mm)
1940 struct vm_area_struct *vma;
1942 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1943 if (!valid_vma(vma, false))
1946 * This is not strictly accurate, we can race with
1947 * uprobe_unregister() and see the already removed
1948 * uprobe if delete_uprobe() was not yet called.
1949 * Or this uprobe can be filtered out.
1951 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1955 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1958 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1961 uprobe_opcode_t opcode;
1964 pagefault_disable();
1965 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
1968 if (likely(result == 0))
1972 * The NULL 'tsk' here ensures that any faults that occur here
1973 * will not be accounted to the task. 'mm' *is* current->mm,
1974 * but we treat this as a 'remote' access since it is
1975 * essentially a kernel access to the memory.
1977 result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
1982 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1985 /* This needs to return true for any variant of the trap insn */
1986 return is_trap_insn(&opcode);
1989 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1991 struct mm_struct *mm = current->mm;
1992 struct uprobe *uprobe = NULL;
1993 struct vm_area_struct *vma;
1995 down_read(&mm->mmap_sem);
1996 vma = find_vma(mm, bp_vaddr);
1997 if (vma && vma->vm_start <= bp_vaddr) {
1998 if (valid_vma(vma, false)) {
1999 struct inode *inode = file_inode(vma->vm_file);
2000 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
2002 uprobe = find_uprobe(inode, offset);
2006 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
2011 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
2012 mmf_recalc_uprobes(mm);
2013 up_read(&mm->mmap_sem);
2018 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
2020 struct uprobe_consumer *uc;
2021 int remove = UPROBE_HANDLER_REMOVE;
2022 bool need_prep = false; /* prepare return uprobe, when needed */
2024 down_read(&uprobe->register_rwsem);
2025 for (uc = uprobe->consumers; uc; uc = uc->next) {
2029 rc = uc->handler(uc, regs);
2030 WARN(rc & ~UPROBE_HANDLER_MASK,
2031 "bad rc=0x%x from %pf()\n", rc, uc->handler);
2034 if (uc->ret_handler)
2040 if (need_prep && !remove)
2041 prepare_uretprobe(uprobe, regs); /* put bp at return */
2043 if (remove && uprobe->consumers) {
2044 WARN_ON(!uprobe_is_active(uprobe));
2045 unapply_uprobe(uprobe, current->mm);
2047 up_read(&uprobe->register_rwsem);
2051 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
2053 struct uprobe *uprobe = ri->uprobe;
2054 struct uprobe_consumer *uc;
2056 down_read(&uprobe->register_rwsem);
2057 for (uc = uprobe->consumers; uc; uc = uc->next) {
2058 if (uc->ret_handler)
2059 uc->ret_handler(uc, ri->func, regs);
2061 up_read(&uprobe->register_rwsem);
2064 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
2069 chained = ri->chained;
2070 ri = ri->next; /* can't be NULL if chained */
2076 static void handle_trampoline(struct pt_regs *regs)
2078 struct uprobe_task *utask;
2079 struct return_instance *ri, *next;
2082 utask = current->utask;
2086 ri = utask->return_instances;
2092 * We should throw out the frames invalidated by longjmp().
2093 * If this chain is valid, then the next one should be alive
2094 * or NULL; the latter case means that nobody but ri->func
2095 * could hit this trampoline on return. TODO: sigaltstack().
2097 next = find_next_ret_chain(ri);
2098 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
2100 instruction_pointer_set(regs, ri->orig_ret_vaddr);
2103 handle_uretprobe_chain(ri, regs);
2104 ri = free_ret_instance(ri);
2106 } while (ri != next);
2109 utask->return_instances = ri;
2113 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
2114 force_sig(SIGILL, current);
2118 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
2123 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
2124 struct pt_regs *regs)
2130 * Run handler and ask thread to singlestep.
2131 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2133 static void handle_swbp(struct pt_regs *regs)
2135 struct uprobe *uprobe;
2136 unsigned long bp_vaddr;
2137 int uninitialized_var(is_swbp);
2139 bp_vaddr = uprobe_get_swbp_addr(regs);
2140 if (bp_vaddr == get_trampoline_vaddr())
2141 return handle_trampoline(regs);
2143 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
2146 /* No matching uprobe; signal SIGTRAP. */
2147 send_sig(SIGTRAP, current, 0);
2150 * Either we raced with uprobe_unregister() or we can't
2151 * access this memory. The latter is only possible if
2152 * another thread plays with our ->mm. In both cases
2153 * we can simply restart. If this vma was unmapped we
2154 * can pretend this insn was not executed yet and get
2155 * the (correct) SIGSEGV after restart.
2157 instruction_pointer_set(regs, bp_vaddr);
2162 /* change it in advance for ->handler() and restart */
2163 instruction_pointer_set(regs, bp_vaddr);
2166 * TODO: move copy_insn/etc into _register and remove this hack.
2167 * After we hit the bp, _unregister + _register can install the
2168 * new and not-yet-analyzed uprobe at the same address, restart.
2170 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
2174 * Pairs with the smp_wmb() in prepare_uprobe().
2176 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2177 * we must also see the stores to &uprobe->arch performed by the
2178 * prepare_uprobe() call.
2182 /* Tracing handlers use ->utask to communicate with fetch methods */
2186 if (arch_uprobe_ignore(&uprobe->arch, regs))
2189 handler_chain(uprobe, regs);
2191 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
2194 if (!pre_ssout(uprobe, regs, bp_vaddr))
2197 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2203 * Perform required fix-ups and disable singlestep.
2204 * Allow pending signals to take effect.
2206 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
2208 struct uprobe *uprobe;
2211 uprobe = utask->active_uprobe;
2212 if (utask->state == UTASK_SSTEP_ACK)
2213 err = arch_uprobe_post_xol(&uprobe->arch, regs);
2214 else if (utask->state == UTASK_SSTEP_TRAPPED)
2215 arch_uprobe_abort_xol(&uprobe->arch, regs);
2220 utask->active_uprobe = NULL;
2221 utask->state = UTASK_RUNNING;
2222 xol_free_insn_slot(current);
2224 spin_lock_irq(¤t->sighand->siglock);
2225 recalc_sigpending(); /* see uprobe_deny_signal() */
2226 spin_unlock_irq(¤t->sighand->siglock);
2228 if (unlikely(err)) {
2229 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
2230 force_sig(SIGILL, current);
2235 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2236 * allows the thread to return from interrupt. After that handle_swbp()
2237 * sets utask->active_uprobe.
2239 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2240 * and allows the thread to return from interrupt.
2242 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2243 * uprobe_notify_resume().
2245 void uprobe_notify_resume(struct pt_regs *regs)
2247 struct uprobe_task *utask;
2249 clear_thread_flag(TIF_UPROBE);
2251 utask = current->utask;
2252 if (utask && utask->active_uprobe)
2253 handle_singlestep(utask, regs);
2259 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2260 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2262 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2267 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2268 (!current->utask || !current->utask->return_instances))
2271 set_thread_flag(TIF_UPROBE);
2276 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2277 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2279 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2281 struct uprobe_task *utask = current->utask;
2283 if (!current->mm || !utask || !utask->active_uprobe)
2284 /* task is currently not uprobed */
2287 utask->state = UTASK_SSTEP_ACK;
2288 set_thread_flag(TIF_UPROBE);
2292 static struct notifier_block uprobe_exception_nb = {
2293 .notifier_call = arch_uprobe_exception_notify,
2294 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2297 static int __init init_uprobes(void)
2301 for (i = 0; i < UPROBES_HASH_SZ; i++)
2302 mutex_init(&uprobes_mmap_mutex[i]);
2304 if (percpu_init_rwsem(&dup_mmap_sem))
2307 return register_die_notifier(&uprobe_exception_nb);
2309 __initcall(init_uprobes);