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;
165 struct mem_cgroup *memcg;
167 mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, addr,
171 err = mem_cgroup_try_charge(new_page, vma->vm_mm, GFP_KERNEL,
177 /* For try_to_free_swap() and munlock_vma_page() below */
180 mmu_notifier_invalidate_range_start(&range);
182 if (!page_vma_mapped_walk(&pvmw)) {
184 mem_cgroup_cancel_charge(new_page, memcg, false);
187 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
191 page_add_new_anon_rmap(new_page, vma, addr, false);
192 mem_cgroup_commit_charge(new_page, memcg, false, false);
193 lru_cache_add_active_or_unevictable(new_page, vma);
195 /* no new page, just dec_mm_counter for old_page */
196 dec_mm_counter(mm, MM_ANONPAGES);
198 if (!PageAnon(old_page)) {
199 dec_mm_counter(mm, mm_counter_file(old_page));
200 inc_mm_counter(mm, MM_ANONPAGES);
203 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
204 ptep_clear_flush_notify(vma, addr, pvmw.pte);
206 set_pte_at_notify(mm, addr, pvmw.pte,
207 mk_pte(new_page, vma->vm_page_prot));
209 page_remove_rmap(old_page, false);
210 if (!page_mapped(old_page))
211 try_to_free_swap(old_page);
212 page_vma_mapped_walk_done(&pvmw);
214 if (vma->vm_flags & VM_LOCKED)
215 munlock_vma_page(old_page);
220 mmu_notifier_invalidate_range_end(&range);
221 unlock_page(old_page);
226 * is_swbp_insn - check if instruction is breakpoint instruction.
227 * @insn: instruction to be checked.
228 * Default implementation of is_swbp_insn
229 * Returns true if @insn is a breakpoint instruction.
231 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
233 return *insn == UPROBE_SWBP_INSN;
237 * is_trap_insn - check if instruction is breakpoint instruction.
238 * @insn: instruction to be checked.
239 * Default implementation of is_trap_insn
240 * Returns true if @insn is a breakpoint instruction.
242 * This function is needed for the case where an architecture has multiple
243 * trap instructions (like powerpc).
245 bool __weak is_trap_insn(uprobe_opcode_t *insn)
247 return is_swbp_insn(insn);
250 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
252 void *kaddr = kmap_atomic(page);
253 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
254 kunmap_atomic(kaddr);
257 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
259 void *kaddr = kmap_atomic(page);
260 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
261 kunmap_atomic(kaddr);
264 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
266 uprobe_opcode_t old_opcode;
270 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
271 * We do not check if it is any other 'trap variant' which could
272 * be conditional trap instruction such as the one powerpc supports.
274 * The logic is that we do not care if the underlying instruction
275 * is a trap variant; uprobes always wins over any other (gdb)
278 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
279 is_swbp = is_swbp_insn(&old_opcode);
281 if (is_swbp_insn(new_opcode)) {
282 if (is_swbp) /* register: already installed? */
285 if (!is_swbp) /* unregister: was it changed by us? */
292 static struct delayed_uprobe *
293 delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm)
295 struct delayed_uprobe *du;
297 list_for_each_entry(du, &delayed_uprobe_list, list)
298 if (du->uprobe == uprobe && du->mm == mm)
303 static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
305 struct delayed_uprobe *du;
307 if (delayed_uprobe_check(uprobe, mm))
310 du = kzalloc(sizeof(*du), GFP_KERNEL);
316 list_add(&du->list, &delayed_uprobe_list);
320 static void delayed_uprobe_delete(struct delayed_uprobe *du)
328 static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm)
330 struct list_head *pos, *q;
331 struct delayed_uprobe *du;
336 list_for_each_safe(pos, q, &delayed_uprobe_list) {
337 du = list_entry(pos, struct delayed_uprobe, list);
339 if (uprobe && du->uprobe != uprobe)
341 if (mm && du->mm != mm)
344 delayed_uprobe_delete(du);
348 static bool valid_ref_ctr_vma(struct uprobe *uprobe,
349 struct vm_area_struct *vma)
351 unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
353 return uprobe->ref_ctr_offset &&
355 file_inode(vma->vm_file) == uprobe->inode &&
356 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
357 vma->vm_start <= vaddr &&
361 static struct vm_area_struct *
362 find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm)
364 struct vm_area_struct *tmp;
366 for (tmp = mm->mmap; tmp; tmp = tmp->vm_next)
367 if (valid_ref_ctr_vma(uprobe, tmp))
374 __update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d)
378 struct vm_area_struct *vma;
385 ret = get_user_pages_remote(NULL, mm, vaddr, 1,
386 FOLL_WRITE, &page, &vma, NULL);
387 if (unlikely(ret <= 0)) {
389 * We are asking for 1 page. If get_user_pages_remote() fails,
390 * it may return 0, in that case we have to return error.
392 return ret == 0 ? -EBUSY : ret;
395 kaddr = kmap_atomic(page);
396 ptr = kaddr + (vaddr & ~PAGE_MASK);
398 if (unlikely(*ptr + d < 0)) {
399 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
400 "curr val: %d, delta: %d\n", vaddr, *ptr, d);
408 kunmap_atomic(kaddr);
413 static void update_ref_ctr_warn(struct uprobe *uprobe,
414 struct mm_struct *mm, short d)
416 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
417 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
418 d > 0 ? "increment" : "decrement", uprobe->inode->i_ino,
419 (unsigned long long) uprobe->offset,
420 (unsigned long long) uprobe->ref_ctr_offset, mm);
423 static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
426 struct vm_area_struct *rc_vma;
427 unsigned long rc_vaddr;
430 rc_vma = find_ref_ctr_vma(uprobe, mm);
433 rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset);
434 ret = __update_ref_ctr(mm, rc_vaddr, d);
436 update_ref_ctr_warn(uprobe, mm, d);
442 mutex_lock(&delayed_uprobe_lock);
444 ret = delayed_uprobe_add(uprobe, mm);
446 delayed_uprobe_remove(uprobe, mm);
447 mutex_unlock(&delayed_uprobe_lock);
454 * Expect the breakpoint instruction to be the smallest size instruction for
455 * the architecture. If an arch has variable length instruction and the
456 * breakpoint instruction is not of the smallest length instruction
457 * supported by that architecture then we need to modify is_trap_at_addr and
458 * uprobe_write_opcode accordingly. This would never be a problem for archs
459 * that have fixed length instructions.
461 * uprobe_write_opcode - write the opcode at a given virtual address.
462 * @mm: the probed process address space.
463 * @vaddr: the virtual address to store the opcode.
464 * @opcode: opcode to be written at @vaddr.
466 * Called with mm->mmap_sem held for write.
467 * Return 0 (success) or a negative errno.
469 int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
470 unsigned long vaddr, uprobe_opcode_t opcode)
472 struct uprobe *uprobe;
473 struct page *old_page, *new_page;
474 struct vm_area_struct *vma;
475 int ret, is_register, ref_ctr_updated = 0;
476 bool orig_page_huge = false;
477 unsigned int gup_flags = FOLL_FORCE;
479 is_register = is_swbp_insn(&opcode);
480 uprobe = container_of(auprobe, struct uprobe, arch);
484 gup_flags |= FOLL_SPLIT_PMD;
485 /* Read the page with vaddr into memory */
486 ret = get_user_pages_remote(NULL, mm, vaddr, 1, gup_flags,
487 &old_page, &vma, NULL);
491 ret = verify_opcode(old_page, vaddr, &opcode);
495 if (WARN(!is_register && PageCompound(old_page),
496 "uprobe unregister should never work on compound page\n")) {
501 /* We are going to replace instruction, update ref_ctr. */
502 if (!ref_ctr_updated && uprobe->ref_ctr_offset) {
503 ret = update_ref_ctr(uprobe, mm, is_register ? 1 : -1);
511 if (!is_register && !PageAnon(old_page))
514 ret = anon_vma_prepare(vma);
519 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
523 __SetPageUptodate(new_page);
524 copy_highpage(new_page, old_page);
525 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
528 struct page *orig_page;
531 VM_BUG_ON_PAGE(!PageAnon(old_page), old_page);
533 index = vaddr_to_offset(vma, vaddr & PAGE_MASK) >> PAGE_SHIFT;
534 orig_page = find_get_page(vma->vm_file->f_inode->i_mapping,
538 if (PageUptodate(orig_page) &&
539 pages_identical(new_page, orig_page)) {
540 /* let go new_page */
544 if (PageCompound(orig_page))
545 orig_page_huge = true;
551 ret = __replace_page(vma, vaddr, old_page, new_page);
557 if (unlikely(ret == -EAGAIN))
560 /* Revert back reference counter if instruction update failed. */
561 if (ret && is_register && ref_ctr_updated)
562 update_ref_ctr(uprobe, mm, -1);
564 /* try collapse pmd for compound page */
565 if (!ret && orig_page_huge)
566 collapse_pte_mapped_thp(mm, vaddr);
572 * set_swbp - store breakpoint at a given address.
573 * @auprobe: arch specific probepoint information.
574 * @mm: the probed process address space.
575 * @vaddr: the virtual address to insert the opcode.
577 * For mm @mm, store the breakpoint instruction at @vaddr.
578 * Return 0 (success) or a negative errno.
580 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
582 return uprobe_write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
586 * set_orig_insn - Restore the original instruction.
587 * @mm: the probed process address space.
588 * @auprobe: arch specific probepoint information.
589 * @vaddr: the virtual address to insert the opcode.
591 * For mm @mm, restore the original opcode (opcode) at @vaddr.
592 * Return 0 (success) or a negative errno.
595 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
597 return uprobe_write_opcode(auprobe, mm, vaddr,
598 *(uprobe_opcode_t *)&auprobe->insn);
601 static struct uprobe *get_uprobe(struct uprobe *uprobe)
603 refcount_inc(&uprobe->ref);
607 static void put_uprobe(struct uprobe *uprobe)
609 if (refcount_dec_and_test(&uprobe->ref)) {
611 * If application munmap(exec_vma) before uprobe_unregister()
612 * gets called, we don't get a chance to remove uprobe from
613 * delayed_uprobe_list from remove_breakpoint(). Do it here.
615 mutex_lock(&delayed_uprobe_lock);
616 delayed_uprobe_remove(uprobe, NULL);
617 mutex_unlock(&delayed_uprobe_lock);
622 static int match_uprobe(struct uprobe *l, struct uprobe *r)
624 if (l->inode < r->inode)
627 if (l->inode > r->inode)
630 if (l->offset < r->offset)
633 if (l->offset > r->offset)
639 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
641 struct uprobe u = { .inode = inode, .offset = offset };
642 struct rb_node *n = uprobes_tree.rb_node;
643 struct uprobe *uprobe;
647 uprobe = rb_entry(n, struct uprobe, rb_node);
648 match = match_uprobe(&u, uprobe);
650 return get_uprobe(uprobe);
661 * Find a uprobe corresponding to a given inode:offset
662 * Acquires uprobes_treelock
664 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
666 struct uprobe *uprobe;
668 spin_lock(&uprobes_treelock);
669 uprobe = __find_uprobe(inode, offset);
670 spin_unlock(&uprobes_treelock);
675 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
677 struct rb_node **p = &uprobes_tree.rb_node;
678 struct rb_node *parent = NULL;
684 u = rb_entry(parent, struct uprobe, rb_node);
685 match = match_uprobe(uprobe, u);
687 return get_uprobe(u);
690 p = &parent->rb_left;
692 p = &parent->rb_right;
697 rb_link_node(&uprobe->rb_node, parent, p);
698 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
699 /* get access + creation ref */
700 refcount_set(&uprobe->ref, 2);
706 * Acquire uprobes_treelock.
707 * Matching uprobe already exists in rbtree;
708 * increment (access refcount) and return the matching uprobe.
710 * No matching uprobe; insert the uprobe in rb_tree;
711 * get a double refcount (access + creation) and return NULL.
713 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
717 spin_lock(&uprobes_treelock);
718 u = __insert_uprobe(uprobe);
719 spin_unlock(&uprobes_treelock);
725 ref_ctr_mismatch_warn(struct uprobe *cur_uprobe, struct uprobe *uprobe)
727 pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx "
728 "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n",
729 uprobe->inode->i_ino, (unsigned long long) uprobe->offset,
730 (unsigned long long) cur_uprobe->ref_ctr_offset,
731 (unsigned long long) uprobe->ref_ctr_offset);
734 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
735 loff_t ref_ctr_offset)
737 struct uprobe *uprobe, *cur_uprobe;
739 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
743 uprobe->inode = inode;
744 uprobe->offset = offset;
745 uprobe->ref_ctr_offset = ref_ctr_offset;
746 init_rwsem(&uprobe->register_rwsem);
747 init_rwsem(&uprobe->consumer_rwsem);
749 /* add to uprobes_tree, sorted on inode:offset */
750 cur_uprobe = insert_uprobe(uprobe);
751 /* a uprobe exists for this inode:offset combination */
753 if (cur_uprobe->ref_ctr_offset != uprobe->ref_ctr_offset) {
754 ref_ctr_mismatch_warn(cur_uprobe, uprobe);
755 put_uprobe(cur_uprobe);
757 return ERR_PTR(-EINVAL);
766 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
768 down_write(&uprobe->consumer_rwsem);
769 uc->next = uprobe->consumers;
770 uprobe->consumers = uc;
771 up_write(&uprobe->consumer_rwsem);
775 * For uprobe @uprobe, delete the consumer @uc.
776 * Return true if the @uc is deleted successfully
779 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
781 struct uprobe_consumer **con;
784 down_write(&uprobe->consumer_rwsem);
785 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
792 up_write(&uprobe->consumer_rwsem);
797 static int __copy_insn(struct address_space *mapping, struct file *filp,
798 void *insn, int nbytes, loff_t offset)
802 * Ensure that the page that has the original instruction is populated
803 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
804 * see uprobe_register().
806 if (mapping->a_ops->readpage)
807 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
809 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
811 return PTR_ERR(page);
813 copy_from_page(page, offset, insn, nbytes);
819 static int copy_insn(struct uprobe *uprobe, struct file *filp)
821 struct address_space *mapping = uprobe->inode->i_mapping;
822 loff_t offs = uprobe->offset;
823 void *insn = &uprobe->arch.insn;
824 int size = sizeof(uprobe->arch.insn);
827 /* Copy only available bytes, -EIO if nothing was read */
829 if (offs >= i_size_read(uprobe->inode))
832 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
833 err = __copy_insn(mapping, filp, insn, len, offs);
845 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
846 struct mm_struct *mm, unsigned long vaddr)
850 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
853 /* TODO: move this into _register, until then we abuse this sem. */
854 down_write(&uprobe->consumer_rwsem);
855 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
858 ret = copy_insn(uprobe, file);
863 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
866 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
870 /* uprobe_write_opcode() assumes we don't cross page boundary */
871 BUG_ON((uprobe->offset & ~PAGE_MASK) +
872 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
874 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
875 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
878 up_write(&uprobe->consumer_rwsem);
883 static inline bool consumer_filter(struct uprobe_consumer *uc,
884 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
886 return !uc->filter || uc->filter(uc, ctx, mm);
889 static bool filter_chain(struct uprobe *uprobe,
890 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
892 struct uprobe_consumer *uc;
895 down_read(&uprobe->consumer_rwsem);
896 for (uc = uprobe->consumers; uc; uc = uc->next) {
897 ret = consumer_filter(uc, ctx, mm);
901 up_read(&uprobe->consumer_rwsem);
907 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
908 struct vm_area_struct *vma, unsigned long vaddr)
913 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
918 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
919 * the task can hit this breakpoint right after __replace_page().
921 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
923 set_bit(MMF_HAS_UPROBES, &mm->flags);
925 ret = set_swbp(&uprobe->arch, mm, vaddr);
927 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
928 else if (first_uprobe)
929 clear_bit(MMF_HAS_UPROBES, &mm->flags);
935 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
937 set_bit(MMF_RECALC_UPROBES, &mm->flags);
938 return set_orig_insn(&uprobe->arch, mm, vaddr);
941 static inline bool uprobe_is_active(struct uprobe *uprobe)
943 return !RB_EMPTY_NODE(&uprobe->rb_node);
946 * There could be threads that have already hit the breakpoint. They
947 * will recheck the current insn and restart if find_uprobe() fails.
948 * See find_active_uprobe().
950 static void delete_uprobe(struct uprobe *uprobe)
952 if (WARN_ON(!uprobe_is_active(uprobe)))
955 spin_lock(&uprobes_treelock);
956 rb_erase(&uprobe->rb_node, &uprobes_tree);
957 spin_unlock(&uprobes_treelock);
958 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
963 struct map_info *next;
964 struct mm_struct *mm;
968 static inline struct map_info *free_map_info(struct map_info *info)
970 struct map_info *next = info->next;
975 static struct map_info *
976 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
978 unsigned long pgoff = offset >> PAGE_SHIFT;
979 struct vm_area_struct *vma;
980 struct map_info *curr = NULL;
981 struct map_info *prev = NULL;
982 struct map_info *info;
986 i_mmap_lock_read(mapping);
987 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
988 if (!valid_vma(vma, is_register))
991 if (!prev && !more) {
993 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
994 * reclaim. This is optimistic, no harm done if it fails.
996 prev = kmalloc(sizeof(struct map_info),
997 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
1006 if (!mmget_not_zero(vma->vm_mm))
1014 info->mm = vma->vm_mm;
1015 info->vaddr = offset_to_vaddr(vma, offset);
1017 i_mmap_unlock_read(mapping);
1029 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
1031 curr = ERR_PTR(-ENOMEM);
1041 prev = free_map_info(prev);
1046 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
1048 bool is_register = !!new;
1049 struct map_info *info;
1052 percpu_down_write(&dup_mmap_sem);
1053 info = build_map_info(uprobe->inode->i_mapping,
1054 uprobe->offset, is_register);
1056 err = PTR_ERR(info);
1061 struct mm_struct *mm = info->mm;
1062 struct vm_area_struct *vma;
1064 if (err && is_register)
1067 down_write(&mm->mmap_sem);
1068 vma = find_vma(mm, info->vaddr);
1069 if (!vma || !valid_vma(vma, is_register) ||
1070 file_inode(vma->vm_file) != uprobe->inode)
1073 if (vma->vm_start > info->vaddr ||
1074 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
1078 /* consult only the "caller", new consumer. */
1079 if (consumer_filter(new,
1080 UPROBE_FILTER_REGISTER, mm))
1081 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
1082 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
1083 if (!filter_chain(uprobe,
1084 UPROBE_FILTER_UNREGISTER, mm))
1085 err |= remove_breakpoint(uprobe, mm, info->vaddr);
1089 up_write(&mm->mmap_sem);
1092 info = free_map_info(info);
1095 percpu_up_write(&dup_mmap_sem);
1100 __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
1104 if (WARN_ON(!consumer_del(uprobe, uc)))
1107 err = register_for_each_vma(uprobe, NULL);
1108 /* TODO : cant unregister? schedule a worker thread */
1109 if (!uprobe->consumers && !err)
1110 delete_uprobe(uprobe);
1114 * uprobe_unregister - unregister an already registered probe.
1115 * @inode: the file in which the probe has to be removed.
1116 * @offset: offset from the start of the file.
1117 * @uc: identify which probe if multiple probes are colocated.
1119 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
1121 struct uprobe *uprobe;
1123 uprobe = find_uprobe(inode, offset);
1124 if (WARN_ON(!uprobe))
1127 down_write(&uprobe->register_rwsem);
1128 __uprobe_unregister(uprobe, uc);
1129 up_write(&uprobe->register_rwsem);
1132 EXPORT_SYMBOL_GPL(uprobe_unregister);
1135 * __uprobe_register - register a probe
1136 * @inode: the file in which the probe has to be placed.
1137 * @offset: offset from the start of the file.
1138 * @uc: information on howto handle the probe..
1140 * Apart from the access refcount, __uprobe_register() takes a creation
1141 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
1142 * inserted into the rbtree (i.e first consumer for a @inode:@offset
1143 * tuple). Creation refcount stops uprobe_unregister from freeing the
1144 * @uprobe even before the register operation is complete. Creation
1145 * refcount is released when the last @uc for the @uprobe
1146 * unregisters. Caller of __uprobe_register() is required to keep @inode
1147 * (and the containing mount) referenced.
1149 * Return errno if it cannot successully install probes
1150 * else return 0 (success)
1152 static int __uprobe_register(struct inode *inode, loff_t offset,
1153 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1155 struct uprobe *uprobe;
1158 /* Uprobe must have at least one set consumer */
1159 if (!uc->handler && !uc->ret_handler)
1162 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
1163 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
1165 /* Racy, just to catch the obvious mistakes */
1166 if (offset > i_size_read(inode))
1170 uprobe = alloc_uprobe(inode, offset, ref_ctr_offset);
1174 return PTR_ERR(uprobe);
1177 * We can race with uprobe_unregister()->delete_uprobe().
1178 * Check uprobe_is_active() and retry if it is false.
1180 down_write(&uprobe->register_rwsem);
1182 if (likely(uprobe_is_active(uprobe))) {
1183 consumer_add(uprobe, uc);
1184 ret = register_for_each_vma(uprobe, uc);
1186 __uprobe_unregister(uprobe, uc);
1188 up_write(&uprobe->register_rwsem);
1191 if (unlikely(ret == -EAGAIN))
1196 int uprobe_register(struct inode *inode, loff_t offset,
1197 struct uprobe_consumer *uc)
1199 return __uprobe_register(inode, offset, 0, uc);
1201 EXPORT_SYMBOL_GPL(uprobe_register);
1203 int uprobe_register_refctr(struct inode *inode, loff_t offset,
1204 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1206 return __uprobe_register(inode, offset, ref_ctr_offset, uc);
1208 EXPORT_SYMBOL_GPL(uprobe_register_refctr);
1211 * uprobe_apply - unregister an already registered probe.
1212 * @inode: the file in which the probe has to be removed.
1213 * @offset: offset from the start of the file.
1214 * @uc: consumer which wants to add more or remove some breakpoints
1215 * @add: add or remove the breakpoints
1217 int uprobe_apply(struct inode *inode, loff_t offset,
1218 struct uprobe_consumer *uc, bool add)
1220 struct uprobe *uprobe;
1221 struct uprobe_consumer *con;
1224 uprobe = find_uprobe(inode, offset);
1225 if (WARN_ON(!uprobe))
1228 down_write(&uprobe->register_rwsem);
1229 for (con = uprobe->consumers; con && con != uc ; con = con->next)
1232 ret = register_for_each_vma(uprobe, add ? uc : NULL);
1233 up_write(&uprobe->register_rwsem);
1239 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
1241 struct vm_area_struct *vma;
1244 down_read(&mm->mmap_sem);
1245 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1246 unsigned long vaddr;
1249 if (!valid_vma(vma, false) ||
1250 file_inode(vma->vm_file) != uprobe->inode)
1253 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
1254 if (uprobe->offset < offset ||
1255 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
1258 vaddr = offset_to_vaddr(vma, uprobe->offset);
1259 err |= remove_breakpoint(uprobe, mm, vaddr);
1261 up_read(&mm->mmap_sem);
1266 static struct rb_node *
1267 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1269 struct rb_node *n = uprobes_tree.rb_node;
1272 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1274 if (inode < u->inode) {
1276 } else if (inode > u->inode) {
1279 if (max < u->offset)
1281 else if (min > u->offset)
1292 * For a given range in vma, build a list of probes that need to be inserted.
1294 static void build_probe_list(struct inode *inode,
1295 struct vm_area_struct *vma,
1296 unsigned long start, unsigned long end,
1297 struct list_head *head)
1300 struct rb_node *n, *t;
1303 INIT_LIST_HEAD(head);
1304 min = vaddr_to_offset(vma, start);
1305 max = min + (end - start) - 1;
1307 spin_lock(&uprobes_treelock);
1308 n = find_node_in_range(inode, min, max);
1310 for (t = n; t; t = rb_prev(t)) {
1311 u = rb_entry(t, struct uprobe, rb_node);
1312 if (u->inode != inode || u->offset < min)
1314 list_add(&u->pending_list, head);
1317 for (t = n; (t = rb_next(t)); ) {
1318 u = rb_entry(t, struct uprobe, rb_node);
1319 if (u->inode != inode || u->offset > max)
1321 list_add(&u->pending_list, head);
1325 spin_unlock(&uprobes_treelock);
1328 /* @vma contains reference counter, not the probed instruction. */
1329 static int delayed_ref_ctr_inc(struct vm_area_struct *vma)
1331 struct list_head *pos, *q;
1332 struct delayed_uprobe *du;
1333 unsigned long vaddr;
1334 int ret = 0, err = 0;
1336 mutex_lock(&delayed_uprobe_lock);
1337 list_for_each_safe(pos, q, &delayed_uprobe_list) {
1338 du = list_entry(pos, struct delayed_uprobe, list);
1340 if (du->mm != vma->vm_mm ||
1341 !valid_ref_ctr_vma(du->uprobe, vma))
1344 vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
1345 ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
1347 update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1);
1351 delayed_uprobe_delete(du);
1353 mutex_unlock(&delayed_uprobe_lock);
1358 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1360 * Currently we ignore all errors and always return 0, the callers
1361 * can't handle the failure anyway.
1363 int uprobe_mmap(struct vm_area_struct *vma)
1365 struct list_head tmp_list;
1366 struct uprobe *uprobe, *u;
1367 struct inode *inode;
1369 if (no_uprobe_events())
1373 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
1374 test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
1375 delayed_ref_ctr_inc(vma);
1377 if (!valid_vma(vma, true))
1380 inode = file_inode(vma->vm_file);
1384 mutex_lock(uprobes_mmap_hash(inode));
1385 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1387 * We can race with uprobe_unregister(), this uprobe can be already
1388 * removed. But in this case filter_chain() must return false, all
1389 * consumers have gone away.
1391 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1392 if (!fatal_signal_pending(current) &&
1393 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1394 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1395 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1399 mutex_unlock(uprobes_mmap_hash(inode));
1405 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1408 struct inode *inode;
1411 inode = file_inode(vma->vm_file);
1413 min = vaddr_to_offset(vma, start);
1414 max = min + (end - start) - 1;
1416 spin_lock(&uprobes_treelock);
1417 n = find_node_in_range(inode, min, max);
1418 spin_unlock(&uprobes_treelock);
1424 * Called in context of a munmap of a vma.
1426 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1428 if (no_uprobe_events() || !valid_vma(vma, false))
1431 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1434 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1435 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1438 if (vma_has_uprobes(vma, start, end))
1439 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1442 /* Slot allocation for XOL */
1443 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1445 struct vm_area_struct *vma;
1448 if (down_write_killable(&mm->mmap_sem))
1451 if (mm->uprobes_state.xol_area) {
1457 /* Try to map as high as possible, this is only a hint. */
1458 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1460 if (IS_ERR_VALUE(area->vaddr)) {
1466 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1467 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1468 &area->xol_mapping);
1475 /* pairs with get_xol_area() */
1476 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1478 up_write(&mm->mmap_sem);
1483 static struct xol_area *__create_xol_area(unsigned long vaddr)
1485 struct mm_struct *mm = current->mm;
1486 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1487 struct xol_area *area;
1489 area = kmalloc(sizeof(*area), GFP_KERNEL);
1490 if (unlikely(!area))
1493 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1498 area->xol_mapping.name = "[uprobes]";
1499 area->xol_mapping.fault = NULL;
1500 area->xol_mapping.pages = area->pages;
1501 area->pages[0] = alloc_page(GFP_HIGHUSER);
1502 if (!area->pages[0])
1504 area->pages[1] = NULL;
1506 area->vaddr = vaddr;
1507 init_waitqueue_head(&area->wq);
1508 /* Reserve the 1st slot for get_trampoline_vaddr() */
1509 set_bit(0, area->bitmap);
1510 atomic_set(&area->slot_count, 1);
1511 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1513 if (!xol_add_vma(mm, area))
1516 __free_page(area->pages[0]);
1518 kfree(area->bitmap);
1526 * get_xol_area - Allocate process's xol_area if necessary.
1527 * This area will be used for storing instructions for execution out of line.
1529 * Returns the allocated area or NULL.
1531 static struct xol_area *get_xol_area(void)
1533 struct mm_struct *mm = current->mm;
1534 struct xol_area *area;
1536 if (!mm->uprobes_state.xol_area)
1537 __create_xol_area(0);
1539 /* Pairs with xol_add_vma() smp_store_release() */
1540 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1545 * uprobe_clear_state - Free the area allocated for slots.
1547 void uprobe_clear_state(struct mm_struct *mm)
1549 struct xol_area *area = mm->uprobes_state.xol_area;
1551 mutex_lock(&delayed_uprobe_lock);
1552 delayed_uprobe_remove(NULL, mm);
1553 mutex_unlock(&delayed_uprobe_lock);
1558 put_page(area->pages[0]);
1559 kfree(area->bitmap);
1563 void uprobe_start_dup_mmap(void)
1565 percpu_down_read(&dup_mmap_sem);
1568 void uprobe_end_dup_mmap(void)
1570 percpu_up_read(&dup_mmap_sem);
1573 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1575 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1576 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1577 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1578 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1583 * - search for a free slot.
1585 static unsigned long xol_take_insn_slot(struct xol_area *area)
1587 unsigned long slot_addr;
1591 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1592 if (slot_nr < UINSNS_PER_PAGE) {
1593 if (!test_and_set_bit(slot_nr, area->bitmap))
1596 slot_nr = UINSNS_PER_PAGE;
1599 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1600 } while (slot_nr >= UINSNS_PER_PAGE);
1602 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1603 atomic_inc(&area->slot_count);
1609 * xol_get_insn_slot - allocate a slot for xol.
1610 * Returns the allocated slot address or 0.
1612 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1614 struct xol_area *area;
1615 unsigned long xol_vaddr;
1617 area = get_xol_area();
1621 xol_vaddr = xol_take_insn_slot(area);
1622 if (unlikely(!xol_vaddr))
1625 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1626 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1632 * xol_free_insn_slot - If slot was earlier allocated by
1633 * @xol_get_insn_slot(), make the slot available for
1634 * subsequent requests.
1636 static void xol_free_insn_slot(struct task_struct *tsk)
1638 struct xol_area *area;
1639 unsigned long vma_end;
1640 unsigned long slot_addr;
1642 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1645 slot_addr = tsk->utask->xol_vaddr;
1646 if (unlikely(!slot_addr))
1649 area = tsk->mm->uprobes_state.xol_area;
1650 vma_end = area->vaddr + PAGE_SIZE;
1651 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1652 unsigned long offset;
1655 offset = slot_addr - area->vaddr;
1656 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1657 if (slot_nr >= UINSNS_PER_PAGE)
1660 clear_bit(slot_nr, area->bitmap);
1661 atomic_dec(&area->slot_count);
1662 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1663 if (waitqueue_active(&area->wq))
1666 tsk->utask->xol_vaddr = 0;
1670 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1671 void *src, unsigned long len)
1673 /* Initialize the slot */
1674 copy_to_page(page, vaddr, src, len);
1677 * We probably need flush_icache_user_range() but it needs vma.
1678 * This should work on most of architectures by default. If
1679 * architecture needs to do something different it can define
1680 * its own version of the function.
1682 flush_dcache_page(page);
1686 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1687 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1689 * Return the address of the breakpoint instruction.
1691 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1693 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1696 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1698 struct uprobe_task *utask = current->utask;
1700 if (unlikely(utask && utask->active_uprobe))
1701 return utask->vaddr;
1703 return instruction_pointer(regs);
1706 static struct return_instance *free_ret_instance(struct return_instance *ri)
1708 struct return_instance *next = ri->next;
1709 put_uprobe(ri->uprobe);
1715 * Called with no locks held.
1716 * Called in context of an exiting or an exec-ing thread.
1718 void uprobe_free_utask(struct task_struct *t)
1720 struct uprobe_task *utask = t->utask;
1721 struct return_instance *ri;
1726 if (utask->active_uprobe)
1727 put_uprobe(utask->active_uprobe);
1729 ri = utask->return_instances;
1731 ri = free_ret_instance(ri);
1733 xol_free_insn_slot(t);
1739 * Allocate a uprobe_task object for the task if if necessary.
1740 * Called when the thread hits a breakpoint.
1743 * - pointer to new uprobe_task on success
1746 static struct uprobe_task *get_utask(void)
1748 if (!current->utask)
1749 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1750 return current->utask;
1753 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1755 struct uprobe_task *n_utask;
1756 struct return_instance **p, *o, *n;
1758 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1763 p = &n_utask->return_instances;
1764 for (o = o_utask->return_instances; o; o = o->next) {
1765 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1770 get_uprobe(n->uprobe);
1781 static void uprobe_warn(struct task_struct *t, const char *msg)
1783 pr_warn("uprobe: %s:%d failed to %s\n",
1784 current->comm, current->pid, msg);
1787 static void dup_xol_work(struct callback_head *work)
1789 if (current->flags & PF_EXITING)
1792 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1793 !fatal_signal_pending(current))
1794 uprobe_warn(current, "dup xol area");
1798 * Called in context of a new clone/fork from copy_process.
1800 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1802 struct uprobe_task *utask = current->utask;
1803 struct mm_struct *mm = current->mm;
1804 struct xol_area *area;
1808 if (!utask || !utask->return_instances)
1811 if (mm == t->mm && !(flags & CLONE_VFORK))
1814 if (dup_utask(t, utask))
1815 return uprobe_warn(t, "dup ret instances");
1817 /* The task can fork() after dup_xol_work() fails */
1818 area = mm->uprobes_state.xol_area;
1820 return uprobe_warn(t, "dup xol area");
1825 t->utask->dup_xol_addr = area->vaddr;
1826 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1827 task_work_add(t, &t->utask->dup_xol_work, true);
1831 * Current area->vaddr notion assume the trampoline address is always
1832 * equal area->vaddr.
1834 * Returns -1 in case the xol_area is not allocated.
1836 static unsigned long get_trampoline_vaddr(void)
1838 struct xol_area *area;
1839 unsigned long trampoline_vaddr = -1;
1841 /* Pairs with xol_add_vma() smp_store_release() */
1842 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1844 trampoline_vaddr = area->vaddr;
1846 return trampoline_vaddr;
1849 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1850 struct pt_regs *regs)
1852 struct return_instance *ri = utask->return_instances;
1853 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1855 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1856 ri = free_ret_instance(ri);
1859 utask->return_instances = ri;
1862 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1864 struct return_instance *ri;
1865 struct uprobe_task *utask;
1866 unsigned long orig_ret_vaddr, trampoline_vaddr;
1869 if (!get_xol_area())
1872 utask = get_utask();
1876 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1877 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1878 " nestedness limit pid/tgid=%d/%d\n",
1879 current->pid, current->tgid);
1883 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1887 trampoline_vaddr = get_trampoline_vaddr();
1888 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1889 if (orig_ret_vaddr == -1)
1892 /* drop the entries invalidated by longjmp() */
1893 chained = (orig_ret_vaddr == trampoline_vaddr);
1894 cleanup_return_instances(utask, chained, regs);
1897 * We don't want to keep trampoline address in stack, rather keep the
1898 * original return address of first caller thru all the consequent
1899 * instances. This also makes breakpoint unwrapping easier.
1902 if (!utask->return_instances) {
1904 * This situation is not possible. Likely we have an
1905 * attack from user-space.
1907 uprobe_warn(current, "handle tail call");
1910 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1913 ri->uprobe = get_uprobe(uprobe);
1914 ri->func = instruction_pointer(regs);
1915 ri->stack = user_stack_pointer(regs);
1916 ri->orig_ret_vaddr = orig_ret_vaddr;
1917 ri->chained = chained;
1920 ri->next = utask->return_instances;
1921 utask->return_instances = ri;
1928 /* Prepare to single-step probed instruction out of line. */
1930 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1932 struct uprobe_task *utask;
1933 unsigned long xol_vaddr;
1936 utask = get_utask();
1940 xol_vaddr = xol_get_insn_slot(uprobe);
1944 utask->xol_vaddr = xol_vaddr;
1945 utask->vaddr = bp_vaddr;
1947 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1948 if (unlikely(err)) {
1949 xol_free_insn_slot(current);
1953 utask->active_uprobe = uprobe;
1954 utask->state = UTASK_SSTEP;
1959 * If we are singlestepping, then ensure this thread is not connected to
1960 * non-fatal signals until completion of singlestep. When xol insn itself
1961 * triggers the signal, restart the original insn even if the task is
1962 * already SIGKILL'ed (since coredump should report the correct ip). This
1963 * is even more important if the task has a handler for SIGSEGV/etc, The
1964 * _same_ instruction should be repeated again after return from the signal
1965 * handler, and SSTEP can never finish in this case.
1967 bool uprobe_deny_signal(void)
1969 struct task_struct *t = current;
1970 struct uprobe_task *utask = t->utask;
1972 if (likely(!utask || !utask->active_uprobe))
1975 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1977 if (signal_pending(t)) {
1978 spin_lock_irq(&t->sighand->siglock);
1979 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1980 spin_unlock_irq(&t->sighand->siglock);
1982 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1983 utask->state = UTASK_SSTEP_TRAPPED;
1984 set_tsk_thread_flag(t, TIF_UPROBE);
1991 static void mmf_recalc_uprobes(struct mm_struct *mm)
1993 struct vm_area_struct *vma;
1995 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1996 if (!valid_vma(vma, false))
1999 * This is not strictly accurate, we can race with
2000 * uprobe_unregister() and see the already removed
2001 * uprobe if delete_uprobe() was not yet called.
2002 * Or this uprobe can be filtered out.
2004 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
2008 clear_bit(MMF_HAS_UPROBES, &mm->flags);
2011 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
2014 uprobe_opcode_t opcode;
2017 pagefault_disable();
2018 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
2021 if (likely(result == 0))
2025 * The NULL 'tsk' here ensures that any faults that occur here
2026 * will not be accounted to the task. 'mm' *is* current->mm,
2027 * but we treat this as a 'remote' access since it is
2028 * essentially a kernel access to the memory.
2030 result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
2035 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
2038 /* This needs to return true for any variant of the trap insn */
2039 return is_trap_insn(&opcode);
2042 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
2044 struct mm_struct *mm = current->mm;
2045 struct uprobe *uprobe = NULL;
2046 struct vm_area_struct *vma;
2048 down_read(&mm->mmap_sem);
2049 vma = find_vma(mm, bp_vaddr);
2050 if (vma && vma->vm_start <= bp_vaddr) {
2051 if (valid_vma(vma, false)) {
2052 struct inode *inode = file_inode(vma->vm_file);
2053 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
2055 uprobe = find_uprobe(inode, offset);
2059 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
2064 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
2065 mmf_recalc_uprobes(mm);
2066 up_read(&mm->mmap_sem);
2071 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
2073 struct uprobe_consumer *uc;
2074 int remove = UPROBE_HANDLER_REMOVE;
2075 bool need_prep = false; /* prepare return uprobe, when needed */
2077 down_read(&uprobe->register_rwsem);
2078 for (uc = uprobe->consumers; uc; uc = uc->next) {
2082 rc = uc->handler(uc, regs);
2083 WARN(rc & ~UPROBE_HANDLER_MASK,
2084 "bad rc=0x%x from %ps()\n", rc, uc->handler);
2087 if (uc->ret_handler)
2093 if (need_prep && !remove)
2094 prepare_uretprobe(uprobe, regs); /* put bp at return */
2096 if (remove && uprobe->consumers) {
2097 WARN_ON(!uprobe_is_active(uprobe));
2098 unapply_uprobe(uprobe, current->mm);
2100 up_read(&uprobe->register_rwsem);
2104 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
2106 struct uprobe *uprobe = ri->uprobe;
2107 struct uprobe_consumer *uc;
2109 down_read(&uprobe->register_rwsem);
2110 for (uc = uprobe->consumers; uc; uc = uc->next) {
2111 if (uc->ret_handler)
2112 uc->ret_handler(uc, ri->func, regs);
2114 up_read(&uprobe->register_rwsem);
2117 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
2122 chained = ri->chained;
2123 ri = ri->next; /* can't be NULL if chained */
2129 static void handle_trampoline(struct pt_regs *regs)
2131 struct uprobe_task *utask;
2132 struct return_instance *ri, *next;
2135 utask = current->utask;
2139 ri = utask->return_instances;
2145 * We should throw out the frames invalidated by longjmp().
2146 * If this chain is valid, then the next one should be alive
2147 * or NULL; the latter case means that nobody but ri->func
2148 * could hit this trampoline on return. TODO: sigaltstack().
2150 next = find_next_ret_chain(ri);
2151 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
2153 instruction_pointer_set(regs, ri->orig_ret_vaddr);
2156 handle_uretprobe_chain(ri, regs);
2157 ri = free_ret_instance(ri);
2159 } while (ri != next);
2162 utask->return_instances = ri;
2166 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
2171 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
2176 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
2177 struct pt_regs *regs)
2183 * Run handler and ask thread to singlestep.
2184 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2186 static void handle_swbp(struct pt_regs *regs)
2188 struct uprobe *uprobe;
2189 unsigned long bp_vaddr;
2190 int uninitialized_var(is_swbp);
2192 bp_vaddr = uprobe_get_swbp_addr(regs);
2193 if (bp_vaddr == get_trampoline_vaddr())
2194 return handle_trampoline(regs);
2196 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
2199 /* No matching uprobe; signal SIGTRAP. */
2200 send_sig(SIGTRAP, current, 0);
2203 * Either we raced with uprobe_unregister() or we can't
2204 * access this memory. The latter is only possible if
2205 * another thread plays with our ->mm. In both cases
2206 * we can simply restart. If this vma was unmapped we
2207 * can pretend this insn was not executed yet and get
2208 * the (correct) SIGSEGV after restart.
2210 instruction_pointer_set(regs, bp_vaddr);
2215 /* change it in advance for ->handler() and restart */
2216 instruction_pointer_set(regs, bp_vaddr);
2219 * TODO: move copy_insn/etc into _register and remove this hack.
2220 * After we hit the bp, _unregister + _register can install the
2221 * new and not-yet-analyzed uprobe at the same address, restart.
2223 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
2227 * Pairs with the smp_wmb() in prepare_uprobe().
2229 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2230 * we must also see the stores to &uprobe->arch performed by the
2231 * prepare_uprobe() call.
2235 /* Tracing handlers use ->utask to communicate with fetch methods */
2239 if (arch_uprobe_ignore(&uprobe->arch, regs))
2242 handler_chain(uprobe, regs);
2244 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
2247 if (!pre_ssout(uprobe, regs, bp_vaddr))
2250 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2256 * Perform required fix-ups and disable singlestep.
2257 * Allow pending signals to take effect.
2259 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
2261 struct uprobe *uprobe;
2264 uprobe = utask->active_uprobe;
2265 if (utask->state == UTASK_SSTEP_ACK)
2266 err = arch_uprobe_post_xol(&uprobe->arch, regs);
2267 else if (utask->state == UTASK_SSTEP_TRAPPED)
2268 arch_uprobe_abort_xol(&uprobe->arch, regs);
2273 utask->active_uprobe = NULL;
2274 utask->state = UTASK_RUNNING;
2275 xol_free_insn_slot(current);
2277 spin_lock_irq(¤t->sighand->siglock);
2278 recalc_sigpending(); /* see uprobe_deny_signal() */
2279 spin_unlock_irq(¤t->sighand->siglock);
2281 if (unlikely(err)) {
2282 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
2288 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2289 * allows the thread to return from interrupt. After that handle_swbp()
2290 * sets utask->active_uprobe.
2292 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2293 * and allows the thread to return from interrupt.
2295 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2296 * uprobe_notify_resume().
2298 void uprobe_notify_resume(struct pt_regs *regs)
2300 struct uprobe_task *utask;
2302 clear_thread_flag(TIF_UPROBE);
2304 utask = current->utask;
2305 if (utask && utask->active_uprobe)
2306 handle_singlestep(utask, regs);
2312 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2313 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2315 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2320 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2321 (!current->utask || !current->utask->return_instances))
2324 set_thread_flag(TIF_UPROBE);
2329 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2330 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2332 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2334 struct uprobe_task *utask = current->utask;
2336 if (!current->mm || !utask || !utask->active_uprobe)
2337 /* task is currently not uprobed */
2340 utask->state = UTASK_SSTEP_ACK;
2341 set_thread_flag(TIF_UPROBE);
2345 static struct notifier_block uprobe_exception_nb = {
2346 .notifier_call = arch_uprobe_exception_notify,
2347 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2350 void __init uprobes_init(void)
2354 for (i = 0; i < UPROBES_HASH_SZ; i++)
2355 mutex_init(&uprobes_mmap_mutex[i]);
2357 BUG_ON(register_die_notifier(&uprobe_exception_nb));