2 * User-space Probes (UProbes)
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright (C) IBM Corporation, 2008-2012
22 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra
25 #include <linux/kernel.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h> /* read_mapping_page */
28 #include <linux/slab.h>
29 #include <linux/sched.h>
30 #include <linux/sched/mm.h>
31 #include <linux/sched/coredump.h>
32 #include <linux/export.h>
33 #include <linux/rmap.h> /* anon_vma_prepare */
34 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
35 #include <linux/swap.h> /* try_to_free_swap */
36 #include <linux/ptrace.h> /* user_enable_single_step */
37 #include <linux/kdebug.h> /* notifier mechanism */
38 #include "../../mm/internal.h" /* munlock_vma_page */
39 #include <linux/percpu-rwsem.h>
40 #include <linux/task_work.h>
41 #include <linux/shmem_fs.h>
43 #include <linux/uprobes.h>
45 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
46 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
48 static struct rb_root uprobes_tree = RB_ROOT;
50 * allows us to skip the uprobe_mmap if there are no uprobe events active
51 * at this time. Probably a fine grained per inode count is better?
53 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
55 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
57 #define UPROBES_HASH_SZ 13
58 /* serialize uprobe->pending_list */
59 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
60 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
62 static struct percpu_rw_semaphore dup_mmap_sem;
64 /* Have a copy of original instruction */
65 #define UPROBE_COPY_INSN 0
68 struct rb_node rb_node; /* node in the rb tree */
70 struct rw_semaphore register_rwsem;
71 struct rw_semaphore consumer_rwsem;
72 struct list_head pending_list;
73 struct uprobe_consumer *consumers;
74 struct inode *inode; /* Also hold a ref to inode */
79 * The generic code assumes that it has two members of unknown type
80 * owned by the arch-specific code:
82 * insn - copy_insn() saves the original instruction here for
83 * arch_uprobe_analyze_insn().
85 * ixol - potentially modified instruction to execute out of
86 * line, copied to xol_area by xol_get_insn_slot().
88 struct arch_uprobe arch;
92 * Execute out of line area: anonymous executable mapping installed
93 * by the probed task to execute the copy of the original instruction
94 * mangled by set_swbp().
96 * On a breakpoint hit, thread contests for a slot. It frees the
97 * slot after singlestep. Currently a fixed number of slots are
101 wait_queue_head_t wq; /* if all slots are busy */
102 atomic_t slot_count; /* number of in-use slots */
103 unsigned long *bitmap; /* 0 = free slot */
105 struct vm_special_mapping xol_mapping;
106 struct page *pages[2];
108 * We keep the vma's vm_start rather than a pointer to the vma
109 * itself. The probed process or a naughty kernel module could make
110 * the vma go away, and we must handle that reasonably gracefully.
112 unsigned long vaddr; /* Page(s) of instruction slots */
116 * valid_vma: Verify if the specified vma is an executable vma
117 * Relax restrictions while unregistering: vm_flags might have
118 * changed after breakpoint was inserted.
119 * - is_register: indicates if we are in register context.
120 * - Return 1 if the specified virtual address is in an
123 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
125 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
130 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
133 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
135 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
138 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
140 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
144 * __replace_page - replace page in vma by new page.
145 * based on replace_page in mm/ksm.c
147 * @vma: vma that holds the pte pointing to page
148 * @addr: address the old @page is mapped at
149 * @page: the cowed page we are replacing by kpage
150 * @kpage: the modified page we replace page by
152 * Returns 0 on success, -EFAULT on failure.
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 = {
164 /* For mmu_notifiers */
165 const unsigned long mmun_start = addr;
166 const unsigned long mmun_end = addr + PAGE_SIZE;
167 struct mem_cgroup *memcg;
169 VM_BUG_ON_PAGE(PageTransHuge(old_page), old_page);
171 err = mem_cgroup_try_charge(new_page, vma->vm_mm, GFP_KERNEL, &memcg,
176 /* For try_to_free_swap() and munlock_vma_page() below */
179 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
181 if (!page_vma_mapped_walk(&pvmw)) {
182 mem_cgroup_cancel_charge(new_page, memcg, false);
185 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
188 page_add_new_anon_rmap(new_page, vma, addr, false);
189 mem_cgroup_commit_charge(new_page, memcg, false, false);
190 lru_cache_add_active_or_unevictable(new_page, vma);
192 if (!PageAnon(old_page)) {
193 dec_mm_counter(mm, mm_counter_file(old_page));
194 inc_mm_counter(mm, MM_ANONPAGES);
197 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
198 ptep_clear_flush_notify(vma, addr, pvmw.pte);
199 set_pte_at_notify(mm, addr, pvmw.pte,
200 mk_pte(new_page, vma->vm_page_prot));
202 page_remove_rmap(old_page, false);
203 if (!page_mapped(old_page))
204 try_to_free_swap(old_page);
205 page_vma_mapped_walk_done(&pvmw);
207 if (vma->vm_flags & VM_LOCKED)
208 munlock_vma_page(old_page);
213 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
214 unlock_page(old_page);
219 * is_swbp_insn - check if instruction is breakpoint instruction.
220 * @insn: instruction to be checked.
221 * Default implementation of is_swbp_insn
222 * Returns true if @insn is a breakpoint instruction.
224 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
226 return *insn == UPROBE_SWBP_INSN;
230 * is_trap_insn - check if instruction is breakpoint instruction.
231 * @insn: instruction to be checked.
232 * Default implementation of is_trap_insn
233 * Returns true if @insn is a breakpoint instruction.
235 * This function is needed for the case where an architecture has multiple
236 * trap instructions (like powerpc).
238 bool __weak is_trap_insn(uprobe_opcode_t *insn)
240 return is_swbp_insn(insn);
243 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
245 void *kaddr = kmap_atomic(page);
246 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
247 kunmap_atomic(kaddr);
250 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
252 void *kaddr = kmap_atomic(page);
253 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
254 kunmap_atomic(kaddr);
257 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
259 uprobe_opcode_t old_opcode;
263 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
264 * We do not check if it is any other 'trap variant' which could
265 * be conditional trap instruction such as the one powerpc supports.
267 * The logic is that we do not care if the underlying instruction
268 * is a trap variant; uprobes always wins over any other (gdb)
271 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
272 is_swbp = is_swbp_insn(&old_opcode);
274 if (is_swbp_insn(new_opcode)) {
275 if (is_swbp) /* register: already installed? */
278 if (!is_swbp) /* unregister: was it changed by us? */
287 * Expect the breakpoint instruction to be the smallest size instruction for
288 * the architecture. If an arch has variable length instruction and the
289 * breakpoint instruction is not of the smallest length instruction
290 * supported by that architecture then we need to modify is_trap_at_addr and
291 * uprobe_write_opcode accordingly. This would never be a problem for archs
292 * that have fixed length instructions.
294 * uprobe_write_opcode - write the opcode at a given virtual address.
295 * @mm: the probed process address space.
296 * @vaddr: the virtual address to store the opcode.
297 * @opcode: opcode to be written at @vaddr.
299 * Called with mm->mmap_sem held for write.
300 * Return 0 (success) or a negative errno.
302 int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr,
303 uprobe_opcode_t opcode)
305 struct page *old_page, *new_page;
306 struct vm_area_struct *vma;
310 /* Read the page with vaddr into memory */
311 ret = get_user_pages_remote(NULL, mm, vaddr, 1,
312 FOLL_FORCE | FOLL_SPLIT, &old_page, &vma, NULL);
316 ret = verify_opcode(old_page, vaddr, &opcode);
320 ret = anon_vma_prepare(vma);
325 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
329 __SetPageUptodate(new_page);
330 copy_highpage(new_page, old_page);
331 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
333 ret = __replace_page(vma, vaddr, old_page, new_page);
338 if (unlikely(ret == -EAGAIN))
344 * set_swbp - store breakpoint at a given address.
345 * @auprobe: arch specific probepoint information.
346 * @mm: the probed process address space.
347 * @vaddr: the virtual address to insert the opcode.
349 * For mm @mm, store the breakpoint instruction at @vaddr.
350 * Return 0 (success) or a negative errno.
352 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
354 return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
358 * set_orig_insn - Restore the original instruction.
359 * @mm: the probed process address space.
360 * @auprobe: arch specific probepoint information.
361 * @vaddr: the virtual address to insert the opcode.
363 * For mm @mm, restore the original opcode (opcode) at @vaddr.
364 * Return 0 (success) or a negative errno.
367 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
369 return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn);
372 static struct uprobe *get_uprobe(struct uprobe *uprobe)
374 atomic_inc(&uprobe->ref);
378 static void put_uprobe(struct uprobe *uprobe)
380 if (atomic_dec_and_test(&uprobe->ref))
384 static int match_uprobe(struct uprobe *l, struct uprobe *r)
386 if (l->inode < r->inode)
389 if (l->inode > r->inode)
392 if (l->offset < r->offset)
395 if (l->offset > r->offset)
401 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
403 struct uprobe u = { .inode = inode, .offset = offset };
404 struct rb_node *n = uprobes_tree.rb_node;
405 struct uprobe *uprobe;
409 uprobe = rb_entry(n, struct uprobe, rb_node);
410 match = match_uprobe(&u, uprobe);
412 return get_uprobe(uprobe);
423 * Find a uprobe corresponding to a given inode:offset
424 * Acquires uprobes_treelock
426 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
428 struct uprobe *uprobe;
430 spin_lock(&uprobes_treelock);
431 uprobe = __find_uprobe(inode, offset);
432 spin_unlock(&uprobes_treelock);
437 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
439 struct rb_node **p = &uprobes_tree.rb_node;
440 struct rb_node *parent = NULL;
446 u = rb_entry(parent, struct uprobe, rb_node);
447 match = match_uprobe(uprobe, u);
449 return get_uprobe(u);
452 p = &parent->rb_left;
454 p = &parent->rb_right;
459 rb_link_node(&uprobe->rb_node, parent, p);
460 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
461 /* get access + creation ref */
462 atomic_set(&uprobe->ref, 2);
468 * Acquire uprobes_treelock.
469 * Matching uprobe already exists in rbtree;
470 * increment (access refcount) and return the matching uprobe.
472 * No matching uprobe; insert the uprobe in rb_tree;
473 * get a double refcount (access + creation) and return NULL.
475 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
479 spin_lock(&uprobes_treelock);
480 u = __insert_uprobe(uprobe);
481 spin_unlock(&uprobes_treelock);
486 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
488 struct uprobe *uprobe, *cur_uprobe;
490 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
494 uprobe->inode = igrab(inode);
495 uprobe->offset = offset;
496 init_rwsem(&uprobe->register_rwsem);
497 init_rwsem(&uprobe->consumer_rwsem);
499 /* add to uprobes_tree, sorted on inode:offset */
500 cur_uprobe = insert_uprobe(uprobe);
501 /* a uprobe exists for this inode:offset combination */
511 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
513 down_write(&uprobe->consumer_rwsem);
514 uc->next = uprobe->consumers;
515 uprobe->consumers = uc;
516 up_write(&uprobe->consumer_rwsem);
520 * For uprobe @uprobe, delete the consumer @uc.
521 * Return true if the @uc is deleted successfully
524 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
526 struct uprobe_consumer **con;
529 down_write(&uprobe->consumer_rwsem);
530 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
537 up_write(&uprobe->consumer_rwsem);
542 static int __copy_insn(struct address_space *mapping, struct file *filp,
543 void *insn, int nbytes, loff_t offset)
547 * Ensure that the page that has the original instruction is populated
548 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
549 * see uprobe_register().
551 if (mapping->a_ops->readpage)
552 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
554 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
556 return PTR_ERR(page);
558 copy_from_page(page, offset, insn, nbytes);
564 static int copy_insn(struct uprobe *uprobe, struct file *filp)
566 struct address_space *mapping = uprobe->inode->i_mapping;
567 loff_t offs = uprobe->offset;
568 void *insn = &uprobe->arch.insn;
569 int size = sizeof(uprobe->arch.insn);
572 /* Copy only available bytes, -EIO if nothing was read */
574 if (offs >= i_size_read(uprobe->inode))
577 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
578 err = __copy_insn(mapping, filp, insn, len, offs);
590 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
591 struct mm_struct *mm, unsigned long vaddr)
595 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
598 /* TODO: move this into _register, until then we abuse this sem. */
599 down_write(&uprobe->consumer_rwsem);
600 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
603 ret = copy_insn(uprobe, file);
608 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
611 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
615 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
616 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
619 up_write(&uprobe->consumer_rwsem);
624 static inline bool consumer_filter(struct uprobe_consumer *uc,
625 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
627 return !uc->filter || uc->filter(uc, ctx, mm);
630 static bool filter_chain(struct uprobe *uprobe,
631 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
633 struct uprobe_consumer *uc;
636 down_read(&uprobe->consumer_rwsem);
637 for (uc = uprobe->consumers; uc; uc = uc->next) {
638 ret = consumer_filter(uc, ctx, mm);
642 up_read(&uprobe->consumer_rwsem);
648 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
649 struct vm_area_struct *vma, unsigned long vaddr)
654 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
659 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
660 * the task can hit this breakpoint right after __replace_page().
662 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
664 set_bit(MMF_HAS_UPROBES, &mm->flags);
666 ret = set_swbp(&uprobe->arch, mm, vaddr);
668 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
669 else if (first_uprobe)
670 clear_bit(MMF_HAS_UPROBES, &mm->flags);
676 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
678 set_bit(MMF_RECALC_UPROBES, &mm->flags);
679 return set_orig_insn(&uprobe->arch, mm, vaddr);
682 static inline bool uprobe_is_active(struct uprobe *uprobe)
684 return !RB_EMPTY_NODE(&uprobe->rb_node);
687 * There could be threads that have already hit the breakpoint. They
688 * will recheck the current insn and restart if find_uprobe() fails.
689 * See find_active_uprobe().
691 static void delete_uprobe(struct uprobe *uprobe)
693 if (WARN_ON(!uprobe_is_active(uprobe)))
696 spin_lock(&uprobes_treelock);
697 rb_erase(&uprobe->rb_node, &uprobes_tree);
698 spin_unlock(&uprobes_treelock);
699 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
705 struct map_info *next;
706 struct mm_struct *mm;
710 static inline struct map_info *free_map_info(struct map_info *info)
712 struct map_info *next = info->next;
717 static struct map_info *
718 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
720 unsigned long pgoff = offset >> PAGE_SHIFT;
721 struct vm_area_struct *vma;
722 struct map_info *curr = NULL;
723 struct map_info *prev = NULL;
724 struct map_info *info;
728 i_mmap_lock_read(mapping);
729 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
730 if (!valid_vma(vma, is_register))
733 if (!prev && !more) {
735 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
736 * reclaim. This is optimistic, no harm done if it fails.
738 prev = kmalloc(sizeof(struct map_info),
739 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
748 if (!mmget_not_zero(vma->vm_mm))
756 info->mm = vma->vm_mm;
757 info->vaddr = offset_to_vaddr(vma, offset);
759 i_mmap_unlock_read(mapping);
771 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
773 curr = ERR_PTR(-ENOMEM);
783 prev = free_map_info(prev);
788 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
790 bool is_register = !!new;
791 struct map_info *info;
794 percpu_down_write(&dup_mmap_sem);
795 info = build_map_info(uprobe->inode->i_mapping,
796 uprobe->offset, is_register);
803 struct mm_struct *mm = info->mm;
804 struct vm_area_struct *vma;
806 if (err && is_register)
809 down_write(&mm->mmap_sem);
810 vma = find_vma(mm, info->vaddr);
811 if (!vma || !valid_vma(vma, is_register) ||
812 file_inode(vma->vm_file) != uprobe->inode)
815 if (vma->vm_start > info->vaddr ||
816 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
820 /* consult only the "caller", new consumer. */
821 if (consumer_filter(new,
822 UPROBE_FILTER_REGISTER, mm))
823 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
824 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
825 if (!filter_chain(uprobe,
826 UPROBE_FILTER_UNREGISTER, mm))
827 err |= remove_breakpoint(uprobe, mm, info->vaddr);
831 up_write(&mm->mmap_sem);
834 info = free_map_info(info);
837 percpu_up_write(&dup_mmap_sem);
841 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
843 consumer_add(uprobe, uc);
844 return register_for_each_vma(uprobe, uc);
847 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
851 if (WARN_ON(!consumer_del(uprobe, uc)))
854 err = register_for_each_vma(uprobe, NULL);
855 /* TODO : cant unregister? schedule a worker thread */
856 if (!uprobe->consumers && !err)
857 delete_uprobe(uprobe);
861 * uprobe_register - register a probe
862 * @inode: the file in which the probe has to be placed.
863 * @offset: offset from the start of the file.
864 * @uc: information on howto handle the probe..
866 * Apart from the access refcount, uprobe_register() takes a creation
867 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
868 * inserted into the rbtree (i.e first consumer for a @inode:@offset
869 * tuple). Creation refcount stops uprobe_unregister from freeing the
870 * @uprobe even before the register operation is complete. Creation
871 * refcount is released when the last @uc for the @uprobe
874 * Return errno if it cannot successully install probes
875 * else return 0 (success)
877 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
879 struct uprobe *uprobe;
882 /* Uprobe must have at least one set consumer */
883 if (!uc->handler && !uc->ret_handler)
886 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
887 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
889 /* Racy, just to catch the obvious mistakes */
890 if (offset > i_size_read(inode))
894 * This ensures that copy_from_page() and copy_to_page()
895 * can't cross page boundary.
897 if (!IS_ALIGNED(offset, UPROBE_SWBP_INSN_SIZE))
901 uprobe = alloc_uprobe(inode, offset);
905 * We can race with uprobe_unregister()->delete_uprobe().
906 * Check uprobe_is_active() and retry if it is false.
908 down_write(&uprobe->register_rwsem);
910 if (likely(uprobe_is_active(uprobe))) {
911 ret = __uprobe_register(uprobe, uc);
913 __uprobe_unregister(uprobe, uc);
915 up_write(&uprobe->register_rwsem);
918 if (unlikely(ret == -EAGAIN))
922 EXPORT_SYMBOL_GPL(uprobe_register);
925 * uprobe_apply - unregister a already registered probe.
926 * @inode: the file in which the probe has to be removed.
927 * @offset: offset from the start of the file.
928 * @uc: consumer which wants to add more or remove some breakpoints
929 * @add: add or remove the breakpoints
931 int uprobe_apply(struct inode *inode, loff_t offset,
932 struct uprobe_consumer *uc, bool add)
934 struct uprobe *uprobe;
935 struct uprobe_consumer *con;
938 uprobe = find_uprobe(inode, offset);
939 if (WARN_ON(!uprobe))
942 down_write(&uprobe->register_rwsem);
943 for (con = uprobe->consumers; con && con != uc ; con = con->next)
946 ret = register_for_each_vma(uprobe, add ? uc : NULL);
947 up_write(&uprobe->register_rwsem);
954 * uprobe_unregister - unregister a already registered probe.
955 * @inode: the file in which the probe has to be removed.
956 * @offset: offset from the start of the file.
957 * @uc: identify which probe if multiple probes are colocated.
959 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
961 struct uprobe *uprobe;
963 uprobe = find_uprobe(inode, offset);
964 if (WARN_ON(!uprobe))
967 down_write(&uprobe->register_rwsem);
968 __uprobe_unregister(uprobe, uc);
969 up_write(&uprobe->register_rwsem);
972 EXPORT_SYMBOL_GPL(uprobe_unregister);
974 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
976 struct vm_area_struct *vma;
979 down_read(&mm->mmap_sem);
980 for (vma = mm->mmap; vma; vma = vma->vm_next) {
984 if (!valid_vma(vma, false) ||
985 file_inode(vma->vm_file) != uprobe->inode)
988 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
989 if (uprobe->offset < offset ||
990 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
993 vaddr = offset_to_vaddr(vma, uprobe->offset);
994 err |= remove_breakpoint(uprobe, mm, vaddr);
996 up_read(&mm->mmap_sem);
1001 static struct rb_node *
1002 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1004 struct rb_node *n = uprobes_tree.rb_node;
1007 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1009 if (inode < u->inode) {
1011 } else if (inode > u->inode) {
1014 if (max < u->offset)
1016 else if (min > u->offset)
1027 * For a given range in vma, build a list of probes that need to be inserted.
1029 static void build_probe_list(struct inode *inode,
1030 struct vm_area_struct *vma,
1031 unsigned long start, unsigned long end,
1032 struct list_head *head)
1035 struct rb_node *n, *t;
1038 INIT_LIST_HEAD(head);
1039 min = vaddr_to_offset(vma, start);
1040 max = min + (end - start) - 1;
1042 spin_lock(&uprobes_treelock);
1043 n = find_node_in_range(inode, min, max);
1045 for (t = n; t; t = rb_prev(t)) {
1046 u = rb_entry(t, struct uprobe, rb_node);
1047 if (u->inode != inode || u->offset < min)
1049 list_add(&u->pending_list, head);
1052 for (t = n; (t = rb_next(t)); ) {
1053 u = rb_entry(t, struct uprobe, rb_node);
1054 if (u->inode != inode || u->offset > max)
1056 list_add(&u->pending_list, head);
1060 spin_unlock(&uprobes_treelock);
1064 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1066 * Currently we ignore all errors and always return 0, the callers
1067 * can't handle the failure anyway.
1069 int uprobe_mmap(struct vm_area_struct *vma)
1071 struct list_head tmp_list;
1072 struct uprobe *uprobe, *u;
1073 struct inode *inode;
1075 if (no_uprobe_events() || !valid_vma(vma, true))
1078 inode = file_inode(vma->vm_file);
1082 mutex_lock(uprobes_mmap_hash(inode));
1083 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1085 * We can race with uprobe_unregister(), this uprobe can be already
1086 * removed. But in this case filter_chain() must return false, all
1087 * consumers have gone away.
1089 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1090 if (!fatal_signal_pending(current) &&
1091 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1092 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1093 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1097 mutex_unlock(uprobes_mmap_hash(inode));
1103 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1106 struct inode *inode;
1109 inode = file_inode(vma->vm_file);
1111 min = vaddr_to_offset(vma, start);
1112 max = min + (end - start) - 1;
1114 spin_lock(&uprobes_treelock);
1115 n = find_node_in_range(inode, min, max);
1116 spin_unlock(&uprobes_treelock);
1122 * Called in context of a munmap of a vma.
1124 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1126 if (no_uprobe_events() || !valid_vma(vma, false))
1129 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1132 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1133 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1136 if (vma_has_uprobes(vma, start, end))
1137 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1140 /* Slot allocation for XOL */
1141 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1143 struct vm_area_struct *vma;
1146 if (down_write_killable(&mm->mmap_sem))
1149 if (mm->uprobes_state.xol_area) {
1155 /* Try to map as high as possible, this is only a hint. */
1156 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1158 if (area->vaddr & ~PAGE_MASK) {
1164 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1165 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1166 &area->xol_mapping);
1173 smp_wmb(); /* pairs with get_xol_area() */
1174 mm->uprobes_state.xol_area = area;
1176 up_write(&mm->mmap_sem);
1181 static struct xol_area *__create_xol_area(unsigned long vaddr)
1183 struct mm_struct *mm = current->mm;
1184 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1185 struct xol_area *area;
1187 area = kmalloc(sizeof(*area), GFP_KERNEL);
1188 if (unlikely(!area))
1191 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1195 area->xol_mapping.name = "[uprobes]";
1196 area->xol_mapping.fault = NULL;
1197 area->xol_mapping.pages = area->pages;
1198 area->pages[0] = alloc_page(GFP_HIGHUSER);
1199 if (!area->pages[0])
1201 area->pages[1] = NULL;
1203 area->vaddr = vaddr;
1204 init_waitqueue_head(&area->wq);
1205 /* Reserve the 1st slot for get_trampoline_vaddr() */
1206 set_bit(0, area->bitmap);
1207 atomic_set(&area->slot_count, 1);
1208 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1210 if (!xol_add_vma(mm, area))
1213 __free_page(area->pages[0]);
1215 kfree(area->bitmap);
1223 * get_xol_area - Allocate process's xol_area if necessary.
1224 * This area will be used for storing instructions for execution out of line.
1226 * Returns the allocated area or NULL.
1228 static struct xol_area *get_xol_area(void)
1230 struct mm_struct *mm = current->mm;
1231 struct xol_area *area;
1233 if (!mm->uprobes_state.xol_area)
1234 __create_xol_area(0);
1236 area = mm->uprobes_state.xol_area;
1237 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1242 * uprobe_clear_state - Free the area allocated for slots.
1244 void uprobe_clear_state(struct mm_struct *mm)
1246 struct xol_area *area = mm->uprobes_state.xol_area;
1251 put_page(area->pages[0]);
1252 kfree(area->bitmap);
1256 void uprobe_start_dup_mmap(void)
1258 percpu_down_read(&dup_mmap_sem);
1261 void uprobe_end_dup_mmap(void)
1263 percpu_up_read(&dup_mmap_sem);
1266 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1268 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1269 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1270 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1271 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1276 * - search for a free slot.
1278 static unsigned long xol_take_insn_slot(struct xol_area *area)
1280 unsigned long slot_addr;
1284 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1285 if (slot_nr < UINSNS_PER_PAGE) {
1286 if (!test_and_set_bit(slot_nr, area->bitmap))
1289 slot_nr = UINSNS_PER_PAGE;
1292 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1293 } while (slot_nr >= UINSNS_PER_PAGE);
1295 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1296 atomic_inc(&area->slot_count);
1302 * xol_get_insn_slot - allocate a slot for xol.
1303 * Returns the allocated slot address or 0.
1305 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1307 struct xol_area *area;
1308 unsigned long xol_vaddr;
1310 area = get_xol_area();
1314 xol_vaddr = xol_take_insn_slot(area);
1315 if (unlikely(!xol_vaddr))
1318 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1319 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1325 * xol_free_insn_slot - If slot was earlier allocated by
1326 * @xol_get_insn_slot(), make the slot available for
1327 * subsequent requests.
1329 static void xol_free_insn_slot(struct task_struct *tsk)
1331 struct xol_area *area;
1332 unsigned long vma_end;
1333 unsigned long slot_addr;
1335 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1338 slot_addr = tsk->utask->xol_vaddr;
1339 if (unlikely(!slot_addr))
1342 area = tsk->mm->uprobes_state.xol_area;
1343 vma_end = area->vaddr + PAGE_SIZE;
1344 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1345 unsigned long offset;
1348 offset = slot_addr - area->vaddr;
1349 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1350 if (slot_nr >= UINSNS_PER_PAGE)
1353 clear_bit(slot_nr, area->bitmap);
1354 atomic_dec(&area->slot_count);
1355 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1356 if (waitqueue_active(&area->wq))
1359 tsk->utask->xol_vaddr = 0;
1363 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1364 void *src, unsigned long len)
1366 /* Initialize the slot */
1367 copy_to_page(page, vaddr, src, len);
1370 * We probably need flush_icache_user_range() but it needs vma.
1371 * This should work on most of architectures by default. If
1372 * architecture needs to do something different it can define
1373 * its own version of the function.
1375 flush_dcache_page(page);
1379 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1380 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1382 * Return the address of the breakpoint instruction.
1384 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1386 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1389 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1391 struct uprobe_task *utask = current->utask;
1393 if (unlikely(utask && utask->active_uprobe))
1394 return utask->vaddr;
1396 return instruction_pointer(regs);
1399 static struct return_instance *free_ret_instance(struct return_instance *ri)
1401 struct return_instance *next = ri->next;
1402 put_uprobe(ri->uprobe);
1408 * Called with no locks held.
1409 * Called in context of a exiting or a exec-ing thread.
1411 void uprobe_free_utask(struct task_struct *t)
1413 struct uprobe_task *utask = t->utask;
1414 struct return_instance *ri;
1419 if (utask->active_uprobe)
1420 put_uprobe(utask->active_uprobe);
1422 ri = utask->return_instances;
1424 ri = free_ret_instance(ri);
1426 xol_free_insn_slot(t);
1432 * Allocate a uprobe_task object for the task if if necessary.
1433 * Called when the thread hits a breakpoint.
1436 * - pointer to new uprobe_task on success
1439 static struct uprobe_task *get_utask(void)
1441 if (!current->utask)
1442 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1443 return current->utask;
1446 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1448 struct uprobe_task *n_utask;
1449 struct return_instance **p, *o, *n;
1451 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1456 p = &n_utask->return_instances;
1457 for (o = o_utask->return_instances; o; o = o->next) {
1458 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1463 get_uprobe(n->uprobe);
1474 static void uprobe_warn(struct task_struct *t, const char *msg)
1476 pr_warn("uprobe: %s:%d failed to %s\n",
1477 current->comm, current->pid, msg);
1480 static void dup_xol_work(struct callback_head *work)
1482 if (current->flags & PF_EXITING)
1485 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1486 !fatal_signal_pending(current))
1487 uprobe_warn(current, "dup xol area");
1491 * Called in context of a new clone/fork from copy_process.
1493 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1495 struct uprobe_task *utask = current->utask;
1496 struct mm_struct *mm = current->mm;
1497 struct xol_area *area;
1501 if (!utask || !utask->return_instances)
1504 if (mm == t->mm && !(flags & CLONE_VFORK))
1507 if (dup_utask(t, utask))
1508 return uprobe_warn(t, "dup ret instances");
1510 /* The task can fork() after dup_xol_work() fails */
1511 area = mm->uprobes_state.xol_area;
1513 return uprobe_warn(t, "dup xol area");
1518 t->utask->dup_xol_addr = area->vaddr;
1519 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1520 task_work_add(t, &t->utask->dup_xol_work, true);
1524 * Current area->vaddr notion assume the trampoline address is always
1525 * equal area->vaddr.
1527 * Returns -1 in case the xol_area is not allocated.
1529 static unsigned long get_trampoline_vaddr(void)
1531 struct xol_area *area;
1532 unsigned long trampoline_vaddr = -1;
1534 area = current->mm->uprobes_state.xol_area;
1535 smp_read_barrier_depends();
1537 trampoline_vaddr = area->vaddr;
1539 return trampoline_vaddr;
1542 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1543 struct pt_regs *regs)
1545 struct return_instance *ri = utask->return_instances;
1546 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1548 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1549 ri = free_ret_instance(ri);
1552 utask->return_instances = ri;
1555 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1557 struct return_instance *ri;
1558 struct uprobe_task *utask;
1559 unsigned long orig_ret_vaddr, trampoline_vaddr;
1562 if (!get_xol_area())
1565 utask = get_utask();
1569 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1570 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1571 " nestedness limit pid/tgid=%d/%d\n",
1572 current->pid, current->tgid);
1576 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1580 trampoline_vaddr = get_trampoline_vaddr();
1581 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1582 if (orig_ret_vaddr == -1)
1585 /* drop the entries invalidated by longjmp() */
1586 chained = (orig_ret_vaddr == trampoline_vaddr);
1587 cleanup_return_instances(utask, chained, regs);
1590 * We don't want to keep trampoline address in stack, rather keep the
1591 * original return address of first caller thru all the consequent
1592 * instances. This also makes breakpoint unwrapping easier.
1595 if (!utask->return_instances) {
1597 * This situation is not possible. Likely we have an
1598 * attack from user-space.
1600 uprobe_warn(current, "handle tail call");
1603 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1606 ri->uprobe = get_uprobe(uprobe);
1607 ri->func = instruction_pointer(regs);
1608 ri->stack = user_stack_pointer(regs);
1609 ri->orig_ret_vaddr = orig_ret_vaddr;
1610 ri->chained = chained;
1613 ri->next = utask->return_instances;
1614 utask->return_instances = ri;
1621 /* Prepare to single-step probed instruction out of line. */
1623 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1625 struct uprobe_task *utask;
1626 unsigned long xol_vaddr;
1629 utask = get_utask();
1633 xol_vaddr = xol_get_insn_slot(uprobe);
1637 utask->xol_vaddr = xol_vaddr;
1638 utask->vaddr = bp_vaddr;
1640 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1641 if (unlikely(err)) {
1642 xol_free_insn_slot(current);
1646 utask->active_uprobe = uprobe;
1647 utask->state = UTASK_SSTEP;
1652 * If we are singlestepping, then ensure this thread is not connected to
1653 * non-fatal signals until completion of singlestep. When xol insn itself
1654 * triggers the signal, restart the original insn even if the task is
1655 * already SIGKILL'ed (since coredump should report the correct ip). This
1656 * is even more important if the task has a handler for SIGSEGV/etc, The
1657 * _same_ instruction should be repeated again after return from the signal
1658 * handler, and SSTEP can never finish in this case.
1660 bool uprobe_deny_signal(void)
1662 struct task_struct *t = current;
1663 struct uprobe_task *utask = t->utask;
1665 if (likely(!utask || !utask->active_uprobe))
1668 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1670 if (signal_pending(t)) {
1671 spin_lock_irq(&t->sighand->siglock);
1672 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1673 spin_unlock_irq(&t->sighand->siglock);
1675 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1676 utask->state = UTASK_SSTEP_TRAPPED;
1677 set_tsk_thread_flag(t, TIF_UPROBE);
1684 static void mmf_recalc_uprobes(struct mm_struct *mm)
1686 struct vm_area_struct *vma;
1688 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1689 if (!valid_vma(vma, false))
1692 * This is not strictly accurate, we can race with
1693 * uprobe_unregister() and see the already removed
1694 * uprobe if delete_uprobe() was not yet called.
1695 * Or this uprobe can be filtered out.
1697 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1701 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1704 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1707 uprobe_opcode_t opcode;
1710 if (WARN_ON_ONCE(!IS_ALIGNED(vaddr, UPROBE_SWBP_INSN_SIZE)))
1713 pagefault_disable();
1714 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
1717 if (likely(result == 0))
1721 * The NULL 'tsk' here ensures that any faults that occur here
1722 * will not be accounted to the task. 'mm' *is* current->mm,
1723 * but we treat this as a 'remote' access since it is
1724 * essentially a kernel access to the memory.
1726 result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
1731 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1734 /* This needs to return true for any variant of the trap insn */
1735 return is_trap_insn(&opcode);
1738 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1740 struct mm_struct *mm = current->mm;
1741 struct uprobe *uprobe = NULL;
1742 struct vm_area_struct *vma;
1744 down_read(&mm->mmap_sem);
1745 vma = find_vma(mm, bp_vaddr);
1746 if (vma && vma->vm_start <= bp_vaddr) {
1747 if (valid_vma(vma, false)) {
1748 struct inode *inode = file_inode(vma->vm_file);
1749 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1751 uprobe = find_uprobe(inode, offset);
1755 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1760 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1761 mmf_recalc_uprobes(mm);
1762 up_read(&mm->mmap_sem);
1767 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1769 struct uprobe_consumer *uc;
1770 int remove = UPROBE_HANDLER_REMOVE;
1771 bool need_prep = false; /* prepare return uprobe, when needed */
1773 down_read(&uprobe->register_rwsem);
1774 for (uc = uprobe->consumers; uc; uc = uc->next) {
1778 rc = uc->handler(uc, regs);
1779 WARN(rc & ~UPROBE_HANDLER_MASK,
1780 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1783 if (uc->ret_handler)
1789 if (need_prep && !remove)
1790 prepare_uretprobe(uprobe, regs); /* put bp at return */
1792 if (remove && uprobe->consumers) {
1793 WARN_ON(!uprobe_is_active(uprobe));
1794 unapply_uprobe(uprobe, current->mm);
1796 up_read(&uprobe->register_rwsem);
1800 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1802 struct uprobe *uprobe = ri->uprobe;
1803 struct uprobe_consumer *uc;
1805 down_read(&uprobe->register_rwsem);
1806 for (uc = uprobe->consumers; uc; uc = uc->next) {
1807 if (uc->ret_handler)
1808 uc->ret_handler(uc, ri->func, regs);
1810 up_read(&uprobe->register_rwsem);
1813 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
1818 chained = ri->chained;
1819 ri = ri->next; /* can't be NULL if chained */
1825 static void handle_trampoline(struct pt_regs *regs)
1827 struct uprobe_task *utask;
1828 struct return_instance *ri, *next;
1831 utask = current->utask;
1835 ri = utask->return_instances;
1841 * We should throw out the frames invalidated by longjmp().
1842 * If this chain is valid, then the next one should be alive
1843 * or NULL; the latter case means that nobody but ri->func
1844 * could hit this trampoline on return. TODO: sigaltstack().
1846 next = find_next_ret_chain(ri);
1847 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
1849 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1852 handle_uretprobe_chain(ri, regs);
1853 ri = free_ret_instance(ri);
1855 } while (ri != next);
1858 utask->return_instances = ri;
1862 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
1863 force_sig(SIGILL, current);
1867 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
1872 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
1873 struct pt_regs *regs)
1879 * Run handler and ask thread to singlestep.
1880 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1882 static void handle_swbp(struct pt_regs *regs)
1884 struct uprobe *uprobe;
1885 unsigned long bp_vaddr;
1888 bp_vaddr = uprobe_get_swbp_addr(regs);
1889 if (bp_vaddr == get_trampoline_vaddr())
1890 return handle_trampoline(regs);
1892 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1895 /* No matching uprobe; signal SIGTRAP. */
1896 force_sig(SIGTRAP, current);
1899 * Either we raced with uprobe_unregister() or we can't
1900 * access this memory. The latter is only possible if
1901 * another thread plays with our ->mm. In both cases
1902 * we can simply restart. If this vma was unmapped we
1903 * can pretend this insn was not executed yet and get
1904 * the (correct) SIGSEGV after restart.
1906 instruction_pointer_set(regs, bp_vaddr);
1911 /* change it in advance for ->handler() and restart */
1912 instruction_pointer_set(regs, bp_vaddr);
1915 * TODO: move copy_insn/etc into _register and remove this hack.
1916 * After we hit the bp, _unregister + _register can install the
1917 * new and not-yet-analyzed uprobe at the same address, restart.
1919 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1923 * Pairs with the smp_wmb() in prepare_uprobe().
1925 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
1926 * we must also see the stores to &uprobe->arch performed by the
1927 * prepare_uprobe() call.
1931 /* Tracing handlers use ->utask to communicate with fetch methods */
1935 if (arch_uprobe_ignore(&uprobe->arch, regs))
1938 handler_chain(uprobe, regs);
1940 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1943 if (!pre_ssout(uprobe, regs, bp_vaddr))
1946 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1952 * Perform required fix-ups and disable singlestep.
1953 * Allow pending signals to take effect.
1955 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1957 struct uprobe *uprobe;
1960 uprobe = utask->active_uprobe;
1961 if (utask->state == UTASK_SSTEP_ACK)
1962 err = arch_uprobe_post_xol(&uprobe->arch, regs);
1963 else if (utask->state == UTASK_SSTEP_TRAPPED)
1964 arch_uprobe_abort_xol(&uprobe->arch, regs);
1969 utask->active_uprobe = NULL;
1970 utask->state = UTASK_RUNNING;
1971 xol_free_insn_slot(current);
1973 spin_lock_irq(¤t->sighand->siglock);
1974 recalc_sigpending(); /* see uprobe_deny_signal() */
1975 spin_unlock_irq(¤t->sighand->siglock);
1977 if (unlikely(err)) {
1978 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
1979 force_sig(SIGILL, current);
1984 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1985 * allows the thread to return from interrupt. After that handle_swbp()
1986 * sets utask->active_uprobe.
1988 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1989 * and allows the thread to return from interrupt.
1991 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1992 * uprobe_notify_resume().
1994 void uprobe_notify_resume(struct pt_regs *regs)
1996 struct uprobe_task *utask;
1998 clear_thread_flag(TIF_UPROBE);
2000 utask = current->utask;
2001 if (utask && utask->active_uprobe)
2002 handle_singlestep(utask, regs);
2008 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2009 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2011 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2016 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2017 (!current->utask || !current->utask->return_instances))
2020 set_thread_flag(TIF_UPROBE);
2025 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2026 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2028 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2030 struct uprobe_task *utask = current->utask;
2032 if (!current->mm || !utask || !utask->active_uprobe)
2033 /* task is currently not uprobed */
2036 utask->state = UTASK_SSTEP_ACK;
2037 set_thread_flag(TIF_UPROBE);
2041 static struct notifier_block uprobe_exception_nb = {
2042 .notifier_call = arch_uprobe_exception_notify,
2043 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2046 static int __init init_uprobes(void)
2050 for (i = 0; i < UPROBES_HASH_SZ; i++)
2051 mutex_init(&uprobes_mmap_mutex[i]);
2053 if (percpu_init_rwsem(&dup_mmap_sem))
2056 return register_die_notifier(&uprobe_exception_nb);
2058 __initcall(init_uprobes);