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 arch_uprobe *auprobe, struct mm_struct *mm,
303 unsigned long vaddr, 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(auprobe, 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(auprobe, mm, vaddr,
370 *(uprobe_opcode_t *)&auprobe->insn);
373 static struct uprobe *get_uprobe(struct uprobe *uprobe)
375 atomic_inc(&uprobe->ref);
379 static void put_uprobe(struct uprobe *uprobe)
381 if (atomic_dec_and_test(&uprobe->ref))
385 static int match_uprobe(struct uprobe *l, struct uprobe *r)
387 if (l->inode < r->inode)
390 if (l->inode > r->inode)
393 if (l->offset < r->offset)
396 if (l->offset > r->offset)
402 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
404 struct uprobe u = { .inode = inode, .offset = offset };
405 struct rb_node *n = uprobes_tree.rb_node;
406 struct uprobe *uprobe;
410 uprobe = rb_entry(n, struct uprobe, rb_node);
411 match = match_uprobe(&u, uprobe);
413 return get_uprobe(uprobe);
424 * Find a uprobe corresponding to a given inode:offset
425 * Acquires uprobes_treelock
427 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
429 struct uprobe *uprobe;
431 spin_lock(&uprobes_treelock);
432 uprobe = __find_uprobe(inode, offset);
433 spin_unlock(&uprobes_treelock);
438 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
440 struct rb_node **p = &uprobes_tree.rb_node;
441 struct rb_node *parent = NULL;
447 u = rb_entry(parent, struct uprobe, rb_node);
448 match = match_uprobe(uprobe, u);
450 return get_uprobe(u);
453 p = &parent->rb_left;
455 p = &parent->rb_right;
460 rb_link_node(&uprobe->rb_node, parent, p);
461 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
462 /* get access + creation ref */
463 atomic_set(&uprobe->ref, 2);
469 * Acquire uprobes_treelock.
470 * Matching uprobe already exists in rbtree;
471 * increment (access refcount) and return the matching uprobe.
473 * No matching uprobe; insert the uprobe in rb_tree;
474 * get a double refcount (access + creation) and return NULL.
476 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
480 spin_lock(&uprobes_treelock);
481 u = __insert_uprobe(uprobe);
482 spin_unlock(&uprobes_treelock);
487 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
489 struct uprobe *uprobe, *cur_uprobe;
491 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
495 uprobe->inode = inode;
496 uprobe->offset = offset;
497 init_rwsem(&uprobe->register_rwsem);
498 init_rwsem(&uprobe->consumer_rwsem);
500 /* add to uprobes_tree, sorted on inode:offset */
501 cur_uprobe = insert_uprobe(uprobe);
502 /* 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() */
704 struct map_info *next;
705 struct mm_struct *mm;
709 static inline struct map_info *free_map_info(struct map_info *info)
711 struct map_info *next = info->next;
716 static struct map_info *
717 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
719 unsigned long pgoff = offset >> PAGE_SHIFT;
720 struct vm_area_struct *vma;
721 struct map_info *curr = NULL;
722 struct map_info *prev = NULL;
723 struct map_info *info;
727 i_mmap_lock_read(mapping);
728 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
729 if (!valid_vma(vma, is_register))
732 if (!prev && !more) {
734 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
735 * reclaim. This is optimistic, no harm done if it fails.
737 prev = kmalloc(sizeof(struct map_info),
738 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
747 if (!mmget_not_zero(vma->vm_mm))
755 info->mm = vma->vm_mm;
756 info->vaddr = offset_to_vaddr(vma, offset);
758 i_mmap_unlock_read(mapping);
770 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
772 curr = ERR_PTR(-ENOMEM);
782 prev = free_map_info(prev);
787 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
789 bool is_register = !!new;
790 struct map_info *info;
793 percpu_down_write(&dup_mmap_sem);
794 info = build_map_info(uprobe->inode->i_mapping,
795 uprobe->offset, is_register);
802 struct mm_struct *mm = info->mm;
803 struct vm_area_struct *vma;
805 if (err && is_register)
808 down_write(&mm->mmap_sem);
809 vma = find_vma(mm, info->vaddr);
810 if (!vma || !valid_vma(vma, is_register) ||
811 file_inode(vma->vm_file) != uprobe->inode)
814 if (vma->vm_start > info->vaddr ||
815 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
819 /* consult only the "caller", new consumer. */
820 if (consumer_filter(new,
821 UPROBE_FILTER_REGISTER, mm))
822 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
823 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
824 if (!filter_chain(uprobe,
825 UPROBE_FILTER_UNREGISTER, mm))
826 err |= remove_breakpoint(uprobe, mm, info->vaddr);
830 up_write(&mm->mmap_sem);
833 info = free_map_info(info);
836 percpu_up_write(&dup_mmap_sem);
841 __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
845 if (WARN_ON(!consumer_del(uprobe, uc)))
848 err = register_for_each_vma(uprobe, NULL);
849 /* TODO : cant unregister? schedule a worker thread */
850 if (!uprobe->consumers && !err)
851 delete_uprobe(uprobe);
855 * uprobe_unregister - unregister an already registered probe.
856 * @inode: the file in which the probe has to be removed.
857 * @offset: offset from the start of the file.
858 * @uc: identify which probe if multiple probes are colocated.
860 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
862 struct uprobe *uprobe;
864 uprobe = find_uprobe(inode, offset);
865 if (WARN_ON(!uprobe))
868 down_write(&uprobe->register_rwsem);
869 __uprobe_unregister(uprobe, uc);
870 up_write(&uprobe->register_rwsem);
873 EXPORT_SYMBOL_GPL(uprobe_unregister);
876 * __uprobe_register - register a probe
877 * @inode: the file in which the probe has to be placed.
878 * @offset: offset from the start of the file.
879 * @uc: information on howto handle the probe..
881 * Apart from the access refcount, __uprobe_register() takes a creation
882 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
883 * inserted into the rbtree (i.e first consumer for a @inode:@offset
884 * tuple). Creation refcount stops uprobe_unregister from freeing the
885 * @uprobe even before the register operation is complete. Creation
886 * refcount is released when the last @uc for the @uprobe
887 * unregisters. Caller of __uprobe_register() is required to keep @inode
888 * (and the containing mount) referenced.
890 * Return errno if it cannot successully install probes
891 * else return 0 (success)
893 static int __uprobe_register(struct inode *inode, loff_t offset,
894 struct uprobe_consumer *uc)
896 struct uprobe *uprobe;
899 /* Uprobe must have at least one set consumer */
900 if (!uc->handler && !uc->ret_handler)
903 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
904 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
906 /* Racy, just to catch the obvious mistakes */
907 if (offset > i_size_read(inode))
911 * This ensures that copy_from_page() and copy_to_page()
912 * can't cross page boundary.
914 if (!IS_ALIGNED(offset, UPROBE_SWBP_INSN_SIZE))
918 uprobe = alloc_uprobe(inode, offset);
922 * We can race with uprobe_unregister()->delete_uprobe().
923 * Check uprobe_is_active() and retry if it is false.
925 down_write(&uprobe->register_rwsem);
927 if (likely(uprobe_is_active(uprobe))) {
928 consumer_add(uprobe, uc);
929 ret = register_for_each_vma(uprobe, uc);
931 __uprobe_unregister(uprobe, uc);
933 up_write(&uprobe->register_rwsem);
936 if (unlikely(ret == -EAGAIN))
941 int uprobe_register(struct inode *inode, loff_t offset,
942 struct uprobe_consumer *uc)
944 return __uprobe_register(inode, offset, uc);
946 EXPORT_SYMBOL_GPL(uprobe_register);
949 * uprobe_apply - unregister an already registered probe.
950 * @inode: the file in which the probe has to be removed.
951 * @offset: offset from the start of the file.
952 * @uc: consumer which wants to add more or remove some breakpoints
953 * @add: add or remove the breakpoints
955 int uprobe_apply(struct inode *inode, loff_t offset,
956 struct uprobe_consumer *uc, bool add)
958 struct uprobe *uprobe;
959 struct uprobe_consumer *con;
962 uprobe = find_uprobe(inode, offset);
963 if (WARN_ON(!uprobe))
966 down_write(&uprobe->register_rwsem);
967 for (con = uprobe->consumers; con && con != uc ; con = con->next)
970 ret = register_for_each_vma(uprobe, add ? uc : NULL);
971 up_write(&uprobe->register_rwsem);
977 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
979 struct vm_area_struct *vma;
982 down_read(&mm->mmap_sem);
983 for (vma = mm->mmap; vma; vma = vma->vm_next) {
987 if (!valid_vma(vma, false) ||
988 file_inode(vma->vm_file) != uprobe->inode)
991 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
992 if (uprobe->offset < offset ||
993 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
996 vaddr = offset_to_vaddr(vma, uprobe->offset);
997 err |= remove_breakpoint(uprobe, mm, vaddr);
999 up_read(&mm->mmap_sem);
1004 static struct rb_node *
1005 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1007 struct rb_node *n = uprobes_tree.rb_node;
1010 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1012 if (inode < u->inode) {
1014 } else if (inode > u->inode) {
1017 if (max < u->offset)
1019 else if (min > u->offset)
1030 * For a given range in vma, build a list of probes that need to be inserted.
1032 static void build_probe_list(struct inode *inode,
1033 struct vm_area_struct *vma,
1034 unsigned long start, unsigned long end,
1035 struct list_head *head)
1038 struct rb_node *n, *t;
1041 INIT_LIST_HEAD(head);
1042 min = vaddr_to_offset(vma, start);
1043 max = min + (end - start) - 1;
1045 spin_lock(&uprobes_treelock);
1046 n = find_node_in_range(inode, min, max);
1048 for (t = n; t; t = rb_prev(t)) {
1049 u = rb_entry(t, struct uprobe, rb_node);
1050 if (u->inode != inode || u->offset < min)
1052 list_add(&u->pending_list, head);
1055 for (t = n; (t = rb_next(t)); ) {
1056 u = rb_entry(t, struct uprobe, rb_node);
1057 if (u->inode != inode || u->offset > max)
1059 list_add(&u->pending_list, head);
1063 spin_unlock(&uprobes_treelock);
1067 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1069 * Currently we ignore all errors and always return 0, the callers
1070 * can't handle the failure anyway.
1072 int uprobe_mmap(struct vm_area_struct *vma)
1074 struct list_head tmp_list;
1075 struct uprobe *uprobe, *u;
1076 struct inode *inode;
1078 if (no_uprobe_events() || !valid_vma(vma, true))
1081 inode = file_inode(vma->vm_file);
1085 mutex_lock(uprobes_mmap_hash(inode));
1086 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1088 * We can race with uprobe_unregister(), this uprobe can be already
1089 * removed. But in this case filter_chain() must return false, all
1090 * consumers have gone away.
1092 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1093 if (!fatal_signal_pending(current) &&
1094 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1095 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1096 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1100 mutex_unlock(uprobes_mmap_hash(inode));
1106 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1109 struct inode *inode;
1112 inode = file_inode(vma->vm_file);
1114 min = vaddr_to_offset(vma, start);
1115 max = min + (end - start) - 1;
1117 spin_lock(&uprobes_treelock);
1118 n = find_node_in_range(inode, min, max);
1119 spin_unlock(&uprobes_treelock);
1125 * Called in context of a munmap of a vma.
1127 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1129 if (no_uprobe_events() || !valid_vma(vma, false))
1132 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1135 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1136 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1139 if (vma_has_uprobes(vma, start, end))
1140 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1143 /* Slot allocation for XOL */
1144 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1146 struct vm_area_struct *vma;
1149 if (down_write_killable(&mm->mmap_sem))
1152 if (mm->uprobes_state.xol_area) {
1158 /* Try to map as high as possible, this is only a hint. */
1159 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1161 if (area->vaddr & ~PAGE_MASK) {
1167 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1168 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1169 &area->xol_mapping);
1176 /* pairs with get_xol_area() */
1177 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1179 up_write(&mm->mmap_sem);
1184 static struct xol_area *__create_xol_area(unsigned long vaddr)
1186 struct mm_struct *mm = current->mm;
1187 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1188 struct xol_area *area;
1190 area = kmalloc(sizeof(*area), GFP_KERNEL);
1191 if (unlikely(!area))
1194 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1199 area->xol_mapping.name = "[uprobes]";
1200 area->xol_mapping.fault = NULL;
1201 area->xol_mapping.pages = area->pages;
1202 area->pages[0] = alloc_page(GFP_HIGHUSER);
1203 if (!area->pages[0])
1205 area->pages[1] = NULL;
1207 area->vaddr = vaddr;
1208 init_waitqueue_head(&area->wq);
1209 /* Reserve the 1st slot for get_trampoline_vaddr() */
1210 set_bit(0, area->bitmap);
1211 atomic_set(&area->slot_count, 1);
1212 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1214 if (!xol_add_vma(mm, area))
1217 __free_page(area->pages[0]);
1219 kfree(area->bitmap);
1227 * get_xol_area - Allocate process's xol_area if necessary.
1228 * This area will be used for storing instructions for execution out of line.
1230 * Returns the allocated area or NULL.
1232 static struct xol_area *get_xol_area(void)
1234 struct mm_struct *mm = current->mm;
1235 struct xol_area *area;
1237 if (!mm->uprobes_state.xol_area)
1238 __create_xol_area(0);
1240 /* Pairs with xol_add_vma() smp_store_release() */
1241 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1246 * uprobe_clear_state - Free the area allocated for slots.
1248 void uprobe_clear_state(struct mm_struct *mm)
1250 struct xol_area *area = mm->uprobes_state.xol_area;
1255 put_page(area->pages[0]);
1256 kfree(area->bitmap);
1260 void uprobe_start_dup_mmap(void)
1262 percpu_down_read(&dup_mmap_sem);
1265 void uprobe_end_dup_mmap(void)
1267 percpu_up_read(&dup_mmap_sem);
1270 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1272 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1273 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1274 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1275 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1280 * - search for a free slot.
1282 static unsigned long xol_take_insn_slot(struct xol_area *area)
1284 unsigned long slot_addr;
1288 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1289 if (slot_nr < UINSNS_PER_PAGE) {
1290 if (!test_and_set_bit(slot_nr, area->bitmap))
1293 slot_nr = UINSNS_PER_PAGE;
1296 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1297 } while (slot_nr >= UINSNS_PER_PAGE);
1299 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1300 atomic_inc(&area->slot_count);
1306 * xol_get_insn_slot - allocate a slot for xol.
1307 * Returns the allocated slot address or 0.
1309 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1311 struct xol_area *area;
1312 unsigned long xol_vaddr;
1314 area = get_xol_area();
1318 xol_vaddr = xol_take_insn_slot(area);
1319 if (unlikely(!xol_vaddr))
1322 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1323 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1329 * xol_free_insn_slot - If slot was earlier allocated by
1330 * @xol_get_insn_slot(), make the slot available for
1331 * subsequent requests.
1333 static void xol_free_insn_slot(struct task_struct *tsk)
1335 struct xol_area *area;
1336 unsigned long vma_end;
1337 unsigned long slot_addr;
1339 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1342 slot_addr = tsk->utask->xol_vaddr;
1343 if (unlikely(!slot_addr))
1346 area = tsk->mm->uprobes_state.xol_area;
1347 vma_end = area->vaddr + PAGE_SIZE;
1348 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1349 unsigned long offset;
1352 offset = slot_addr - area->vaddr;
1353 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1354 if (slot_nr >= UINSNS_PER_PAGE)
1357 clear_bit(slot_nr, area->bitmap);
1358 atomic_dec(&area->slot_count);
1359 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1360 if (waitqueue_active(&area->wq))
1363 tsk->utask->xol_vaddr = 0;
1367 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1368 void *src, unsigned long len)
1370 /* Initialize the slot */
1371 copy_to_page(page, vaddr, src, len);
1374 * We probably need flush_icache_user_range() but it needs vma.
1375 * This should work on most of architectures by default. If
1376 * architecture needs to do something different it can define
1377 * its own version of the function.
1379 flush_dcache_page(page);
1383 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1384 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1386 * Return the address of the breakpoint instruction.
1388 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1390 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1393 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1395 struct uprobe_task *utask = current->utask;
1397 if (unlikely(utask && utask->active_uprobe))
1398 return utask->vaddr;
1400 return instruction_pointer(regs);
1403 static struct return_instance *free_ret_instance(struct return_instance *ri)
1405 struct return_instance *next = ri->next;
1406 put_uprobe(ri->uprobe);
1412 * Called with no locks held.
1413 * Called in context of an exiting or an exec-ing thread.
1415 void uprobe_free_utask(struct task_struct *t)
1417 struct uprobe_task *utask = t->utask;
1418 struct return_instance *ri;
1423 if (utask->active_uprobe)
1424 put_uprobe(utask->active_uprobe);
1426 ri = utask->return_instances;
1428 ri = free_ret_instance(ri);
1430 xol_free_insn_slot(t);
1436 * Allocate a uprobe_task object for the task if if necessary.
1437 * Called when the thread hits a breakpoint.
1440 * - pointer to new uprobe_task on success
1443 static struct uprobe_task *get_utask(void)
1445 if (!current->utask)
1446 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1447 return current->utask;
1450 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1452 struct uprobe_task *n_utask;
1453 struct return_instance **p, *o, *n;
1455 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1460 p = &n_utask->return_instances;
1461 for (o = o_utask->return_instances; o; o = o->next) {
1462 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1467 get_uprobe(n->uprobe);
1478 static void uprobe_warn(struct task_struct *t, const char *msg)
1480 pr_warn("uprobe: %s:%d failed to %s\n",
1481 current->comm, current->pid, msg);
1484 static void dup_xol_work(struct callback_head *work)
1486 if (current->flags & PF_EXITING)
1489 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1490 !fatal_signal_pending(current))
1491 uprobe_warn(current, "dup xol area");
1495 * Called in context of a new clone/fork from copy_process.
1497 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1499 struct uprobe_task *utask = current->utask;
1500 struct mm_struct *mm = current->mm;
1501 struct xol_area *area;
1505 if (!utask || !utask->return_instances)
1508 if (mm == t->mm && !(flags & CLONE_VFORK))
1511 if (dup_utask(t, utask))
1512 return uprobe_warn(t, "dup ret instances");
1514 /* The task can fork() after dup_xol_work() fails */
1515 area = mm->uprobes_state.xol_area;
1517 return uprobe_warn(t, "dup xol area");
1522 t->utask->dup_xol_addr = area->vaddr;
1523 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1524 task_work_add(t, &t->utask->dup_xol_work, true);
1528 * Current area->vaddr notion assume the trampoline address is always
1529 * equal area->vaddr.
1531 * Returns -1 in case the xol_area is not allocated.
1533 static unsigned long get_trampoline_vaddr(void)
1535 struct xol_area *area;
1536 unsigned long trampoline_vaddr = -1;
1538 /* Pairs with xol_add_vma() smp_store_release() */
1539 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1541 trampoline_vaddr = area->vaddr;
1543 return trampoline_vaddr;
1546 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1547 struct pt_regs *regs)
1549 struct return_instance *ri = utask->return_instances;
1550 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1552 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1553 ri = free_ret_instance(ri);
1556 utask->return_instances = ri;
1559 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1561 struct return_instance *ri;
1562 struct uprobe_task *utask;
1563 unsigned long orig_ret_vaddr, trampoline_vaddr;
1566 if (!get_xol_area())
1569 utask = get_utask();
1573 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1574 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1575 " nestedness limit pid/tgid=%d/%d\n",
1576 current->pid, current->tgid);
1580 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1584 trampoline_vaddr = get_trampoline_vaddr();
1585 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1586 if (orig_ret_vaddr == -1)
1589 /* drop the entries invalidated by longjmp() */
1590 chained = (orig_ret_vaddr == trampoline_vaddr);
1591 cleanup_return_instances(utask, chained, regs);
1594 * We don't want to keep trampoline address in stack, rather keep the
1595 * original return address of first caller thru all the consequent
1596 * instances. This also makes breakpoint unwrapping easier.
1599 if (!utask->return_instances) {
1601 * This situation is not possible. Likely we have an
1602 * attack from user-space.
1604 uprobe_warn(current, "handle tail call");
1607 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1610 ri->uprobe = get_uprobe(uprobe);
1611 ri->func = instruction_pointer(regs);
1612 ri->stack = user_stack_pointer(regs);
1613 ri->orig_ret_vaddr = orig_ret_vaddr;
1614 ri->chained = chained;
1617 ri->next = utask->return_instances;
1618 utask->return_instances = ri;
1625 /* Prepare to single-step probed instruction out of line. */
1627 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1629 struct uprobe_task *utask;
1630 unsigned long xol_vaddr;
1633 utask = get_utask();
1637 xol_vaddr = xol_get_insn_slot(uprobe);
1641 utask->xol_vaddr = xol_vaddr;
1642 utask->vaddr = bp_vaddr;
1644 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1645 if (unlikely(err)) {
1646 xol_free_insn_slot(current);
1650 utask->active_uprobe = uprobe;
1651 utask->state = UTASK_SSTEP;
1656 * If we are singlestepping, then ensure this thread is not connected to
1657 * non-fatal signals until completion of singlestep. When xol insn itself
1658 * triggers the signal, restart the original insn even if the task is
1659 * already SIGKILL'ed (since coredump should report the correct ip). This
1660 * is even more important if the task has a handler for SIGSEGV/etc, The
1661 * _same_ instruction should be repeated again after return from the signal
1662 * handler, and SSTEP can never finish in this case.
1664 bool uprobe_deny_signal(void)
1666 struct task_struct *t = current;
1667 struct uprobe_task *utask = t->utask;
1669 if (likely(!utask || !utask->active_uprobe))
1672 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1674 if (signal_pending(t)) {
1675 spin_lock_irq(&t->sighand->siglock);
1676 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1677 spin_unlock_irq(&t->sighand->siglock);
1679 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1680 utask->state = UTASK_SSTEP_TRAPPED;
1681 set_tsk_thread_flag(t, TIF_UPROBE);
1688 static void mmf_recalc_uprobes(struct mm_struct *mm)
1690 struct vm_area_struct *vma;
1692 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1693 if (!valid_vma(vma, false))
1696 * This is not strictly accurate, we can race with
1697 * uprobe_unregister() and see the already removed
1698 * uprobe if delete_uprobe() was not yet called.
1699 * Or this uprobe can be filtered out.
1701 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1705 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1708 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1711 uprobe_opcode_t opcode;
1714 if (WARN_ON_ONCE(!IS_ALIGNED(vaddr, UPROBE_SWBP_INSN_SIZE)))
1717 pagefault_disable();
1718 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
1721 if (likely(result == 0))
1725 * The NULL 'tsk' here ensures that any faults that occur here
1726 * will not be accounted to the task. 'mm' *is* current->mm,
1727 * but we treat this as a 'remote' access since it is
1728 * essentially a kernel access to the memory.
1730 result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
1735 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1738 /* This needs to return true for any variant of the trap insn */
1739 return is_trap_insn(&opcode);
1742 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1744 struct mm_struct *mm = current->mm;
1745 struct uprobe *uprobe = NULL;
1746 struct vm_area_struct *vma;
1748 down_read(&mm->mmap_sem);
1749 vma = find_vma(mm, bp_vaddr);
1750 if (vma && vma->vm_start <= bp_vaddr) {
1751 if (valid_vma(vma, false)) {
1752 struct inode *inode = file_inode(vma->vm_file);
1753 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1755 uprobe = find_uprobe(inode, offset);
1759 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1764 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1765 mmf_recalc_uprobes(mm);
1766 up_read(&mm->mmap_sem);
1771 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1773 struct uprobe_consumer *uc;
1774 int remove = UPROBE_HANDLER_REMOVE;
1775 bool need_prep = false; /* prepare return uprobe, when needed */
1777 down_read(&uprobe->register_rwsem);
1778 for (uc = uprobe->consumers; uc; uc = uc->next) {
1782 rc = uc->handler(uc, regs);
1783 WARN(rc & ~UPROBE_HANDLER_MASK,
1784 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1787 if (uc->ret_handler)
1793 if (need_prep && !remove)
1794 prepare_uretprobe(uprobe, regs); /* put bp at return */
1796 if (remove && uprobe->consumers) {
1797 WARN_ON(!uprobe_is_active(uprobe));
1798 unapply_uprobe(uprobe, current->mm);
1800 up_read(&uprobe->register_rwsem);
1804 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1806 struct uprobe *uprobe = ri->uprobe;
1807 struct uprobe_consumer *uc;
1809 down_read(&uprobe->register_rwsem);
1810 for (uc = uprobe->consumers; uc; uc = uc->next) {
1811 if (uc->ret_handler)
1812 uc->ret_handler(uc, ri->func, regs);
1814 up_read(&uprobe->register_rwsem);
1817 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
1822 chained = ri->chained;
1823 ri = ri->next; /* can't be NULL if chained */
1829 static void handle_trampoline(struct pt_regs *regs)
1831 struct uprobe_task *utask;
1832 struct return_instance *ri, *next;
1835 utask = current->utask;
1839 ri = utask->return_instances;
1845 * We should throw out the frames invalidated by longjmp().
1846 * If this chain is valid, then the next one should be alive
1847 * or NULL; the latter case means that nobody but ri->func
1848 * could hit this trampoline on return. TODO: sigaltstack().
1850 next = find_next_ret_chain(ri);
1851 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
1853 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1856 handle_uretprobe_chain(ri, regs);
1857 ri = free_ret_instance(ri);
1859 } while (ri != next);
1862 utask->return_instances = ri;
1866 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
1867 force_sig(SIGILL, current);
1871 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
1876 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
1877 struct pt_regs *regs)
1883 * Run handler and ask thread to singlestep.
1884 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1886 static void handle_swbp(struct pt_regs *regs)
1888 struct uprobe *uprobe;
1889 unsigned long bp_vaddr;
1890 int uninitialized_var(is_swbp);
1892 bp_vaddr = uprobe_get_swbp_addr(regs);
1893 if (bp_vaddr == get_trampoline_vaddr())
1894 return handle_trampoline(regs);
1896 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1899 /* No matching uprobe; signal SIGTRAP. */
1900 force_sig(SIGTRAP, current);
1903 * Either we raced with uprobe_unregister() or we can't
1904 * access this memory. The latter is only possible if
1905 * another thread plays with our ->mm. In both cases
1906 * we can simply restart. If this vma was unmapped we
1907 * can pretend this insn was not executed yet and get
1908 * the (correct) SIGSEGV after restart.
1910 instruction_pointer_set(regs, bp_vaddr);
1915 /* change it in advance for ->handler() and restart */
1916 instruction_pointer_set(regs, bp_vaddr);
1919 * TODO: move copy_insn/etc into _register and remove this hack.
1920 * After we hit the bp, _unregister + _register can install the
1921 * new and not-yet-analyzed uprobe at the same address, restart.
1923 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1927 * Pairs with the smp_wmb() in prepare_uprobe().
1929 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
1930 * we must also see the stores to &uprobe->arch performed by the
1931 * prepare_uprobe() call.
1935 /* Tracing handlers use ->utask to communicate with fetch methods */
1939 if (arch_uprobe_ignore(&uprobe->arch, regs))
1942 handler_chain(uprobe, regs);
1944 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1947 if (!pre_ssout(uprobe, regs, bp_vaddr))
1950 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1956 * Perform required fix-ups and disable singlestep.
1957 * Allow pending signals to take effect.
1959 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1961 struct uprobe *uprobe;
1964 uprobe = utask->active_uprobe;
1965 if (utask->state == UTASK_SSTEP_ACK)
1966 err = arch_uprobe_post_xol(&uprobe->arch, regs);
1967 else if (utask->state == UTASK_SSTEP_TRAPPED)
1968 arch_uprobe_abort_xol(&uprobe->arch, regs);
1973 utask->active_uprobe = NULL;
1974 utask->state = UTASK_RUNNING;
1975 xol_free_insn_slot(current);
1977 spin_lock_irq(¤t->sighand->siglock);
1978 recalc_sigpending(); /* see uprobe_deny_signal() */
1979 spin_unlock_irq(¤t->sighand->siglock);
1981 if (unlikely(err)) {
1982 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
1983 force_sig(SIGILL, current);
1988 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1989 * allows the thread to return from interrupt. After that handle_swbp()
1990 * sets utask->active_uprobe.
1992 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1993 * and allows the thread to return from interrupt.
1995 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1996 * uprobe_notify_resume().
1998 void uprobe_notify_resume(struct pt_regs *regs)
2000 struct uprobe_task *utask;
2002 clear_thread_flag(TIF_UPROBE);
2004 utask = current->utask;
2005 if (utask && utask->active_uprobe)
2006 handle_singlestep(utask, regs);
2012 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2013 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2015 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2020 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2021 (!current->utask || !current->utask->return_instances))
2024 set_thread_flag(TIF_UPROBE);
2029 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2030 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2032 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2034 struct uprobe_task *utask = current->utask;
2036 if (!current->mm || !utask || !utask->active_uprobe)
2037 /* task is currently not uprobed */
2040 utask->state = UTASK_SSTEP_ACK;
2041 set_thread_flag(TIF_UPROBE);
2045 static struct notifier_block uprobe_exception_nb = {
2046 .notifier_call = arch_uprobe_exception_notify,
2047 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2050 static int __init init_uprobes(void)
2054 for (i = 0; i < UPROBES_HASH_SZ; i++)
2055 mutex_init(&uprobes_mmap_mutex[i]);
2057 if (percpu_init_rwsem(&dup_mmap_sem))
2060 return register_die_notifier(&uprobe_exception_nb);
2062 __initcall(init_uprobes);