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 <pzijlstr@redhat.com>
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/export.h>
31 #include <linux/rmap.h> /* anon_vma_prepare */
32 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
33 #include <linux/swap.h> /* try_to_free_swap */
34 #include <linux/ptrace.h> /* user_enable_single_step */
35 #include <linux/kdebug.h> /* notifier mechanism */
36 #include "../../mm/internal.h" /* munlock_vma_page */
37 #include <linux/percpu-rwsem.h>
39 #include <linux/uprobes.h>
41 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
42 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
44 static struct rb_root uprobes_tree = RB_ROOT;
46 * allows us to skip the uprobe_mmap if there are no uprobe events active
47 * at this time. Probably a fine grained per inode count is better?
49 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
51 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
53 #define UPROBES_HASH_SZ 13
54 /* serialize uprobe->pending_list */
55 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
56 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
58 static struct percpu_rw_semaphore dup_mmap_sem;
60 /* Have a copy of original instruction */
61 #define UPROBE_COPY_INSN 0
62 /* Can skip singlestep */
63 #define UPROBE_SKIP_SSTEP 1
66 struct rb_node rb_node; /* node in the rb tree */
68 struct rw_semaphore register_rwsem;
69 struct rw_semaphore consumer_rwsem;
70 struct list_head pending_list;
71 struct uprobe_consumer *consumers;
72 struct inode *inode; /* Also hold a ref to inode */
75 struct arch_uprobe arch;
79 * valid_vma: Verify if the specified vma is an executable vma
80 * Relax restrictions while unregistering: vm_flags might have
81 * changed after breakpoint was inserted.
82 * - is_register: indicates if we are in register context.
83 * - Return 1 if the specified virtual address is in an
86 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
88 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
93 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
96 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
98 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
101 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
103 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
107 * __replace_page - replace page in vma by new page.
108 * based on replace_page in mm/ksm.c
110 * @vma: vma that holds the pte pointing to page
111 * @addr: address the old @page is mapped at
112 * @page: the cowed page we are replacing by kpage
113 * @kpage: the modified page we replace page by
115 * Returns 0 on success, -EFAULT on failure.
117 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
118 struct page *page, struct page *kpage)
120 struct mm_struct *mm = vma->vm_mm;
124 /* For mmu_notifiers */
125 const unsigned long mmun_start = addr;
126 const unsigned long mmun_end = addr + PAGE_SIZE;
128 /* For try_to_free_swap() and munlock_vma_page() below */
131 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
133 ptep = page_check_address(page, mm, addr, &ptl, 0);
138 page_add_new_anon_rmap(kpage, vma, addr);
140 if (!PageAnon(page)) {
141 dec_mm_counter(mm, MM_FILEPAGES);
142 inc_mm_counter(mm, MM_ANONPAGES);
145 flush_cache_page(vma, addr, pte_pfn(*ptep));
146 ptep_clear_flush(vma, addr, ptep);
147 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
149 page_remove_rmap(page);
150 if (!page_mapped(page))
151 try_to_free_swap(page);
152 pte_unmap_unlock(ptep, ptl);
154 if (vma->vm_flags & VM_LOCKED)
155 munlock_vma_page(page);
160 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
166 * is_swbp_insn - check if instruction is breakpoint instruction.
167 * @insn: instruction to be checked.
168 * Default implementation of is_swbp_insn
169 * Returns true if @insn is a breakpoint instruction.
171 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
173 return *insn == UPROBE_SWBP_INSN;
177 * is_trap_insn - check if instruction is breakpoint instruction.
178 * @insn: instruction to be checked.
179 * Default implementation of is_trap_insn
180 * Returns true if @insn is a breakpoint instruction.
182 * This function is needed for the case where an architecture has multiple
183 * trap instructions (like powerpc).
185 bool __weak is_trap_insn(uprobe_opcode_t *insn)
187 return is_swbp_insn(insn);
190 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
192 void *kaddr = kmap_atomic(page);
193 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
194 kunmap_atomic(kaddr);
197 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
199 uprobe_opcode_t old_opcode;
203 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
204 * We do not check if it is any other 'trap variant' which could
205 * be conditional trap instruction such as the one powerpc supports.
207 * The logic is that we do not care if the underlying instruction
208 * is a trap variant; uprobes always wins over any other (gdb)
211 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
212 is_swbp = is_swbp_insn(&old_opcode);
214 if (is_swbp_insn(new_opcode)) {
215 if (is_swbp) /* register: already installed? */
218 if (!is_swbp) /* unregister: was it changed by us? */
227 * Expect the breakpoint instruction to be the smallest size instruction for
228 * the architecture. If an arch has variable length instruction and the
229 * breakpoint instruction is not of the smallest length instruction
230 * supported by that architecture then we need to modify is_trap_at_addr and
231 * write_opcode accordingly. This would never be a problem for archs that
232 * have fixed length instructions.
236 * write_opcode - write the opcode at a given virtual address.
237 * @mm: the probed process address space.
238 * @vaddr: the virtual address to store the opcode.
239 * @opcode: opcode to be written at @vaddr.
241 * Called with mm->mmap_sem held (for read and with a reference to
244 * For mm @mm, write the opcode at @vaddr.
245 * Return 0 (success) or a negative errno.
247 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
248 uprobe_opcode_t opcode)
250 struct page *old_page, *new_page;
251 void *vaddr_old, *vaddr_new;
252 struct vm_area_struct *vma;
256 /* Read the page with vaddr into memory */
257 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
261 ret = verify_opcode(old_page, vaddr, &opcode);
266 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
270 __SetPageUptodate(new_page);
272 /* copy the page now that we've got it stable */
273 vaddr_old = kmap_atomic(old_page);
274 vaddr_new = kmap_atomic(new_page);
276 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
277 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
279 kunmap_atomic(vaddr_new);
280 kunmap_atomic(vaddr_old);
282 ret = anon_vma_prepare(vma);
286 ret = __replace_page(vma, vaddr, old_page, new_page);
289 page_cache_release(new_page);
293 if (unlikely(ret == -EAGAIN))
299 * set_swbp - store breakpoint at a given address.
300 * @auprobe: arch specific probepoint information.
301 * @mm: the probed process address space.
302 * @vaddr: the virtual address to insert the opcode.
304 * For mm @mm, store the breakpoint instruction at @vaddr.
305 * Return 0 (success) or a negative errno.
307 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
309 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
313 * set_orig_insn - Restore the original instruction.
314 * @mm: the probed process address space.
315 * @auprobe: arch specific probepoint information.
316 * @vaddr: the virtual address to insert the opcode.
318 * For mm @mm, restore the original opcode (opcode) at @vaddr.
319 * Return 0 (success) or a negative errno.
322 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
324 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
327 static int match_uprobe(struct uprobe *l, struct uprobe *r)
329 if (l->inode < r->inode)
332 if (l->inode > r->inode)
335 if (l->offset < r->offset)
338 if (l->offset > r->offset)
344 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
346 struct uprobe u = { .inode = inode, .offset = offset };
347 struct rb_node *n = uprobes_tree.rb_node;
348 struct uprobe *uprobe;
352 uprobe = rb_entry(n, struct uprobe, rb_node);
353 match = match_uprobe(&u, uprobe);
355 atomic_inc(&uprobe->ref);
368 * Find a uprobe corresponding to a given inode:offset
369 * Acquires uprobes_treelock
371 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
373 struct uprobe *uprobe;
375 spin_lock(&uprobes_treelock);
376 uprobe = __find_uprobe(inode, offset);
377 spin_unlock(&uprobes_treelock);
382 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
384 struct rb_node **p = &uprobes_tree.rb_node;
385 struct rb_node *parent = NULL;
391 u = rb_entry(parent, struct uprobe, rb_node);
392 match = match_uprobe(uprobe, u);
399 p = &parent->rb_left;
401 p = &parent->rb_right;
406 rb_link_node(&uprobe->rb_node, parent, p);
407 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
408 /* get access + creation ref */
409 atomic_set(&uprobe->ref, 2);
415 * Acquire uprobes_treelock.
416 * Matching uprobe already exists in rbtree;
417 * increment (access refcount) and return the matching uprobe.
419 * No matching uprobe; insert the uprobe in rb_tree;
420 * get a double refcount (access + creation) and return NULL.
422 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
426 spin_lock(&uprobes_treelock);
427 u = __insert_uprobe(uprobe);
428 spin_unlock(&uprobes_treelock);
433 static void put_uprobe(struct uprobe *uprobe)
435 if (atomic_dec_and_test(&uprobe->ref))
439 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
441 struct uprobe *uprobe, *cur_uprobe;
443 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
447 uprobe->inode = igrab(inode);
448 uprobe->offset = offset;
449 init_rwsem(&uprobe->register_rwsem);
450 init_rwsem(&uprobe->consumer_rwsem);
451 /* For now assume that the instruction need not be single-stepped */
452 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
454 /* add to uprobes_tree, sorted on inode:offset */
455 cur_uprobe = insert_uprobe(uprobe);
457 /* a uprobe exists for this inode:offset combination */
467 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
469 down_write(&uprobe->consumer_rwsem);
470 uc->next = uprobe->consumers;
471 uprobe->consumers = uc;
472 up_write(&uprobe->consumer_rwsem);
476 * For uprobe @uprobe, delete the consumer @uc.
477 * Return true if the @uc is deleted successfully
480 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
482 struct uprobe_consumer **con;
485 down_write(&uprobe->consumer_rwsem);
486 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
493 up_write(&uprobe->consumer_rwsem);
499 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
500 unsigned long nbytes, loff_t offset)
510 if (!mapping->a_ops->readpage)
513 idx = offset >> PAGE_CACHE_SHIFT;
514 off = offset & ~PAGE_MASK;
517 * Ensure that the page that has the original instruction is
518 * populated and in page-cache.
520 page = read_mapping_page(mapping, idx, filp);
522 return PTR_ERR(page);
524 vaddr = kmap_atomic(page);
525 memcpy(insn, vaddr + off, nbytes);
526 kunmap_atomic(vaddr);
527 page_cache_release(page);
532 static int copy_insn(struct uprobe *uprobe, struct file *filp)
534 struct address_space *mapping;
535 unsigned long nbytes;
538 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
539 mapping = uprobe->inode->i_mapping;
541 /* Instruction at end of binary; copy only available bytes */
542 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
543 bytes = uprobe->inode->i_size - uprobe->offset;
545 bytes = MAX_UINSN_BYTES;
547 /* Instruction at the page-boundary; copy bytes in second page */
548 if (nbytes < bytes) {
549 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
550 bytes - nbytes, uprobe->offset + nbytes);
555 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
558 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
559 struct mm_struct *mm, unsigned long vaddr)
563 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
566 /* TODO: move this into _register, until then we abuse this sem. */
567 down_write(&uprobe->consumer_rwsem);
568 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
571 ret = copy_insn(uprobe, file);
576 if (is_trap_insn((uprobe_opcode_t *)uprobe->arch.insn))
579 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
583 /* write_opcode() assumes we don't cross page boundary */
584 BUG_ON((uprobe->offset & ~PAGE_MASK) +
585 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
587 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
588 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
591 up_write(&uprobe->consumer_rwsem);
596 static inline bool consumer_filter(struct uprobe_consumer *uc,
597 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
599 return !uc->filter || uc->filter(uc, ctx, mm);
602 static bool filter_chain(struct uprobe *uprobe,
603 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
605 struct uprobe_consumer *uc;
608 down_read(&uprobe->consumer_rwsem);
609 for (uc = uprobe->consumers; uc; uc = uc->next) {
610 ret = consumer_filter(uc, ctx, mm);
614 up_read(&uprobe->consumer_rwsem);
620 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
621 struct vm_area_struct *vma, unsigned long vaddr)
626 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
631 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
632 * the task can hit this breakpoint right after __replace_page().
634 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
636 set_bit(MMF_HAS_UPROBES, &mm->flags);
638 ret = set_swbp(&uprobe->arch, mm, vaddr);
640 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
641 else if (first_uprobe)
642 clear_bit(MMF_HAS_UPROBES, &mm->flags);
648 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
650 set_bit(MMF_RECALC_UPROBES, &mm->flags);
651 return set_orig_insn(&uprobe->arch, mm, vaddr);
654 static inline bool uprobe_is_active(struct uprobe *uprobe)
656 return !RB_EMPTY_NODE(&uprobe->rb_node);
659 * There could be threads that have already hit the breakpoint. They
660 * will recheck the current insn and restart if find_uprobe() fails.
661 * See find_active_uprobe().
663 static void delete_uprobe(struct uprobe *uprobe)
665 if (WARN_ON(!uprobe_is_active(uprobe)))
668 spin_lock(&uprobes_treelock);
669 rb_erase(&uprobe->rb_node, &uprobes_tree);
670 spin_unlock(&uprobes_treelock);
671 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
677 struct map_info *next;
678 struct mm_struct *mm;
682 static inline struct map_info *free_map_info(struct map_info *info)
684 struct map_info *next = info->next;
689 static struct map_info *
690 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
692 unsigned long pgoff = offset >> PAGE_SHIFT;
693 struct vm_area_struct *vma;
694 struct map_info *curr = NULL;
695 struct map_info *prev = NULL;
696 struct map_info *info;
700 mutex_lock(&mapping->i_mmap_mutex);
701 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
702 if (!valid_vma(vma, is_register))
705 if (!prev && !more) {
707 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
708 * reclaim. This is optimistic, no harm done if it fails.
710 prev = kmalloc(sizeof(struct map_info),
711 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
720 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
728 info->mm = vma->vm_mm;
729 info->vaddr = offset_to_vaddr(vma, offset);
731 mutex_unlock(&mapping->i_mmap_mutex);
743 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
745 curr = ERR_PTR(-ENOMEM);
755 prev = free_map_info(prev);
760 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
762 bool is_register = !!new;
763 struct map_info *info;
766 percpu_down_write(&dup_mmap_sem);
767 info = build_map_info(uprobe->inode->i_mapping,
768 uprobe->offset, is_register);
775 struct mm_struct *mm = info->mm;
776 struct vm_area_struct *vma;
778 if (err && is_register)
781 down_write(&mm->mmap_sem);
782 vma = find_vma(mm, info->vaddr);
783 if (!vma || !valid_vma(vma, is_register) ||
784 file_inode(vma->vm_file) != uprobe->inode)
787 if (vma->vm_start > info->vaddr ||
788 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
792 /* consult only the "caller", new consumer. */
793 if (consumer_filter(new,
794 UPROBE_FILTER_REGISTER, mm))
795 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
796 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
797 if (!filter_chain(uprobe,
798 UPROBE_FILTER_UNREGISTER, mm))
799 err |= remove_breakpoint(uprobe, mm, info->vaddr);
803 up_write(&mm->mmap_sem);
806 info = free_map_info(info);
809 percpu_up_write(&dup_mmap_sem);
813 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
815 consumer_add(uprobe, uc);
816 return register_for_each_vma(uprobe, uc);
819 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
823 if (!consumer_del(uprobe, uc)) /* WARN? */
826 err = register_for_each_vma(uprobe, NULL);
827 /* TODO : cant unregister? schedule a worker thread */
828 if (!uprobe->consumers && !err)
829 delete_uprobe(uprobe);
833 * uprobe_register - register a probe
834 * @inode: the file in which the probe has to be placed.
835 * @offset: offset from the start of the file.
836 * @uc: information on howto handle the probe..
838 * Apart from the access refcount, uprobe_register() takes a creation
839 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
840 * inserted into the rbtree (i.e first consumer for a @inode:@offset
841 * tuple). Creation refcount stops uprobe_unregister from freeing the
842 * @uprobe even before the register operation is complete. Creation
843 * refcount is released when the last @uc for the @uprobe
846 * Return errno if it cannot successully install probes
847 * else return 0 (success)
849 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
851 struct uprobe *uprobe;
854 /* Racy, just to catch the obvious mistakes */
855 if (offset > i_size_read(inode))
859 uprobe = alloc_uprobe(inode, offset);
863 * We can race with uprobe_unregister()->delete_uprobe().
864 * Check uprobe_is_active() and retry if it is false.
866 down_write(&uprobe->register_rwsem);
868 if (likely(uprobe_is_active(uprobe))) {
869 ret = __uprobe_register(uprobe, uc);
871 __uprobe_unregister(uprobe, uc);
873 up_write(&uprobe->register_rwsem);
876 if (unlikely(ret == -EAGAIN))
880 EXPORT_SYMBOL_GPL(uprobe_register);
883 * uprobe_apply - unregister a already registered probe.
884 * @inode: the file in which the probe has to be removed.
885 * @offset: offset from the start of the file.
886 * @uc: consumer which wants to add more or remove some breakpoints
887 * @add: add or remove the breakpoints
889 int uprobe_apply(struct inode *inode, loff_t offset,
890 struct uprobe_consumer *uc, bool add)
892 struct uprobe *uprobe;
893 struct uprobe_consumer *con;
896 uprobe = find_uprobe(inode, offset);
900 down_write(&uprobe->register_rwsem);
901 for (con = uprobe->consumers; con && con != uc ; con = con->next)
904 ret = register_for_each_vma(uprobe, add ? uc : NULL);
905 up_write(&uprobe->register_rwsem);
912 * uprobe_unregister - unregister a already registered probe.
913 * @inode: the file in which the probe has to be removed.
914 * @offset: offset from the start of the file.
915 * @uc: identify which probe if multiple probes are colocated.
917 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
919 struct uprobe *uprobe;
921 uprobe = find_uprobe(inode, offset);
925 down_write(&uprobe->register_rwsem);
926 __uprobe_unregister(uprobe, uc);
927 up_write(&uprobe->register_rwsem);
930 EXPORT_SYMBOL_GPL(uprobe_unregister);
932 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
934 struct vm_area_struct *vma;
937 down_read(&mm->mmap_sem);
938 for (vma = mm->mmap; vma; vma = vma->vm_next) {
942 if (!valid_vma(vma, false) ||
943 file_inode(vma->vm_file) != uprobe->inode)
946 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
947 if (uprobe->offset < offset ||
948 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
951 vaddr = offset_to_vaddr(vma, uprobe->offset);
952 err |= remove_breakpoint(uprobe, mm, vaddr);
954 up_read(&mm->mmap_sem);
959 static struct rb_node *
960 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
962 struct rb_node *n = uprobes_tree.rb_node;
965 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
967 if (inode < u->inode) {
969 } else if (inode > u->inode) {
974 else if (min > u->offset)
985 * For a given range in vma, build a list of probes that need to be inserted.
987 static void build_probe_list(struct inode *inode,
988 struct vm_area_struct *vma,
989 unsigned long start, unsigned long end,
990 struct list_head *head)
993 struct rb_node *n, *t;
996 INIT_LIST_HEAD(head);
997 min = vaddr_to_offset(vma, start);
998 max = min + (end - start) - 1;
1000 spin_lock(&uprobes_treelock);
1001 n = find_node_in_range(inode, min, max);
1003 for (t = n; t; t = rb_prev(t)) {
1004 u = rb_entry(t, struct uprobe, rb_node);
1005 if (u->inode != inode || u->offset < min)
1007 list_add(&u->pending_list, head);
1008 atomic_inc(&u->ref);
1010 for (t = n; (t = rb_next(t)); ) {
1011 u = rb_entry(t, struct uprobe, rb_node);
1012 if (u->inode != inode || u->offset > max)
1014 list_add(&u->pending_list, head);
1015 atomic_inc(&u->ref);
1018 spin_unlock(&uprobes_treelock);
1022 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1024 * Currently we ignore all errors and always return 0, the callers
1025 * can't handle the failure anyway.
1027 int uprobe_mmap(struct vm_area_struct *vma)
1029 struct list_head tmp_list;
1030 struct uprobe *uprobe, *u;
1031 struct inode *inode;
1033 if (no_uprobe_events() || !valid_vma(vma, true))
1036 inode = file_inode(vma->vm_file);
1040 mutex_lock(uprobes_mmap_hash(inode));
1041 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1043 * We can race with uprobe_unregister(), this uprobe can be already
1044 * removed. But in this case filter_chain() must return false, all
1045 * consumers have gone away.
1047 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1048 if (!fatal_signal_pending(current) &&
1049 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1050 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1051 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1055 mutex_unlock(uprobes_mmap_hash(inode));
1061 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1064 struct inode *inode;
1067 inode = file_inode(vma->vm_file);
1069 min = vaddr_to_offset(vma, start);
1070 max = min + (end - start) - 1;
1072 spin_lock(&uprobes_treelock);
1073 n = find_node_in_range(inode, min, max);
1074 spin_unlock(&uprobes_treelock);
1080 * Called in context of a munmap of a vma.
1082 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1084 if (no_uprobe_events() || !valid_vma(vma, false))
1087 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1090 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1091 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1094 if (vma_has_uprobes(vma, start, end))
1095 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1098 /* Slot allocation for XOL */
1099 static int xol_add_vma(struct xol_area *area)
1101 struct mm_struct *mm = current->mm;
1102 int ret = -EALREADY;
1104 down_write(&mm->mmap_sem);
1105 if (mm->uprobes_state.xol_area)
1109 /* Try to map as high as possible, this is only a hint. */
1110 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1111 if (area->vaddr & ~PAGE_MASK) {
1116 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1117 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1121 smp_wmb(); /* pairs with get_xol_area() */
1122 mm->uprobes_state.xol_area = area;
1125 up_write(&mm->mmap_sem);
1131 * get_xol_area - Allocate process's xol_area if necessary.
1132 * This area will be used for storing instructions for execution out of line.
1134 * Returns the allocated area or NULL.
1136 static struct xol_area *get_xol_area(void)
1138 struct mm_struct *mm = current->mm;
1139 struct xol_area *area;
1141 area = mm->uprobes_state.xol_area;
1145 area = kzalloc(sizeof(*area), GFP_KERNEL);
1146 if (unlikely(!area))
1149 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1153 area->page = alloc_page(GFP_HIGHUSER);
1157 init_waitqueue_head(&area->wq);
1158 if (!xol_add_vma(area))
1161 __free_page(area->page);
1163 kfree(area->bitmap);
1167 area = mm->uprobes_state.xol_area;
1169 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1174 * uprobe_clear_state - Free the area allocated for slots.
1176 void uprobe_clear_state(struct mm_struct *mm)
1178 struct xol_area *area = mm->uprobes_state.xol_area;
1183 put_page(area->page);
1184 kfree(area->bitmap);
1188 void uprobe_start_dup_mmap(void)
1190 percpu_down_read(&dup_mmap_sem);
1193 void uprobe_end_dup_mmap(void)
1195 percpu_up_read(&dup_mmap_sem);
1198 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1200 newmm->uprobes_state.xol_area = NULL;
1202 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1203 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1204 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1205 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1210 * - search for a free slot.
1212 static unsigned long xol_take_insn_slot(struct xol_area *area)
1214 unsigned long slot_addr;
1218 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1219 if (slot_nr < UINSNS_PER_PAGE) {
1220 if (!test_and_set_bit(slot_nr, area->bitmap))
1223 slot_nr = UINSNS_PER_PAGE;
1226 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1227 } while (slot_nr >= UINSNS_PER_PAGE);
1229 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1230 atomic_inc(&area->slot_count);
1236 * xol_get_insn_slot - allocate a slot for xol.
1237 * Returns the allocated slot address or 0.
1239 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1241 struct xol_area *area;
1242 unsigned long offset;
1243 unsigned long xol_vaddr;
1246 area = get_xol_area();
1250 xol_vaddr = xol_take_insn_slot(area);
1251 if (unlikely(!xol_vaddr))
1254 /* Initialize the slot */
1255 offset = xol_vaddr & ~PAGE_MASK;
1256 vaddr = kmap_atomic(area->page);
1257 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1258 kunmap_atomic(vaddr);
1260 * We probably need flush_icache_user_range() but it needs vma.
1261 * This should work on supported architectures too.
1263 flush_dcache_page(area->page);
1269 * xol_free_insn_slot - If slot was earlier allocated by
1270 * @xol_get_insn_slot(), make the slot available for
1271 * subsequent requests.
1273 static void xol_free_insn_slot(struct task_struct *tsk)
1275 struct xol_area *area;
1276 unsigned long vma_end;
1277 unsigned long slot_addr;
1279 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1282 slot_addr = tsk->utask->xol_vaddr;
1283 if (unlikely(!slot_addr))
1286 area = tsk->mm->uprobes_state.xol_area;
1287 vma_end = area->vaddr + PAGE_SIZE;
1288 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1289 unsigned long offset;
1292 offset = slot_addr - area->vaddr;
1293 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1294 if (slot_nr >= UINSNS_PER_PAGE)
1297 clear_bit(slot_nr, area->bitmap);
1298 atomic_dec(&area->slot_count);
1299 if (waitqueue_active(&area->wq))
1302 tsk->utask->xol_vaddr = 0;
1307 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1308 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1310 * Return the address of the breakpoint instruction.
1312 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1314 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1318 * Called with no locks held.
1319 * Called in context of a exiting or a exec-ing thread.
1321 void uprobe_free_utask(struct task_struct *t)
1323 struct uprobe_task *utask = t->utask;
1328 if (utask->active_uprobe)
1329 put_uprobe(utask->active_uprobe);
1331 xol_free_insn_slot(t);
1337 * Called in context of a new clone/fork from copy_process.
1339 void uprobe_copy_process(struct task_struct *t)
1345 * Allocate a uprobe_task object for the task if if necessary.
1346 * Called when the thread hits a breakpoint.
1349 * - pointer to new uprobe_task on success
1352 static struct uprobe_task *get_utask(void)
1354 if (!current->utask)
1355 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1356 return current->utask;
1359 /* Prepare to single-step probed instruction out of line. */
1361 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1363 struct uprobe_task *utask;
1364 unsigned long xol_vaddr;
1367 utask = get_utask();
1371 xol_vaddr = xol_get_insn_slot(uprobe);
1375 utask->xol_vaddr = xol_vaddr;
1376 utask->vaddr = bp_vaddr;
1378 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1379 if (unlikely(err)) {
1380 xol_free_insn_slot(current);
1384 utask->active_uprobe = uprobe;
1385 utask->state = UTASK_SSTEP;
1390 * If we are singlestepping, then ensure this thread is not connected to
1391 * non-fatal signals until completion of singlestep. When xol insn itself
1392 * triggers the signal, restart the original insn even if the task is
1393 * already SIGKILL'ed (since coredump should report the correct ip). This
1394 * is even more important if the task has a handler for SIGSEGV/etc, The
1395 * _same_ instruction should be repeated again after return from the signal
1396 * handler, and SSTEP can never finish in this case.
1398 bool uprobe_deny_signal(void)
1400 struct task_struct *t = current;
1401 struct uprobe_task *utask = t->utask;
1403 if (likely(!utask || !utask->active_uprobe))
1406 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1408 if (signal_pending(t)) {
1409 spin_lock_irq(&t->sighand->siglock);
1410 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1411 spin_unlock_irq(&t->sighand->siglock);
1413 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1414 utask->state = UTASK_SSTEP_TRAPPED;
1415 set_tsk_thread_flag(t, TIF_UPROBE);
1416 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1424 * Avoid singlestepping the original instruction if the original instruction
1425 * is a NOP or can be emulated.
1427 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1429 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1430 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1432 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1437 static void mmf_recalc_uprobes(struct mm_struct *mm)
1439 struct vm_area_struct *vma;
1441 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1442 if (!valid_vma(vma, false))
1445 * This is not strictly accurate, we can race with
1446 * uprobe_unregister() and see the already removed
1447 * uprobe if delete_uprobe() was not yet called.
1448 * Or this uprobe can be filtered out.
1450 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1454 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1457 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1460 uprobe_opcode_t opcode;
1463 pagefault_disable();
1464 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1468 if (likely(result == 0))
1471 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1475 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1478 /* This needs to return true for any variant of the trap insn */
1479 return is_trap_insn(&opcode);
1482 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1484 struct mm_struct *mm = current->mm;
1485 struct uprobe *uprobe = NULL;
1486 struct vm_area_struct *vma;
1488 down_read(&mm->mmap_sem);
1489 vma = find_vma(mm, bp_vaddr);
1490 if (vma && vma->vm_start <= bp_vaddr) {
1491 if (valid_vma(vma, false)) {
1492 struct inode *inode = file_inode(vma->vm_file);
1493 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1495 uprobe = find_uprobe(inode, offset);
1499 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1504 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1505 mmf_recalc_uprobes(mm);
1506 up_read(&mm->mmap_sem);
1511 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1513 struct uprobe_consumer *uc;
1514 int remove = UPROBE_HANDLER_REMOVE;
1516 down_read(&uprobe->register_rwsem);
1517 for (uc = uprobe->consumers; uc; uc = uc->next) {
1518 int rc = uc->handler(uc, regs);
1520 WARN(rc & ~UPROBE_HANDLER_MASK,
1521 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1525 if (remove && uprobe->consumers) {
1526 WARN_ON(!uprobe_is_active(uprobe));
1527 unapply_uprobe(uprobe, current->mm);
1529 up_read(&uprobe->register_rwsem);
1533 * Run handler and ask thread to singlestep.
1534 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1536 static void handle_swbp(struct pt_regs *regs)
1538 struct uprobe *uprobe;
1539 unsigned long bp_vaddr;
1540 int uninitialized_var(is_swbp);
1542 bp_vaddr = uprobe_get_swbp_addr(regs);
1543 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1547 /* No matching uprobe; signal SIGTRAP. */
1548 send_sig(SIGTRAP, current, 0);
1551 * Either we raced with uprobe_unregister() or we can't
1552 * access this memory. The latter is only possible if
1553 * another thread plays with our ->mm. In both cases
1554 * we can simply restart. If this vma was unmapped we
1555 * can pretend this insn was not executed yet and get
1556 * the (correct) SIGSEGV after restart.
1558 instruction_pointer_set(regs, bp_vaddr);
1563 /* change it in advance for ->handler() and restart */
1564 instruction_pointer_set(regs, bp_vaddr);
1567 * TODO: move copy_insn/etc into _register and remove this hack.
1568 * After we hit the bp, _unregister + _register can install the
1569 * new and not-yet-analyzed uprobe at the same address, restart.
1571 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1572 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1575 handler_chain(uprobe, regs);
1576 if (can_skip_sstep(uprobe, regs))
1579 if (!pre_ssout(uprobe, regs, bp_vaddr))
1582 /* can_skip_sstep() succeeded, or restart if can't singlestep */
1588 * Perform required fix-ups and disable singlestep.
1589 * Allow pending signals to take effect.
1591 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1593 struct uprobe *uprobe;
1595 uprobe = utask->active_uprobe;
1596 if (utask->state == UTASK_SSTEP_ACK)
1597 arch_uprobe_post_xol(&uprobe->arch, regs);
1598 else if (utask->state == UTASK_SSTEP_TRAPPED)
1599 arch_uprobe_abort_xol(&uprobe->arch, regs);
1604 utask->active_uprobe = NULL;
1605 utask->state = UTASK_RUNNING;
1606 xol_free_insn_slot(current);
1608 spin_lock_irq(¤t->sighand->siglock);
1609 recalc_sigpending(); /* see uprobe_deny_signal() */
1610 spin_unlock_irq(¤t->sighand->siglock);
1614 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1615 * allows the thread to return from interrupt. After that handle_swbp()
1616 * sets utask->active_uprobe.
1618 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1619 * and allows the thread to return from interrupt.
1621 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1622 * uprobe_notify_resume().
1624 void uprobe_notify_resume(struct pt_regs *regs)
1626 struct uprobe_task *utask;
1628 clear_thread_flag(TIF_UPROBE);
1630 utask = current->utask;
1631 if (utask && utask->active_uprobe)
1632 handle_singlestep(utask, regs);
1638 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1639 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1641 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1643 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1646 set_thread_flag(TIF_UPROBE);
1651 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1652 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1654 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1656 struct uprobe_task *utask = current->utask;
1658 if (!current->mm || !utask || !utask->active_uprobe)
1659 /* task is currently not uprobed */
1662 utask->state = UTASK_SSTEP_ACK;
1663 set_thread_flag(TIF_UPROBE);
1667 static struct notifier_block uprobe_exception_nb = {
1668 .notifier_call = arch_uprobe_exception_notify,
1669 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1672 static int __init init_uprobes(void)
1676 for (i = 0; i < UPROBES_HASH_SZ; i++)
1677 mutex_init(&uprobes_mmap_mutex[i]);
1679 if (percpu_init_rwsem(&dup_mmap_sem))
1682 return register_die_notifier(&uprobe_exception_nb);
1684 module_init(init_uprobes);
1686 static void __exit exit_uprobes(void)
1689 module_exit(exit_uprobes);