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;
176 static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode)
178 void *kaddr = kmap_atomic(page);
179 memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE);
180 kunmap_atomic(kaddr);
183 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
185 uprobe_opcode_t old_opcode;
188 copy_opcode(page, vaddr, &old_opcode);
189 is_swbp = is_swbp_insn(&old_opcode);
191 if (is_swbp_insn(new_opcode)) {
192 if (is_swbp) /* register: already installed? */
195 if (!is_swbp) /* unregister: was it changed by us? */
204 * Expect the breakpoint instruction to be the smallest size instruction for
205 * the architecture. If an arch has variable length instruction and the
206 * breakpoint instruction is not of the smallest length instruction
207 * supported by that architecture then we need to modify is_swbp_at_addr and
208 * write_opcode accordingly. This would never be a problem for archs that
209 * have fixed length instructions.
213 * write_opcode - write the opcode at a given virtual address.
214 * @mm: the probed process address space.
215 * @vaddr: the virtual address to store the opcode.
216 * @opcode: opcode to be written at @vaddr.
218 * Called with mm->mmap_sem held (for read and with a reference to
221 * For mm @mm, write the opcode at @vaddr.
222 * Return 0 (success) or a negative errno.
224 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
225 uprobe_opcode_t opcode)
227 struct page *old_page, *new_page;
228 void *vaddr_old, *vaddr_new;
229 struct vm_area_struct *vma;
233 /* Read the page with vaddr into memory */
234 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
238 ret = verify_opcode(old_page, vaddr, &opcode);
243 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
247 __SetPageUptodate(new_page);
249 /* copy the page now that we've got it stable */
250 vaddr_old = kmap_atomic(old_page);
251 vaddr_new = kmap_atomic(new_page);
253 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
254 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
256 kunmap_atomic(vaddr_new);
257 kunmap_atomic(vaddr_old);
259 ret = anon_vma_prepare(vma);
263 ret = __replace_page(vma, vaddr, old_page, new_page);
266 page_cache_release(new_page);
270 if (unlikely(ret == -EAGAIN))
276 * set_swbp - store breakpoint at a given address.
277 * @auprobe: arch specific probepoint information.
278 * @mm: the probed process address space.
279 * @vaddr: the virtual address to insert the opcode.
281 * For mm @mm, store the breakpoint instruction at @vaddr.
282 * Return 0 (success) or a negative errno.
284 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
286 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
290 * set_orig_insn - Restore the original instruction.
291 * @mm: the probed process address space.
292 * @auprobe: arch specific probepoint information.
293 * @vaddr: the virtual address to insert the opcode.
295 * For mm @mm, restore the original opcode (opcode) at @vaddr.
296 * Return 0 (success) or a negative errno.
299 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
301 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
304 static int match_uprobe(struct uprobe *l, struct uprobe *r)
306 if (l->inode < r->inode)
309 if (l->inode > r->inode)
312 if (l->offset < r->offset)
315 if (l->offset > r->offset)
321 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
323 struct uprobe u = { .inode = inode, .offset = offset };
324 struct rb_node *n = uprobes_tree.rb_node;
325 struct uprobe *uprobe;
329 uprobe = rb_entry(n, struct uprobe, rb_node);
330 match = match_uprobe(&u, uprobe);
332 atomic_inc(&uprobe->ref);
345 * Find a uprobe corresponding to a given inode:offset
346 * Acquires uprobes_treelock
348 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
350 struct uprobe *uprobe;
352 spin_lock(&uprobes_treelock);
353 uprobe = __find_uprobe(inode, offset);
354 spin_unlock(&uprobes_treelock);
359 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
361 struct rb_node **p = &uprobes_tree.rb_node;
362 struct rb_node *parent = NULL;
368 u = rb_entry(parent, struct uprobe, rb_node);
369 match = match_uprobe(uprobe, u);
376 p = &parent->rb_left;
378 p = &parent->rb_right;
383 rb_link_node(&uprobe->rb_node, parent, p);
384 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
385 /* get access + creation ref */
386 atomic_set(&uprobe->ref, 2);
392 * Acquire uprobes_treelock.
393 * Matching uprobe already exists in rbtree;
394 * increment (access refcount) and return the matching uprobe.
396 * No matching uprobe; insert the uprobe in rb_tree;
397 * get a double refcount (access + creation) and return NULL.
399 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
403 spin_lock(&uprobes_treelock);
404 u = __insert_uprobe(uprobe);
405 spin_unlock(&uprobes_treelock);
410 static void put_uprobe(struct uprobe *uprobe)
412 if (atomic_dec_and_test(&uprobe->ref))
416 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
418 struct uprobe *uprobe, *cur_uprobe;
420 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
424 uprobe->inode = igrab(inode);
425 uprobe->offset = offset;
426 init_rwsem(&uprobe->register_rwsem);
427 init_rwsem(&uprobe->consumer_rwsem);
428 /* For now assume that the instruction need not be single-stepped */
429 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
431 /* add to uprobes_tree, sorted on inode:offset */
432 cur_uprobe = insert_uprobe(uprobe);
434 /* a uprobe exists for this inode:offset combination */
444 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
446 down_write(&uprobe->consumer_rwsem);
447 uc->next = uprobe->consumers;
448 uprobe->consumers = uc;
449 up_write(&uprobe->consumer_rwsem);
453 * For uprobe @uprobe, delete the consumer @uc.
454 * Return true if the @uc is deleted successfully
457 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
459 struct uprobe_consumer **con;
462 down_write(&uprobe->consumer_rwsem);
463 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
470 up_write(&uprobe->consumer_rwsem);
476 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
477 unsigned long nbytes, loff_t offset)
487 if (!mapping->a_ops->readpage)
490 idx = offset >> PAGE_CACHE_SHIFT;
491 off = offset & ~PAGE_MASK;
494 * Ensure that the page that has the original instruction is
495 * populated and in page-cache.
497 page = read_mapping_page(mapping, idx, filp);
499 return PTR_ERR(page);
501 vaddr = kmap_atomic(page);
502 memcpy(insn, vaddr + off, nbytes);
503 kunmap_atomic(vaddr);
504 page_cache_release(page);
509 static int copy_insn(struct uprobe *uprobe, struct file *filp)
511 struct address_space *mapping;
512 unsigned long nbytes;
515 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
516 mapping = uprobe->inode->i_mapping;
518 /* Instruction at end of binary; copy only available bytes */
519 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
520 bytes = uprobe->inode->i_size - uprobe->offset;
522 bytes = MAX_UINSN_BYTES;
524 /* Instruction at the page-boundary; copy bytes in second page */
525 if (nbytes < bytes) {
526 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
527 bytes - nbytes, uprobe->offset + nbytes);
532 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
535 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
536 struct mm_struct *mm, unsigned long vaddr)
540 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
543 /* TODO: move this into _register, until then we abuse this sem. */
544 down_write(&uprobe->consumer_rwsem);
545 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
548 ret = copy_insn(uprobe, file);
553 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
556 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
560 /* write_opcode() assumes we don't cross page boundary */
561 BUG_ON((uprobe->offset & ~PAGE_MASK) +
562 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
564 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
565 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
568 up_write(&uprobe->consumer_rwsem);
573 static inline bool consumer_filter(struct uprobe_consumer *uc,
574 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
576 return !uc->filter || uc->filter(uc, ctx, mm);
579 static bool filter_chain(struct uprobe *uprobe,
580 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
582 struct uprobe_consumer *uc;
585 down_read(&uprobe->consumer_rwsem);
586 for (uc = uprobe->consumers; uc; uc = uc->next) {
587 ret = consumer_filter(uc, ctx, mm);
591 up_read(&uprobe->consumer_rwsem);
597 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
598 struct vm_area_struct *vma, unsigned long vaddr)
603 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
608 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
609 * the task can hit this breakpoint right after __replace_page().
611 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
613 set_bit(MMF_HAS_UPROBES, &mm->flags);
615 ret = set_swbp(&uprobe->arch, mm, vaddr);
617 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
618 else if (first_uprobe)
619 clear_bit(MMF_HAS_UPROBES, &mm->flags);
625 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
627 set_bit(MMF_RECALC_UPROBES, &mm->flags);
628 return set_orig_insn(&uprobe->arch, mm, vaddr);
631 static inline bool uprobe_is_active(struct uprobe *uprobe)
633 return !RB_EMPTY_NODE(&uprobe->rb_node);
636 * There could be threads that have already hit the breakpoint. They
637 * will recheck the current insn and restart if find_uprobe() fails.
638 * See find_active_uprobe().
640 static void delete_uprobe(struct uprobe *uprobe)
642 if (WARN_ON(!uprobe_is_active(uprobe)))
645 spin_lock(&uprobes_treelock);
646 rb_erase(&uprobe->rb_node, &uprobes_tree);
647 spin_unlock(&uprobes_treelock);
648 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
654 struct map_info *next;
655 struct mm_struct *mm;
659 static inline struct map_info *free_map_info(struct map_info *info)
661 struct map_info *next = info->next;
666 static struct map_info *
667 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
669 unsigned long pgoff = offset >> PAGE_SHIFT;
670 struct vm_area_struct *vma;
671 struct map_info *curr = NULL;
672 struct map_info *prev = NULL;
673 struct map_info *info;
677 mutex_lock(&mapping->i_mmap_mutex);
678 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
679 if (!valid_vma(vma, is_register))
682 if (!prev && !more) {
684 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
685 * reclaim. This is optimistic, no harm done if it fails.
687 prev = kmalloc(sizeof(struct map_info),
688 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
697 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
705 info->mm = vma->vm_mm;
706 info->vaddr = offset_to_vaddr(vma, offset);
708 mutex_unlock(&mapping->i_mmap_mutex);
720 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
722 curr = ERR_PTR(-ENOMEM);
732 prev = free_map_info(prev);
737 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
739 bool is_register = !!new;
740 struct map_info *info;
743 percpu_down_write(&dup_mmap_sem);
744 info = build_map_info(uprobe->inode->i_mapping,
745 uprobe->offset, is_register);
752 struct mm_struct *mm = info->mm;
753 struct vm_area_struct *vma;
755 if (err && is_register)
758 down_write(&mm->mmap_sem);
759 vma = find_vma(mm, info->vaddr);
760 if (!vma || !valid_vma(vma, is_register) ||
761 file_inode(vma->vm_file) != uprobe->inode)
764 if (vma->vm_start > info->vaddr ||
765 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
769 /* consult only the "caller", new consumer. */
770 if (consumer_filter(new,
771 UPROBE_FILTER_REGISTER, mm))
772 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
773 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
774 if (!filter_chain(uprobe,
775 UPROBE_FILTER_UNREGISTER, mm))
776 err |= remove_breakpoint(uprobe, mm, info->vaddr);
780 up_write(&mm->mmap_sem);
783 info = free_map_info(info);
786 percpu_up_write(&dup_mmap_sem);
790 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
792 consumer_add(uprobe, uc);
793 return register_for_each_vma(uprobe, uc);
796 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
800 if (!consumer_del(uprobe, uc)) /* WARN? */
803 err = register_for_each_vma(uprobe, NULL);
804 /* TODO : cant unregister? schedule a worker thread */
805 if (!uprobe->consumers && !err)
806 delete_uprobe(uprobe);
810 * uprobe_register - register a probe
811 * @inode: the file in which the probe has to be placed.
812 * @offset: offset from the start of the file.
813 * @uc: information on howto handle the probe..
815 * Apart from the access refcount, uprobe_register() takes a creation
816 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
817 * inserted into the rbtree (i.e first consumer for a @inode:@offset
818 * tuple). Creation refcount stops uprobe_unregister from freeing the
819 * @uprobe even before the register operation is complete. Creation
820 * refcount is released when the last @uc for the @uprobe
823 * Return errno if it cannot successully install probes
824 * else return 0 (success)
826 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
828 struct uprobe *uprobe;
831 /* Racy, just to catch the obvious mistakes */
832 if (offset > i_size_read(inode))
836 uprobe = alloc_uprobe(inode, offset);
840 * We can race with uprobe_unregister()->delete_uprobe().
841 * Check uprobe_is_active() and retry if it is false.
843 down_write(&uprobe->register_rwsem);
845 if (likely(uprobe_is_active(uprobe))) {
846 ret = __uprobe_register(uprobe, uc);
848 __uprobe_unregister(uprobe, uc);
850 up_write(&uprobe->register_rwsem);
853 if (unlikely(ret == -EAGAIN))
857 EXPORT_SYMBOL_GPL(uprobe_register);
860 * uprobe_apply - unregister a already registered probe.
861 * @inode: the file in which the probe has to be removed.
862 * @offset: offset from the start of the file.
863 * @uc: consumer which wants to add more or remove some breakpoints
864 * @add: add or remove the breakpoints
866 int uprobe_apply(struct inode *inode, loff_t offset,
867 struct uprobe_consumer *uc, bool add)
869 struct uprobe *uprobe;
870 struct uprobe_consumer *con;
873 uprobe = find_uprobe(inode, offset);
877 down_write(&uprobe->register_rwsem);
878 for (con = uprobe->consumers; con && con != uc ; con = con->next)
881 ret = register_for_each_vma(uprobe, add ? uc : NULL);
882 up_write(&uprobe->register_rwsem);
889 * uprobe_unregister - unregister a already registered probe.
890 * @inode: the file in which the probe has to be removed.
891 * @offset: offset from the start of the file.
892 * @uc: identify which probe if multiple probes are colocated.
894 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
896 struct uprobe *uprobe;
898 uprobe = find_uprobe(inode, offset);
902 down_write(&uprobe->register_rwsem);
903 __uprobe_unregister(uprobe, uc);
904 up_write(&uprobe->register_rwsem);
907 EXPORT_SYMBOL_GPL(uprobe_unregister);
909 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
911 struct vm_area_struct *vma;
914 down_read(&mm->mmap_sem);
915 for (vma = mm->mmap; vma; vma = vma->vm_next) {
919 if (!valid_vma(vma, false) ||
920 file_inode(vma->vm_file) != uprobe->inode)
923 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
924 if (uprobe->offset < offset ||
925 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
928 vaddr = offset_to_vaddr(vma, uprobe->offset);
929 err |= remove_breakpoint(uprobe, mm, vaddr);
931 up_read(&mm->mmap_sem);
936 static struct rb_node *
937 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
939 struct rb_node *n = uprobes_tree.rb_node;
942 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
944 if (inode < u->inode) {
946 } else if (inode > u->inode) {
951 else if (min > u->offset)
962 * For a given range in vma, build a list of probes that need to be inserted.
964 static void build_probe_list(struct inode *inode,
965 struct vm_area_struct *vma,
966 unsigned long start, unsigned long end,
967 struct list_head *head)
970 struct rb_node *n, *t;
973 INIT_LIST_HEAD(head);
974 min = vaddr_to_offset(vma, start);
975 max = min + (end - start) - 1;
977 spin_lock(&uprobes_treelock);
978 n = find_node_in_range(inode, min, max);
980 for (t = n; t; t = rb_prev(t)) {
981 u = rb_entry(t, struct uprobe, rb_node);
982 if (u->inode != inode || u->offset < min)
984 list_add(&u->pending_list, head);
987 for (t = n; (t = rb_next(t)); ) {
988 u = rb_entry(t, struct uprobe, rb_node);
989 if (u->inode != inode || u->offset > max)
991 list_add(&u->pending_list, head);
995 spin_unlock(&uprobes_treelock);
999 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1001 * Currently we ignore all errors and always return 0, the callers
1002 * can't handle the failure anyway.
1004 int uprobe_mmap(struct vm_area_struct *vma)
1006 struct list_head tmp_list;
1007 struct uprobe *uprobe, *u;
1008 struct inode *inode;
1010 if (no_uprobe_events() || !valid_vma(vma, true))
1013 inode = file_inode(vma->vm_file);
1017 mutex_lock(uprobes_mmap_hash(inode));
1018 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1020 * We can race with uprobe_unregister(), this uprobe can be already
1021 * removed. But in this case filter_chain() must return false, all
1022 * consumers have gone away.
1024 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1025 if (!fatal_signal_pending(current) &&
1026 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1027 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1028 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1032 mutex_unlock(uprobes_mmap_hash(inode));
1038 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1041 struct inode *inode;
1044 inode = file_inode(vma->vm_file);
1046 min = vaddr_to_offset(vma, start);
1047 max = min + (end - start) - 1;
1049 spin_lock(&uprobes_treelock);
1050 n = find_node_in_range(inode, min, max);
1051 spin_unlock(&uprobes_treelock);
1057 * Called in context of a munmap of a vma.
1059 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1061 if (no_uprobe_events() || !valid_vma(vma, false))
1064 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1067 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1068 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1071 if (vma_has_uprobes(vma, start, end))
1072 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1075 /* Slot allocation for XOL */
1076 static int xol_add_vma(struct xol_area *area)
1078 struct mm_struct *mm = current->mm;
1079 int ret = -EALREADY;
1081 down_write(&mm->mmap_sem);
1082 if (mm->uprobes_state.xol_area)
1086 /* Try to map as high as possible, this is only a hint. */
1087 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1088 if (area->vaddr & ~PAGE_MASK) {
1093 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1094 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1098 smp_wmb(); /* pairs with get_xol_area() */
1099 mm->uprobes_state.xol_area = area;
1102 up_write(&mm->mmap_sem);
1108 * get_xol_area - Allocate process's xol_area if necessary.
1109 * This area will be used for storing instructions for execution out of line.
1111 * Returns the allocated area or NULL.
1113 static struct xol_area *get_xol_area(void)
1115 struct mm_struct *mm = current->mm;
1116 struct xol_area *area;
1118 area = mm->uprobes_state.xol_area;
1122 area = kzalloc(sizeof(*area), GFP_KERNEL);
1123 if (unlikely(!area))
1126 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1130 area->page = alloc_page(GFP_HIGHUSER);
1134 init_waitqueue_head(&area->wq);
1135 if (!xol_add_vma(area))
1138 __free_page(area->page);
1140 kfree(area->bitmap);
1144 area = mm->uprobes_state.xol_area;
1146 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1151 * uprobe_clear_state - Free the area allocated for slots.
1153 void uprobe_clear_state(struct mm_struct *mm)
1155 struct xol_area *area = mm->uprobes_state.xol_area;
1160 put_page(area->page);
1161 kfree(area->bitmap);
1165 void uprobe_start_dup_mmap(void)
1167 percpu_down_read(&dup_mmap_sem);
1170 void uprobe_end_dup_mmap(void)
1172 percpu_up_read(&dup_mmap_sem);
1175 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1177 newmm->uprobes_state.xol_area = NULL;
1179 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1180 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1181 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1182 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1187 * - search for a free slot.
1189 static unsigned long xol_take_insn_slot(struct xol_area *area)
1191 unsigned long slot_addr;
1195 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1196 if (slot_nr < UINSNS_PER_PAGE) {
1197 if (!test_and_set_bit(slot_nr, area->bitmap))
1200 slot_nr = UINSNS_PER_PAGE;
1203 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1204 } while (slot_nr >= UINSNS_PER_PAGE);
1206 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1207 atomic_inc(&area->slot_count);
1213 * xol_get_insn_slot - allocate a slot for xol.
1214 * Returns the allocated slot address or 0.
1216 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1218 struct xol_area *area;
1219 unsigned long offset;
1220 unsigned long xol_vaddr;
1223 area = get_xol_area();
1227 xol_vaddr = xol_take_insn_slot(area);
1228 if (unlikely(!xol_vaddr))
1231 /* Initialize the slot */
1232 offset = xol_vaddr & ~PAGE_MASK;
1233 vaddr = kmap_atomic(area->page);
1234 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1235 kunmap_atomic(vaddr);
1237 * We probably need flush_icache_user_range() but it needs vma.
1238 * This should work on supported architectures too.
1240 flush_dcache_page(area->page);
1246 * xol_free_insn_slot - If slot was earlier allocated by
1247 * @xol_get_insn_slot(), make the slot available for
1248 * subsequent requests.
1250 static void xol_free_insn_slot(struct task_struct *tsk)
1252 struct xol_area *area;
1253 unsigned long vma_end;
1254 unsigned long slot_addr;
1256 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1259 slot_addr = tsk->utask->xol_vaddr;
1260 if (unlikely(!slot_addr))
1263 area = tsk->mm->uprobes_state.xol_area;
1264 vma_end = area->vaddr + PAGE_SIZE;
1265 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1266 unsigned long offset;
1269 offset = slot_addr - area->vaddr;
1270 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1271 if (slot_nr >= UINSNS_PER_PAGE)
1274 clear_bit(slot_nr, area->bitmap);
1275 atomic_dec(&area->slot_count);
1276 if (waitqueue_active(&area->wq))
1279 tsk->utask->xol_vaddr = 0;
1284 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1285 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1287 * Return the address of the breakpoint instruction.
1289 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1291 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1295 * Called with no locks held.
1296 * Called in context of a exiting or a exec-ing thread.
1298 void uprobe_free_utask(struct task_struct *t)
1300 struct uprobe_task *utask = t->utask;
1305 if (utask->active_uprobe)
1306 put_uprobe(utask->active_uprobe);
1308 xol_free_insn_slot(t);
1314 * Called in context of a new clone/fork from copy_process.
1316 void uprobe_copy_process(struct task_struct *t)
1322 * Allocate a uprobe_task object for the task if if necessary.
1323 * Called when the thread hits a breakpoint.
1326 * - pointer to new uprobe_task on success
1329 static struct uprobe_task *get_utask(void)
1331 if (!current->utask)
1332 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1333 return current->utask;
1336 /* Prepare to single-step probed instruction out of line. */
1338 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1340 struct uprobe_task *utask;
1341 unsigned long xol_vaddr;
1344 utask = get_utask();
1348 xol_vaddr = xol_get_insn_slot(uprobe);
1352 utask->xol_vaddr = xol_vaddr;
1353 utask->vaddr = bp_vaddr;
1355 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1356 if (unlikely(err)) {
1357 xol_free_insn_slot(current);
1361 utask->active_uprobe = uprobe;
1362 utask->state = UTASK_SSTEP;
1367 * If we are singlestepping, then ensure this thread is not connected to
1368 * non-fatal signals until completion of singlestep. When xol insn itself
1369 * triggers the signal, restart the original insn even if the task is
1370 * already SIGKILL'ed (since coredump should report the correct ip). This
1371 * is even more important if the task has a handler for SIGSEGV/etc, The
1372 * _same_ instruction should be repeated again after return from the signal
1373 * handler, and SSTEP can never finish in this case.
1375 bool uprobe_deny_signal(void)
1377 struct task_struct *t = current;
1378 struct uprobe_task *utask = t->utask;
1380 if (likely(!utask || !utask->active_uprobe))
1383 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1385 if (signal_pending(t)) {
1386 spin_lock_irq(&t->sighand->siglock);
1387 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1388 spin_unlock_irq(&t->sighand->siglock);
1390 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1391 utask->state = UTASK_SSTEP_TRAPPED;
1392 set_tsk_thread_flag(t, TIF_UPROBE);
1393 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1401 * Avoid singlestepping the original instruction if the original instruction
1402 * is a NOP or can be emulated.
1404 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1406 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1407 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1409 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1414 static void mmf_recalc_uprobes(struct mm_struct *mm)
1416 struct vm_area_struct *vma;
1418 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1419 if (!valid_vma(vma, false))
1422 * This is not strictly accurate, we can race with
1423 * uprobe_unregister() and see the already removed
1424 * uprobe if delete_uprobe() was not yet called.
1425 * Or this uprobe can be filtered out.
1427 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1431 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1434 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1437 uprobe_opcode_t opcode;
1440 pagefault_disable();
1441 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1445 if (likely(result == 0))
1448 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1452 copy_opcode(page, vaddr, &opcode);
1455 return is_swbp_insn(&opcode);
1458 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1460 struct mm_struct *mm = current->mm;
1461 struct uprobe *uprobe = NULL;
1462 struct vm_area_struct *vma;
1464 down_read(&mm->mmap_sem);
1465 vma = find_vma(mm, bp_vaddr);
1466 if (vma && vma->vm_start <= bp_vaddr) {
1467 if (valid_vma(vma, false)) {
1468 struct inode *inode = file_inode(vma->vm_file);
1469 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1471 uprobe = find_uprobe(inode, offset);
1475 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1480 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1481 mmf_recalc_uprobes(mm);
1482 up_read(&mm->mmap_sem);
1487 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1489 struct uprobe_consumer *uc;
1490 int remove = UPROBE_HANDLER_REMOVE;
1492 down_read(&uprobe->register_rwsem);
1493 for (uc = uprobe->consumers; uc; uc = uc->next) {
1494 int rc = uc->handler(uc, regs);
1496 WARN(rc & ~UPROBE_HANDLER_MASK,
1497 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1501 if (remove && uprobe->consumers) {
1502 WARN_ON(!uprobe_is_active(uprobe));
1503 unapply_uprobe(uprobe, current->mm);
1505 up_read(&uprobe->register_rwsem);
1509 * Run handler and ask thread to singlestep.
1510 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1512 static void handle_swbp(struct pt_regs *regs)
1514 struct uprobe *uprobe;
1515 unsigned long bp_vaddr;
1516 int uninitialized_var(is_swbp);
1518 bp_vaddr = uprobe_get_swbp_addr(regs);
1519 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1523 /* No matching uprobe; signal SIGTRAP. */
1524 send_sig(SIGTRAP, current, 0);
1527 * Either we raced with uprobe_unregister() or we can't
1528 * access this memory. The latter is only possible if
1529 * another thread plays with our ->mm. In both cases
1530 * we can simply restart. If this vma was unmapped we
1531 * can pretend this insn was not executed yet and get
1532 * the (correct) SIGSEGV after restart.
1534 instruction_pointer_set(regs, bp_vaddr);
1539 /* change it in advance for ->handler() and restart */
1540 instruction_pointer_set(regs, bp_vaddr);
1543 * TODO: move copy_insn/etc into _register and remove this hack.
1544 * After we hit the bp, _unregister + _register can install the
1545 * new and not-yet-analyzed uprobe at the same address, restart.
1547 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1548 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1551 handler_chain(uprobe, regs);
1552 if (can_skip_sstep(uprobe, regs))
1555 if (!pre_ssout(uprobe, regs, bp_vaddr))
1558 /* can_skip_sstep() succeeded, or restart if can't singlestep */
1564 * Perform required fix-ups and disable singlestep.
1565 * Allow pending signals to take effect.
1567 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1569 struct uprobe *uprobe;
1571 uprobe = utask->active_uprobe;
1572 if (utask->state == UTASK_SSTEP_ACK)
1573 arch_uprobe_post_xol(&uprobe->arch, regs);
1574 else if (utask->state == UTASK_SSTEP_TRAPPED)
1575 arch_uprobe_abort_xol(&uprobe->arch, regs);
1580 utask->active_uprobe = NULL;
1581 utask->state = UTASK_RUNNING;
1582 xol_free_insn_slot(current);
1584 spin_lock_irq(¤t->sighand->siglock);
1585 recalc_sigpending(); /* see uprobe_deny_signal() */
1586 spin_unlock_irq(¤t->sighand->siglock);
1590 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1591 * allows the thread to return from interrupt. After that handle_swbp()
1592 * sets utask->active_uprobe.
1594 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1595 * and allows the thread to return from interrupt.
1597 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1598 * uprobe_notify_resume().
1600 void uprobe_notify_resume(struct pt_regs *regs)
1602 struct uprobe_task *utask;
1604 clear_thread_flag(TIF_UPROBE);
1606 utask = current->utask;
1607 if (utask && utask->active_uprobe)
1608 handle_singlestep(utask, regs);
1614 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1615 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1617 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1619 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1622 set_thread_flag(TIF_UPROBE);
1627 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1628 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1630 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1632 struct uprobe_task *utask = current->utask;
1634 if (!current->mm || !utask || !utask->active_uprobe)
1635 /* task is currently not uprobed */
1638 utask->state = UTASK_SSTEP_ACK;
1639 set_thread_flag(TIF_UPROBE);
1643 static struct notifier_block uprobe_exception_nb = {
1644 .notifier_call = arch_uprobe_exception_notify,
1645 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1648 static int __init init_uprobes(void)
1652 for (i = 0; i < UPROBES_HASH_SZ; i++)
1653 mutex_init(&uprobes_mmap_mutex[i]);
1655 if (percpu_init_rwsem(&dup_mmap_sem))
1658 return register_die_notifier(&uprobe_exception_nb);
1660 module_init(init_uprobes);
1662 static void __exit exit_uprobes(void)
1665 module_exit(exit_uprobes);