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 void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
199 void *kaddr = kmap_atomic(page);
200 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
201 kunmap_atomic(kaddr);
204 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
206 uprobe_opcode_t old_opcode;
210 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
211 * We do not check if it is any other 'trap variant' which could
212 * be conditional trap instruction such as the one powerpc supports.
214 * The logic is that we do not care if the underlying instruction
215 * is a trap variant; uprobes always wins over any other (gdb)
218 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
219 is_swbp = is_swbp_insn(&old_opcode);
221 if (is_swbp_insn(new_opcode)) {
222 if (is_swbp) /* register: already installed? */
225 if (!is_swbp) /* unregister: was it changed by us? */
234 * Expect the breakpoint instruction to be the smallest size instruction for
235 * the architecture. If an arch has variable length instruction and the
236 * breakpoint instruction is not of the smallest length instruction
237 * supported by that architecture then we need to modify is_trap_at_addr and
238 * write_opcode accordingly. This would never be a problem for archs that
239 * have fixed length instructions.
243 * write_opcode - write the opcode at a given virtual address.
244 * @mm: the probed process address space.
245 * @vaddr: the virtual address to store the opcode.
246 * @opcode: opcode to be written at @vaddr.
248 * Called with mm->mmap_sem held (for read and with a reference to
251 * For mm @mm, write the opcode at @vaddr.
252 * Return 0 (success) or a negative errno.
254 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
255 uprobe_opcode_t opcode)
257 struct page *old_page, *new_page;
258 struct vm_area_struct *vma;
262 /* Read the page with vaddr into memory */
263 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
267 ret = verify_opcode(old_page, vaddr, &opcode);
272 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
276 __SetPageUptodate(new_page);
278 copy_highpage(new_page, old_page);
279 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
281 ret = anon_vma_prepare(vma);
285 ret = __replace_page(vma, vaddr, old_page, new_page);
288 page_cache_release(new_page);
292 if (unlikely(ret == -EAGAIN))
298 * set_swbp - store breakpoint at a given address.
299 * @auprobe: arch specific probepoint information.
300 * @mm: the probed process address space.
301 * @vaddr: the virtual address to insert the opcode.
303 * For mm @mm, store the breakpoint instruction at @vaddr.
304 * Return 0 (success) or a negative errno.
306 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
308 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
312 * set_orig_insn - Restore the original instruction.
313 * @mm: the probed process address space.
314 * @auprobe: arch specific probepoint information.
315 * @vaddr: the virtual address to insert the opcode.
317 * For mm @mm, restore the original opcode (opcode) at @vaddr.
318 * Return 0 (success) or a negative errno.
321 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
323 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
326 static int match_uprobe(struct uprobe *l, struct uprobe *r)
328 if (l->inode < r->inode)
331 if (l->inode > r->inode)
334 if (l->offset < r->offset)
337 if (l->offset > r->offset)
343 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
345 struct uprobe u = { .inode = inode, .offset = offset };
346 struct rb_node *n = uprobes_tree.rb_node;
347 struct uprobe *uprobe;
351 uprobe = rb_entry(n, struct uprobe, rb_node);
352 match = match_uprobe(&u, uprobe);
354 atomic_inc(&uprobe->ref);
367 * Find a uprobe corresponding to a given inode:offset
368 * Acquires uprobes_treelock
370 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
372 struct uprobe *uprobe;
374 spin_lock(&uprobes_treelock);
375 uprobe = __find_uprobe(inode, offset);
376 spin_unlock(&uprobes_treelock);
381 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
383 struct rb_node **p = &uprobes_tree.rb_node;
384 struct rb_node *parent = NULL;
390 u = rb_entry(parent, struct uprobe, rb_node);
391 match = match_uprobe(uprobe, u);
398 p = &parent->rb_left;
400 p = &parent->rb_right;
405 rb_link_node(&uprobe->rb_node, parent, p);
406 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
407 /* get access + creation ref */
408 atomic_set(&uprobe->ref, 2);
414 * Acquire uprobes_treelock.
415 * Matching uprobe already exists in rbtree;
416 * increment (access refcount) and return the matching uprobe.
418 * No matching uprobe; insert the uprobe in rb_tree;
419 * get a double refcount (access + creation) and return NULL.
421 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
425 spin_lock(&uprobes_treelock);
426 u = __insert_uprobe(uprobe);
427 spin_unlock(&uprobes_treelock);
432 static void put_uprobe(struct uprobe *uprobe)
434 if (atomic_dec_and_test(&uprobe->ref))
438 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
440 struct uprobe *uprobe, *cur_uprobe;
442 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
446 uprobe->inode = igrab(inode);
447 uprobe->offset = offset;
448 init_rwsem(&uprobe->register_rwsem);
449 init_rwsem(&uprobe->consumer_rwsem);
450 /* For now assume that the instruction need not be single-stepped */
451 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
453 /* add to uprobes_tree, sorted on inode:offset */
454 cur_uprobe = insert_uprobe(uprobe);
456 /* a uprobe exists for this inode:offset combination */
466 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
468 down_write(&uprobe->consumer_rwsem);
469 uc->next = uprobe->consumers;
470 uprobe->consumers = uc;
471 up_write(&uprobe->consumer_rwsem);
475 * For uprobe @uprobe, delete the consumer @uc.
476 * Return true if the @uc is deleted successfully
479 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
481 struct uprobe_consumer **con;
484 down_write(&uprobe->consumer_rwsem);
485 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
492 up_write(&uprobe->consumer_rwsem);
498 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
499 unsigned long nbytes, loff_t offset)
503 if (!mapping->a_ops->readpage)
506 * Ensure that the page that has the original instruction is
507 * populated and in page-cache.
509 page = read_mapping_page(mapping, offset >> PAGE_CACHE_SHIFT, filp);
511 return PTR_ERR(page);
513 copy_from_page(page, offset, insn, nbytes);
514 page_cache_release(page);
519 static int copy_insn(struct uprobe *uprobe, struct file *filp)
521 struct address_space *mapping;
522 unsigned long nbytes;
525 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
526 mapping = uprobe->inode->i_mapping;
528 /* Instruction at end of binary; copy only available bytes */
529 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
530 bytes = uprobe->inode->i_size - uprobe->offset;
532 bytes = MAX_UINSN_BYTES;
534 /* Instruction at the page-boundary; copy bytes in second page */
535 if (nbytes < bytes) {
536 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
537 bytes - nbytes, uprobe->offset + nbytes);
542 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
545 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
546 struct mm_struct *mm, unsigned long vaddr)
550 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
553 /* TODO: move this into _register, until then we abuse this sem. */
554 down_write(&uprobe->consumer_rwsem);
555 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
558 ret = copy_insn(uprobe, file);
563 if (is_trap_insn((uprobe_opcode_t *)uprobe->arch.insn))
566 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
570 /* write_opcode() assumes we don't cross page boundary */
571 BUG_ON((uprobe->offset & ~PAGE_MASK) +
572 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
574 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
575 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
578 up_write(&uprobe->consumer_rwsem);
583 static inline bool consumer_filter(struct uprobe_consumer *uc,
584 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
586 return !uc->filter || uc->filter(uc, ctx, mm);
589 static bool filter_chain(struct uprobe *uprobe,
590 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
592 struct uprobe_consumer *uc;
595 down_read(&uprobe->consumer_rwsem);
596 for (uc = uprobe->consumers; uc; uc = uc->next) {
597 ret = consumer_filter(uc, ctx, mm);
601 up_read(&uprobe->consumer_rwsem);
607 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
608 struct vm_area_struct *vma, unsigned long vaddr)
613 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
618 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
619 * the task can hit this breakpoint right after __replace_page().
621 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
623 set_bit(MMF_HAS_UPROBES, &mm->flags);
625 ret = set_swbp(&uprobe->arch, mm, vaddr);
627 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
628 else if (first_uprobe)
629 clear_bit(MMF_HAS_UPROBES, &mm->flags);
635 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
637 set_bit(MMF_RECALC_UPROBES, &mm->flags);
638 return set_orig_insn(&uprobe->arch, mm, vaddr);
641 static inline bool uprobe_is_active(struct uprobe *uprobe)
643 return !RB_EMPTY_NODE(&uprobe->rb_node);
646 * There could be threads that have already hit the breakpoint. They
647 * will recheck the current insn and restart if find_uprobe() fails.
648 * See find_active_uprobe().
650 static void delete_uprobe(struct uprobe *uprobe)
652 if (WARN_ON(!uprobe_is_active(uprobe)))
655 spin_lock(&uprobes_treelock);
656 rb_erase(&uprobe->rb_node, &uprobes_tree);
657 spin_unlock(&uprobes_treelock);
658 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
664 struct map_info *next;
665 struct mm_struct *mm;
669 static inline struct map_info *free_map_info(struct map_info *info)
671 struct map_info *next = info->next;
676 static struct map_info *
677 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
679 unsigned long pgoff = offset >> PAGE_SHIFT;
680 struct vm_area_struct *vma;
681 struct map_info *curr = NULL;
682 struct map_info *prev = NULL;
683 struct map_info *info;
687 mutex_lock(&mapping->i_mmap_mutex);
688 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
689 if (!valid_vma(vma, is_register))
692 if (!prev && !more) {
694 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
695 * reclaim. This is optimistic, no harm done if it fails.
697 prev = kmalloc(sizeof(struct map_info),
698 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
707 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
715 info->mm = vma->vm_mm;
716 info->vaddr = offset_to_vaddr(vma, offset);
718 mutex_unlock(&mapping->i_mmap_mutex);
730 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
732 curr = ERR_PTR(-ENOMEM);
742 prev = free_map_info(prev);
747 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
749 bool is_register = !!new;
750 struct map_info *info;
753 percpu_down_write(&dup_mmap_sem);
754 info = build_map_info(uprobe->inode->i_mapping,
755 uprobe->offset, is_register);
762 struct mm_struct *mm = info->mm;
763 struct vm_area_struct *vma;
765 if (err && is_register)
768 down_write(&mm->mmap_sem);
769 vma = find_vma(mm, info->vaddr);
770 if (!vma || !valid_vma(vma, is_register) ||
771 file_inode(vma->vm_file) != uprobe->inode)
774 if (vma->vm_start > info->vaddr ||
775 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
779 /* consult only the "caller", new consumer. */
780 if (consumer_filter(new,
781 UPROBE_FILTER_REGISTER, mm))
782 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
783 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
784 if (!filter_chain(uprobe,
785 UPROBE_FILTER_UNREGISTER, mm))
786 err |= remove_breakpoint(uprobe, mm, info->vaddr);
790 up_write(&mm->mmap_sem);
793 info = free_map_info(info);
796 percpu_up_write(&dup_mmap_sem);
800 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
802 consumer_add(uprobe, uc);
803 return register_for_each_vma(uprobe, uc);
806 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
810 if (!consumer_del(uprobe, uc)) /* WARN? */
813 err = register_for_each_vma(uprobe, NULL);
814 /* TODO : cant unregister? schedule a worker thread */
815 if (!uprobe->consumers && !err)
816 delete_uprobe(uprobe);
820 * uprobe_register - register a probe
821 * @inode: the file in which the probe has to be placed.
822 * @offset: offset from the start of the file.
823 * @uc: information on howto handle the probe..
825 * Apart from the access refcount, uprobe_register() takes a creation
826 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
827 * inserted into the rbtree (i.e first consumer for a @inode:@offset
828 * tuple). Creation refcount stops uprobe_unregister from freeing the
829 * @uprobe even before the register operation is complete. Creation
830 * refcount is released when the last @uc for the @uprobe
833 * Return errno if it cannot successully install probes
834 * else return 0 (success)
836 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
838 struct uprobe *uprobe;
841 /* Racy, just to catch the obvious mistakes */
842 if (offset > i_size_read(inode))
846 uprobe = alloc_uprobe(inode, offset);
850 * We can race with uprobe_unregister()->delete_uprobe().
851 * Check uprobe_is_active() and retry if it is false.
853 down_write(&uprobe->register_rwsem);
855 if (likely(uprobe_is_active(uprobe))) {
856 ret = __uprobe_register(uprobe, uc);
858 __uprobe_unregister(uprobe, uc);
860 up_write(&uprobe->register_rwsem);
863 if (unlikely(ret == -EAGAIN))
867 EXPORT_SYMBOL_GPL(uprobe_register);
870 * uprobe_apply - unregister a already registered probe.
871 * @inode: the file in which the probe has to be removed.
872 * @offset: offset from the start of the file.
873 * @uc: consumer which wants to add more or remove some breakpoints
874 * @add: add or remove the breakpoints
876 int uprobe_apply(struct inode *inode, loff_t offset,
877 struct uprobe_consumer *uc, bool add)
879 struct uprobe *uprobe;
880 struct uprobe_consumer *con;
883 uprobe = find_uprobe(inode, offset);
887 down_write(&uprobe->register_rwsem);
888 for (con = uprobe->consumers; con && con != uc ; con = con->next)
891 ret = register_for_each_vma(uprobe, add ? uc : NULL);
892 up_write(&uprobe->register_rwsem);
899 * uprobe_unregister - unregister a already registered probe.
900 * @inode: the file in which the probe has to be removed.
901 * @offset: offset from the start of the file.
902 * @uc: identify which probe if multiple probes are colocated.
904 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
906 struct uprobe *uprobe;
908 uprobe = find_uprobe(inode, offset);
912 down_write(&uprobe->register_rwsem);
913 __uprobe_unregister(uprobe, uc);
914 up_write(&uprobe->register_rwsem);
917 EXPORT_SYMBOL_GPL(uprobe_unregister);
919 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
921 struct vm_area_struct *vma;
924 down_read(&mm->mmap_sem);
925 for (vma = mm->mmap; vma; vma = vma->vm_next) {
929 if (!valid_vma(vma, false) ||
930 file_inode(vma->vm_file) != uprobe->inode)
933 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
934 if (uprobe->offset < offset ||
935 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
938 vaddr = offset_to_vaddr(vma, uprobe->offset);
939 err |= remove_breakpoint(uprobe, mm, vaddr);
941 up_read(&mm->mmap_sem);
946 static struct rb_node *
947 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
949 struct rb_node *n = uprobes_tree.rb_node;
952 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
954 if (inode < u->inode) {
956 } else if (inode > u->inode) {
961 else if (min > u->offset)
972 * For a given range in vma, build a list of probes that need to be inserted.
974 static void build_probe_list(struct inode *inode,
975 struct vm_area_struct *vma,
976 unsigned long start, unsigned long end,
977 struct list_head *head)
980 struct rb_node *n, *t;
983 INIT_LIST_HEAD(head);
984 min = vaddr_to_offset(vma, start);
985 max = min + (end - start) - 1;
987 spin_lock(&uprobes_treelock);
988 n = find_node_in_range(inode, min, max);
990 for (t = n; t; t = rb_prev(t)) {
991 u = rb_entry(t, struct uprobe, rb_node);
992 if (u->inode != inode || u->offset < min)
994 list_add(&u->pending_list, head);
997 for (t = n; (t = rb_next(t)); ) {
998 u = rb_entry(t, struct uprobe, rb_node);
999 if (u->inode != inode || u->offset > max)
1001 list_add(&u->pending_list, head);
1002 atomic_inc(&u->ref);
1005 spin_unlock(&uprobes_treelock);
1009 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1011 * Currently we ignore all errors and always return 0, the callers
1012 * can't handle the failure anyway.
1014 int uprobe_mmap(struct vm_area_struct *vma)
1016 struct list_head tmp_list;
1017 struct uprobe *uprobe, *u;
1018 struct inode *inode;
1020 if (no_uprobe_events() || !valid_vma(vma, true))
1023 inode = file_inode(vma->vm_file);
1027 mutex_lock(uprobes_mmap_hash(inode));
1028 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1030 * We can race with uprobe_unregister(), this uprobe can be already
1031 * removed. But in this case filter_chain() must return false, all
1032 * consumers have gone away.
1034 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1035 if (!fatal_signal_pending(current) &&
1036 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1037 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1038 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1042 mutex_unlock(uprobes_mmap_hash(inode));
1048 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1051 struct inode *inode;
1054 inode = file_inode(vma->vm_file);
1056 min = vaddr_to_offset(vma, start);
1057 max = min + (end - start) - 1;
1059 spin_lock(&uprobes_treelock);
1060 n = find_node_in_range(inode, min, max);
1061 spin_unlock(&uprobes_treelock);
1067 * Called in context of a munmap of a vma.
1069 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1071 if (no_uprobe_events() || !valid_vma(vma, false))
1074 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1077 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1078 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1081 if (vma_has_uprobes(vma, start, end))
1082 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1085 /* Slot allocation for XOL */
1086 static int xol_add_vma(struct xol_area *area)
1088 struct mm_struct *mm = current->mm;
1089 int ret = -EALREADY;
1091 down_write(&mm->mmap_sem);
1092 if (mm->uprobes_state.xol_area)
1096 /* Try to map as high as possible, this is only a hint. */
1097 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1098 if (area->vaddr & ~PAGE_MASK) {
1103 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1104 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1108 smp_wmb(); /* pairs with get_xol_area() */
1109 mm->uprobes_state.xol_area = area;
1112 up_write(&mm->mmap_sem);
1118 * get_xol_area - Allocate process's xol_area if necessary.
1119 * This area will be used for storing instructions for execution out of line.
1121 * Returns the allocated area or NULL.
1123 static struct xol_area *get_xol_area(void)
1125 struct mm_struct *mm = current->mm;
1126 struct xol_area *area;
1128 area = mm->uprobes_state.xol_area;
1132 area = kzalloc(sizeof(*area), GFP_KERNEL);
1133 if (unlikely(!area))
1136 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1140 area->page = alloc_page(GFP_HIGHUSER);
1144 init_waitqueue_head(&area->wq);
1145 if (!xol_add_vma(area))
1148 __free_page(area->page);
1150 kfree(area->bitmap);
1154 area = mm->uprobes_state.xol_area;
1156 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1161 * uprobe_clear_state - Free the area allocated for slots.
1163 void uprobe_clear_state(struct mm_struct *mm)
1165 struct xol_area *area = mm->uprobes_state.xol_area;
1170 put_page(area->page);
1171 kfree(area->bitmap);
1175 void uprobe_start_dup_mmap(void)
1177 percpu_down_read(&dup_mmap_sem);
1180 void uprobe_end_dup_mmap(void)
1182 percpu_up_read(&dup_mmap_sem);
1185 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1187 newmm->uprobes_state.xol_area = NULL;
1189 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1190 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1191 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1192 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1197 * - search for a free slot.
1199 static unsigned long xol_take_insn_slot(struct xol_area *area)
1201 unsigned long slot_addr;
1205 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1206 if (slot_nr < UINSNS_PER_PAGE) {
1207 if (!test_and_set_bit(slot_nr, area->bitmap))
1210 slot_nr = UINSNS_PER_PAGE;
1213 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1214 } while (slot_nr >= UINSNS_PER_PAGE);
1216 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1217 atomic_inc(&area->slot_count);
1223 * xol_get_insn_slot - allocate a slot for xol.
1224 * Returns the allocated slot address or 0.
1226 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1228 struct xol_area *area;
1229 unsigned long xol_vaddr;
1231 area = get_xol_area();
1235 xol_vaddr = xol_take_insn_slot(area);
1236 if (unlikely(!xol_vaddr))
1239 /* Initialize the slot */
1240 copy_to_page(area->page, xol_vaddr, uprobe->arch.insn, MAX_UINSN_BYTES);
1242 * We probably need flush_icache_user_range() but it needs vma.
1243 * This should work on supported architectures too.
1245 flush_dcache_page(area->page);
1251 * xol_free_insn_slot - If slot was earlier allocated by
1252 * @xol_get_insn_slot(), make the slot available for
1253 * subsequent requests.
1255 static void xol_free_insn_slot(struct task_struct *tsk)
1257 struct xol_area *area;
1258 unsigned long vma_end;
1259 unsigned long slot_addr;
1261 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1264 slot_addr = tsk->utask->xol_vaddr;
1265 if (unlikely(!slot_addr))
1268 area = tsk->mm->uprobes_state.xol_area;
1269 vma_end = area->vaddr + PAGE_SIZE;
1270 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1271 unsigned long offset;
1274 offset = slot_addr - area->vaddr;
1275 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1276 if (slot_nr >= UINSNS_PER_PAGE)
1279 clear_bit(slot_nr, area->bitmap);
1280 atomic_dec(&area->slot_count);
1281 if (waitqueue_active(&area->wq))
1284 tsk->utask->xol_vaddr = 0;
1289 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1290 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1292 * Return the address of the breakpoint instruction.
1294 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1296 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1300 * Called with no locks held.
1301 * Called in context of a exiting or a exec-ing thread.
1303 void uprobe_free_utask(struct task_struct *t)
1305 struct uprobe_task *utask = t->utask;
1310 if (utask->active_uprobe)
1311 put_uprobe(utask->active_uprobe);
1313 xol_free_insn_slot(t);
1319 * Called in context of a new clone/fork from copy_process.
1321 void uprobe_copy_process(struct task_struct *t)
1327 * Allocate a uprobe_task object for the task if if necessary.
1328 * Called when the thread hits a breakpoint.
1331 * - pointer to new uprobe_task on success
1334 static struct uprobe_task *get_utask(void)
1336 if (!current->utask)
1337 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1338 return current->utask;
1341 /* Prepare to single-step probed instruction out of line. */
1343 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1345 struct uprobe_task *utask;
1346 unsigned long xol_vaddr;
1349 utask = get_utask();
1353 xol_vaddr = xol_get_insn_slot(uprobe);
1357 utask->xol_vaddr = xol_vaddr;
1358 utask->vaddr = bp_vaddr;
1360 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1361 if (unlikely(err)) {
1362 xol_free_insn_slot(current);
1366 utask->active_uprobe = uprobe;
1367 utask->state = UTASK_SSTEP;
1372 * If we are singlestepping, then ensure this thread is not connected to
1373 * non-fatal signals until completion of singlestep. When xol insn itself
1374 * triggers the signal, restart the original insn even if the task is
1375 * already SIGKILL'ed (since coredump should report the correct ip). This
1376 * is even more important if the task has a handler for SIGSEGV/etc, The
1377 * _same_ instruction should be repeated again after return from the signal
1378 * handler, and SSTEP can never finish in this case.
1380 bool uprobe_deny_signal(void)
1382 struct task_struct *t = current;
1383 struct uprobe_task *utask = t->utask;
1385 if (likely(!utask || !utask->active_uprobe))
1388 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1390 if (signal_pending(t)) {
1391 spin_lock_irq(&t->sighand->siglock);
1392 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1393 spin_unlock_irq(&t->sighand->siglock);
1395 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1396 utask->state = UTASK_SSTEP_TRAPPED;
1397 set_tsk_thread_flag(t, TIF_UPROBE);
1398 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1406 * Avoid singlestepping the original instruction if the original instruction
1407 * is a NOP or can be emulated.
1409 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1411 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1412 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1414 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1419 static void mmf_recalc_uprobes(struct mm_struct *mm)
1421 struct vm_area_struct *vma;
1423 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1424 if (!valid_vma(vma, false))
1427 * This is not strictly accurate, we can race with
1428 * uprobe_unregister() and see the already removed
1429 * uprobe if delete_uprobe() was not yet called.
1430 * Or this uprobe can be filtered out.
1432 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1436 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1439 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1442 uprobe_opcode_t opcode;
1445 pagefault_disable();
1446 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1450 if (likely(result == 0))
1453 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1457 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1460 /* This needs to return true for any variant of the trap insn */
1461 return is_trap_insn(&opcode);
1464 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1466 struct mm_struct *mm = current->mm;
1467 struct uprobe *uprobe = NULL;
1468 struct vm_area_struct *vma;
1470 down_read(&mm->mmap_sem);
1471 vma = find_vma(mm, bp_vaddr);
1472 if (vma && vma->vm_start <= bp_vaddr) {
1473 if (valid_vma(vma, false)) {
1474 struct inode *inode = file_inode(vma->vm_file);
1475 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1477 uprobe = find_uprobe(inode, offset);
1481 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1486 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1487 mmf_recalc_uprobes(mm);
1488 up_read(&mm->mmap_sem);
1493 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1495 struct uprobe_consumer *uc;
1496 int remove = UPROBE_HANDLER_REMOVE;
1498 down_read(&uprobe->register_rwsem);
1499 for (uc = uprobe->consumers; uc; uc = uc->next) {
1500 int rc = uc->handler(uc, regs);
1502 WARN(rc & ~UPROBE_HANDLER_MASK,
1503 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1507 if (remove && uprobe->consumers) {
1508 WARN_ON(!uprobe_is_active(uprobe));
1509 unapply_uprobe(uprobe, current->mm);
1511 up_read(&uprobe->register_rwsem);
1515 * Run handler and ask thread to singlestep.
1516 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1518 static void handle_swbp(struct pt_regs *regs)
1520 struct uprobe *uprobe;
1521 unsigned long bp_vaddr;
1522 int uninitialized_var(is_swbp);
1524 bp_vaddr = uprobe_get_swbp_addr(regs);
1525 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1529 /* No matching uprobe; signal SIGTRAP. */
1530 send_sig(SIGTRAP, current, 0);
1533 * Either we raced with uprobe_unregister() or we can't
1534 * access this memory. The latter is only possible if
1535 * another thread plays with our ->mm. In both cases
1536 * we can simply restart. If this vma was unmapped we
1537 * can pretend this insn was not executed yet and get
1538 * the (correct) SIGSEGV after restart.
1540 instruction_pointer_set(regs, bp_vaddr);
1545 /* change it in advance for ->handler() and restart */
1546 instruction_pointer_set(regs, bp_vaddr);
1549 * TODO: move copy_insn/etc into _register and remove this hack.
1550 * After we hit the bp, _unregister + _register can install the
1551 * new and not-yet-analyzed uprobe at the same address, restart.
1553 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1554 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1557 handler_chain(uprobe, regs);
1558 if (can_skip_sstep(uprobe, regs))
1561 if (!pre_ssout(uprobe, regs, bp_vaddr))
1564 /* can_skip_sstep() succeeded, or restart if can't singlestep */
1570 * Perform required fix-ups and disable singlestep.
1571 * Allow pending signals to take effect.
1573 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1575 struct uprobe *uprobe;
1577 uprobe = utask->active_uprobe;
1578 if (utask->state == UTASK_SSTEP_ACK)
1579 arch_uprobe_post_xol(&uprobe->arch, regs);
1580 else if (utask->state == UTASK_SSTEP_TRAPPED)
1581 arch_uprobe_abort_xol(&uprobe->arch, regs);
1586 utask->active_uprobe = NULL;
1587 utask->state = UTASK_RUNNING;
1588 xol_free_insn_slot(current);
1590 spin_lock_irq(¤t->sighand->siglock);
1591 recalc_sigpending(); /* see uprobe_deny_signal() */
1592 spin_unlock_irq(¤t->sighand->siglock);
1596 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1597 * allows the thread to return from interrupt. After that handle_swbp()
1598 * sets utask->active_uprobe.
1600 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1601 * and allows the thread to return from interrupt.
1603 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1604 * uprobe_notify_resume().
1606 void uprobe_notify_resume(struct pt_regs *regs)
1608 struct uprobe_task *utask;
1610 clear_thread_flag(TIF_UPROBE);
1612 utask = current->utask;
1613 if (utask && utask->active_uprobe)
1614 handle_singlestep(utask, regs);
1620 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1621 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1623 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1625 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1628 set_thread_flag(TIF_UPROBE);
1633 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1634 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1636 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1638 struct uprobe_task *utask = current->utask;
1640 if (!current->mm || !utask || !utask->active_uprobe)
1641 /* task is currently not uprobed */
1644 utask->state = UTASK_SSTEP_ACK;
1645 set_thread_flag(TIF_UPROBE);
1649 static struct notifier_block uprobe_exception_nb = {
1650 .notifier_call = arch_uprobe_exception_notify,
1651 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1654 static int __init init_uprobes(void)
1658 for (i = 0; i < UPROBES_HASH_SZ; i++)
1659 mutex_init(&uprobes_mmap_mutex[i]);
1661 if (percpu_init_rwsem(&dup_mmap_sem))
1664 return register_die_notifier(&uprobe_exception_nb);
1666 module_init(init_uprobes);
1668 static void __exit exit_uprobes(void)
1671 module_exit(exit_uprobes);