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/rmap.h> /* anon_vma_prepare */
31 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
32 #include <linux/swap.h> /* try_to_free_swap */
33 #include <linux/ptrace.h> /* user_enable_single_step */
34 #include <linux/kdebug.h> /* notifier mechanism */
35 #include "../../mm/internal.h" /* munlock_vma_page */
36 #include <linux/percpu-rwsem.h>
38 #include <linux/uprobes.h>
40 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
41 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
43 static struct rb_root uprobes_tree = RB_ROOT;
45 * allows us to skip the uprobe_mmap if there are no uprobe events active
46 * at this time. Probably a fine grained per inode count is better?
48 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
50 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
52 #define UPROBES_HASH_SZ 13
53 /* serialize uprobe->pending_list */
54 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
55 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
57 static struct percpu_rw_semaphore dup_mmap_sem;
59 /* Have a copy of original instruction */
60 #define UPROBE_COPY_INSN 0
61 /* Can skip singlestep */
62 #define UPROBE_SKIP_SSTEP 1
65 struct rb_node rb_node; /* node in the rb tree */
67 struct rw_semaphore register_rwsem;
68 struct rw_semaphore consumer_rwsem;
69 struct list_head pending_list;
70 struct uprobe_consumer *consumers;
71 struct inode *inode; /* Also hold a ref to inode */
74 struct arch_uprobe arch;
78 * valid_vma: Verify if the specified vma is an executable vma
79 * Relax restrictions while unregistering: vm_flags might have
80 * changed after breakpoint was inserted.
81 * - is_register: indicates if we are in register context.
82 * - Return 1 if the specified virtual address is in an
85 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
87 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
92 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
95 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
97 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
100 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
102 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
106 * __replace_page - replace page in vma by new page.
107 * based on replace_page in mm/ksm.c
109 * @vma: vma that holds the pte pointing to page
110 * @addr: address the old @page is mapped at
111 * @page: the cowed page we are replacing by kpage
112 * @kpage: the modified page we replace page by
114 * Returns 0 on success, -EFAULT on failure.
116 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
117 struct page *page, struct page *kpage)
119 struct mm_struct *mm = vma->vm_mm;
123 /* For mmu_notifiers */
124 const unsigned long mmun_start = addr;
125 const unsigned long mmun_end = addr + PAGE_SIZE;
127 /* For try_to_free_swap() and munlock_vma_page() below */
130 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
132 ptep = page_check_address(page, mm, addr, &ptl, 0);
137 page_add_new_anon_rmap(kpage, vma, addr);
139 if (!PageAnon(page)) {
140 dec_mm_counter(mm, MM_FILEPAGES);
141 inc_mm_counter(mm, MM_ANONPAGES);
144 flush_cache_page(vma, addr, pte_pfn(*ptep));
145 ptep_clear_flush(vma, addr, ptep);
146 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
148 page_remove_rmap(page);
149 if (!page_mapped(page))
150 try_to_free_swap(page);
151 pte_unmap_unlock(ptep, ptl);
153 if (vma->vm_flags & VM_LOCKED)
154 munlock_vma_page(page);
159 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
165 * is_swbp_insn - check if instruction is breakpoint instruction.
166 * @insn: instruction to be checked.
167 * Default implementation of is_swbp_insn
168 * Returns true if @insn is a breakpoint instruction.
170 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
172 return *insn == UPROBE_SWBP_INSN;
175 static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode)
177 void *kaddr = kmap_atomic(page);
178 memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE);
179 kunmap_atomic(kaddr);
182 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
184 uprobe_opcode_t old_opcode;
187 copy_opcode(page, vaddr, &old_opcode);
188 is_swbp = is_swbp_insn(&old_opcode);
190 if (is_swbp_insn(new_opcode)) {
191 if (is_swbp) /* register: already installed? */
194 if (!is_swbp) /* unregister: was it changed by us? */
203 * Expect the breakpoint instruction to be the smallest size instruction for
204 * the architecture. If an arch has variable length instruction and the
205 * breakpoint instruction is not of the smallest length instruction
206 * supported by that architecture then we need to modify is_swbp_at_addr and
207 * write_opcode accordingly. This would never be a problem for archs that
208 * have fixed length instructions.
212 * write_opcode - write the opcode at a given virtual address.
213 * @mm: the probed process address space.
214 * @vaddr: the virtual address to store the opcode.
215 * @opcode: opcode to be written at @vaddr.
217 * Called with mm->mmap_sem held (for read and with a reference to
220 * For mm @mm, write the opcode at @vaddr.
221 * Return 0 (success) or a negative errno.
223 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
224 uprobe_opcode_t opcode)
226 struct page *old_page, *new_page;
227 void *vaddr_old, *vaddr_new;
228 struct vm_area_struct *vma;
232 /* Read the page with vaddr into memory */
233 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
237 ret = verify_opcode(old_page, vaddr, &opcode);
242 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
246 __SetPageUptodate(new_page);
248 /* copy the page now that we've got it stable */
249 vaddr_old = kmap_atomic(old_page);
250 vaddr_new = kmap_atomic(new_page);
252 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
253 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
255 kunmap_atomic(vaddr_new);
256 kunmap_atomic(vaddr_old);
258 ret = anon_vma_prepare(vma);
262 ret = __replace_page(vma, vaddr, old_page, new_page);
265 page_cache_release(new_page);
269 if (unlikely(ret == -EAGAIN))
275 * set_swbp - store breakpoint at a given address.
276 * @auprobe: arch specific probepoint information.
277 * @mm: the probed process address space.
278 * @vaddr: the virtual address to insert the opcode.
280 * For mm @mm, store the breakpoint instruction at @vaddr.
281 * Return 0 (success) or a negative errno.
283 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
285 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
289 * set_orig_insn - Restore the original instruction.
290 * @mm: the probed process address space.
291 * @auprobe: arch specific probepoint information.
292 * @vaddr: the virtual address to insert the opcode.
294 * For mm @mm, restore the original opcode (opcode) at @vaddr.
295 * Return 0 (success) or a negative errno.
298 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
300 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
303 static int match_uprobe(struct uprobe *l, struct uprobe *r)
305 if (l->inode < r->inode)
308 if (l->inode > r->inode)
311 if (l->offset < r->offset)
314 if (l->offset > r->offset)
320 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
322 struct uprobe u = { .inode = inode, .offset = offset };
323 struct rb_node *n = uprobes_tree.rb_node;
324 struct uprobe *uprobe;
328 uprobe = rb_entry(n, struct uprobe, rb_node);
329 match = match_uprobe(&u, uprobe);
331 atomic_inc(&uprobe->ref);
344 * Find a uprobe corresponding to a given inode:offset
345 * Acquires uprobes_treelock
347 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
349 struct uprobe *uprobe;
351 spin_lock(&uprobes_treelock);
352 uprobe = __find_uprobe(inode, offset);
353 spin_unlock(&uprobes_treelock);
358 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
360 struct rb_node **p = &uprobes_tree.rb_node;
361 struct rb_node *parent = NULL;
367 u = rb_entry(parent, struct uprobe, rb_node);
368 match = match_uprobe(uprobe, u);
375 p = &parent->rb_left;
377 p = &parent->rb_right;
382 rb_link_node(&uprobe->rb_node, parent, p);
383 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
384 /* get access + creation ref */
385 atomic_set(&uprobe->ref, 2);
391 * Acquire uprobes_treelock.
392 * Matching uprobe already exists in rbtree;
393 * increment (access refcount) and return the matching uprobe.
395 * No matching uprobe; insert the uprobe in rb_tree;
396 * get a double refcount (access + creation) and return NULL.
398 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
402 spin_lock(&uprobes_treelock);
403 u = __insert_uprobe(uprobe);
404 spin_unlock(&uprobes_treelock);
409 static void put_uprobe(struct uprobe *uprobe)
411 if (atomic_dec_and_test(&uprobe->ref))
415 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
417 struct uprobe *uprobe, *cur_uprobe;
419 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
423 uprobe->inode = igrab(inode);
424 uprobe->offset = offset;
425 init_rwsem(&uprobe->register_rwsem);
426 init_rwsem(&uprobe->consumer_rwsem);
427 /* For now assume that the instruction need not be single-stepped */
428 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
430 /* add to uprobes_tree, sorted on inode:offset */
431 cur_uprobe = insert_uprobe(uprobe);
433 /* a uprobe exists for this inode:offset combination */
443 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
445 struct uprobe_consumer *uc;
447 down_read(&uprobe->register_rwsem);
448 for (uc = uprobe->consumers; uc; uc = uc->next)
449 uc->handler(uc, regs);
450 up_read(&uprobe->register_rwsem);
453 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
455 down_write(&uprobe->consumer_rwsem);
456 uc->next = uprobe->consumers;
457 uprobe->consumers = uc;
458 up_write(&uprobe->consumer_rwsem);
462 * For uprobe @uprobe, delete the consumer @uc.
463 * Return true if the @uc is deleted successfully
466 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
468 struct uprobe_consumer **con;
471 down_write(&uprobe->consumer_rwsem);
472 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
479 up_write(&uprobe->consumer_rwsem);
485 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
486 unsigned long nbytes, loff_t offset)
496 if (!mapping->a_ops->readpage)
499 idx = offset >> PAGE_CACHE_SHIFT;
500 off = offset & ~PAGE_MASK;
503 * Ensure that the page that has the original instruction is
504 * populated and in page-cache.
506 page = read_mapping_page(mapping, idx, filp);
508 return PTR_ERR(page);
510 vaddr = kmap_atomic(page);
511 memcpy(insn, vaddr + off, nbytes);
512 kunmap_atomic(vaddr);
513 page_cache_release(page);
518 static int copy_insn(struct uprobe *uprobe, struct file *filp)
520 struct address_space *mapping;
521 unsigned long nbytes;
524 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
525 mapping = uprobe->inode->i_mapping;
527 /* Instruction at end of binary; copy only available bytes */
528 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
529 bytes = uprobe->inode->i_size - uprobe->offset;
531 bytes = MAX_UINSN_BYTES;
533 /* Instruction at the page-boundary; copy bytes in second page */
534 if (nbytes < bytes) {
535 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
536 bytes - nbytes, uprobe->offset + nbytes);
541 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
544 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
545 struct mm_struct *mm, unsigned long vaddr)
549 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
552 /* TODO: move this into _register, until then we abuse this sem. */
553 down_write(&uprobe->consumer_rwsem);
554 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
557 ret = copy_insn(uprobe, file);
562 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
565 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
569 /* write_opcode() assumes we don't cross page boundary */
570 BUG_ON((uprobe->offset & ~PAGE_MASK) +
571 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
573 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
574 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
577 up_write(&uprobe->consumer_rwsem);
582 static inline bool consumer_filter(struct uprobe_consumer *uc,
583 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
585 return !uc->filter || uc->filter(uc, ctx, mm);
588 static bool filter_chain(struct uprobe *uprobe,
589 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
591 struct uprobe_consumer *uc;
594 down_read(&uprobe->consumer_rwsem);
595 for (uc = uprobe->consumers; uc; uc = uc->next) {
596 ret = consumer_filter(uc, ctx, mm);
600 up_read(&uprobe->consumer_rwsem);
606 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
607 struct vm_area_struct *vma, unsigned long vaddr)
612 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
617 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
618 * the task can hit this breakpoint right after __replace_page().
620 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
622 set_bit(MMF_HAS_UPROBES, &mm->flags);
624 ret = set_swbp(&uprobe->arch, mm, vaddr);
626 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
627 else if (first_uprobe)
628 clear_bit(MMF_HAS_UPROBES, &mm->flags);
634 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
636 set_bit(MMF_RECALC_UPROBES, &mm->flags);
637 return set_orig_insn(&uprobe->arch, mm, vaddr);
640 static inline bool uprobe_is_active(struct uprobe *uprobe)
642 return !RB_EMPTY_NODE(&uprobe->rb_node);
645 * There could be threads that have already hit the breakpoint. They
646 * will recheck the current insn and restart if find_uprobe() fails.
647 * See find_active_uprobe().
649 static void delete_uprobe(struct uprobe *uprobe)
651 if (WARN_ON(!uprobe_is_active(uprobe)))
654 spin_lock(&uprobes_treelock);
655 rb_erase(&uprobe->rb_node, &uprobes_tree);
656 spin_unlock(&uprobes_treelock);
657 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
663 struct map_info *next;
664 struct mm_struct *mm;
668 static inline struct map_info *free_map_info(struct map_info *info)
670 struct map_info *next = info->next;
675 static struct map_info *
676 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
678 unsigned long pgoff = offset >> PAGE_SHIFT;
679 struct vm_area_struct *vma;
680 struct map_info *curr = NULL;
681 struct map_info *prev = NULL;
682 struct map_info *info;
686 mutex_lock(&mapping->i_mmap_mutex);
687 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
688 if (!valid_vma(vma, is_register))
691 if (!prev && !more) {
693 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
694 * reclaim. This is optimistic, no harm done if it fails.
696 prev = kmalloc(sizeof(struct map_info),
697 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
706 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
714 info->mm = vma->vm_mm;
715 info->vaddr = offset_to_vaddr(vma, offset);
717 mutex_unlock(&mapping->i_mmap_mutex);
729 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
731 curr = ERR_PTR(-ENOMEM);
741 prev = free_map_info(prev);
745 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
747 struct map_info *info;
750 percpu_down_write(&dup_mmap_sem);
751 info = build_map_info(uprobe->inode->i_mapping,
752 uprobe->offset, is_register);
759 struct mm_struct *mm = info->mm;
760 struct vm_area_struct *vma;
762 if (err && is_register)
765 down_write(&mm->mmap_sem);
766 vma = find_vma(mm, info->vaddr);
767 if (!vma || !valid_vma(vma, is_register) ||
768 vma->vm_file->f_mapping->host != uprobe->inode)
771 if (vma->vm_start > info->vaddr ||
772 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
776 /* consult only the "caller", new consumer. */
777 if (consumer_filter(uprobe->consumers,
778 UPROBE_FILTER_REGISTER, mm))
779 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
780 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
781 if (!filter_chain(uprobe,
782 UPROBE_FILTER_UNREGISTER, mm))
783 err |= remove_breakpoint(uprobe, mm, info->vaddr);
787 up_write(&mm->mmap_sem);
790 info = free_map_info(info);
793 percpu_up_write(&dup_mmap_sem);
797 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
799 consumer_add(uprobe, uc);
800 return register_for_each_vma(uprobe, true);
803 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
807 if (!consumer_del(uprobe, uc)) /* WARN? */
810 err = register_for_each_vma(uprobe, false);
811 /* TODO : cant unregister? schedule a worker thread */
812 if (!uprobe->consumers && !err)
813 delete_uprobe(uprobe);
817 * uprobe_register - register a probe
818 * @inode: the file in which the probe has to be placed.
819 * @offset: offset from the start of the file.
820 * @uc: information on howto handle the probe..
822 * Apart from the access refcount, uprobe_register() takes a creation
823 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
824 * inserted into the rbtree (i.e first consumer for a @inode:@offset
825 * tuple). Creation refcount stops uprobe_unregister from freeing the
826 * @uprobe even before the register operation is complete. Creation
827 * refcount is released when the last @uc for the @uprobe
830 * Return errno if it cannot successully install probes
831 * else return 0 (success)
833 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
835 struct uprobe *uprobe;
838 /* Racy, just to catch the obvious mistakes */
839 if (offset > i_size_read(inode))
843 uprobe = alloc_uprobe(inode, offset);
847 * We can race with uprobe_unregister()->delete_uprobe().
848 * Check uprobe_is_active() and retry if it is false.
850 down_write(&uprobe->register_rwsem);
852 if (likely(uprobe_is_active(uprobe))) {
853 ret = __uprobe_register(uprobe, uc);
855 __uprobe_unregister(uprobe, uc);
857 up_write(&uprobe->register_rwsem);
860 if (unlikely(ret == -EAGAIN))
866 * uprobe_unregister - unregister a already registered probe.
867 * @inode: the file in which the probe has to be removed.
868 * @offset: offset from the start of the file.
869 * @uc: identify which probe if multiple probes are colocated.
871 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
873 struct uprobe *uprobe;
875 uprobe = find_uprobe(inode, offset);
879 down_write(&uprobe->register_rwsem);
880 __uprobe_unregister(uprobe, uc);
881 up_write(&uprobe->register_rwsem);
885 static struct rb_node *
886 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
888 struct rb_node *n = uprobes_tree.rb_node;
891 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
893 if (inode < u->inode) {
895 } else if (inode > u->inode) {
900 else if (min > u->offset)
911 * For a given range in vma, build a list of probes that need to be inserted.
913 static void build_probe_list(struct inode *inode,
914 struct vm_area_struct *vma,
915 unsigned long start, unsigned long end,
916 struct list_head *head)
919 struct rb_node *n, *t;
922 INIT_LIST_HEAD(head);
923 min = vaddr_to_offset(vma, start);
924 max = min + (end - start) - 1;
926 spin_lock(&uprobes_treelock);
927 n = find_node_in_range(inode, min, max);
929 for (t = n; t; t = rb_prev(t)) {
930 u = rb_entry(t, struct uprobe, rb_node);
931 if (u->inode != inode || u->offset < min)
933 list_add(&u->pending_list, head);
936 for (t = n; (t = rb_next(t)); ) {
937 u = rb_entry(t, struct uprobe, rb_node);
938 if (u->inode != inode || u->offset > max)
940 list_add(&u->pending_list, head);
944 spin_unlock(&uprobes_treelock);
948 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
950 * Currently we ignore all errors and always return 0, the callers
951 * can't handle the failure anyway.
953 int uprobe_mmap(struct vm_area_struct *vma)
955 struct list_head tmp_list;
956 struct uprobe *uprobe, *u;
959 if (no_uprobe_events() || !valid_vma(vma, true))
962 inode = vma->vm_file->f_mapping->host;
966 mutex_lock(uprobes_mmap_hash(inode));
967 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
969 * We can race with uprobe_unregister(), this uprobe can be already
970 * removed. But in this case filter_chain() must return false, all
971 * consumers have gone away.
973 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
974 if (!fatal_signal_pending(current) &&
975 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
976 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
977 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
981 mutex_unlock(uprobes_mmap_hash(inode));
987 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
993 inode = vma->vm_file->f_mapping->host;
995 min = vaddr_to_offset(vma, start);
996 max = min + (end - start) - 1;
998 spin_lock(&uprobes_treelock);
999 n = find_node_in_range(inode, min, max);
1000 spin_unlock(&uprobes_treelock);
1006 * Called in context of a munmap of a vma.
1008 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1010 if (no_uprobe_events() || !valid_vma(vma, false))
1013 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1016 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1017 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1020 if (vma_has_uprobes(vma, start, end))
1021 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1024 /* Slot allocation for XOL */
1025 static int xol_add_vma(struct xol_area *area)
1027 struct mm_struct *mm;
1030 area->page = alloc_page(GFP_HIGHUSER);
1037 down_write(&mm->mmap_sem);
1038 if (mm->uprobes_state.xol_area)
1043 /* Try to map as high as possible, this is only a hint. */
1044 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1045 if (area->vaddr & ~PAGE_MASK) {
1050 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1051 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1055 smp_wmb(); /* pairs with get_xol_area() */
1056 mm->uprobes_state.xol_area = area;
1060 up_write(&mm->mmap_sem);
1062 __free_page(area->page);
1067 static struct xol_area *get_xol_area(struct mm_struct *mm)
1069 struct xol_area *area;
1071 area = mm->uprobes_state.xol_area;
1072 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1078 * xol_alloc_area - Allocate process's xol_area.
1079 * This area will be used for storing instructions for execution out of
1082 * Returns the allocated area or NULL.
1084 static struct xol_area *xol_alloc_area(void)
1086 struct xol_area *area;
1088 area = kzalloc(sizeof(*area), GFP_KERNEL);
1089 if (unlikely(!area))
1092 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1097 init_waitqueue_head(&area->wq);
1098 if (!xol_add_vma(area))
1102 kfree(area->bitmap);
1105 return get_xol_area(current->mm);
1109 * uprobe_clear_state - Free the area allocated for slots.
1111 void uprobe_clear_state(struct mm_struct *mm)
1113 struct xol_area *area = mm->uprobes_state.xol_area;
1118 put_page(area->page);
1119 kfree(area->bitmap);
1123 void uprobe_start_dup_mmap(void)
1125 percpu_down_read(&dup_mmap_sem);
1128 void uprobe_end_dup_mmap(void)
1130 percpu_up_read(&dup_mmap_sem);
1133 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1135 newmm->uprobes_state.xol_area = NULL;
1137 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1138 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1139 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1140 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1145 * - search for a free slot.
1147 static unsigned long xol_take_insn_slot(struct xol_area *area)
1149 unsigned long slot_addr;
1153 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1154 if (slot_nr < UINSNS_PER_PAGE) {
1155 if (!test_and_set_bit(slot_nr, area->bitmap))
1158 slot_nr = UINSNS_PER_PAGE;
1161 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1162 } while (slot_nr >= UINSNS_PER_PAGE);
1164 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1165 atomic_inc(&area->slot_count);
1171 * xol_get_insn_slot - If was not allocated a slot, then
1173 * Returns the allocated slot address or 0.
1175 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1177 struct xol_area *area;
1178 unsigned long offset;
1181 area = get_xol_area(current->mm);
1183 area = xol_alloc_area();
1187 current->utask->xol_vaddr = xol_take_insn_slot(area);
1190 * Initialize the slot if xol_vaddr points to valid
1193 if (unlikely(!current->utask->xol_vaddr))
1196 current->utask->vaddr = slot_addr;
1197 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1198 vaddr = kmap_atomic(area->page);
1199 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1200 kunmap_atomic(vaddr);
1202 * We probably need flush_icache_user_range() but it needs vma.
1203 * This should work on supported architectures too.
1205 flush_dcache_page(area->page);
1207 return current->utask->xol_vaddr;
1211 * xol_free_insn_slot - If slot was earlier allocated by
1212 * @xol_get_insn_slot(), make the slot available for
1213 * subsequent requests.
1215 static void xol_free_insn_slot(struct task_struct *tsk)
1217 struct xol_area *area;
1218 unsigned long vma_end;
1219 unsigned long slot_addr;
1221 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1224 slot_addr = tsk->utask->xol_vaddr;
1226 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1229 area = tsk->mm->uprobes_state.xol_area;
1230 vma_end = area->vaddr + PAGE_SIZE;
1231 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1232 unsigned long offset;
1235 offset = slot_addr - area->vaddr;
1236 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1237 if (slot_nr >= UINSNS_PER_PAGE)
1240 clear_bit(slot_nr, area->bitmap);
1241 atomic_dec(&area->slot_count);
1242 if (waitqueue_active(&area->wq))
1245 tsk->utask->xol_vaddr = 0;
1250 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1251 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1253 * Return the address of the breakpoint instruction.
1255 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1257 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1261 * Called with no locks held.
1262 * Called in context of a exiting or a exec-ing thread.
1264 void uprobe_free_utask(struct task_struct *t)
1266 struct uprobe_task *utask = t->utask;
1271 if (utask->active_uprobe)
1272 put_uprobe(utask->active_uprobe);
1274 xol_free_insn_slot(t);
1280 * Called in context of a new clone/fork from copy_process.
1282 void uprobe_copy_process(struct task_struct *t)
1288 * Allocate a uprobe_task object for the task.
1289 * Called when the thread hits a breakpoint for the first time.
1292 * - pointer to new uprobe_task on success
1295 static struct uprobe_task *add_utask(void)
1297 struct uprobe_task *utask;
1299 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1300 if (unlikely(!utask))
1303 current->utask = utask;
1307 /* Prepare to single-step probed instruction out of line. */
1309 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1311 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1318 * If we are singlestepping, then ensure this thread is not connected to
1319 * non-fatal signals until completion of singlestep. When xol insn itself
1320 * triggers the signal, restart the original insn even if the task is
1321 * already SIGKILL'ed (since coredump should report the correct ip). This
1322 * is even more important if the task has a handler for SIGSEGV/etc, The
1323 * _same_ instruction should be repeated again after return from the signal
1324 * handler, and SSTEP can never finish in this case.
1326 bool uprobe_deny_signal(void)
1328 struct task_struct *t = current;
1329 struct uprobe_task *utask = t->utask;
1331 if (likely(!utask || !utask->active_uprobe))
1334 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1336 if (signal_pending(t)) {
1337 spin_lock_irq(&t->sighand->siglock);
1338 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1339 spin_unlock_irq(&t->sighand->siglock);
1341 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1342 utask->state = UTASK_SSTEP_TRAPPED;
1343 set_tsk_thread_flag(t, TIF_UPROBE);
1344 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1352 * Avoid singlestepping the original instruction if the original instruction
1353 * is a NOP or can be emulated.
1355 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1357 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1358 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1360 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1365 static void mmf_recalc_uprobes(struct mm_struct *mm)
1367 struct vm_area_struct *vma;
1369 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1370 if (!valid_vma(vma, false))
1373 * This is not strictly accurate, we can race with
1374 * uprobe_unregister() and see the already removed
1375 * uprobe if delete_uprobe() was not yet called.
1376 * Or this uprobe can be filtered out.
1378 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1382 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1385 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1388 uprobe_opcode_t opcode;
1391 pagefault_disable();
1392 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1396 if (likely(result == 0))
1399 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1403 copy_opcode(page, vaddr, &opcode);
1406 return is_swbp_insn(&opcode);
1409 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1411 struct mm_struct *mm = current->mm;
1412 struct uprobe *uprobe = NULL;
1413 struct vm_area_struct *vma;
1415 down_read(&mm->mmap_sem);
1416 vma = find_vma(mm, bp_vaddr);
1417 if (vma && vma->vm_start <= bp_vaddr) {
1418 if (valid_vma(vma, false)) {
1419 struct inode *inode = vma->vm_file->f_mapping->host;
1420 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1422 uprobe = find_uprobe(inode, offset);
1426 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1431 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1432 mmf_recalc_uprobes(mm);
1433 up_read(&mm->mmap_sem);
1439 * Run handler and ask thread to singlestep.
1440 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1442 static void handle_swbp(struct pt_regs *regs)
1444 struct uprobe_task *utask;
1445 struct uprobe *uprobe;
1446 unsigned long bp_vaddr;
1447 int uninitialized_var(is_swbp);
1449 bp_vaddr = uprobe_get_swbp_addr(regs);
1450 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1454 /* No matching uprobe; signal SIGTRAP. */
1455 send_sig(SIGTRAP, current, 0);
1458 * Either we raced with uprobe_unregister() or we can't
1459 * access this memory. The latter is only possible if
1460 * another thread plays with our ->mm. In both cases
1461 * we can simply restart. If this vma was unmapped we
1462 * can pretend this insn was not executed yet and get
1463 * the (correct) SIGSEGV after restart.
1465 instruction_pointer_set(regs, bp_vaddr);
1470 * TODO: move copy_insn/etc into _register and remove this hack.
1471 * After we hit the bp, _unregister + _register can install the
1472 * new and not-yet-analyzed uprobe at the same address, restart.
1474 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1475 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1478 utask = current->utask;
1480 utask = add_utask();
1481 /* Cannot allocate; re-execute the instruction. */
1486 handler_chain(uprobe, regs);
1487 if (can_skip_sstep(uprobe, regs))
1490 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1491 utask->active_uprobe = uprobe;
1492 utask->state = UTASK_SSTEP;
1498 * cannot singlestep; cannot skip instruction;
1499 * re-execute the instruction.
1501 instruction_pointer_set(regs, bp_vaddr);
1507 * Perform required fix-ups and disable singlestep.
1508 * Allow pending signals to take effect.
1510 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1512 struct uprobe *uprobe;
1514 uprobe = utask->active_uprobe;
1515 if (utask->state == UTASK_SSTEP_ACK)
1516 arch_uprobe_post_xol(&uprobe->arch, regs);
1517 else if (utask->state == UTASK_SSTEP_TRAPPED)
1518 arch_uprobe_abort_xol(&uprobe->arch, regs);
1523 utask->active_uprobe = NULL;
1524 utask->state = UTASK_RUNNING;
1525 xol_free_insn_slot(current);
1527 spin_lock_irq(¤t->sighand->siglock);
1528 recalc_sigpending(); /* see uprobe_deny_signal() */
1529 spin_unlock_irq(¤t->sighand->siglock);
1533 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1534 * allows the thread to return from interrupt. After that handle_swbp()
1535 * sets utask->active_uprobe.
1537 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1538 * and allows the thread to return from interrupt.
1540 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1541 * uprobe_notify_resume().
1543 void uprobe_notify_resume(struct pt_regs *regs)
1545 struct uprobe_task *utask;
1547 clear_thread_flag(TIF_UPROBE);
1549 utask = current->utask;
1550 if (utask && utask->active_uprobe)
1551 handle_singlestep(utask, regs);
1557 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1558 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1560 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1562 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1565 set_thread_flag(TIF_UPROBE);
1570 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1571 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1573 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1575 struct uprobe_task *utask = current->utask;
1577 if (!current->mm || !utask || !utask->active_uprobe)
1578 /* task is currently not uprobed */
1581 utask->state = UTASK_SSTEP_ACK;
1582 set_thread_flag(TIF_UPROBE);
1586 static struct notifier_block uprobe_exception_nb = {
1587 .notifier_call = arch_uprobe_exception_notify,
1588 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1591 static int __init init_uprobes(void)
1595 for (i = 0; i < UPROBES_HASH_SZ; i++)
1596 mutex_init(&uprobes_mmap_mutex[i]);
1598 if (percpu_init_rwsem(&dup_mmap_sem))
1601 return register_die_notifier(&uprobe_exception_nb);
1603 module_init(init_uprobes);
1605 static void __exit exit_uprobes(void)
1608 module_exit(exit_uprobes);