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 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
47 #define UPROBES_HASH_SZ 13
50 * We need separate register/unregister and mmap/munmap lock hashes because
51 * of mmap_sem nesting.
53 * uprobe_register() needs to install probes on (potentially) all processes
54 * and thus needs to acquire multiple mmap_sems (consequtively, not
55 * concurrently), whereas uprobe_mmap() is called while holding mmap_sem
56 * for the particular process doing the mmap.
58 * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem
59 * because of lock order against i_mmap_mutex. This means there's a hole in
60 * the register vma iteration where a mmap() can happen.
62 * Thus uprobe_register() can race with uprobe_mmap() and we can try and
63 * install a probe where one is already installed.
66 /* serialize (un)register */
67 static struct mutex uprobes_mutex[UPROBES_HASH_SZ];
69 #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
71 /* serialize uprobe->pending_list */
72 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
73 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
75 static struct percpu_rw_semaphore dup_mmap_sem;
78 * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe
79 * events active at this time. Probably a fine grained per inode count is
82 static atomic_t uprobe_events = ATOMIC_INIT(0);
84 /* Have a copy of original instruction */
85 #define UPROBE_COPY_INSN 0
86 /* Can skip singlestep */
87 #define UPROBE_SKIP_SSTEP 1
90 struct rb_node rb_node; /* node in the rb tree */
92 struct rw_semaphore register_rwsem;
93 struct rw_semaphore consumer_rwsem;
94 struct list_head pending_list;
95 struct uprobe_consumer *consumers;
96 struct inode *inode; /* Also hold a ref to inode */
99 struct arch_uprobe arch;
103 * valid_vma: Verify if the specified vma is an executable vma
104 * Relax restrictions while unregistering: vm_flags might have
105 * changed after breakpoint was inserted.
106 * - is_register: indicates if we are in register context.
107 * - Return 1 if the specified virtual address is in an
110 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
112 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
117 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
120 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
122 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
125 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
127 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
131 * __replace_page - replace page in vma by new page.
132 * based on replace_page in mm/ksm.c
134 * @vma: vma that holds the pte pointing to page
135 * @addr: address the old @page is mapped at
136 * @page: the cowed page we are replacing by kpage
137 * @kpage: the modified page we replace page by
139 * Returns 0 on success, -EFAULT on failure.
141 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
142 struct page *page, struct page *kpage)
144 struct mm_struct *mm = vma->vm_mm;
148 /* For mmu_notifiers */
149 const unsigned long mmun_start = addr;
150 const unsigned long mmun_end = addr + PAGE_SIZE;
152 /* For try_to_free_swap() and munlock_vma_page() below */
155 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
157 ptep = page_check_address(page, mm, addr, &ptl, 0);
162 page_add_new_anon_rmap(kpage, vma, addr);
164 if (!PageAnon(page)) {
165 dec_mm_counter(mm, MM_FILEPAGES);
166 inc_mm_counter(mm, MM_ANONPAGES);
169 flush_cache_page(vma, addr, pte_pfn(*ptep));
170 ptep_clear_flush(vma, addr, ptep);
171 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
173 page_remove_rmap(page);
174 if (!page_mapped(page))
175 try_to_free_swap(page);
176 pte_unmap_unlock(ptep, ptl);
178 if (vma->vm_flags & VM_LOCKED)
179 munlock_vma_page(page);
184 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
190 * is_swbp_insn - check if instruction is breakpoint instruction.
191 * @insn: instruction to be checked.
192 * Default implementation of is_swbp_insn
193 * Returns true if @insn is a breakpoint instruction.
195 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
197 return *insn == UPROBE_SWBP_INSN;
200 static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode)
202 void *kaddr = kmap_atomic(page);
203 memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE);
204 kunmap_atomic(kaddr);
207 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
209 uprobe_opcode_t old_opcode;
212 copy_opcode(page, vaddr, &old_opcode);
213 is_swbp = is_swbp_insn(&old_opcode);
215 if (is_swbp_insn(new_opcode)) {
216 if (is_swbp) /* register: already installed? */
219 if (!is_swbp) /* unregister: was it changed by us? */
228 * Expect the breakpoint instruction to be the smallest size instruction for
229 * the architecture. If an arch has variable length instruction and the
230 * breakpoint instruction is not of the smallest length instruction
231 * supported by that architecture then we need to modify is_swbp_at_addr and
232 * write_opcode accordingly. This would never be a problem for archs that
233 * have fixed length instructions.
237 * write_opcode - write the opcode at a given virtual address.
238 * @mm: the probed process address space.
239 * @vaddr: the virtual address to store the opcode.
240 * @opcode: opcode to be written at @vaddr.
242 * Called with mm->mmap_sem held (for read and with a reference to
245 * For mm @mm, write the opcode at @vaddr.
246 * Return 0 (success) or a negative errno.
248 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
249 uprobe_opcode_t opcode)
251 struct page *old_page, *new_page;
252 void *vaddr_old, *vaddr_new;
253 struct vm_area_struct *vma;
257 /* Read the page with vaddr into memory */
258 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
262 ret = verify_opcode(old_page, vaddr, &opcode);
267 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
271 __SetPageUptodate(new_page);
273 /* copy the page now that we've got it stable */
274 vaddr_old = kmap_atomic(old_page);
275 vaddr_new = kmap_atomic(new_page);
277 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
278 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
280 kunmap_atomic(vaddr_new);
281 kunmap_atomic(vaddr_old);
283 ret = anon_vma_prepare(vma);
287 ret = __replace_page(vma, vaddr, old_page, new_page);
290 page_cache_release(new_page);
294 if (unlikely(ret == -EAGAIN))
300 * set_swbp - store breakpoint at a given address.
301 * @auprobe: arch specific probepoint information.
302 * @mm: the probed process address space.
303 * @vaddr: the virtual address to insert the opcode.
305 * For mm @mm, store the breakpoint instruction at @vaddr.
306 * Return 0 (success) or a negative errno.
308 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
310 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
314 * set_orig_insn - Restore the original instruction.
315 * @mm: the probed process address space.
316 * @auprobe: arch specific probepoint information.
317 * @vaddr: the virtual address to insert the opcode.
319 * For mm @mm, restore the original opcode (opcode) at @vaddr.
320 * Return 0 (success) or a negative errno.
323 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
325 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
328 static int match_uprobe(struct uprobe *l, struct uprobe *r)
330 if (l->inode < r->inode)
333 if (l->inode > r->inode)
336 if (l->offset < r->offset)
339 if (l->offset > r->offset)
345 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
347 struct uprobe u = { .inode = inode, .offset = offset };
348 struct rb_node *n = uprobes_tree.rb_node;
349 struct uprobe *uprobe;
353 uprobe = rb_entry(n, struct uprobe, rb_node);
354 match = match_uprobe(&u, uprobe);
356 atomic_inc(&uprobe->ref);
369 * Find a uprobe corresponding to a given inode:offset
370 * Acquires uprobes_treelock
372 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
374 struct uprobe *uprobe;
376 spin_lock(&uprobes_treelock);
377 uprobe = __find_uprobe(inode, offset);
378 spin_unlock(&uprobes_treelock);
383 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
385 struct rb_node **p = &uprobes_tree.rb_node;
386 struct rb_node *parent = NULL;
392 u = rb_entry(parent, struct uprobe, rb_node);
393 match = match_uprobe(uprobe, u);
400 p = &parent->rb_left;
402 p = &parent->rb_right;
407 rb_link_node(&uprobe->rb_node, parent, p);
408 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
409 /* get access + creation ref */
410 atomic_set(&uprobe->ref, 2);
416 * Acquire uprobes_treelock.
417 * Matching uprobe already exists in rbtree;
418 * increment (access refcount) and return the matching uprobe.
420 * No matching uprobe; insert the uprobe in rb_tree;
421 * get a double refcount (access + creation) and return NULL.
423 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
427 spin_lock(&uprobes_treelock);
428 u = __insert_uprobe(uprobe);
429 spin_unlock(&uprobes_treelock);
434 static void put_uprobe(struct uprobe *uprobe)
436 if (atomic_dec_and_test(&uprobe->ref))
440 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
442 struct uprobe *uprobe, *cur_uprobe;
444 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
448 uprobe->inode = igrab(inode);
449 uprobe->offset = offset;
450 init_rwsem(&uprobe->register_rwsem);
451 init_rwsem(&uprobe->consumer_rwsem);
452 /* For now assume that the instruction need not be single-stepped */
453 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
455 /* add to uprobes_tree, sorted on inode:offset */
456 cur_uprobe = insert_uprobe(uprobe);
458 /* a uprobe exists for this inode:offset combination */
464 atomic_inc(&uprobe_events);
470 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
472 struct uprobe_consumer *uc;
474 down_read(&uprobe->register_rwsem);
475 for (uc = uprobe->consumers; uc; uc = uc->next)
476 uc->handler(uc, regs);
477 up_read(&uprobe->register_rwsem);
480 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
482 down_write(&uprobe->consumer_rwsem);
483 uc->next = uprobe->consumers;
484 uprobe->consumers = uc;
485 up_write(&uprobe->consumer_rwsem);
489 * For uprobe @uprobe, delete the consumer @uc.
490 * Return true if the @uc is deleted successfully
493 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
495 struct uprobe_consumer **con;
498 down_write(&uprobe->consumer_rwsem);
499 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
506 up_write(&uprobe->consumer_rwsem);
512 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
513 unsigned long nbytes, loff_t offset)
523 if (!mapping->a_ops->readpage)
526 idx = offset >> PAGE_CACHE_SHIFT;
527 off = offset & ~PAGE_MASK;
530 * Ensure that the page that has the original instruction is
531 * populated and in page-cache.
533 page = read_mapping_page(mapping, idx, filp);
535 return PTR_ERR(page);
537 vaddr = kmap_atomic(page);
538 memcpy(insn, vaddr + off, nbytes);
539 kunmap_atomic(vaddr);
540 page_cache_release(page);
545 static int copy_insn(struct uprobe *uprobe, struct file *filp)
547 struct address_space *mapping;
548 unsigned long nbytes;
551 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
552 mapping = uprobe->inode->i_mapping;
554 /* Instruction at end of binary; copy only available bytes */
555 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
556 bytes = uprobe->inode->i_size - uprobe->offset;
558 bytes = MAX_UINSN_BYTES;
560 /* Instruction at the page-boundary; copy bytes in second page */
561 if (nbytes < bytes) {
562 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
563 bytes - nbytes, uprobe->offset + nbytes);
568 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
571 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
572 struct mm_struct *mm, unsigned long vaddr)
576 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
579 /* TODO: move this into _register, until then we abuse this sem. */
580 down_write(&uprobe->consumer_rwsem);
581 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
584 ret = copy_insn(uprobe, file);
589 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
592 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
596 /* write_opcode() assumes we don't cross page boundary */
597 BUG_ON((uprobe->offset & ~PAGE_MASK) +
598 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
600 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
601 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
604 up_write(&uprobe->consumer_rwsem);
609 static bool filter_chain(struct uprobe *uprobe)
611 struct uprobe_consumer *uc;
614 down_read(&uprobe->consumer_rwsem);
615 for (uc = uprobe->consumers; uc; uc = uc->next) {
616 /* TODO: ret = uc->filter(...) */
621 up_read(&uprobe->consumer_rwsem);
627 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
628 struct vm_area_struct *vma, unsigned long vaddr)
634 * If probe is being deleted, unregister thread could be done with
635 * the vma-rmap-walk through. Adding a probe now can be fatal since
636 * nobody will be able to cleanup. But in this case filter_chain()
637 * must return false, all consumers have gone away.
639 if (!filter_chain(uprobe))
642 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
647 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
648 * the task can hit this breakpoint right after __replace_page().
650 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
652 set_bit(MMF_HAS_UPROBES, &mm->flags);
654 ret = set_swbp(&uprobe->arch, mm, vaddr);
656 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
657 else if (first_uprobe)
658 clear_bit(MMF_HAS_UPROBES, &mm->flags);
664 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
666 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
669 if (filter_chain(uprobe))
672 set_bit(MMF_RECALC_UPROBES, &mm->flags);
673 return set_orig_insn(&uprobe->arch, mm, vaddr);
677 * There could be threads that have already hit the breakpoint. They
678 * will recheck the current insn and restart if find_uprobe() fails.
679 * See find_active_uprobe().
681 static void delete_uprobe(struct uprobe *uprobe)
683 spin_lock(&uprobes_treelock);
684 rb_erase(&uprobe->rb_node, &uprobes_tree);
685 spin_unlock(&uprobes_treelock);
688 atomic_dec(&uprobe_events);
692 struct map_info *next;
693 struct mm_struct *mm;
697 static inline struct map_info *free_map_info(struct map_info *info)
699 struct map_info *next = info->next;
704 static struct map_info *
705 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
707 unsigned long pgoff = offset >> PAGE_SHIFT;
708 struct vm_area_struct *vma;
709 struct map_info *curr = NULL;
710 struct map_info *prev = NULL;
711 struct map_info *info;
715 mutex_lock(&mapping->i_mmap_mutex);
716 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
717 if (!valid_vma(vma, is_register))
720 if (!prev && !more) {
722 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
723 * reclaim. This is optimistic, no harm done if it fails.
725 prev = kmalloc(sizeof(struct map_info),
726 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
735 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
743 info->mm = vma->vm_mm;
744 info->vaddr = offset_to_vaddr(vma, offset);
746 mutex_unlock(&mapping->i_mmap_mutex);
758 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
760 curr = ERR_PTR(-ENOMEM);
770 prev = free_map_info(prev);
774 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
776 struct map_info *info;
779 percpu_down_write(&dup_mmap_sem);
780 info = build_map_info(uprobe->inode->i_mapping,
781 uprobe->offset, is_register);
788 struct mm_struct *mm = info->mm;
789 struct vm_area_struct *vma;
791 if (err && is_register)
794 down_write(&mm->mmap_sem);
795 vma = find_vma(mm, info->vaddr);
796 if (!vma || !valid_vma(vma, is_register) ||
797 vma->vm_file->f_mapping->host != uprobe->inode)
800 if (vma->vm_start > info->vaddr ||
801 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
805 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
807 err |= remove_breakpoint(uprobe, mm, info->vaddr);
810 up_write(&mm->mmap_sem);
813 info = free_map_info(info);
816 percpu_up_write(&dup_mmap_sem);
820 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
822 consumer_add(uprobe, uc);
823 return register_for_each_vma(uprobe, true);
826 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
830 if (!consumer_del(uprobe, uc)) /* WARN? */
833 err = register_for_each_vma(uprobe, false);
834 /* TODO : cant unregister? schedule a worker thread */
835 if (!uprobe->consumers && !err)
836 delete_uprobe(uprobe);
840 * uprobe_register - register a probe
841 * @inode: the file in which the probe has to be placed.
842 * @offset: offset from the start of the file.
843 * @uc: information on howto handle the probe..
845 * Apart from the access refcount, uprobe_register() takes a creation
846 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
847 * inserted into the rbtree (i.e first consumer for a @inode:@offset
848 * tuple). Creation refcount stops uprobe_unregister from freeing the
849 * @uprobe even before the register operation is complete. Creation
850 * refcount is released when the last @uc for the @uprobe
853 * Return errno if it cannot successully install probes
854 * else return 0 (success)
856 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
858 struct uprobe *uprobe;
861 /* Racy, just to catch the obvious mistakes */
862 if (offset > i_size_read(inode))
866 mutex_lock(uprobes_hash(inode));
867 uprobe = alloc_uprobe(inode, offset);
869 down_write(&uprobe->register_rwsem);
870 ret = __uprobe_register(uprobe, uc);
872 __uprobe_unregister(uprobe, uc);
873 up_write(&uprobe->register_rwsem);
875 mutex_unlock(uprobes_hash(inode));
883 * uprobe_unregister - unregister a already registered probe.
884 * @inode: the file in which the probe has to be removed.
885 * @offset: offset from the start of the file.
886 * @uc: identify which probe if multiple probes are colocated.
888 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
890 struct uprobe *uprobe;
892 uprobe = find_uprobe(inode, offset);
896 mutex_lock(uprobes_hash(inode));
897 down_write(&uprobe->register_rwsem);
898 __uprobe_unregister(uprobe, uc);
899 up_write(&uprobe->register_rwsem);
900 mutex_unlock(uprobes_hash(inode));
904 static struct rb_node *
905 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
907 struct rb_node *n = uprobes_tree.rb_node;
910 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
912 if (inode < u->inode) {
914 } else if (inode > u->inode) {
919 else if (min > u->offset)
930 * For a given range in vma, build a list of probes that need to be inserted.
932 static void build_probe_list(struct inode *inode,
933 struct vm_area_struct *vma,
934 unsigned long start, unsigned long end,
935 struct list_head *head)
938 struct rb_node *n, *t;
941 INIT_LIST_HEAD(head);
942 min = vaddr_to_offset(vma, start);
943 max = min + (end - start) - 1;
945 spin_lock(&uprobes_treelock);
946 n = find_node_in_range(inode, min, max);
948 for (t = n; t; t = rb_prev(t)) {
949 u = rb_entry(t, struct uprobe, rb_node);
950 if (u->inode != inode || u->offset < min)
952 list_add(&u->pending_list, head);
955 for (t = n; (t = rb_next(t)); ) {
956 u = rb_entry(t, struct uprobe, rb_node);
957 if (u->inode != inode || u->offset > max)
959 list_add(&u->pending_list, head);
963 spin_unlock(&uprobes_treelock);
967 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
969 * Currently we ignore all errors and always return 0, the callers
970 * can't handle the failure anyway.
972 int uprobe_mmap(struct vm_area_struct *vma)
974 struct list_head tmp_list;
975 struct uprobe *uprobe, *u;
978 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
981 inode = vma->vm_file->f_mapping->host;
985 mutex_lock(uprobes_mmap_hash(inode));
986 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
988 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
989 if (!fatal_signal_pending(current)) {
990 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
991 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
995 mutex_unlock(uprobes_mmap_hash(inode));
1001 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1004 struct inode *inode;
1007 inode = vma->vm_file->f_mapping->host;
1009 min = vaddr_to_offset(vma, start);
1010 max = min + (end - start) - 1;
1012 spin_lock(&uprobes_treelock);
1013 n = find_node_in_range(inode, min, max);
1014 spin_unlock(&uprobes_treelock);
1020 * Called in context of a munmap of a vma.
1022 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1024 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1027 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1030 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1031 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1034 if (vma_has_uprobes(vma, start, end))
1035 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1038 /* Slot allocation for XOL */
1039 static int xol_add_vma(struct xol_area *area)
1041 struct mm_struct *mm;
1044 area->page = alloc_page(GFP_HIGHUSER);
1051 down_write(&mm->mmap_sem);
1052 if (mm->uprobes_state.xol_area)
1057 /* Try to map as high as possible, this is only a hint. */
1058 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1059 if (area->vaddr & ~PAGE_MASK) {
1064 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1065 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1069 smp_wmb(); /* pairs with get_xol_area() */
1070 mm->uprobes_state.xol_area = area;
1074 up_write(&mm->mmap_sem);
1076 __free_page(area->page);
1081 static struct xol_area *get_xol_area(struct mm_struct *mm)
1083 struct xol_area *area;
1085 area = mm->uprobes_state.xol_area;
1086 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1092 * xol_alloc_area - Allocate process's xol_area.
1093 * This area will be used for storing instructions for execution out of
1096 * Returns the allocated area or NULL.
1098 static struct xol_area *xol_alloc_area(void)
1100 struct xol_area *area;
1102 area = kzalloc(sizeof(*area), GFP_KERNEL);
1103 if (unlikely(!area))
1106 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1111 init_waitqueue_head(&area->wq);
1112 if (!xol_add_vma(area))
1116 kfree(area->bitmap);
1119 return get_xol_area(current->mm);
1123 * uprobe_clear_state - Free the area allocated for slots.
1125 void uprobe_clear_state(struct mm_struct *mm)
1127 struct xol_area *area = mm->uprobes_state.xol_area;
1132 put_page(area->page);
1133 kfree(area->bitmap);
1137 void uprobe_start_dup_mmap(void)
1139 percpu_down_read(&dup_mmap_sem);
1142 void uprobe_end_dup_mmap(void)
1144 percpu_up_read(&dup_mmap_sem);
1147 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1149 newmm->uprobes_state.xol_area = NULL;
1151 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1152 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1153 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1154 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1159 * - search for a free slot.
1161 static unsigned long xol_take_insn_slot(struct xol_area *area)
1163 unsigned long slot_addr;
1167 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1168 if (slot_nr < UINSNS_PER_PAGE) {
1169 if (!test_and_set_bit(slot_nr, area->bitmap))
1172 slot_nr = UINSNS_PER_PAGE;
1175 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1176 } while (slot_nr >= UINSNS_PER_PAGE);
1178 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1179 atomic_inc(&area->slot_count);
1185 * xol_get_insn_slot - If was not allocated a slot, then
1187 * Returns the allocated slot address or 0.
1189 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1191 struct xol_area *area;
1192 unsigned long offset;
1195 area = get_xol_area(current->mm);
1197 area = xol_alloc_area();
1201 current->utask->xol_vaddr = xol_take_insn_slot(area);
1204 * Initialize the slot if xol_vaddr points to valid
1207 if (unlikely(!current->utask->xol_vaddr))
1210 current->utask->vaddr = slot_addr;
1211 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1212 vaddr = kmap_atomic(area->page);
1213 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1214 kunmap_atomic(vaddr);
1216 * We probably need flush_icache_user_range() but it needs vma.
1217 * This should work on supported architectures too.
1219 flush_dcache_page(area->page);
1221 return current->utask->xol_vaddr;
1225 * xol_free_insn_slot - If slot was earlier allocated by
1226 * @xol_get_insn_slot(), make the slot available for
1227 * subsequent requests.
1229 static void xol_free_insn_slot(struct task_struct *tsk)
1231 struct xol_area *area;
1232 unsigned long vma_end;
1233 unsigned long slot_addr;
1235 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1238 slot_addr = tsk->utask->xol_vaddr;
1240 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1243 area = tsk->mm->uprobes_state.xol_area;
1244 vma_end = area->vaddr + PAGE_SIZE;
1245 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1246 unsigned long offset;
1249 offset = slot_addr - area->vaddr;
1250 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1251 if (slot_nr >= UINSNS_PER_PAGE)
1254 clear_bit(slot_nr, area->bitmap);
1255 atomic_dec(&area->slot_count);
1256 if (waitqueue_active(&area->wq))
1259 tsk->utask->xol_vaddr = 0;
1264 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1265 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1267 * Return the address of the breakpoint instruction.
1269 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1271 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1275 * Called with no locks held.
1276 * Called in context of a exiting or a exec-ing thread.
1278 void uprobe_free_utask(struct task_struct *t)
1280 struct uprobe_task *utask = t->utask;
1285 if (utask->active_uprobe)
1286 put_uprobe(utask->active_uprobe);
1288 xol_free_insn_slot(t);
1294 * Called in context of a new clone/fork from copy_process.
1296 void uprobe_copy_process(struct task_struct *t)
1302 * Allocate a uprobe_task object for the task.
1303 * Called when the thread hits a breakpoint for the first time.
1306 * - pointer to new uprobe_task on success
1309 static struct uprobe_task *add_utask(void)
1311 struct uprobe_task *utask;
1313 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1314 if (unlikely(!utask))
1317 current->utask = utask;
1321 /* Prepare to single-step probed instruction out of line. */
1323 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1325 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1332 * If we are singlestepping, then ensure this thread is not connected to
1333 * non-fatal signals until completion of singlestep. When xol insn itself
1334 * triggers the signal, restart the original insn even if the task is
1335 * already SIGKILL'ed (since coredump should report the correct ip). This
1336 * is even more important if the task has a handler for SIGSEGV/etc, The
1337 * _same_ instruction should be repeated again after return from the signal
1338 * handler, and SSTEP can never finish in this case.
1340 bool uprobe_deny_signal(void)
1342 struct task_struct *t = current;
1343 struct uprobe_task *utask = t->utask;
1345 if (likely(!utask || !utask->active_uprobe))
1348 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1350 if (signal_pending(t)) {
1351 spin_lock_irq(&t->sighand->siglock);
1352 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1353 spin_unlock_irq(&t->sighand->siglock);
1355 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1356 utask->state = UTASK_SSTEP_TRAPPED;
1357 set_tsk_thread_flag(t, TIF_UPROBE);
1358 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1366 * Avoid singlestepping the original instruction if the original instruction
1367 * is a NOP or can be emulated.
1369 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1371 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1372 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1374 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1379 static void mmf_recalc_uprobes(struct mm_struct *mm)
1381 struct vm_area_struct *vma;
1383 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1384 if (!valid_vma(vma, false))
1387 * This is not strictly accurate, we can race with
1388 * uprobe_unregister() and see the already removed
1389 * uprobe if delete_uprobe() was not yet called.
1390 * Or this uprobe can be filtered out.
1392 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1396 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1399 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1402 uprobe_opcode_t opcode;
1405 pagefault_disable();
1406 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1410 if (likely(result == 0))
1413 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1417 copy_opcode(page, vaddr, &opcode);
1420 return is_swbp_insn(&opcode);
1423 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1425 struct mm_struct *mm = current->mm;
1426 struct uprobe *uprobe = NULL;
1427 struct vm_area_struct *vma;
1429 down_read(&mm->mmap_sem);
1430 vma = find_vma(mm, bp_vaddr);
1431 if (vma && vma->vm_start <= bp_vaddr) {
1432 if (valid_vma(vma, false)) {
1433 struct inode *inode = vma->vm_file->f_mapping->host;
1434 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1436 uprobe = find_uprobe(inode, offset);
1440 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1445 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1446 mmf_recalc_uprobes(mm);
1447 up_read(&mm->mmap_sem);
1453 * Run handler and ask thread to singlestep.
1454 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1456 static void handle_swbp(struct pt_regs *regs)
1458 struct uprobe_task *utask;
1459 struct uprobe *uprobe;
1460 unsigned long bp_vaddr;
1461 int uninitialized_var(is_swbp);
1463 bp_vaddr = uprobe_get_swbp_addr(regs);
1464 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1468 /* No matching uprobe; signal SIGTRAP. */
1469 send_sig(SIGTRAP, current, 0);
1472 * Either we raced with uprobe_unregister() or we can't
1473 * access this memory. The latter is only possible if
1474 * another thread plays with our ->mm. In both cases
1475 * we can simply restart. If this vma was unmapped we
1476 * can pretend this insn was not executed yet and get
1477 * the (correct) SIGSEGV after restart.
1479 instruction_pointer_set(regs, bp_vaddr);
1484 * TODO: move copy_insn/etc into _register and remove this hack.
1485 * After we hit the bp, _unregister + _register can install the
1486 * new and not-yet-analyzed uprobe at the same address, restart.
1488 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1489 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1492 utask = current->utask;
1494 utask = add_utask();
1495 /* Cannot allocate; re-execute the instruction. */
1500 handler_chain(uprobe, regs);
1501 if (can_skip_sstep(uprobe, regs))
1504 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1505 utask->active_uprobe = uprobe;
1506 utask->state = UTASK_SSTEP;
1512 * cannot singlestep; cannot skip instruction;
1513 * re-execute the instruction.
1515 instruction_pointer_set(regs, bp_vaddr);
1521 * Perform required fix-ups and disable singlestep.
1522 * Allow pending signals to take effect.
1524 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1526 struct uprobe *uprobe;
1528 uprobe = utask->active_uprobe;
1529 if (utask->state == UTASK_SSTEP_ACK)
1530 arch_uprobe_post_xol(&uprobe->arch, regs);
1531 else if (utask->state == UTASK_SSTEP_TRAPPED)
1532 arch_uprobe_abort_xol(&uprobe->arch, regs);
1537 utask->active_uprobe = NULL;
1538 utask->state = UTASK_RUNNING;
1539 xol_free_insn_slot(current);
1541 spin_lock_irq(¤t->sighand->siglock);
1542 recalc_sigpending(); /* see uprobe_deny_signal() */
1543 spin_unlock_irq(¤t->sighand->siglock);
1547 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1548 * allows the thread to return from interrupt. After that handle_swbp()
1549 * sets utask->active_uprobe.
1551 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1552 * and allows the thread to return from interrupt.
1554 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1555 * uprobe_notify_resume().
1557 void uprobe_notify_resume(struct pt_regs *regs)
1559 struct uprobe_task *utask;
1561 clear_thread_flag(TIF_UPROBE);
1563 utask = current->utask;
1564 if (utask && utask->active_uprobe)
1565 handle_singlestep(utask, regs);
1571 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1572 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1574 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1576 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1579 set_thread_flag(TIF_UPROBE);
1584 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1585 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1587 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1589 struct uprobe_task *utask = current->utask;
1591 if (!current->mm || !utask || !utask->active_uprobe)
1592 /* task is currently not uprobed */
1595 utask->state = UTASK_SSTEP_ACK;
1596 set_thread_flag(TIF_UPROBE);
1600 static struct notifier_block uprobe_exception_nb = {
1601 .notifier_call = arch_uprobe_exception_notify,
1602 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1605 static int __init init_uprobes(void)
1609 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1610 mutex_init(&uprobes_mutex[i]);
1611 mutex_init(&uprobes_mmap_mutex[i]);
1614 if (percpu_init_rwsem(&dup_mmap_sem))
1617 return register_die_notifier(&uprobe_exception_nb);
1619 module_init(init_uprobes);
1621 static void __exit exit_uprobes(void)
1624 module_exit(exit_uprobes);