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 /* Dont run handlers when first register/ last unregister in progress*/
87 #define UPROBE_RUN_HANDLER 1
88 /* Can skip singlestep */
89 #define UPROBE_SKIP_SSTEP 2
92 struct rb_node rb_node; /* node in the rb tree */
94 struct rw_semaphore consumer_rwsem;
95 struct mutex copy_mutex; /* TODO: kill me and UPROBE_COPY_INSN */
96 struct list_head pending_list;
97 struct uprobe_consumer *consumers;
98 struct inode *inode; /* Also hold a ref to inode */
101 struct arch_uprobe arch;
105 * valid_vma: Verify if the specified vma is an executable vma
106 * Relax restrictions while unregistering: vm_flags might have
107 * changed after breakpoint was inserted.
108 * - is_register: indicates if we are in register context.
109 * - Return 1 if the specified virtual address is in an
112 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
114 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
119 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
122 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
124 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
127 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
129 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
133 * __replace_page - replace page in vma by new page.
134 * based on replace_page in mm/ksm.c
136 * @vma: vma that holds the pte pointing to page
137 * @addr: address the old @page is mapped at
138 * @page: the cowed page we are replacing by kpage
139 * @kpage: the modified page we replace page by
141 * Returns 0 on success, -EFAULT on failure.
143 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
144 struct page *page, struct page *kpage)
146 struct mm_struct *mm = vma->vm_mm;
150 /* For mmu_notifiers */
151 const unsigned long mmun_start = addr;
152 const unsigned long mmun_end = addr + PAGE_SIZE;
154 /* For try_to_free_swap() and munlock_vma_page() below */
157 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
159 ptep = page_check_address(page, mm, addr, &ptl, 0);
164 page_add_new_anon_rmap(kpage, vma, addr);
166 if (!PageAnon(page)) {
167 dec_mm_counter(mm, MM_FILEPAGES);
168 inc_mm_counter(mm, MM_ANONPAGES);
171 flush_cache_page(vma, addr, pte_pfn(*ptep));
172 ptep_clear_flush(vma, addr, ptep);
173 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
175 page_remove_rmap(page);
176 if (!page_mapped(page))
177 try_to_free_swap(page);
178 pte_unmap_unlock(ptep, ptl);
180 if (vma->vm_flags & VM_LOCKED)
181 munlock_vma_page(page);
186 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
192 * is_swbp_insn - check if instruction is breakpoint instruction.
193 * @insn: instruction to be checked.
194 * Default implementation of is_swbp_insn
195 * Returns true if @insn is a breakpoint instruction.
197 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
199 return *insn == UPROBE_SWBP_INSN;
202 static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode)
204 void *kaddr = kmap_atomic(page);
205 memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE);
206 kunmap_atomic(kaddr);
209 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
211 uprobe_opcode_t old_opcode;
214 copy_opcode(page, vaddr, &old_opcode);
215 is_swbp = is_swbp_insn(&old_opcode);
217 if (is_swbp_insn(new_opcode)) {
218 if (is_swbp) /* register: already installed? */
221 if (!is_swbp) /* unregister: was it changed by us? */
230 * Expect the breakpoint instruction to be the smallest size instruction for
231 * the architecture. If an arch has variable length instruction and the
232 * breakpoint instruction is not of the smallest length instruction
233 * supported by that architecture then we need to modify is_swbp_at_addr and
234 * write_opcode accordingly. This would never be a problem for archs that
235 * have fixed length instructions.
239 * write_opcode - write the opcode at a given virtual address.
240 * @mm: the probed process address space.
241 * @vaddr: the virtual address to store the opcode.
242 * @opcode: opcode to be written at @vaddr.
244 * Called with mm->mmap_sem held (for read and with a reference to
247 * For mm @mm, write the opcode at @vaddr.
248 * Return 0 (success) or a negative errno.
250 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
251 uprobe_opcode_t opcode)
253 struct page *old_page, *new_page;
254 void *vaddr_old, *vaddr_new;
255 struct vm_area_struct *vma;
259 /* Read the page with vaddr into memory */
260 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
264 ret = verify_opcode(old_page, vaddr, &opcode);
269 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
273 __SetPageUptodate(new_page);
275 /* copy the page now that we've got it stable */
276 vaddr_old = kmap_atomic(old_page);
277 vaddr_new = kmap_atomic(new_page);
279 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
280 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
282 kunmap_atomic(vaddr_new);
283 kunmap_atomic(vaddr_old);
285 ret = anon_vma_prepare(vma);
289 ret = __replace_page(vma, vaddr, old_page, new_page);
292 page_cache_release(new_page);
296 if (unlikely(ret == -EAGAIN))
302 * set_swbp - store breakpoint at a given address.
303 * @auprobe: arch specific probepoint information.
304 * @mm: the probed process address space.
305 * @vaddr: the virtual address to insert the opcode.
307 * For mm @mm, store the breakpoint instruction at @vaddr.
308 * Return 0 (success) or a negative errno.
310 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
312 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
316 * set_orig_insn - Restore the original instruction.
317 * @mm: the probed process address space.
318 * @auprobe: arch specific probepoint information.
319 * @vaddr: the virtual address to insert the opcode.
321 * For mm @mm, restore the original opcode (opcode) at @vaddr.
322 * Return 0 (success) or a negative errno.
325 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
327 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
330 static int match_uprobe(struct uprobe *l, struct uprobe *r)
332 if (l->inode < r->inode)
335 if (l->inode > r->inode)
338 if (l->offset < r->offset)
341 if (l->offset > r->offset)
347 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
349 struct uprobe u = { .inode = inode, .offset = offset };
350 struct rb_node *n = uprobes_tree.rb_node;
351 struct uprobe *uprobe;
355 uprobe = rb_entry(n, struct uprobe, rb_node);
356 match = match_uprobe(&u, uprobe);
358 atomic_inc(&uprobe->ref);
371 * Find a uprobe corresponding to a given inode:offset
372 * Acquires uprobes_treelock
374 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
376 struct uprobe *uprobe;
378 spin_lock(&uprobes_treelock);
379 uprobe = __find_uprobe(inode, offset);
380 spin_unlock(&uprobes_treelock);
385 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
387 struct rb_node **p = &uprobes_tree.rb_node;
388 struct rb_node *parent = NULL;
394 u = rb_entry(parent, struct uprobe, rb_node);
395 match = match_uprobe(uprobe, u);
402 p = &parent->rb_left;
404 p = &parent->rb_right;
409 rb_link_node(&uprobe->rb_node, parent, p);
410 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
411 /* get access + creation ref */
412 atomic_set(&uprobe->ref, 2);
418 * Acquire uprobes_treelock.
419 * Matching uprobe already exists in rbtree;
420 * increment (access refcount) and return the matching uprobe.
422 * No matching uprobe; insert the uprobe in rb_tree;
423 * get a double refcount (access + creation) and return NULL.
425 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
429 spin_lock(&uprobes_treelock);
430 u = __insert_uprobe(uprobe);
431 spin_unlock(&uprobes_treelock);
433 /* For now assume that the instruction need not be single-stepped */
434 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
439 static void put_uprobe(struct uprobe *uprobe)
441 if (atomic_dec_and_test(&uprobe->ref))
445 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
447 struct uprobe *uprobe, *cur_uprobe;
449 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
453 uprobe->inode = igrab(inode);
454 uprobe->offset = offset;
455 init_rwsem(&uprobe->consumer_rwsem);
456 mutex_init(&uprobe->copy_mutex);
458 /* add to uprobes_tree, sorted on inode:offset */
459 cur_uprobe = insert_uprobe(uprobe);
461 /* a uprobe exists for this inode:offset combination */
467 atomic_inc(&uprobe_events);
473 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
475 struct uprobe_consumer *uc;
477 if (!test_bit(UPROBE_RUN_HANDLER, &uprobe->flags))
480 down_read(&uprobe->consumer_rwsem);
481 for (uc = uprobe->consumers; uc; uc = uc->next) {
482 if (!uc->filter || uc->filter(uc, current))
483 uc->handler(uc, regs);
485 up_read(&uprobe->consumer_rwsem);
488 /* Returns the previous consumer */
489 static struct uprobe_consumer *
490 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
492 down_write(&uprobe->consumer_rwsem);
493 uc->next = uprobe->consumers;
494 uprobe->consumers = uc;
495 up_write(&uprobe->consumer_rwsem);
501 * For uprobe @uprobe, delete the consumer @uc.
502 * Return true if the @uc is deleted successfully
505 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
507 struct uprobe_consumer **con;
510 down_write(&uprobe->consumer_rwsem);
511 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
518 up_write(&uprobe->consumer_rwsem);
524 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
525 unsigned long nbytes, loff_t offset)
535 if (!mapping->a_ops->readpage)
538 idx = offset >> PAGE_CACHE_SHIFT;
539 off = offset & ~PAGE_MASK;
542 * Ensure that the page that has the original instruction is
543 * populated and in page-cache.
545 page = read_mapping_page(mapping, idx, filp);
547 return PTR_ERR(page);
549 vaddr = kmap_atomic(page);
550 memcpy(insn, vaddr + off, nbytes);
551 kunmap_atomic(vaddr);
552 page_cache_release(page);
557 static int copy_insn(struct uprobe *uprobe, struct file *filp)
559 struct address_space *mapping;
560 unsigned long nbytes;
563 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
564 mapping = uprobe->inode->i_mapping;
566 /* Instruction at end of binary; copy only available bytes */
567 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
568 bytes = uprobe->inode->i_size - uprobe->offset;
570 bytes = MAX_UINSN_BYTES;
572 /* Instruction at the page-boundary; copy bytes in second page */
573 if (nbytes < bytes) {
574 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
575 bytes - nbytes, uprobe->offset + nbytes);
580 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
583 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
584 struct mm_struct *mm, unsigned long vaddr)
588 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
591 mutex_lock(&uprobe->copy_mutex);
592 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
595 ret = copy_insn(uprobe, file);
600 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
603 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
607 /* write_opcode() assumes we don't cross page boundary */
608 BUG_ON((uprobe->offset & ~PAGE_MASK) +
609 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
611 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
612 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
615 mutex_unlock(&uprobe->copy_mutex);
621 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
622 struct vm_area_struct *vma, unsigned long vaddr)
628 * If probe is being deleted, unregister thread could be done with
629 * the vma-rmap-walk through. Adding a probe now can be fatal since
630 * nobody will be able to cleanup. Also we could be from fork or
631 * mremap path, where the probe might have already been inserted.
632 * Hence behave as if probe already existed.
634 if (!uprobe->consumers)
637 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
642 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
643 * the task can hit this breakpoint right after __replace_page().
645 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
647 set_bit(MMF_HAS_UPROBES, &mm->flags);
649 ret = set_swbp(&uprobe->arch, mm, vaddr);
651 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
652 else if (first_uprobe)
653 clear_bit(MMF_HAS_UPROBES, &mm->flags);
659 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
661 /* can happen if uprobe_register() fails */
662 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
665 set_bit(MMF_RECALC_UPROBES, &mm->flags);
666 return set_orig_insn(&uprobe->arch, mm, vaddr);
670 * There could be threads that have already hit the breakpoint. They
671 * will recheck the current insn and restart if find_uprobe() fails.
672 * See find_active_uprobe().
674 static void delete_uprobe(struct uprobe *uprobe)
676 spin_lock(&uprobes_treelock);
677 rb_erase(&uprobe->rb_node, &uprobes_tree);
678 spin_unlock(&uprobes_treelock);
681 atomic_dec(&uprobe_events);
685 struct map_info *next;
686 struct mm_struct *mm;
690 static inline struct map_info *free_map_info(struct map_info *info)
692 struct map_info *next = info->next;
697 static struct map_info *
698 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
700 unsigned long pgoff = offset >> PAGE_SHIFT;
701 struct vm_area_struct *vma;
702 struct map_info *curr = NULL;
703 struct map_info *prev = NULL;
704 struct map_info *info;
708 mutex_lock(&mapping->i_mmap_mutex);
709 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
710 if (!valid_vma(vma, is_register))
713 if (!prev && !more) {
715 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
716 * reclaim. This is optimistic, no harm done if it fails.
718 prev = kmalloc(sizeof(struct map_info),
719 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
728 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
736 info->mm = vma->vm_mm;
737 info->vaddr = offset_to_vaddr(vma, offset);
739 mutex_unlock(&mapping->i_mmap_mutex);
751 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
753 curr = ERR_PTR(-ENOMEM);
763 prev = free_map_info(prev);
767 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
769 struct map_info *info;
772 percpu_down_write(&dup_mmap_sem);
773 info = build_map_info(uprobe->inode->i_mapping,
774 uprobe->offset, is_register);
781 struct mm_struct *mm = info->mm;
782 struct vm_area_struct *vma;
784 if (err && is_register)
787 down_write(&mm->mmap_sem);
788 vma = find_vma(mm, info->vaddr);
789 if (!vma || !valid_vma(vma, is_register) ||
790 vma->vm_file->f_mapping->host != uprobe->inode)
793 if (vma->vm_start > info->vaddr ||
794 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
798 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
800 err |= remove_breakpoint(uprobe, mm, info->vaddr);
803 up_write(&mm->mmap_sem);
806 info = free_map_info(info);
809 percpu_up_write(&dup_mmap_sem);
813 static int __uprobe_register(struct uprobe *uprobe)
815 return register_for_each_vma(uprobe, true);
818 static void __uprobe_unregister(struct uprobe *uprobe)
820 if (!register_for_each_vma(uprobe, false))
821 delete_uprobe(uprobe);
823 /* TODO : cant unregister? schedule a worker thread */
827 * uprobe_register - register a probe
828 * @inode: the file in which the probe has to be placed.
829 * @offset: offset from the start of the file.
830 * @uc: information on howto handle the probe..
832 * Apart from the access refcount, uprobe_register() takes a creation
833 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
834 * inserted into the rbtree (i.e first consumer for a @inode:@offset
835 * tuple). Creation refcount stops uprobe_unregister from freeing the
836 * @uprobe even before the register operation is complete. Creation
837 * refcount is released when the last @uc for the @uprobe
840 * Return errno if it cannot successully install probes
841 * else return 0 (success)
843 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
845 struct uprobe *uprobe;
848 if (!inode || !uc || uc->next)
851 if (offset > i_size_read(inode))
855 mutex_lock(uprobes_hash(inode));
856 uprobe = alloc_uprobe(inode, offset);
860 } else if (!consumer_add(uprobe, uc)) {
861 ret = __uprobe_register(uprobe);
863 uprobe->consumers = NULL;
864 __uprobe_unregister(uprobe);
866 set_bit(UPROBE_RUN_HANDLER, &uprobe->flags);
870 mutex_unlock(uprobes_hash(inode));
878 * uprobe_unregister - unregister a already registered probe.
879 * @inode: the file in which the probe has to be removed.
880 * @offset: offset from the start of the file.
881 * @uc: identify which probe if multiple probes are colocated.
883 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
885 struct uprobe *uprobe;
890 uprobe = find_uprobe(inode, offset);
894 mutex_lock(uprobes_hash(inode));
896 if (consumer_del(uprobe, uc)) {
897 if (!uprobe->consumers) {
898 __uprobe_unregister(uprobe);
899 clear_bit(UPROBE_RUN_HANDLER, &uprobe->flags);
903 mutex_unlock(uprobes_hash(inode));
907 static struct rb_node *
908 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
910 struct rb_node *n = uprobes_tree.rb_node;
913 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
915 if (inode < u->inode) {
917 } else if (inode > u->inode) {
922 else if (min > u->offset)
933 * For a given range in vma, build a list of probes that need to be inserted.
935 static void build_probe_list(struct inode *inode,
936 struct vm_area_struct *vma,
937 unsigned long start, unsigned long end,
938 struct list_head *head)
941 struct rb_node *n, *t;
944 INIT_LIST_HEAD(head);
945 min = vaddr_to_offset(vma, start);
946 max = min + (end - start) - 1;
948 spin_lock(&uprobes_treelock);
949 n = find_node_in_range(inode, min, max);
951 for (t = n; t; t = rb_prev(t)) {
952 u = rb_entry(t, struct uprobe, rb_node);
953 if (u->inode != inode || u->offset < min)
955 list_add(&u->pending_list, head);
958 for (t = n; (t = rb_next(t)); ) {
959 u = rb_entry(t, struct uprobe, rb_node);
960 if (u->inode != inode || u->offset > max)
962 list_add(&u->pending_list, head);
966 spin_unlock(&uprobes_treelock);
970 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
972 * Currently we ignore all errors and always return 0, the callers
973 * can't handle the failure anyway.
975 int uprobe_mmap(struct vm_area_struct *vma)
977 struct list_head tmp_list;
978 struct uprobe *uprobe, *u;
981 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
984 inode = vma->vm_file->f_mapping->host;
988 mutex_lock(uprobes_mmap_hash(inode));
989 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
991 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
992 if (!fatal_signal_pending(current)) {
993 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
994 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
998 mutex_unlock(uprobes_mmap_hash(inode));
1004 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1007 struct inode *inode;
1010 inode = vma->vm_file->f_mapping->host;
1012 min = vaddr_to_offset(vma, start);
1013 max = min + (end - start) - 1;
1015 spin_lock(&uprobes_treelock);
1016 n = find_node_in_range(inode, min, max);
1017 spin_unlock(&uprobes_treelock);
1023 * Called in context of a munmap of a vma.
1025 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1027 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1030 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1033 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1034 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1037 if (vma_has_uprobes(vma, start, end))
1038 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1041 /* Slot allocation for XOL */
1042 static int xol_add_vma(struct xol_area *area)
1044 struct mm_struct *mm;
1047 area->page = alloc_page(GFP_HIGHUSER);
1054 down_write(&mm->mmap_sem);
1055 if (mm->uprobes_state.xol_area)
1060 /* Try to map as high as possible, this is only a hint. */
1061 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1062 if (area->vaddr & ~PAGE_MASK) {
1067 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1068 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1072 smp_wmb(); /* pairs with get_xol_area() */
1073 mm->uprobes_state.xol_area = area;
1077 up_write(&mm->mmap_sem);
1079 __free_page(area->page);
1084 static struct xol_area *get_xol_area(struct mm_struct *mm)
1086 struct xol_area *area;
1088 area = mm->uprobes_state.xol_area;
1089 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1095 * xol_alloc_area - Allocate process's xol_area.
1096 * This area will be used for storing instructions for execution out of
1099 * Returns the allocated area or NULL.
1101 static struct xol_area *xol_alloc_area(void)
1103 struct xol_area *area;
1105 area = kzalloc(sizeof(*area), GFP_KERNEL);
1106 if (unlikely(!area))
1109 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1114 init_waitqueue_head(&area->wq);
1115 if (!xol_add_vma(area))
1119 kfree(area->bitmap);
1122 return get_xol_area(current->mm);
1126 * uprobe_clear_state - Free the area allocated for slots.
1128 void uprobe_clear_state(struct mm_struct *mm)
1130 struct xol_area *area = mm->uprobes_state.xol_area;
1135 put_page(area->page);
1136 kfree(area->bitmap);
1140 void uprobe_start_dup_mmap(void)
1142 percpu_down_read(&dup_mmap_sem);
1145 void uprobe_end_dup_mmap(void)
1147 percpu_up_read(&dup_mmap_sem);
1150 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1152 newmm->uprobes_state.xol_area = NULL;
1154 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1155 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1156 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1157 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1162 * - search for a free slot.
1164 static unsigned long xol_take_insn_slot(struct xol_area *area)
1166 unsigned long slot_addr;
1170 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1171 if (slot_nr < UINSNS_PER_PAGE) {
1172 if (!test_and_set_bit(slot_nr, area->bitmap))
1175 slot_nr = UINSNS_PER_PAGE;
1178 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1179 } while (slot_nr >= UINSNS_PER_PAGE);
1181 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1182 atomic_inc(&area->slot_count);
1188 * xol_get_insn_slot - If was not allocated a slot, then
1190 * Returns the allocated slot address or 0.
1192 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1194 struct xol_area *area;
1195 unsigned long offset;
1198 area = get_xol_area(current->mm);
1200 area = xol_alloc_area();
1204 current->utask->xol_vaddr = xol_take_insn_slot(area);
1207 * Initialize the slot if xol_vaddr points to valid
1210 if (unlikely(!current->utask->xol_vaddr))
1213 current->utask->vaddr = slot_addr;
1214 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1215 vaddr = kmap_atomic(area->page);
1216 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1217 kunmap_atomic(vaddr);
1219 * We probably need flush_icache_user_range() but it needs vma.
1220 * This should work on supported architectures too.
1222 flush_dcache_page(area->page);
1224 return current->utask->xol_vaddr;
1228 * xol_free_insn_slot - If slot was earlier allocated by
1229 * @xol_get_insn_slot(), make the slot available for
1230 * subsequent requests.
1232 static void xol_free_insn_slot(struct task_struct *tsk)
1234 struct xol_area *area;
1235 unsigned long vma_end;
1236 unsigned long slot_addr;
1238 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1241 slot_addr = tsk->utask->xol_vaddr;
1243 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1246 area = tsk->mm->uprobes_state.xol_area;
1247 vma_end = area->vaddr + PAGE_SIZE;
1248 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1249 unsigned long offset;
1252 offset = slot_addr - area->vaddr;
1253 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1254 if (slot_nr >= UINSNS_PER_PAGE)
1257 clear_bit(slot_nr, area->bitmap);
1258 atomic_dec(&area->slot_count);
1259 if (waitqueue_active(&area->wq))
1262 tsk->utask->xol_vaddr = 0;
1267 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1268 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1270 * Return the address of the breakpoint instruction.
1272 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1274 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1278 * Called with no locks held.
1279 * Called in context of a exiting or a exec-ing thread.
1281 void uprobe_free_utask(struct task_struct *t)
1283 struct uprobe_task *utask = t->utask;
1288 if (utask->active_uprobe)
1289 put_uprobe(utask->active_uprobe);
1291 xol_free_insn_slot(t);
1297 * Called in context of a new clone/fork from copy_process.
1299 void uprobe_copy_process(struct task_struct *t)
1305 * Allocate a uprobe_task object for the task.
1306 * Called when the thread hits a breakpoint for the first time.
1309 * - pointer to new uprobe_task on success
1312 static struct uprobe_task *add_utask(void)
1314 struct uprobe_task *utask;
1316 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1317 if (unlikely(!utask))
1320 current->utask = utask;
1324 /* Prepare to single-step probed instruction out of line. */
1326 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1328 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1335 * If we are singlestepping, then ensure this thread is not connected to
1336 * non-fatal signals until completion of singlestep. When xol insn itself
1337 * triggers the signal, restart the original insn even if the task is
1338 * already SIGKILL'ed (since coredump should report the correct ip). This
1339 * is even more important if the task has a handler for SIGSEGV/etc, The
1340 * _same_ instruction should be repeated again after return from the signal
1341 * handler, and SSTEP can never finish in this case.
1343 bool uprobe_deny_signal(void)
1345 struct task_struct *t = current;
1346 struct uprobe_task *utask = t->utask;
1348 if (likely(!utask || !utask->active_uprobe))
1351 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1353 if (signal_pending(t)) {
1354 spin_lock_irq(&t->sighand->siglock);
1355 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1356 spin_unlock_irq(&t->sighand->siglock);
1358 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1359 utask->state = UTASK_SSTEP_TRAPPED;
1360 set_tsk_thread_flag(t, TIF_UPROBE);
1361 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1369 * Avoid singlestepping the original instruction if the original instruction
1370 * is a NOP or can be emulated.
1372 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1374 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1375 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1377 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1382 static void mmf_recalc_uprobes(struct mm_struct *mm)
1384 struct vm_area_struct *vma;
1386 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1387 if (!valid_vma(vma, false))
1390 * This is not strictly accurate, we can race with
1391 * uprobe_unregister() and see the already removed
1392 * uprobe if delete_uprobe() was not yet called.
1394 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1398 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1401 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1404 uprobe_opcode_t opcode;
1407 pagefault_disable();
1408 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1412 if (likely(result == 0))
1415 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1419 copy_opcode(page, vaddr, &opcode);
1422 return is_swbp_insn(&opcode);
1425 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1427 struct mm_struct *mm = current->mm;
1428 struct uprobe *uprobe = NULL;
1429 struct vm_area_struct *vma;
1431 down_read(&mm->mmap_sem);
1432 vma = find_vma(mm, bp_vaddr);
1433 if (vma && vma->vm_start <= bp_vaddr) {
1434 if (valid_vma(vma, false)) {
1435 struct inode *inode = vma->vm_file->f_mapping->host;
1436 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1438 uprobe = find_uprobe(inode, offset);
1442 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1447 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1448 mmf_recalc_uprobes(mm);
1449 up_read(&mm->mmap_sem);
1455 * Run handler and ask thread to singlestep.
1456 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1458 static void handle_swbp(struct pt_regs *regs)
1460 struct uprobe_task *utask;
1461 struct uprobe *uprobe;
1462 unsigned long bp_vaddr;
1463 int uninitialized_var(is_swbp);
1465 bp_vaddr = uprobe_get_swbp_addr(regs);
1466 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1470 /* No matching uprobe; signal SIGTRAP. */
1471 send_sig(SIGTRAP, current, 0);
1474 * Either we raced with uprobe_unregister() or we can't
1475 * access this memory. The latter is only possible if
1476 * another thread plays with our ->mm. In both cases
1477 * we can simply restart. If this vma was unmapped we
1478 * can pretend this insn was not executed yet and get
1479 * the (correct) SIGSEGV after restart.
1481 instruction_pointer_set(regs, bp_vaddr);
1486 * TODO: move copy_insn/etc into _register and remove this hack.
1487 * After we hit the bp, _unregister + _register can install the
1488 * new and not-yet-analyzed uprobe at the same address, restart.
1490 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1491 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1494 utask = current->utask;
1496 utask = add_utask();
1497 /* Cannot allocate; re-execute the instruction. */
1502 handler_chain(uprobe, regs);
1503 if (can_skip_sstep(uprobe, regs))
1506 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1507 utask->active_uprobe = uprobe;
1508 utask->state = UTASK_SSTEP;
1514 * cannot singlestep; cannot skip instruction;
1515 * re-execute the instruction.
1517 instruction_pointer_set(regs, bp_vaddr);
1523 * Perform required fix-ups and disable singlestep.
1524 * Allow pending signals to take effect.
1526 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1528 struct uprobe *uprobe;
1530 uprobe = utask->active_uprobe;
1531 if (utask->state == UTASK_SSTEP_ACK)
1532 arch_uprobe_post_xol(&uprobe->arch, regs);
1533 else if (utask->state == UTASK_SSTEP_TRAPPED)
1534 arch_uprobe_abort_xol(&uprobe->arch, regs);
1539 utask->active_uprobe = NULL;
1540 utask->state = UTASK_RUNNING;
1541 xol_free_insn_slot(current);
1543 spin_lock_irq(¤t->sighand->siglock);
1544 recalc_sigpending(); /* see uprobe_deny_signal() */
1545 spin_unlock_irq(¤t->sighand->siglock);
1549 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1550 * allows the thread to return from interrupt. After that handle_swbp()
1551 * sets utask->active_uprobe.
1553 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1554 * and allows the thread to return from interrupt.
1556 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1557 * uprobe_notify_resume().
1559 void uprobe_notify_resume(struct pt_regs *regs)
1561 struct uprobe_task *utask;
1563 clear_thread_flag(TIF_UPROBE);
1565 utask = current->utask;
1566 if (utask && utask->active_uprobe)
1567 handle_singlestep(utask, regs);
1573 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1574 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1576 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1578 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1581 set_thread_flag(TIF_UPROBE);
1586 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1587 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1589 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1591 struct uprobe_task *utask = current->utask;
1593 if (!current->mm || !utask || !utask->active_uprobe)
1594 /* task is currently not uprobed */
1597 utask->state = UTASK_SSTEP_ACK;
1598 set_thread_flag(TIF_UPROBE);
1602 static struct notifier_block uprobe_exception_nb = {
1603 .notifier_call = arch_uprobe_exception_notify,
1604 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1607 static int __init init_uprobes(void)
1611 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1612 mutex_init(&uprobes_mutex[i]);
1613 mutex_init(&uprobes_mmap_mutex[i]);
1616 if (percpu_init_rwsem(&dup_mmap_sem))
1619 return register_die_notifier(&uprobe_exception_nb);
1621 module_init(init_uprobes);
1623 static void __exit exit_uprobes(void)
1626 module_exit(exit_uprobes);