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 mutex copy_mutex; /* TODO: kill me and UPROBE_COPY_INSN */
95 struct list_head pending_list;
96 struct uprobe_consumer *consumers;
97 struct inode *inode; /* Also hold a ref to inode */
100 struct arch_uprobe arch;
104 * valid_vma: Verify if the specified vma is an executable vma
105 * Relax restrictions while unregistering: vm_flags might have
106 * changed after breakpoint was inserted.
107 * - is_register: indicates if we are in register context.
108 * - Return 1 if the specified virtual address is in an
111 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
113 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
118 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
121 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
123 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
126 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
128 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
132 * __replace_page - replace page in vma by new page.
133 * based on replace_page in mm/ksm.c
135 * @vma: vma that holds the pte pointing to page
136 * @addr: address the old @page is mapped at
137 * @page: the cowed page we are replacing by kpage
138 * @kpage: the modified page we replace page by
140 * Returns 0 on success, -EFAULT on failure.
142 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
143 struct page *page, struct page *kpage)
145 struct mm_struct *mm = vma->vm_mm;
149 /* For mmu_notifiers */
150 const unsigned long mmun_start = addr;
151 const unsigned long mmun_end = addr + PAGE_SIZE;
153 /* For try_to_free_swap() and munlock_vma_page() below */
156 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
158 ptep = page_check_address(page, mm, addr, &ptl, 0);
163 page_add_new_anon_rmap(kpage, vma, addr);
165 if (!PageAnon(page)) {
166 dec_mm_counter(mm, MM_FILEPAGES);
167 inc_mm_counter(mm, MM_ANONPAGES);
170 flush_cache_page(vma, addr, pte_pfn(*ptep));
171 ptep_clear_flush(vma, addr, ptep);
172 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
174 page_remove_rmap(page);
175 if (!page_mapped(page))
176 try_to_free_swap(page);
177 pte_unmap_unlock(ptep, ptl);
179 if (vma->vm_flags & VM_LOCKED)
180 munlock_vma_page(page);
185 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
191 * is_swbp_insn - check if instruction is breakpoint instruction.
192 * @insn: instruction to be checked.
193 * Default implementation of is_swbp_insn
194 * Returns true if @insn is a breakpoint instruction.
196 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
198 return *insn == UPROBE_SWBP_INSN;
201 static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode)
203 void *kaddr = kmap_atomic(page);
204 memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE);
205 kunmap_atomic(kaddr);
208 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
210 uprobe_opcode_t old_opcode;
213 copy_opcode(page, vaddr, &old_opcode);
214 is_swbp = is_swbp_insn(&old_opcode);
216 if (is_swbp_insn(new_opcode)) {
217 if (is_swbp) /* register: already installed? */
220 if (!is_swbp) /* unregister: was it changed by us? */
229 * Expect the breakpoint instruction to be the smallest size instruction for
230 * the architecture. If an arch has variable length instruction and the
231 * breakpoint instruction is not of the smallest length instruction
232 * supported by that architecture then we need to modify is_swbp_at_addr and
233 * write_opcode accordingly. This would never be a problem for archs that
234 * have fixed length instructions.
238 * write_opcode - write the opcode at a given virtual address.
239 * @mm: the probed process address space.
240 * @vaddr: the virtual address to store the opcode.
241 * @opcode: opcode to be written at @vaddr.
243 * Called with mm->mmap_sem held (for read and with a reference to
246 * For mm @mm, write the opcode at @vaddr.
247 * Return 0 (success) or a negative errno.
249 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
250 uprobe_opcode_t opcode)
252 struct page *old_page, *new_page;
253 void *vaddr_old, *vaddr_new;
254 struct vm_area_struct *vma;
258 /* Read the page with vaddr into memory */
259 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
263 ret = verify_opcode(old_page, vaddr, &opcode);
268 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
272 __SetPageUptodate(new_page);
274 /* copy the page now that we've got it stable */
275 vaddr_old = kmap_atomic(old_page);
276 vaddr_new = kmap_atomic(new_page);
278 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
279 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
281 kunmap_atomic(vaddr_new);
282 kunmap_atomic(vaddr_old);
284 ret = anon_vma_prepare(vma);
288 ret = __replace_page(vma, vaddr, old_page, new_page);
291 page_cache_release(new_page);
295 if (unlikely(ret == -EAGAIN))
301 * set_swbp - store breakpoint at a given address.
302 * @auprobe: arch specific probepoint information.
303 * @mm: the probed process address space.
304 * @vaddr: the virtual address to insert the opcode.
306 * For mm @mm, store the breakpoint instruction at @vaddr.
307 * Return 0 (success) or a negative errno.
309 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
311 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
315 * set_orig_insn - Restore the original instruction.
316 * @mm: the probed process address space.
317 * @auprobe: arch specific probepoint information.
318 * @vaddr: the virtual address to insert the opcode.
320 * For mm @mm, restore the original opcode (opcode) at @vaddr.
321 * Return 0 (success) or a negative errno.
324 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
326 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
329 static int match_uprobe(struct uprobe *l, struct uprobe *r)
331 if (l->inode < r->inode)
334 if (l->inode > r->inode)
337 if (l->offset < r->offset)
340 if (l->offset > r->offset)
346 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
348 struct uprobe u = { .inode = inode, .offset = offset };
349 struct rb_node *n = uprobes_tree.rb_node;
350 struct uprobe *uprobe;
354 uprobe = rb_entry(n, struct uprobe, rb_node);
355 match = match_uprobe(&u, uprobe);
357 atomic_inc(&uprobe->ref);
370 * Find a uprobe corresponding to a given inode:offset
371 * Acquires uprobes_treelock
373 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
375 struct uprobe *uprobe;
377 spin_lock(&uprobes_treelock);
378 uprobe = __find_uprobe(inode, offset);
379 spin_unlock(&uprobes_treelock);
384 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
386 struct rb_node **p = &uprobes_tree.rb_node;
387 struct rb_node *parent = NULL;
393 u = rb_entry(parent, struct uprobe, rb_node);
394 match = match_uprobe(uprobe, u);
401 p = &parent->rb_left;
403 p = &parent->rb_right;
408 rb_link_node(&uprobe->rb_node, parent, p);
409 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
410 /* get access + creation ref */
411 atomic_set(&uprobe->ref, 2);
417 * Acquire uprobes_treelock.
418 * Matching uprobe already exists in rbtree;
419 * increment (access refcount) and return the matching uprobe.
421 * No matching uprobe; insert the uprobe in rb_tree;
422 * get a double refcount (access + creation) and return NULL.
424 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
428 spin_lock(&uprobes_treelock);
429 u = __insert_uprobe(uprobe);
430 spin_unlock(&uprobes_treelock);
435 static void put_uprobe(struct uprobe *uprobe)
437 if (atomic_dec_and_test(&uprobe->ref))
441 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
443 struct uprobe *uprobe, *cur_uprobe;
445 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
449 uprobe->inode = igrab(inode);
450 uprobe->offset = offset;
451 init_rwsem(&uprobe->register_rwsem);
452 init_rwsem(&uprobe->consumer_rwsem);
453 mutex_init(&uprobe->copy_mutex);
454 /* For now assume that the instruction need not be single-stepped */
455 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
457 /* add to uprobes_tree, sorted on inode:offset */
458 cur_uprobe = insert_uprobe(uprobe);
460 /* a uprobe exists for this inode:offset combination */
466 atomic_inc(&uprobe_events);
472 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
474 struct uprobe_consumer *uc;
476 down_read(&uprobe->register_rwsem);
477 for (uc = uprobe->consumers; uc; uc = uc->next)
478 uc->handler(uc, regs);
479 up_read(&uprobe->register_rwsem);
482 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
484 down_write(&uprobe->consumer_rwsem);
485 uc->next = uprobe->consumers;
486 uprobe->consumers = uc;
487 up_write(&uprobe->consumer_rwsem);
491 * For uprobe @uprobe, delete the consumer @uc.
492 * Return true if the @uc is deleted successfully
495 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
497 struct uprobe_consumer **con;
500 down_write(&uprobe->consumer_rwsem);
501 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
508 up_write(&uprobe->consumer_rwsem);
514 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
515 unsigned long nbytes, loff_t offset)
525 if (!mapping->a_ops->readpage)
528 idx = offset >> PAGE_CACHE_SHIFT;
529 off = offset & ~PAGE_MASK;
532 * Ensure that the page that has the original instruction is
533 * populated and in page-cache.
535 page = read_mapping_page(mapping, idx, filp);
537 return PTR_ERR(page);
539 vaddr = kmap_atomic(page);
540 memcpy(insn, vaddr + off, nbytes);
541 kunmap_atomic(vaddr);
542 page_cache_release(page);
547 static int copy_insn(struct uprobe *uprobe, struct file *filp)
549 struct address_space *mapping;
550 unsigned long nbytes;
553 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
554 mapping = uprobe->inode->i_mapping;
556 /* Instruction at end of binary; copy only available bytes */
557 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
558 bytes = uprobe->inode->i_size - uprobe->offset;
560 bytes = MAX_UINSN_BYTES;
562 /* Instruction at the page-boundary; copy bytes in second page */
563 if (nbytes < bytes) {
564 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
565 bytes - nbytes, uprobe->offset + nbytes);
570 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
573 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
574 struct mm_struct *mm, unsigned long vaddr)
578 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
581 mutex_lock(&uprobe->copy_mutex);
582 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
585 ret = copy_insn(uprobe, file);
590 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
593 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
597 /* write_opcode() assumes we don't cross page boundary */
598 BUG_ON((uprobe->offset & ~PAGE_MASK) +
599 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
601 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
602 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
605 mutex_unlock(&uprobe->copy_mutex);
610 static bool filter_chain(struct uprobe *uprobe)
612 struct uprobe_consumer *uc;
615 down_read(&uprobe->consumer_rwsem);
616 for (uc = uprobe->consumers; uc; uc = uc->next) {
617 /* TODO: ret = uc->filter(...) */
622 up_read(&uprobe->consumer_rwsem);
628 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
629 struct vm_area_struct *vma, unsigned long vaddr)
635 * If probe is being deleted, unregister thread could be done with
636 * the vma-rmap-walk through. Adding a probe now can be fatal since
637 * nobody will be able to cleanup. But in this case filter_chain()
638 * must return false, all consumers have gone away.
640 if (!filter_chain(uprobe))
643 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
648 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
649 * the task can hit this breakpoint right after __replace_page().
651 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
653 set_bit(MMF_HAS_UPROBES, &mm->flags);
655 ret = set_swbp(&uprobe->arch, mm, vaddr);
657 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
658 else if (first_uprobe)
659 clear_bit(MMF_HAS_UPROBES, &mm->flags);
665 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
667 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
670 if (filter_chain(uprobe))
673 set_bit(MMF_RECALC_UPROBES, &mm->flags);
674 return set_orig_insn(&uprobe->arch, mm, vaddr);
678 * There could be threads that have already hit the breakpoint. They
679 * will recheck the current insn and restart if find_uprobe() fails.
680 * See find_active_uprobe().
682 static void delete_uprobe(struct uprobe *uprobe)
684 spin_lock(&uprobes_treelock);
685 rb_erase(&uprobe->rb_node, &uprobes_tree);
686 spin_unlock(&uprobes_treelock);
689 atomic_dec(&uprobe_events);
693 struct map_info *next;
694 struct mm_struct *mm;
698 static inline struct map_info *free_map_info(struct map_info *info)
700 struct map_info *next = info->next;
705 static struct map_info *
706 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
708 unsigned long pgoff = offset >> PAGE_SHIFT;
709 struct vm_area_struct *vma;
710 struct map_info *curr = NULL;
711 struct map_info *prev = NULL;
712 struct map_info *info;
716 mutex_lock(&mapping->i_mmap_mutex);
717 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
718 if (!valid_vma(vma, is_register))
721 if (!prev && !more) {
723 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
724 * reclaim. This is optimistic, no harm done if it fails.
726 prev = kmalloc(sizeof(struct map_info),
727 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
736 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
744 info->mm = vma->vm_mm;
745 info->vaddr = offset_to_vaddr(vma, offset);
747 mutex_unlock(&mapping->i_mmap_mutex);
759 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
761 curr = ERR_PTR(-ENOMEM);
771 prev = free_map_info(prev);
775 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
777 struct map_info *info;
780 percpu_down_write(&dup_mmap_sem);
781 info = build_map_info(uprobe->inode->i_mapping,
782 uprobe->offset, is_register);
789 struct mm_struct *mm = info->mm;
790 struct vm_area_struct *vma;
792 if (err && is_register)
795 down_write(&mm->mmap_sem);
796 vma = find_vma(mm, info->vaddr);
797 if (!vma || !valid_vma(vma, is_register) ||
798 vma->vm_file->f_mapping->host != uprobe->inode)
801 if (vma->vm_start > info->vaddr ||
802 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
806 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
808 err |= remove_breakpoint(uprobe, mm, info->vaddr);
811 up_write(&mm->mmap_sem);
814 info = free_map_info(info);
817 percpu_up_write(&dup_mmap_sem);
821 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
823 consumer_add(uprobe, uc);
824 return register_for_each_vma(uprobe, true);
827 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
831 if (!consumer_del(uprobe, uc)) /* WARN? */
834 err = register_for_each_vma(uprobe, false);
835 /* TODO : cant unregister? schedule a worker thread */
836 if (!uprobe->consumers && !err)
837 delete_uprobe(uprobe);
841 * uprobe_register - register a probe
842 * @inode: the file in which the probe has to be placed.
843 * @offset: offset from the start of the file.
844 * @uc: information on howto handle the probe..
846 * Apart from the access refcount, uprobe_register() takes a creation
847 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
848 * inserted into the rbtree (i.e first consumer for a @inode:@offset
849 * tuple). Creation refcount stops uprobe_unregister from freeing the
850 * @uprobe even before the register operation is complete. Creation
851 * refcount is released when the last @uc for the @uprobe
854 * Return errno if it cannot successully install probes
855 * else return 0 (success)
857 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
859 struct uprobe *uprobe;
862 /* Racy, just to catch the obvious mistakes */
863 if (offset > i_size_read(inode))
867 mutex_lock(uprobes_hash(inode));
868 uprobe = alloc_uprobe(inode, offset);
870 down_write(&uprobe->register_rwsem);
871 ret = __uprobe_register(uprobe, uc);
873 __uprobe_unregister(uprobe, uc);
874 up_write(&uprobe->register_rwsem);
876 mutex_unlock(uprobes_hash(inode));
884 * uprobe_unregister - unregister a already registered probe.
885 * @inode: the file in which the probe has to be removed.
886 * @offset: offset from the start of the file.
887 * @uc: identify which probe if multiple probes are colocated.
889 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
891 struct uprobe *uprobe;
893 uprobe = find_uprobe(inode, offset);
897 mutex_lock(uprobes_hash(inode));
898 down_write(&uprobe->register_rwsem);
899 __uprobe_unregister(uprobe, uc);
900 up_write(&uprobe->register_rwsem);
901 mutex_unlock(uprobes_hash(inode));
905 static struct rb_node *
906 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
908 struct rb_node *n = uprobes_tree.rb_node;
911 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
913 if (inode < u->inode) {
915 } else if (inode > u->inode) {
920 else if (min > u->offset)
931 * For a given range in vma, build a list of probes that need to be inserted.
933 static void build_probe_list(struct inode *inode,
934 struct vm_area_struct *vma,
935 unsigned long start, unsigned long end,
936 struct list_head *head)
939 struct rb_node *n, *t;
942 INIT_LIST_HEAD(head);
943 min = vaddr_to_offset(vma, start);
944 max = min + (end - start) - 1;
946 spin_lock(&uprobes_treelock);
947 n = find_node_in_range(inode, min, max);
949 for (t = n; t; t = rb_prev(t)) {
950 u = rb_entry(t, struct uprobe, rb_node);
951 if (u->inode != inode || u->offset < min)
953 list_add(&u->pending_list, head);
956 for (t = n; (t = rb_next(t)); ) {
957 u = rb_entry(t, struct uprobe, rb_node);
958 if (u->inode != inode || u->offset > max)
960 list_add(&u->pending_list, head);
964 spin_unlock(&uprobes_treelock);
968 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
970 * Currently we ignore all errors and always return 0, the callers
971 * can't handle the failure anyway.
973 int uprobe_mmap(struct vm_area_struct *vma)
975 struct list_head tmp_list;
976 struct uprobe *uprobe, *u;
979 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
982 inode = vma->vm_file->f_mapping->host;
986 mutex_lock(uprobes_mmap_hash(inode));
987 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
989 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
990 if (!fatal_signal_pending(current)) {
991 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
992 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
996 mutex_unlock(uprobes_mmap_hash(inode));
1002 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1005 struct inode *inode;
1008 inode = vma->vm_file->f_mapping->host;
1010 min = vaddr_to_offset(vma, start);
1011 max = min + (end - start) - 1;
1013 spin_lock(&uprobes_treelock);
1014 n = find_node_in_range(inode, min, max);
1015 spin_unlock(&uprobes_treelock);
1021 * Called in context of a munmap of a vma.
1023 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1025 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1028 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1031 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1032 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1035 if (vma_has_uprobes(vma, start, end))
1036 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1039 /* Slot allocation for XOL */
1040 static int xol_add_vma(struct xol_area *area)
1042 struct mm_struct *mm;
1045 area->page = alloc_page(GFP_HIGHUSER);
1052 down_write(&mm->mmap_sem);
1053 if (mm->uprobes_state.xol_area)
1058 /* Try to map as high as possible, this is only a hint. */
1059 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1060 if (area->vaddr & ~PAGE_MASK) {
1065 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1066 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1070 smp_wmb(); /* pairs with get_xol_area() */
1071 mm->uprobes_state.xol_area = area;
1075 up_write(&mm->mmap_sem);
1077 __free_page(area->page);
1082 static struct xol_area *get_xol_area(struct mm_struct *mm)
1084 struct xol_area *area;
1086 area = mm->uprobes_state.xol_area;
1087 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1093 * xol_alloc_area - Allocate process's xol_area.
1094 * This area will be used for storing instructions for execution out of
1097 * Returns the allocated area or NULL.
1099 static struct xol_area *xol_alloc_area(void)
1101 struct xol_area *area;
1103 area = kzalloc(sizeof(*area), GFP_KERNEL);
1104 if (unlikely(!area))
1107 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1112 init_waitqueue_head(&area->wq);
1113 if (!xol_add_vma(area))
1117 kfree(area->bitmap);
1120 return get_xol_area(current->mm);
1124 * uprobe_clear_state - Free the area allocated for slots.
1126 void uprobe_clear_state(struct mm_struct *mm)
1128 struct xol_area *area = mm->uprobes_state.xol_area;
1133 put_page(area->page);
1134 kfree(area->bitmap);
1138 void uprobe_start_dup_mmap(void)
1140 percpu_down_read(&dup_mmap_sem);
1143 void uprobe_end_dup_mmap(void)
1145 percpu_up_read(&dup_mmap_sem);
1148 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1150 newmm->uprobes_state.xol_area = NULL;
1152 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1153 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1154 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1155 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1160 * - search for a free slot.
1162 static unsigned long xol_take_insn_slot(struct xol_area *area)
1164 unsigned long slot_addr;
1168 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1169 if (slot_nr < UINSNS_PER_PAGE) {
1170 if (!test_and_set_bit(slot_nr, area->bitmap))
1173 slot_nr = UINSNS_PER_PAGE;
1176 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1177 } while (slot_nr >= UINSNS_PER_PAGE);
1179 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1180 atomic_inc(&area->slot_count);
1186 * xol_get_insn_slot - If was not allocated a slot, then
1188 * Returns the allocated slot address or 0.
1190 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1192 struct xol_area *area;
1193 unsigned long offset;
1196 area = get_xol_area(current->mm);
1198 area = xol_alloc_area();
1202 current->utask->xol_vaddr = xol_take_insn_slot(area);
1205 * Initialize the slot if xol_vaddr points to valid
1208 if (unlikely(!current->utask->xol_vaddr))
1211 current->utask->vaddr = slot_addr;
1212 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1213 vaddr = kmap_atomic(area->page);
1214 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1215 kunmap_atomic(vaddr);
1217 * We probably need flush_icache_user_range() but it needs vma.
1218 * This should work on supported architectures too.
1220 flush_dcache_page(area->page);
1222 return current->utask->xol_vaddr;
1226 * xol_free_insn_slot - If slot was earlier allocated by
1227 * @xol_get_insn_slot(), make the slot available for
1228 * subsequent requests.
1230 static void xol_free_insn_slot(struct task_struct *tsk)
1232 struct xol_area *area;
1233 unsigned long vma_end;
1234 unsigned long slot_addr;
1236 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1239 slot_addr = tsk->utask->xol_vaddr;
1241 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1244 area = tsk->mm->uprobes_state.xol_area;
1245 vma_end = area->vaddr + PAGE_SIZE;
1246 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1247 unsigned long offset;
1250 offset = slot_addr - area->vaddr;
1251 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1252 if (slot_nr >= UINSNS_PER_PAGE)
1255 clear_bit(slot_nr, area->bitmap);
1256 atomic_dec(&area->slot_count);
1257 if (waitqueue_active(&area->wq))
1260 tsk->utask->xol_vaddr = 0;
1265 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1266 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1268 * Return the address of the breakpoint instruction.
1270 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1272 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1276 * Called with no locks held.
1277 * Called in context of a exiting or a exec-ing thread.
1279 void uprobe_free_utask(struct task_struct *t)
1281 struct uprobe_task *utask = t->utask;
1286 if (utask->active_uprobe)
1287 put_uprobe(utask->active_uprobe);
1289 xol_free_insn_slot(t);
1295 * Called in context of a new clone/fork from copy_process.
1297 void uprobe_copy_process(struct task_struct *t)
1303 * Allocate a uprobe_task object for the task.
1304 * Called when the thread hits a breakpoint for the first time.
1307 * - pointer to new uprobe_task on success
1310 static struct uprobe_task *add_utask(void)
1312 struct uprobe_task *utask;
1314 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1315 if (unlikely(!utask))
1318 current->utask = utask;
1322 /* Prepare to single-step probed instruction out of line. */
1324 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1326 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1333 * If we are singlestepping, then ensure this thread is not connected to
1334 * non-fatal signals until completion of singlestep. When xol insn itself
1335 * triggers the signal, restart the original insn even if the task is
1336 * already SIGKILL'ed (since coredump should report the correct ip). This
1337 * is even more important if the task has a handler for SIGSEGV/etc, The
1338 * _same_ instruction should be repeated again after return from the signal
1339 * handler, and SSTEP can never finish in this case.
1341 bool uprobe_deny_signal(void)
1343 struct task_struct *t = current;
1344 struct uprobe_task *utask = t->utask;
1346 if (likely(!utask || !utask->active_uprobe))
1349 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1351 if (signal_pending(t)) {
1352 spin_lock_irq(&t->sighand->siglock);
1353 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1354 spin_unlock_irq(&t->sighand->siglock);
1356 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1357 utask->state = UTASK_SSTEP_TRAPPED;
1358 set_tsk_thread_flag(t, TIF_UPROBE);
1359 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1367 * Avoid singlestepping the original instruction if the original instruction
1368 * is a NOP or can be emulated.
1370 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1372 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1373 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1375 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1380 static void mmf_recalc_uprobes(struct mm_struct *mm)
1382 struct vm_area_struct *vma;
1384 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1385 if (!valid_vma(vma, false))
1388 * This is not strictly accurate, we can race with
1389 * uprobe_unregister() and see the already removed
1390 * uprobe if delete_uprobe() was not yet called.
1391 * Or this uprobe can be filtered out.
1393 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1397 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1400 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1403 uprobe_opcode_t opcode;
1406 pagefault_disable();
1407 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1411 if (likely(result == 0))
1414 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1418 copy_opcode(page, vaddr, &opcode);
1421 return is_swbp_insn(&opcode);
1424 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1426 struct mm_struct *mm = current->mm;
1427 struct uprobe *uprobe = NULL;
1428 struct vm_area_struct *vma;
1430 down_read(&mm->mmap_sem);
1431 vma = find_vma(mm, bp_vaddr);
1432 if (vma && vma->vm_start <= bp_vaddr) {
1433 if (valid_vma(vma, false)) {
1434 struct inode *inode = vma->vm_file->f_mapping->host;
1435 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1437 uprobe = find_uprobe(inode, offset);
1441 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1446 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1447 mmf_recalc_uprobes(mm);
1448 up_read(&mm->mmap_sem);
1454 * Run handler and ask thread to singlestep.
1455 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1457 static void handle_swbp(struct pt_regs *regs)
1459 struct uprobe_task *utask;
1460 struct uprobe *uprobe;
1461 unsigned long bp_vaddr;
1462 int uninitialized_var(is_swbp);
1464 bp_vaddr = uprobe_get_swbp_addr(regs);
1465 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1469 /* No matching uprobe; signal SIGTRAP. */
1470 send_sig(SIGTRAP, current, 0);
1473 * Either we raced with uprobe_unregister() or we can't
1474 * access this memory. The latter is only possible if
1475 * another thread plays with our ->mm. In both cases
1476 * we can simply restart. If this vma was unmapped we
1477 * can pretend this insn was not executed yet and get
1478 * the (correct) SIGSEGV after restart.
1480 instruction_pointer_set(regs, bp_vaddr);
1485 * TODO: move copy_insn/etc into _register and remove this hack.
1486 * After we hit the bp, _unregister + _register can install the
1487 * new and not-yet-analyzed uprobe at the same address, restart.
1489 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1490 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1493 utask = current->utask;
1495 utask = add_utask();
1496 /* Cannot allocate; re-execute the instruction. */
1501 handler_chain(uprobe, regs);
1502 if (can_skip_sstep(uprobe, regs))
1505 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1506 utask->active_uprobe = uprobe;
1507 utask->state = UTASK_SSTEP;
1513 * cannot singlestep; cannot skip instruction;
1514 * re-execute the instruction.
1516 instruction_pointer_set(regs, bp_vaddr);
1522 * Perform required fix-ups and disable singlestep.
1523 * Allow pending signals to take effect.
1525 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1527 struct uprobe *uprobe;
1529 uprobe = utask->active_uprobe;
1530 if (utask->state == UTASK_SSTEP_ACK)
1531 arch_uprobe_post_xol(&uprobe->arch, regs);
1532 else if (utask->state == UTASK_SSTEP_TRAPPED)
1533 arch_uprobe_abort_xol(&uprobe->arch, regs);
1538 utask->active_uprobe = NULL;
1539 utask->state = UTASK_RUNNING;
1540 xol_free_insn_slot(current);
1542 spin_lock_irq(¤t->sighand->siglock);
1543 recalc_sigpending(); /* see uprobe_deny_signal() */
1544 spin_unlock_irq(¤t->sighand->siglock);
1548 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1549 * allows the thread to return from interrupt. After that handle_swbp()
1550 * sets utask->active_uprobe.
1552 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1553 * and allows the thread to return from interrupt.
1555 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1556 * uprobe_notify_resume().
1558 void uprobe_notify_resume(struct pt_regs *regs)
1560 struct uprobe_task *utask;
1562 clear_thread_flag(TIF_UPROBE);
1564 utask = current->utask;
1565 if (utask && utask->active_uprobe)
1566 handle_singlestep(utask, regs);
1572 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1573 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1575 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1577 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1580 set_thread_flag(TIF_UPROBE);
1585 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1586 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1588 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1590 struct uprobe_task *utask = current->utask;
1592 if (!current->mm || !utask || !utask->active_uprobe)
1593 /* task is currently not uprobed */
1596 utask->state = UTASK_SSTEP_ACK;
1597 set_thread_flag(TIF_UPROBE);
1601 static struct notifier_block uprobe_exception_nb = {
1602 .notifier_call = arch_uprobe_exception_notify,
1603 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1606 static int __init init_uprobes(void)
1610 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1611 mutex_init(&uprobes_mutex[i]);
1612 mutex_init(&uprobes_mmap_mutex[i]);
1615 if (percpu_init_rwsem(&dup_mmap_sem))
1618 return register_die_notifier(&uprobe_exception_nb);
1620 module_init(init_uprobes);
1622 static void __exit exit_uprobes(void)
1625 module_exit(exit_uprobes);