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);
436 static void put_uprobe(struct uprobe *uprobe)
438 if (atomic_dec_and_test(&uprobe->ref))
442 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
444 struct uprobe *uprobe, *cur_uprobe;
446 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
450 uprobe->inode = igrab(inode);
451 uprobe->offset = offset;
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 if (!test_bit(UPROBE_RUN_HANDLER, &uprobe->flags))
479 down_read(&uprobe->consumer_rwsem);
480 for (uc = uprobe->consumers; uc; uc = uc->next) {
481 if (!uc->filter || uc->filter(uc, current))
482 uc->handler(uc, regs);
484 up_read(&uprobe->consumer_rwsem);
487 /* Returns the previous consumer */
488 static struct uprobe_consumer *
489 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
491 down_write(&uprobe->consumer_rwsem);
492 uc->next = uprobe->consumers;
493 uprobe->consumers = uc;
494 up_write(&uprobe->consumer_rwsem);
500 * For uprobe @uprobe, delete the consumer @uc.
501 * Return true if the @uc is deleted successfully
504 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
506 struct uprobe_consumer **con;
509 down_write(&uprobe->consumer_rwsem);
510 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
517 up_write(&uprobe->consumer_rwsem);
523 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
524 unsigned long nbytes, loff_t offset)
534 if (!mapping->a_ops->readpage)
537 idx = offset >> PAGE_CACHE_SHIFT;
538 off = offset & ~PAGE_MASK;
541 * Ensure that the page that has the original instruction is
542 * populated and in page-cache.
544 page = read_mapping_page(mapping, idx, filp);
546 return PTR_ERR(page);
548 vaddr = kmap_atomic(page);
549 memcpy(insn, vaddr + off, nbytes);
550 kunmap_atomic(vaddr);
551 page_cache_release(page);
556 static int copy_insn(struct uprobe *uprobe, struct file *filp)
558 struct address_space *mapping;
559 unsigned long nbytes;
562 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
563 mapping = uprobe->inode->i_mapping;
565 /* Instruction at end of binary; copy only available bytes */
566 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
567 bytes = uprobe->inode->i_size - uprobe->offset;
569 bytes = MAX_UINSN_BYTES;
571 /* Instruction at the page-boundary; copy bytes in second page */
572 if (nbytes < bytes) {
573 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
574 bytes - nbytes, uprobe->offset + nbytes);
579 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
582 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
583 struct mm_struct *mm, unsigned long vaddr)
587 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
590 mutex_lock(&uprobe->copy_mutex);
591 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
594 ret = copy_insn(uprobe, file);
599 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
602 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
606 /* write_opcode() assumes we don't cross page boundary */
607 BUG_ON((uprobe->offset & ~PAGE_MASK) +
608 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
610 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
611 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
614 mutex_unlock(&uprobe->copy_mutex);
620 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
621 struct vm_area_struct *vma, unsigned long vaddr)
627 * If probe is being deleted, unregister thread could be done with
628 * the vma-rmap-walk through. Adding a probe now can be fatal since
629 * nobody will be able to cleanup. Also we could be from fork or
630 * mremap path, where the probe might have already been inserted.
631 * Hence behave as if probe already existed.
633 if (!uprobe->consumers)
636 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
641 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
642 * the task can hit this breakpoint right after __replace_page().
644 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
646 set_bit(MMF_HAS_UPROBES, &mm->flags);
648 ret = set_swbp(&uprobe->arch, mm, vaddr);
650 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
651 else if (first_uprobe)
652 clear_bit(MMF_HAS_UPROBES, &mm->flags);
658 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
660 /* can happen if uprobe_register() fails */
661 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
664 set_bit(MMF_RECALC_UPROBES, &mm->flags);
665 return set_orig_insn(&uprobe->arch, mm, vaddr);
669 * There could be threads that have already hit the breakpoint. They
670 * will recheck the current insn and restart if find_uprobe() fails.
671 * See find_active_uprobe().
673 static void delete_uprobe(struct uprobe *uprobe)
675 spin_lock(&uprobes_treelock);
676 rb_erase(&uprobe->rb_node, &uprobes_tree);
677 spin_unlock(&uprobes_treelock);
680 atomic_dec(&uprobe_events);
684 struct map_info *next;
685 struct mm_struct *mm;
689 static inline struct map_info *free_map_info(struct map_info *info)
691 struct map_info *next = info->next;
696 static struct map_info *
697 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
699 unsigned long pgoff = offset >> PAGE_SHIFT;
700 struct vm_area_struct *vma;
701 struct map_info *curr = NULL;
702 struct map_info *prev = NULL;
703 struct map_info *info;
707 mutex_lock(&mapping->i_mmap_mutex);
708 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
709 if (!valid_vma(vma, is_register))
712 if (!prev && !more) {
714 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
715 * reclaim. This is optimistic, no harm done if it fails.
717 prev = kmalloc(sizeof(struct map_info),
718 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
727 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
735 info->mm = vma->vm_mm;
736 info->vaddr = offset_to_vaddr(vma, offset);
738 mutex_unlock(&mapping->i_mmap_mutex);
750 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
752 curr = ERR_PTR(-ENOMEM);
762 prev = free_map_info(prev);
766 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
768 struct map_info *info;
771 percpu_down_write(&dup_mmap_sem);
772 info = build_map_info(uprobe->inode->i_mapping,
773 uprobe->offset, is_register);
780 struct mm_struct *mm = info->mm;
781 struct vm_area_struct *vma;
783 if (err && is_register)
786 down_write(&mm->mmap_sem);
787 vma = find_vma(mm, info->vaddr);
788 if (!vma || !valid_vma(vma, is_register) ||
789 vma->vm_file->f_mapping->host != uprobe->inode)
792 if (vma->vm_start > info->vaddr ||
793 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
797 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
799 err |= remove_breakpoint(uprobe, mm, info->vaddr);
802 up_write(&mm->mmap_sem);
805 info = free_map_info(info);
808 percpu_up_write(&dup_mmap_sem);
812 static int __uprobe_register(struct uprobe *uprobe)
814 return register_for_each_vma(uprobe, true);
817 static void __uprobe_unregister(struct uprobe *uprobe)
819 if (!register_for_each_vma(uprobe, false))
820 delete_uprobe(uprobe);
822 /* TODO : cant unregister? schedule a worker thread */
826 * uprobe_register - register a probe
827 * @inode: the file in which the probe has to be placed.
828 * @offset: offset from the start of the file.
829 * @uc: information on howto handle the probe..
831 * Apart from the access refcount, uprobe_register() takes a creation
832 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
833 * inserted into the rbtree (i.e first consumer for a @inode:@offset
834 * tuple). Creation refcount stops uprobe_unregister from freeing the
835 * @uprobe even before the register operation is complete. Creation
836 * refcount is released when the last @uc for the @uprobe
839 * Return errno if it cannot successully install probes
840 * else return 0 (success)
842 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
844 struct uprobe *uprobe;
847 /* Racy, just to catch the obvious mistakes */
848 if (offset > i_size_read(inode))
852 mutex_lock(uprobes_hash(inode));
853 uprobe = alloc_uprobe(inode, offset);
857 } else if (!consumer_add(uprobe, uc)) {
858 ret = __uprobe_register(uprobe);
860 uprobe->consumers = NULL;
861 __uprobe_unregister(uprobe);
863 set_bit(UPROBE_RUN_HANDLER, &uprobe->flags);
867 mutex_unlock(uprobes_hash(inode));
875 * uprobe_unregister - unregister a already registered probe.
876 * @inode: the file in which the probe has to be removed.
877 * @offset: offset from the start of the file.
878 * @uc: identify which probe if multiple probes are colocated.
880 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
882 struct uprobe *uprobe;
884 uprobe = find_uprobe(inode, offset);
888 mutex_lock(uprobes_hash(inode));
890 if (consumer_del(uprobe, uc)) {
891 if (!uprobe->consumers) {
892 __uprobe_unregister(uprobe);
893 clear_bit(UPROBE_RUN_HANDLER, &uprobe->flags);
897 mutex_unlock(uprobes_hash(inode));
901 static struct rb_node *
902 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
904 struct rb_node *n = uprobes_tree.rb_node;
907 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
909 if (inode < u->inode) {
911 } else if (inode > u->inode) {
916 else if (min > u->offset)
927 * For a given range in vma, build a list of probes that need to be inserted.
929 static void build_probe_list(struct inode *inode,
930 struct vm_area_struct *vma,
931 unsigned long start, unsigned long end,
932 struct list_head *head)
935 struct rb_node *n, *t;
938 INIT_LIST_HEAD(head);
939 min = vaddr_to_offset(vma, start);
940 max = min + (end - start) - 1;
942 spin_lock(&uprobes_treelock);
943 n = find_node_in_range(inode, min, max);
945 for (t = n; t; t = rb_prev(t)) {
946 u = rb_entry(t, struct uprobe, rb_node);
947 if (u->inode != inode || u->offset < min)
949 list_add(&u->pending_list, head);
952 for (t = n; (t = rb_next(t)); ) {
953 u = rb_entry(t, struct uprobe, rb_node);
954 if (u->inode != inode || u->offset > max)
956 list_add(&u->pending_list, head);
960 spin_unlock(&uprobes_treelock);
964 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
966 * Currently we ignore all errors and always return 0, the callers
967 * can't handle the failure anyway.
969 int uprobe_mmap(struct vm_area_struct *vma)
971 struct list_head tmp_list;
972 struct uprobe *uprobe, *u;
975 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
978 inode = vma->vm_file->f_mapping->host;
982 mutex_lock(uprobes_mmap_hash(inode));
983 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
985 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
986 if (!fatal_signal_pending(current)) {
987 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
988 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
992 mutex_unlock(uprobes_mmap_hash(inode));
998 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1001 struct inode *inode;
1004 inode = vma->vm_file->f_mapping->host;
1006 min = vaddr_to_offset(vma, start);
1007 max = min + (end - start) - 1;
1009 spin_lock(&uprobes_treelock);
1010 n = find_node_in_range(inode, min, max);
1011 spin_unlock(&uprobes_treelock);
1017 * Called in context of a munmap of a vma.
1019 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1021 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1024 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1027 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1028 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1031 if (vma_has_uprobes(vma, start, end))
1032 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1035 /* Slot allocation for XOL */
1036 static int xol_add_vma(struct xol_area *area)
1038 struct mm_struct *mm;
1041 area->page = alloc_page(GFP_HIGHUSER);
1048 down_write(&mm->mmap_sem);
1049 if (mm->uprobes_state.xol_area)
1054 /* Try to map as high as possible, this is only a hint. */
1055 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1056 if (area->vaddr & ~PAGE_MASK) {
1061 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1062 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1066 smp_wmb(); /* pairs with get_xol_area() */
1067 mm->uprobes_state.xol_area = area;
1071 up_write(&mm->mmap_sem);
1073 __free_page(area->page);
1078 static struct xol_area *get_xol_area(struct mm_struct *mm)
1080 struct xol_area *area;
1082 area = mm->uprobes_state.xol_area;
1083 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1089 * xol_alloc_area - Allocate process's xol_area.
1090 * This area will be used for storing instructions for execution out of
1093 * Returns the allocated area or NULL.
1095 static struct xol_area *xol_alloc_area(void)
1097 struct xol_area *area;
1099 area = kzalloc(sizeof(*area), GFP_KERNEL);
1100 if (unlikely(!area))
1103 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1108 init_waitqueue_head(&area->wq);
1109 if (!xol_add_vma(area))
1113 kfree(area->bitmap);
1116 return get_xol_area(current->mm);
1120 * uprobe_clear_state - Free the area allocated for slots.
1122 void uprobe_clear_state(struct mm_struct *mm)
1124 struct xol_area *area = mm->uprobes_state.xol_area;
1129 put_page(area->page);
1130 kfree(area->bitmap);
1134 void uprobe_start_dup_mmap(void)
1136 percpu_down_read(&dup_mmap_sem);
1139 void uprobe_end_dup_mmap(void)
1141 percpu_up_read(&dup_mmap_sem);
1144 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1146 newmm->uprobes_state.xol_area = NULL;
1148 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1149 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1150 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1151 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1156 * - search for a free slot.
1158 static unsigned long xol_take_insn_slot(struct xol_area *area)
1160 unsigned long slot_addr;
1164 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1165 if (slot_nr < UINSNS_PER_PAGE) {
1166 if (!test_and_set_bit(slot_nr, area->bitmap))
1169 slot_nr = UINSNS_PER_PAGE;
1172 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1173 } while (slot_nr >= UINSNS_PER_PAGE);
1175 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1176 atomic_inc(&area->slot_count);
1182 * xol_get_insn_slot - If was not allocated a slot, then
1184 * Returns the allocated slot address or 0.
1186 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1188 struct xol_area *area;
1189 unsigned long offset;
1192 area = get_xol_area(current->mm);
1194 area = xol_alloc_area();
1198 current->utask->xol_vaddr = xol_take_insn_slot(area);
1201 * Initialize the slot if xol_vaddr points to valid
1204 if (unlikely(!current->utask->xol_vaddr))
1207 current->utask->vaddr = slot_addr;
1208 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1209 vaddr = kmap_atomic(area->page);
1210 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1211 kunmap_atomic(vaddr);
1213 * We probably need flush_icache_user_range() but it needs vma.
1214 * This should work on supported architectures too.
1216 flush_dcache_page(area->page);
1218 return current->utask->xol_vaddr;
1222 * xol_free_insn_slot - If slot was earlier allocated by
1223 * @xol_get_insn_slot(), make the slot available for
1224 * subsequent requests.
1226 static void xol_free_insn_slot(struct task_struct *tsk)
1228 struct xol_area *area;
1229 unsigned long vma_end;
1230 unsigned long slot_addr;
1232 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1235 slot_addr = tsk->utask->xol_vaddr;
1237 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1240 area = tsk->mm->uprobes_state.xol_area;
1241 vma_end = area->vaddr + PAGE_SIZE;
1242 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1243 unsigned long offset;
1246 offset = slot_addr - area->vaddr;
1247 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1248 if (slot_nr >= UINSNS_PER_PAGE)
1251 clear_bit(slot_nr, area->bitmap);
1252 atomic_dec(&area->slot_count);
1253 if (waitqueue_active(&area->wq))
1256 tsk->utask->xol_vaddr = 0;
1261 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1262 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1264 * Return the address of the breakpoint instruction.
1266 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1268 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1272 * Called with no locks held.
1273 * Called in context of a exiting or a exec-ing thread.
1275 void uprobe_free_utask(struct task_struct *t)
1277 struct uprobe_task *utask = t->utask;
1282 if (utask->active_uprobe)
1283 put_uprobe(utask->active_uprobe);
1285 xol_free_insn_slot(t);
1291 * Called in context of a new clone/fork from copy_process.
1293 void uprobe_copy_process(struct task_struct *t)
1299 * Allocate a uprobe_task object for the task.
1300 * Called when the thread hits a breakpoint for the first time.
1303 * - pointer to new uprobe_task on success
1306 static struct uprobe_task *add_utask(void)
1308 struct uprobe_task *utask;
1310 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1311 if (unlikely(!utask))
1314 current->utask = utask;
1318 /* Prepare to single-step probed instruction out of line. */
1320 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1322 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1329 * If we are singlestepping, then ensure this thread is not connected to
1330 * non-fatal signals until completion of singlestep. When xol insn itself
1331 * triggers the signal, restart the original insn even if the task is
1332 * already SIGKILL'ed (since coredump should report the correct ip). This
1333 * is even more important if the task has a handler for SIGSEGV/etc, The
1334 * _same_ instruction should be repeated again after return from the signal
1335 * handler, and SSTEP can never finish in this case.
1337 bool uprobe_deny_signal(void)
1339 struct task_struct *t = current;
1340 struct uprobe_task *utask = t->utask;
1342 if (likely(!utask || !utask->active_uprobe))
1345 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1347 if (signal_pending(t)) {
1348 spin_lock_irq(&t->sighand->siglock);
1349 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1350 spin_unlock_irq(&t->sighand->siglock);
1352 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1353 utask->state = UTASK_SSTEP_TRAPPED;
1354 set_tsk_thread_flag(t, TIF_UPROBE);
1355 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1363 * Avoid singlestepping the original instruction if the original instruction
1364 * is a NOP or can be emulated.
1366 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1368 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1369 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1371 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1376 static void mmf_recalc_uprobes(struct mm_struct *mm)
1378 struct vm_area_struct *vma;
1380 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1381 if (!valid_vma(vma, false))
1384 * This is not strictly accurate, we can race with
1385 * uprobe_unregister() and see the already removed
1386 * uprobe if delete_uprobe() was not yet called.
1388 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1392 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1395 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1398 uprobe_opcode_t opcode;
1401 pagefault_disable();
1402 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1406 if (likely(result == 0))
1409 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1413 copy_opcode(page, vaddr, &opcode);
1416 return is_swbp_insn(&opcode);
1419 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1421 struct mm_struct *mm = current->mm;
1422 struct uprobe *uprobe = NULL;
1423 struct vm_area_struct *vma;
1425 down_read(&mm->mmap_sem);
1426 vma = find_vma(mm, bp_vaddr);
1427 if (vma && vma->vm_start <= bp_vaddr) {
1428 if (valid_vma(vma, false)) {
1429 struct inode *inode = vma->vm_file->f_mapping->host;
1430 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1432 uprobe = find_uprobe(inode, offset);
1436 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1441 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1442 mmf_recalc_uprobes(mm);
1443 up_read(&mm->mmap_sem);
1449 * Run handler and ask thread to singlestep.
1450 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1452 static void handle_swbp(struct pt_regs *regs)
1454 struct uprobe_task *utask;
1455 struct uprobe *uprobe;
1456 unsigned long bp_vaddr;
1457 int uninitialized_var(is_swbp);
1459 bp_vaddr = uprobe_get_swbp_addr(regs);
1460 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1464 /* No matching uprobe; signal SIGTRAP. */
1465 send_sig(SIGTRAP, current, 0);
1468 * Either we raced with uprobe_unregister() or we can't
1469 * access this memory. The latter is only possible if
1470 * another thread plays with our ->mm. In both cases
1471 * we can simply restart. If this vma was unmapped we
1472 * can pretend this insn was not executed yet and get
1473 * the (correct) SIGSEGV after restart.
1475 instruction_pointer_set(regs, bp_vaddr);
1480 * TODO: move copy_insn/etc into _register and remove this hack.
1481 * After we hit the bp, _unregister + _register can install the
1482 * new and not-yet-analyzed uprobe at the same address, restart.
1484 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1485 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1488 utask = current->utask;
1490 utask = add_utask();
1491 /* Cannot allocate; re-execute the instruction. */
1496 handler_chain(uprobe, regs);
1497 if (can_skip_sstep(uprobe, regs))
1500 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1501 utask->active_uprobe = uprobe;
1502 utask->state = UTASK_SSTEP;
1508 * cannot singlestep; cannot skip instruction;
1509 * re-execute the instruction.
1511 instruction_pointer_set(regs, bp_vaddr);
1517 * Perform required fix-ups and disable singlestep.
1518 * Allow pending signals to take effect.
1520 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1522 struct uprobe *uprobe;
1524 uprobe = utask->active_uprobe;
1525 if (utask->state == UTASK_SSTEP_ACK)
1526 arch_uprobe_post_xol(&uprobe->arch, regs);
1527 else if (utask->state == UTASK_SSTEP_TRAPPED)
1528 arch_uprobe_abort_xol(&uprobe->arch, regs);
1533 utask->active_uprobe = NULL;
1534 utask->state = UTASK_RUNNING;
1535 xol_free_insn_slot(current);
1537 spin_lock_irq(¤t->sighand->siglock);
1538 recalc_sigpending(); /* see uprobe_deny_signal() */
1539 spin_unlock_irq(¤t->sighand->siglock);
1543 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1544 * allows the thread to return from interrupt. After that handle_swbp()
1545 * sets utask->active_uprobe.
1547 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1548 * and allows the thread to return from interrupt.
1550 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1551 * uprobe_notify_resume().
1553 void uprobe_notify_resume(struct pt_regs *regs)
1555 struct uprobe_task *utask;
1557 clear_thread_flag(TIF_UPROBE);
1559 utask = current->utask;
1560 if (utask && utask->active_uprobe)
1561 handle_singlestep(utask, regs);
1567 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1568 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1570 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1572 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1575 set_thread_flag(TIF_UPROBE);
1580 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1581 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1583 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1585 struct uprobe_task *utask = current->utask;
1587 if (!current->mm || !utask || !utask->active_uprobe)
1588 /* task is currently not uprobed */
1591 utask->state = UTASK_SSTEP_ACK;
1592 set_thread_flag(TIF_UPROBE);
1596 static struct notifier_block uprobe_exception_nb = {
1597 .notifier_call = arch_uprobe_exception_notify,
1598 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1601 static int __init init_uprobes(void)
1605 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1606 mutex_init(&uprobes_mutex[i]);
1607 mutex_init(&uprobes_mmap_mutex[i]);
1610 if (percpu_init_rwsem(&dup_mmap_sem))
1613 return register_die_notifier(&uprobe_exception_nb);
1615 module_init(init_uprobes);
1617 static void __exit exit_uprobes(void)
1620 module_exit(exit_uprobes);