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 */
37 #include <linux/uprobes.h>
39 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
40 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
42 static struct rb_root uprobes_tree = RB_ROOT;
44 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
46 #define UPROBES_HASH_SZ 13
49 * We need separate register/unregister and mmap/munmap lock hashes because
50 * of mmap_sem nesting.
52 * uprobe_register() needs to install probes on (potentially) all processes
53 * and thus needs to acquire multiple mmap_sems (consequtively, not
54 * concurrently), whereas uprobe_mmap() is called while holding mmap_sem
55 * for the particular process doing the mmap.
57 * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem
58 * because of lock order against i_mmap_mutex. This means there's a hole in
59 * the register vma iteration where a mmap() can happen.
61 * Thus uprobe_register() can race with uprobe_mmap() and we can try and
62 * install a probe where one is already installed.
65 /* serialize (un)register */
66 static struct mutex uprobes_mutex[UPROBES_HASH_SZ];
68 #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
70 /* serialize uprobe->pending_list */
71 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
72 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
75 * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe
76 * events active at this time. Probably a fine grained per inode count is
79 static atomic_t uprobe_events = ATOMIC_INIT(0);
82 struct rb_node rb_node; /* node in the rb tree */
84 struct rw_semaphore consumer_rwsem;
85 struct list_head pending_list;
86 struct uprobe_consumer *consumers;
87 struct inode *inode; /* Also hold a ref to inode */
90 struct arch_uprobe arch;
94 * valid_vma: Verify if the specified vma is an executable vma
95 * Relax restrictions while unregistering: vm_flags might have
96 * changed after breakpoint was inserted.
97 * - is_register: indicates if we are in register context.
98 * - Return 1 if the specified virtual address is in an
101 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
109 if ((vma->vm_flags & (VM_HUGETLB|VM_READ|VM_WRITE|VM_EXEC|VM_SHARED))
110 == (VM_READ|VM_EXEC))
116 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
118 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
121 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
123 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
127 * __replace_page - replace page in vma by new page.
128 * based on replace_page in mm/ksm.c
130 * @vma: vma that holds the pte pointing to page
131 * @addr: address the old @page is mapped at
132 * @page: the cowed page we are replacing by kpage
133 * @kpage: the modified page we replace page by
135 * Returns 0 on success, -EFAULT on failure.
137 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
138 struct page *page, struct page *kpage)
140 struct mm_struct *mm = vma->vm_mm;
144 /* For mmu_notifiers */
145 const unsigned long mmun_start = addr;
146 const unsigned long mmun_end = addr + PAGE_SIZE;
148 /* For try_to_free_swap() and munlock_vma_page() below */
151 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
153 ptep = page_check_address(page, mm, addr, &ptl, 0);
158 page_add_new_anon_rmap(kpage, vma, addr);
160 if (!PageAnon(page)) {
161 dec_mm_counter(mm, MM_FILEPAGES);
162 inc_mm_counter(mm, MM_ANONPAGES);
165 flush_cache_page(vma, addr, pte_pfn(*ptep));
166 ptep_clear_flush(vma, addr, ptep);
167 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
169 page_remove_rmap(page);
170 if (!page_mapped(page))
171 try_to_free_swap(page);
172 pte_unmap_unlock(ptep, ptl);
174 if (vma->vm_flags & VM_LOCKED)
175 munlock_vma_page(page);
180 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
186 * is_swbp_insn - check if instruction is breakpoint instruction.
187 * @insn: instruction to be checked.
188 * Default implementation of is_swbp_insn
189 * Returns true if @insn is a breakpoint instruction.
191 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
193 return *insn == UPROBE_SWBP_INSN;
198 * Expect the breakpoint instruction to be the smallest size instruction for
199 * the architecture. If an arch has variable length instruction and the
200 * breakpoint instruction is not of the smallest length instruction
201 * supported by that architecture then we need to modify read_opcode /
202 * write_opcode accordingly. This would never be a problem for archs that
203 * have fixed length instructions.
207 * write_opcode - write the opcode at a given virtual address.
208 * @auprobe: arch breakpointing information.
209 * @mm: the probed process address space.
210 * @vaddr: the virtual address to store the opcode.
211 * @opcode: opcode to be written at @vaddr.
213 * Called with mm->mmap_sem held (for read and with a reference to
216 * For mm @mm, write the opcode at @vaddr.
217 * Return 0 (success) or a negative errno.
219 static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
220 unsigned long vaddr, uprobe_opcode_t opcode)
222 struct page *old_page, *new_page;
223 void *vaddr_old, *vaddr_new;
224 struct vm_area_struct *vma;
228 /* Read the page with vaddr into memory */
229 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
234 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
238 __SetPageUptodate(new_page);
240 /* copy the page now that we've got it stable */
241 vaddr_old = kmap_atomic(old_page);
242 vaddr_new = kmap_atomic(new_page);
244 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
245 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
247 kunmap_atomic(vaddr_new);
248 kunmap_atomic(vaddr_old);
250 ret = anon_vma_prepare(vma);
254 ret = __replace_page(vma, vaddr, old_page, new_page);
257 page_cache_release(new_page);
261 if (unlikely(ret == -EAGAIN))
267 * read_opcode - read the opcode at a given virtual address.
268 * @mm: the probed process address space.
269 * @vaddr: the virtual address to read the opcode.
270 * @opcode: location to store the read opcode.
272 * Called with mm->mmap_sem held (for read and with a reference to
275 * For mm @mm, read the opcode at @vaddr and store it in @opcode.
276 * Return 0 (success) or a negative errno.
278 static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode)
284 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
288 vaddr_new = kmap_atomic(page);
290 memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE);
291 kunmap_atomic(vaddr_new);
298 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
300 uprobe_opcode_t opcode;
303 if (current->mm == mm) {
305 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
309 if (likely(result == 0))
313 result = read_opcode(mm, vaddr, &opcode);
317 if (is_swbp_insn(&opcode))
324 * set_swbp - store breakpoint at a given address.
325 * @auprobe: arch specific probepoint information.
326 * @mm: the probed process address space.
327 * @vaddr: the virtual address to insert the opcode.
329 * For mm @mm, store the breakpoint instruction at @vaddr.
330 * Return 0 (success) or a negative errno.
332 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
336 * See the comment near uprobes_hash().
338 result = is_swbp_at_addr(mm, vaddr);
345 return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
349 * set_orig_insn - Restore the original instruction.
350 * @mm: the probed process address space.
351 * @auprobe: arch specific probepoint information.
352 * @vaddr: the virtual address to insert the opcode.
354 * For mm @mm, restore the original opcode (opcode) at @vaddr.
355 * Return 0 (success) or a negative errno.
358 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
362 result = is_swbp_at_addr(mm, vaddr);
369 return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
372 static int match_uprobe(struct uprobe *l, struct uprobe *r)
374 if (l->inode < r->inode)
377 if (l->inode > r->inode)
380 if (l->offset < r->offset)
383 if (l->offset > r->offset)
389 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
391 struct uprobe u = { .inode = inode, .offset = offset };
392 struct rb_node *n = uprobes_tree.rb_node;
393 struct uprobe *uprobe;
397 uprobe = rb_entry(n, struct uprobe, rb_node);
398 match = match_uprobe(&u, uprobe);
400 atomic_inc(&uprobe->ref);
413 * Find a uprobe corresponding to a given inode:offset
414 * Acquires uprobes_treelock
416 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
418 struct uprobe *uprobe;
420 spin_lock(&uprobes_treelock);
421 uprobe = __find_uprobe(inode, offset);
422 spin_unlock(&uprobes_treelock);
427 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
429 struct rb_node **p = &uprobes_tree.rb_node;
430 struct rb_node *parent = NULL;
436 u = rb_entry(parent, struct uprobe, rb_node);
437 match = match_uprobe(uprobe, u);
444 p = &parent->rb_left;
446 p = &parent->rb_right;
451 rb_link_node(&uprobe->rb_node, parent, p);
452 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
453 /* get access + creation ref */
454 atomic_set(&uprobe->ref, 2);
460 * Acquire uprobes_treelock.
461 * Matching uprobe already exists in rbtree;
462 * increment (access refcount) and return the matching uprobe.
464 * No matching uprobe; insert the uprobe in rb_tree;
465 * get a double refcount (access + creation) and return NULL.
467 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
471 spin_lock(&uprobes_treelock);
472 u = __insert_uprobe(uprobe);
473 spin_unlock(&uprobes_treelock);
475 /* For now assume that the instruction need not be single-stepped */
476 uprobe->flags |= UPROBE_SKIP_SSTEP;
481 static void put_uprobe(struct uprobe *uprobe)
483 if (atomic_dec_and_test(&uprobe->ref))
487 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
489 struct uprobe *uprobe, *cur_uprobe;
491 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
495 uprobe->inode = igrab(inode);
496 uprobe->offset = offset;
497 init_rwsem(&uprobe->consumer_rwsem);
499 /* add to uprobes_tree, sorted on inode:offset */
500 cur_uprobe = insert_uprobe(uprobe);
502 /* a uprobe exists for this inode:offset combination */
508 atomic_inc(&uprobe_events);
514 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
516 struct uprobe_consumer *uc;
518 if (!(uprobe->flags & UPROBE_RUN_HANDLER))
521 down_read(&uprobe->consumer_rwsem);
522 for (uc = uprobe->consumers; uc; uc = uc->next) {
523 if (!uc->filter || uc->filter(uc, current))
524 uc->handler(uc, regs);
526 up_read(&uprobe->consumer_rwsem);
529 /* Returns the previous consumer */
530 static struct uprobe_consumer *
531 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
533 down_write(&uprobe->consumer_rwsem);
534 uc->next = uprobe->consumers;
535 uprobe->consumers = uc;
536 up_write(&uprobe->consumer_rwsem);
542 * For uprobe @uprobe, delete the consumer @uc.
543 * Return true if the @uc is deleted successfully
546 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
548 struct uprobe_consumer **con;
551 down_write(&uprobe->consumer_rwsem);
552 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
559 up_write(&uprobe->consumer_rwsem);
565 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
566 unsigned long nbytes, loff_t offset)
576 if (!mapping->a_ops->readpage)
579 idx = offset >> PAGE_CACHE_SHIFT;
580 off = offset & ~PAGE_MASK;
583 * Ensure that the page that has the original instruction is
584 * populated and in page-cache.
586 page = read_mapping_page(mapping, idx, filp);
588 return PTR_ERR(page);
590 vaddr = kmap_atomic(page);
591 memcpy(insn, vaddr + off, nbytes);
592 kunmap_atomic(vaddr);
593 page_cache_release(page);
598 static int copy_insn(struct uprobe *uprobe, struct file *filp)
600 struct address_space *mapping;
601 unsigned long nbytes;
604 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
605 mapping = uprobe->inode->i_mapping;
607 /* Instruction at end of binary; copy only available bytes */
608 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
609 bytes = uprobe->inode->i_size - uprobe->offset;
611 bytes = MAX_UINSN_BYTES;
613 /* Instruction at the page-boundary; copy bytes in second page */
614 if (nbytes < bytes) {
615 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
616 bytes - nbytes, uprobe->offset + nbytes);
621 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
625 * How mm->uprobes_state.count gets updated
626 * uprobe_mmap() increments the count if
627 * - it successfully adds a breakpoint.
628 * - it cannot add a breakpoint, but sees that there is a underlying
629 * breakpoint (via a is_swbp_at_addr()).
631 * uprobe_munmap() decrements the count if
632 * - it sees a underlying breakpoint, (via is_swbp_at_addr)
633 * (Subsequent uprobe_unregister wouldnt find the breakpoint
634 * unless a uprobe_mmap kicks in, since the old vma would be
635 * dropped just after uprobe_munmap.)
637 * uprobe_register increments the count if:
638 * - it successfully adds a breakpoint.
640 * uprobe_unregister decrements the count if:
641 * - it sees a underlying breakpoint and removes successfully.
642 * (via is_swbp_at_addr)
643 * (Subsequent uprobe_munmap wouldnt find the breakpoint
644 * since there is no underlying breakpoint after the
645 * breakpoint removal.)
648 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
649 struct vm_area_struct *vma, unsigned long vaddr)
655 * If probe is being deleted, unregister thread could be done with
656 * the vma-rmap-walk through. Adding a probe now can be fatal since
657 * nobody will be able to cleanup. Also we could be from fork or
658 * mremap path, where the probe might have already been inserted.
659 * Hence behave as if probe already existed.
661 if (!uprobe->consumers)
664 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
665 ret = copy_insn(uprobe, vma->vm_file);
669 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
672 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
676 /* write_opcode() assumes we don't cross page boundary */
677 BUG_ON((uprobe->offset & ~PAGE_MASK) +
678 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
680 uprobe->flags |= UPROBE_COPY_INSN;
684 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
685 * the task can hit this breakpoint right after __replace_page().
687 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
689 set_bit(MMF_HAS_UPROBES, &mm->flags);
691 ret = set_swbp(&uprobe->arch, mm, vaddr);
693 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
694 else if (first_uprobe)
695 clear_bit(MMF_HAS_UPROBES, &mm->flags);
701 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
703 /* can happen if uprobe_register() fails */
704 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
707 set_bit(MMF_RECALC_UPROBES, &mm->flags);
708 set_orig_insn(&uprobe->arch, mm, vaddr);
712 * There could be threads that have already hit the breakpoint. They
713 * will recheck the current insn and restart if find_uprobe() fails.
714 * See find_active_uprobe().
716 static void delete_uprobe(struct uprobe *uprobe)
718 spin_lock(&uprobes_treelock);
719 rb_erase(&uprobe->rb_node, &uprobes_tree);
720 spin_unlock(&uprobes_treelock);
723 atomic_dec(&uprobe_events);
727 struct map_info *next;
728 struct mm_struct *mm;
732 static inline struct map_info *free_map_info(struct map_info *info)
734 struct map_info *next = info->next;
739 static struct map_info *
740 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
742 unsigned long pgoff = offset >> PAGE_SHIFT;
743 struct vm_area_struct *vma;
744 struct map_info *curr = NULL;
745 struct map_info *prev = NULL;
746 struct map_info *info;
750 mutex_lock(&mapping->i_mmap_mutex);
751 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
752 if (!valid_vma(vma, is_register))
755 if (!prev && !more) {
757 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
758 * reclaim. This is optimistic, no harm done if it fails.
760 prev = kmalloc(sizeof(struct map_info),
761 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
770 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
778 info->mm = vma->vm_mm;
779 info->vaddr = offset_to_vaddr(vma, offset);
781 mutex_unlock(&mapping->i_mmap_mutex);
793 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
795 curr = ERR_PTR(-ENOMEM);
805 prev = free_map_info(prev);
809 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
811 struct map_info *info;
814 info = build_map_info(uprobe->inode->i_mapping,
815 uprobe->offset, is_register);
817 return PTR_ERR(info);
820 struct mm_struct *mm = info->mm;
821 struct vm_area_struct *vma;
826 down_write(&mm->mmap_sem);
827 vma = find_vma(mm, info->vaddr);
828 if (!vma || !valid_vma(vma, is_register) ||
829 vma->vm_file->f_mapping->host != uprobe->inode)
832 if (vma->vm_start > info->vaddr ||
833 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
837 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
839 remove_breakpoint(uprobe, mm, info->vaddr);
842 up_write(&mm->mmap_sem);
845 info = free_map_info(info);
851 static int __uprobe_register(struct uprobe *uprobe)
853 return register_for_each_vma(uprobe, true);
856 static void __uprobe_unregister(struct uprobe *uprobe)
858 if (!register_for_each_vma(uprobe, false))
859 delete_uprobe(uprobe);
861 /* TODO : cant unregister? schedule a worker thread */
865 * uprobe_register - register a probe
866 * @inode: the file in which the probe has to be placed.
867 * @offset: offset from the start of the file.
868 * @uc: information on howto handle the probe..
870 * Apart from the access refcount, uprobe_register() takes a creation
871 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
872 * inserted into the rbtree (i.e first consumer for a @inode:@offset
873 * tuple). Creation refcount stops uprobe_unregister from freeing the
874 * @uprobe even before the register operation is complete. Creation
875 * refcount is released when the last @uc for the @uprobe
878 * Return errno if it cannot successully install probes
879 * else return 0 (success)
881 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
883 struct uprobe *uprobe;
886 if (!inode || !uc || uc->next)
889 if (offset > i_size_read(inode))
893 mutex_lock(uprobes_hash(inode));
894 uprobe = alloc_uprobe(inode, offset);
896 if (uprobe && !consumer_add(uprobe, uc)) {
897 ret = __uprobe_register(uprobe);
899 uprobe->consumers = NULL;
900 __uprobe_unregister(uprobe);
902 uprobe->flags |= UPROBE_RUN_HANDLER;
906 mutex_unlock(uprobes_hash(inode));
914 * uprobe_unregister - unregister a already registered probe.
915 * @inode: the file in which the probe has to be removed.
916 * @offset: offset from the start of the file.
917 * @uc: identify which probe if multiple probes are colocated.
919 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
921 struct uprobe *uprobe;
926 uprobe = find_uprobe(inode, offset);
930 mutex_lock(uprobes_hash(inode));
932 if (consumer_del(uprobe, uc)) {
933 if (!uprobe->consumers) {
934 __uprobe_unregister(uprobe);
935 uprobe->flags &= ~UPROBE_RUN_HANDLER;
939 mutex_unlock(uprobes_hash(inode));
944 static struct rb_node *
945 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
947 struct rb_node *n = uprobes_tree.rb_node;
950 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
952 if (inode < u->inode) {
954 } else if (inode > u->inode) {
959 else if (min > u->offset)
970 * For a given range in vma, build a list of probes that need to be inserted.
972 static void build_probe_list(struct inode *inode,
973 struct vm_area_struct *vma,
974 unsigned long start, unsigned long end,
975 struct list_head *head)
978 struct rb_node *n, *t;
981 INIT_LIST_HEAD(head);
982 min = vaddr_to_offset(vma, start);
983 max = min + (end - start) - 1;
985 spin_lock(&uprobes_treelock);
986 n = find_node_in_range(inode, min, max);
988 for (t = n; t; t = rb_prev(t)) {
989 u = rb_entry(t, struct uprobe, rb_node);
990 if (u->inode != inode || u->offset < min)
992 list_add(&u->pending_list, head);
995 for (t = n; (t = rb_next(t)); ) {
996 u = rb_entry(t, struct uprobe, rb_node);
997 if (u->inode != inode || u->offset > max)
999 list_add(&u->pending_list, head);
1000 atomic_inc(&u->ref);
1003 spin_unlock(&uprobes_treelock);
1007 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1009 * Currently we ignore all errors and always return 0, the callers
1010 * can't handle the failure anyway.
1012 int uprobe_mmap(struct vm_area_struct *vma)
1014 struct list_head tmp_list;
1015 struct uprobe *uprobe, *u;
1016 struct inode *inode;
1018 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1021 inode = vma->vm_file->f_mapping->host;
1025 mutex_lock(uprobes_mmap_hash(inode));
1026 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1028 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1029 if (!fatal_signal_pending(current)) {
1030 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1031 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1035 mutex_unlock(uprobes_mmap_hash(inode));
1041 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1044 struct inode *inode;
1047 inode = vma->vm_file->f_mapping->host;
1049 min = vaddr_to_offset(vma, start);
1050 max = min + (end - start) - 1;
1052 spin_lock(&uprobes_treelock);
1053 n = find_node_in_range(inode, min, max);
1054 spin_unlock(&uprobes_treelock);
1060 * Called in context of a munmap of a vma.
1062 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1064 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1067 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1070 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1071 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1074 if (vma_has_uprobes(vma, start, end))
1075 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1078 /* Slot allocation for XOL */
1079 static int xol_add_vma(struct xol_area *area)
1081 struct mm_struct *mm;
1084 area->page = alloc_page(GFP_HIGHUSER);
1091 down_write(&mm->mmap_sem);
1092 if (mm->uprobes_state.xol_area)
1097 /* Try to map as high as possible, this is only a hint. */
1098 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1099 if (area->vaddr & ~PAGE_MASK) {
1104 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1105 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1109 smp_wmb(); /* pairs with get_xol_area() */
1110 mm->uprobes_state.xol_area = area;
1114 up_write(&mm->mmap_sem);
1116 __free_page(area->page);
1121 static struct xol_area *get_xol_area(struct mm_struct *mm)
1123 struct xol_area *area;
1125 area = mm->uprobes_state.xol_area;
1126 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1132 * xol_alloc_area - Allocate process's xol_area.
1133 * This area will be used for storing instructions for execution out of
1136 * Returns the allocated area or NULL.
1138 static struct xol_area *xol_alloc_area(void)
1140 struct xol_area *area;
1142 area = kzalloc(sizeof(*area), GFP_KERNEL);
1143 if (unlikely(!area))
1146 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1151 init_waitqueue_head(&area->wq);
1152 if (!xol_add_vma(area))
1156 kfree(area->bitmap);
1159 return get_xol_area(current->mm);
1163 * uprobe_clear_state - Free the area allocated for slots.
1165 void uprobe_clear_state(struct mm_struct *mm)
1167 struct xol_area *area = mm->uprobes_state.xol_area;
1172 put_page(area->page);
1173 kfree(area->bitmap);
1177 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1179 newmm->uprobes_state.xol_area = NULL;
1181 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1182 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1183 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1184 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1189 * - search for a free slot.
1191 static unsigned long xol_take_insn_slot(struct xol_area *area)
1193 unsigned long slot_addr;
1197 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1198 if (slot_nr < UINSNS_PER_PAGE) {
1199 if (!test_and_set_bit(slot_nr, area->bitmap))
1202 slot_nr = UINSNS_PER_PAGE;
1205 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1206 } while (slot_nr >= UINSNS_PER_PAGE);
1208 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1209 atomic_inc(&area->slot_count);
1215 * xol_get_insn_slot - If was not allocated a slot, then
1217 * Returns the allocated slot address or 0.
1219 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1221 struct xol_area *area;
1222 unsigned long offset;
1225 area = get_xol_area(current->mm);
1227 area = xol_alloc_area();
1231 current->utask->xol_vaddr = xol_take_insn_slot(area);
1234 * Initialize the slot if xol_vaddr points to valid
1237 if (unlikely(!current->utask->xol_vaddr))
1240 current->utask->vaddr = slot_addr;
1241 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1242 vaddr = kmap_atomic(area->page);
1243 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1244 kunmap_atomic(vaddr);
1246 return current->utask->xol_vaddr;
1250 * xol_free_insn_slot - If slot was earlier allocated by
1251 * @xol_get_insn_slot(), make the slot available for
1252 * subsequent requests.
1254 static void xol_free_insn_slot(struct task_struct *tsk)
1256 struct xol_area *area;
1257 unsigned long vma_end;
1258 unsigned long slot_addr;
1260 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1263 slot_addr = tsk->utask->xol_vaddr;
1265 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1268 area = tsk->mm->uprobes_state.xol_area;
1269 vma_end = area->vaddr + PAGE_SIZE;
1270 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1271 unsigned long offset;
1274 offset = slot_addr - area->vaddr;
1275 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1276 if (slot_nr >= UINSNS_PER_PAGE)
1279 clear_bit(slot_nr, area->bitmap);
1280 atomic_dec(&area->slot_count);
1281 if (waitqueue_active(&area->wq))
1284 tsk->utask->xol_vaddr = 0;
1289 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1290 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1292 * Return the address of the breakpoint instruction.
1294 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1296 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1300 * Called with no locks held.
1301 * Called in context of a exiting or a exec-ing thread.
1303 void uprobe_free_utask(struct task_struct *t)
1305 struct uprobe_task *utask = t->utask;
1310 if (utask->active_uprobe)
1311 put_uprobe(utask->active_uprobe);
1313 xol_free_insn_slot(t);
1319 * Called in context of a new clone/fork from copy_process.
1321 void uprobe_copy_process(struct task_struct *t)
1327 * Allocate a uprobe_task object for the task.
1328 * Called when the thread hits a breakpoint for the first time.
1331 * - pointer to new uprobe_task on success
1334 static struct uprobe_task *add_utask(void)
1336 struct uprobe_task *utask;
1338 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1339 if (unlikely(!utask))
1342 current->utask = utask;
1346 /* Prepare to single-step probed instruction out of line. */
1348 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1350 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1357 * If we are singlestepping, then ensure this thread is not connected to
1358 * non-fatal signals until completion of singlestep. When xol insn itself
1359 * triggers the signal, restart the original insn even if the task is
1360 * already SIGKILL'ed (since coredump should report the correct ip). This
1361 * is even more important if the task has a handler for SIGSEGV/etc, The
1362 * _same_ instruction should be repeated again after return from the signal
1363 * handler, and SSTEP can never finish in this case.
1365 bool uprobe_deny_signal(void)
1367 struct task_struct *t = current;
1368 struct uprobe_task *utask = t->utask;
1370 if (likely(!utask || !utask->active_uprobe))
1373 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1375 if (signal_pending(t)) {
1376 spin_lock_irq(&t->sighand->siglock);
1377 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1378 spin_unlock_irq(&t->sighand->siglock);
1380 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1381 utask->state = UTASK_SSTEP_TRAPPED;
1382 set_tsk_thread_flag(t, TIF_UPROBE);
1383 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1391 * Avoid singlestepping the original instruction if the original instruction
1392 * is a NOP or can be emulated.
1394 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1396 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1399 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1403 static void mmf_recalc_uprobes(struct mm_struct *mm)
1405 struct vm_area_struct *vma;
1407 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1408 if (!valid_vma(vma, false))
1411 * This is not strictly accurate, we can race with
1412 * uprobe_unregister() and see the already removed
1413 * uprobe if delete_uprobe() was not yet called.
1415 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1419 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1422 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1424 struct mm_struct *mm = current->mm;
1425 struct uprobe *uprobe = NULL;
1426 struct vm_area_struct *vma;
1428 down_read(&mm->mmap_sem);
1429 vma = find_vma(mm, bp_vaddr);
1430 if (vma && vma->vm_start <= bp_vaddr) {
1431 if (valid_vma(vma, false)) {
1432 struct inode *inode = vma->vm_file->f_mapping->host;
1433 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1435 uprobe = find_uprobe(inode, offset);
1439 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1444 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1445 mmf_recalc_uprobes(mm);
1446 up_read(&mm->mmap_sem);
1451 void __weak arch_uprobe_enable_step(struct arch_uprobe *arch)
1453 user_enable_single_step(current);
1456 void __weak arch_uprobe_disable_step(struct arch_uprobe *arch)
1458 user_disable_single_step(current);
1462 * Run handler and ask thread to singlestep.
1463 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1465 static void handle_swbp(struct pt_regs *regs)
1467 struct uprobe_task *utask;
1468 struct uprobe *uprobe;
1469 unsigned long bp_vaddr;
1470 int uninitialized_var(is_swbp);
1472 bp_vaddr = uprobe_get_swbp_addr(regs);
1473 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1477 /* No matching uprobe; signal SIGTRAP. */
1478 send_sig(SIGTRAP, current, 0);
1481 * Either we raced with uprobe_unregister() or we can't
1482 * access this memory. The latter is only possible if
1483 * another thread plays with our ->mm. In both cases
1484 * we can simply restart. If this vma was unmapped we
1485 * can pretend this insn was not executed yet and get
1486 * the (correct) SIGSEGV after restart.
1488 instruction_pointer_set(regs, bp_vaddr);
1493 utask = current->utask;
1495 utask = add_utask();
1496 /* Cannot allocate; re-execute the instruction. */
1500 utask->active_uprobe = uprobe;
1501 handler_chain(uprobe, regs);
1502 if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs))
1505 utask->state = UTASK_SSTEP;
1506 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1507 arch_uprobe_enable_step(&uprobe->arch);
1513 utask->active_uprobe = NULL;
1514 utask->state = UTASK_RUNNING;
1516 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1519 * cannot singlestep; cannot skip instruction;
1520 * re-execute the instruction.
1522 instruction_pointer_set(regs, bp_vaddr);
1528 * Perform required fix-ups and disable singlestep.
1529 * Allow pending signals to take effect.
1531 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1533 struct uprobe *uprobe;
1535 uprobe = utask->active_uprobe;
1536 if (utask->state == UTASK_SSTEP_ACK)
1537 arch_uprobe_post_xol(&uprobe->arch, regs);
1538 else if (utask->state == UTASK_SSTEP_TRAPPED)
1539 arch_uprobe_abort_xol(&uprobe->arch, regs);
1543 arch_uprobe_disable_step(&uprobe->arch);
1545 utask->active_uprobe = NULL;
1546 utask->state = UTASK_RUNNING;
1547 xol_free_insn_slot(current);
1549 spin_lock_irq(¤t->sighand->siglock);
1550 recalc_sigpending(); /* see uprobe_deny_signal() */
1551 spin_unlock_irq(¤t->sighand->siglock);
1555 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on
1556 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1557 * allows the thread to return from interrupt.
1559 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1560 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1563 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1564 * uprobe_notify_resume().
1566 void uprobe_notify_resume(struct pt_regs *regs)
1568 struct uprobe_task *utask;
1570 utask = current->utask;
1571 if (!utask || utask->state == UTASK_BP_HIT)
1574 handle_singlestep(utask, regs);
1578 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1579 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1581 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1583 struct uprobe_task *utask;
1585 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1588 utask = current->utask;
1590 utask->state = UTASK_BP_HIT;
1592 set_thread_flag(TIF_UPROBE);
1598 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1599 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1601 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1603 struct uprobe_task *utask = current->utask;
1605 if (!current->mm || !utask || !utask->active_uprobe)
1606 /* task is currently not uprobed */
1609 utask->state = UTASK_SSTEP_ACK;
1610 set_thread_flag(TIF_UPROBE);
1614 static struct notifier_block uprobe_exception_nb = {
1615 .notifier_call = arch_uprobe_exception_notify,
1616 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1619 static int __init init_uprobes(void)
1623 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1624 mutex_init(&uprobes_mutex[i]);
1625 mutex_init(&uprobes_mmap_mutex[i]);
1628 return register_die_notifier(&uprobe_exception_nb);
1630 module_init(init_uprobes);
1632 static void __exit exit_uprobes(void)
1635 module_exit(exit_uprobes);