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;
145 /* For try_to_free_swap() and munlock_vma_page() below */
149 ptep = page_check_address(page, mm, addr, &ptl, 0);
154 page_add_new_anon_rmap(kpage, vma, addr);
156 if (!PageAnon(page)) {
157 dec_mm_counter(mm, MM_FILEPAGES);
158 inc_mm_counter(mm, MM_ANONPAGES);
161 flush_cache_page(vma, addr, pte_pfn(*ptep));
162 ptep_clear_flush(vma, addr, ptep);
163 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
165 page_remove_rmap(page);
166 if (!page_mapped(page))
167 try_to_free_swap(page);
168 pte_unmap_unlock(ptep, ptl);
170 if (vma->vm_flags & VM_LOCKED)
171 munlock_vma_page(page);
181 * is_swbp_insn - check if instruction is breakpoint instruction.
182 * @insn: instruction to be checked.
183 * Default implementation of is_swbp_insn
184 * Returns true if @insn is a breakpoint instruction.
186 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
188 return *insn == UPROBE_SWBP_INSN;
193 * Expect the breakpoint instruction to be the smallest size instruction for
194 * the architecture. If an arch has variable length instruction and the
195 * breakpoint instruction is not of the smallest length instruction
196 * supported by that architecture then we need to modify read_opcode /
197 * write_opcode accordingly. This would never be a problem for archs that
198 * have fixed length instructions.
202 * write_opcode - write the opcode at a given virtual address.
203 * @auprobe: arch breakpointing information.
204 * @mm: the probed process address space.
205 * @vaddr: the virtual address to store the opcode.
206 * @opcode: opcode to be written at @vaddr.
208 * Called with mm->mmap_sem held (for read and with a reference to
211 * For mm @mm, write the opcode at @vaddr.
212 * Return 0 (success) or a negative errno.
214 static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
215 unsigned long vaddr, uprobe_opcode_t opcode)
217 struct page *old_page, *new_page;
218 void *vaddr_old, *vaddr_new;
219 struct vm_area_struct *vma;
223 /* Read the page with vaddr into memory */
224 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
229 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
233 __SetPageUptodate(new_page);
235 /* copy the page now that we've got it stable */
236 vaddr_old = kmap_atomic(old_page);
237 vaddr_new = kmap_atomic(new_page);
239 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
240 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
242 kunmap_atomic(vaddr_new);
243 kunmap_atomic(vaddr_old);
245 ret = anon_vma_prepare(vma);
249 ret = __replace_page(vma, vaddr, old_page, new_page);
252 page_cache_release(new_page);
256 if (unlikely(ret == -EAGAIN))
262 * read_opcode - read the opcode at a given virtual address.
263 * @mm: the probed process address space.
264 * @vaddr: the virtual address to read the opcode.
265 * @opcode: location to store the read opcode.
267 * Called with mm->mmap_sem held (for read and with a reference to
270 * For mm @mm, read the opcode at @vaddr and store it in @opcode.
271 * Return 0 (success) or a negative errno.
273 static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode)
279 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
283 vaddr_new = kmap_atomic(page);
285 memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE);
286 kunmap_atomic(vaddr_new);
293 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
295 uprobe_opcode_t opcode;
298 if (current->mm == mm) {
300 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
304 if (likely(result == 0))
308 result = read_opcode(mm, vaddr, &opcode);
312 if (is_swbp_insn(&opcode))
319 * set_swbp - store breakpoint at a given address.
320 * @auprobe: arch specific probepoint information.
321 * @mm: the probed process address space.
322 * @vaddr: the virtual address to insert the opcode.
324 * For mm @mm, store the breakpoint instruction at @vaddr.
325 * Return 0 (success) or a negative errno.
327 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
331 * See the comment near uprobes_hash().
333 result = is_swbp_at_addr(mm, vaddr);
340 return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
344 * set_orig_insn - Restore the original instruction.
345 * @mm: the probed process address space.
346 * @auprobe: arch specific probepoint information.
347 * @vaddr: the virtual address to insert the opcode.
349 * For mm @mm, restore the original opcode (opcode) at @vaddr.
350 * Return 0 (success) or a negative errno.
353 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
357 result = is_swbp_at_addr(mm, vaddr);
364 return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
367 static int match_uprobe(struct uprobe *l, struct uprobe *r)
369 if (l->inode < r->inode)
372 if (l->inode > r->inode)
375 if (l->offset < r->offset)
378 if (l->offset > r->offset)
384 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
386 struct uprobe u = { .inode = inode, .offset = offset };
387 struct rb_node *n = uprobes_tree.rb_node;
388 struct uprobe *uprobe;
392 uprobe = rb_entry(n, struct uprobe, rb_node);
393 match = match_uprobe(&u, uprobe);
395 atomic_inc(&uprobe->ref);
408 * Find a uprobe corresponding to a given inode:offset
409 * Acquires uprobes_treelock
411 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
413 struct uprobe *uprobe;
415 spin_lock(&uprobes_treelock);
416 uprobe = __find_uprobe(inode, offset);
417 spin_unlock(&uprobes_treelock);
422 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
424 struct rb_node **p = &uprobes_tree.rb_node;
425 struct rb_node *parent = NULL;
431 u = rb_entry(parent, struct uprobe, rb_node);
432 match = match_uprobe(uprobe, u);
439 p = &parent->rb_left;
441 p = &parent->rb_right;
446 rb_link_node(&uprobe->rb_node, parent, p);
447 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
448 /* get access + creation ref */
449 atomic_set(&uprobe->ref, 2);
455 * Acquire uprobes_treelock.
456 * Matching uprobe already exists in rbtree;
457 * increment (access refcount) and return the matching uprobe.
459 * No matching uprobe; insert the uprobe in rb_tree;
460 * get a double refcount (access + creation) and return NULL.
462 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
466 spin_lock(&uprobes_treelock);
467 u = __insert_uprobe(uprobe);
468 spin_unlock(&uprobes_treelock);
470 /* For now assume that the instruction need not be single-stepped */
471 uprobe->flags |= UPROBE_SKIP_SSTEP;
476 static void put_uprobe(struct uprobe *uprobe)
478 if (atomic_dec_and_test(&uprobe->ref))
482 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
484 struct uprobe *uprobe, *cur_uprobe;
486 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
490 uprobe->inode = igrab(inode);
491 uprobe->offset = offset;
492 init_rwsem(&uprobe->consumer_rwsem);
494 /* add to uprobes_tree, sorted on inode:offset */
495 cur_uprobe = insert_uprobe(uprobe);
497 /* a uprobe exists for this inode:offset combination */
503 atomic_inc(&uprobe_events);
509 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
511 struct uprobe_consumer *uc;
513 if (!(uprobe->flags & UPROBE_RUN_HANDLER))
516 down_read(&uprobe->consumer_rwsem);
517 for (uc = uprobe->consumers; uc; uc = uc->next) {
518 if (!uc->filter || uc->filter(uc, current))
519 uc->handler(uc, regs);
521 up_read(&uprobe->consumer_rwsem);
524 /* Returns the previous consumer */
525 static struct uprobe_consumer *
526 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
528 down_write(&uprobe->consumer_rwsem);
529 uc->next = uprobe->consumers;
530 uprobe->consumers = uc;
531 up_write(&uprobe->consumer_rwsem);
537 * For uprobe @uprobe, delete the consumer @uc.
538 * Return true if the @uc is deleted successfully
541 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
543 struct uprobe_consumer **con;
546 down_write(&uprobe->consumer_rwsem);
547 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
554 up_write(&uprobe->consumer_rwsem);
560 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
561 unsigned long nbytes, loff_t offset)
571 if (!mapping->a_ops->readpage)
574 idx = offset >> PAGE_CACHE_SHIFT;
575 off = offset & ~PAGE_MASK;
578 * Ensure that the page that has the original instruction is
579 * populated and in page-cache.
581 page = read_mapping_page(mapping, idx, filp);
583 return PTR_ERR(page);
585 vaddr = kmap_atomic(page);
586 memcpy(insn, vaddr + off, nbytes);
587 kunmap_atomic(vaddr);
588 page_cache_release(page);
593 static int copy_insn(struct uprobe *uprobe, struct file *filp)
595 struct address_space *mapping;
596 unsigned long nbytes;
599 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
600 mapping = uprobe->inode->i_mapping;
602 /* Instruction at end of binary; copy only available bytes */
603 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
604 bytes = uprobe->inode->i_size - uprobe->offset;
606 bytes = MAX_UINSN_BYTES;
608 /* Instruction at the page-boundary; copy bytes in second page */
609 if (nbytes < bytes) {
610 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
611 bytes - nbytes, uprobe->offset + nbytes);
616 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
620 * How mm->uprobes_state.count gets updated
621 * uprobe_mmap() increments the count if
622 * - it successfully adds a breakpoint.
623 * - it cannot add a breakpoint, but sees that there is a underlying
624 * breakpoint (via a is_swbp_at_addr()).
626 * uprobe_munmap() decrements the count if
627 * - it sees a underlying breakpoint, (via is_swbp_at_addr)
628 * (Subsequent uprobe_unregister wouldnt find the breakpoint
629 * unless a uprobe_mmap kicks in, since the old vma would be
630 * dropped just after uprobe_munmap.)
632 * uprobe_register increments the count if:
633 * - it successfully adds a breakpoint.
635 * uprobe_unregister decrements the count if:
636 * - it sees a underlying breakpoint and removes successfully.
637 * (via is_swbp_at_addr)
638 * (Subsequent uprobe_munmap wouldnt find the breakpoint
639 * since there is no underlying breakpoint after the
640 * breakpoint removal.)
643 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
644 struct vm_area_struct *vma, unsigned long vaddr)
650 * If probe is being deleted, unregister thread could be done with
651 * the vma-rmap-walk through. Adding a probe now can be fatal since
652 * nobody will be able to cleanup. Also we could be from fork or
653 * mremap path, where the probe might have already been inserted.
654 * Hence behave as if probe already existed.
656 if (!uprobe->consumers)
659 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
660 ret = copy_insn(uprobe, vma->vm_file);
664 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
667 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
671 /* write_opcode() assumes we don't cross page boundary */
672 BUG_ON((uprobe->offset & ~PAGE_MASK) +
673 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
675 uprobe->flags |= UPROBE_COPY_INSN;
679 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
680 * the task can hit this breakpoint right after __replace_page().
682 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
684 set_bit(MMF_HAS_UPROBES, &mm->flags);
686 ret = set_swbp(&uprobe->arch, mm, vaddr);
688 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
689 else if (first_uprobe)
690 clear_bit(MMF_HAS_UPROBES, &mm->flags);
696 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
698 /* can happen if uprobe_register() fails */
699 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
702 set_bit(MMF_RECALC_UPROBES, &mm->flags);
703 set_orig_insn(&uprobe->arch, mm, vaddr);
707 * There could be threads that have already hit the breakpoint. They
708 * will recheck the current insn and restart if find_uprobe() fails.
709 * See find_active_uprobe().
711 static void delete_uprobe(struct uprobe *uprobe)
713 spin_lock(&uprobes_treelock);
714 rb_erase(&uprobe->rb_node, &uprobes_tree);
715 spin_unlock(&uprobes_treelock);
718 atomic_dec(&uprobe_events);
722 struct map_info *next;
723 struct mm_struct *mm;
727 static inline struct map_info *free_map_info(struct map_info *info)
729 struct map_info *next = info->next;
734 static struct map_info *
735 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
737 unsigned long pgoff = offset >> PAGE_SHIFT;
738 struct prio_tree_iter iter;
739 struct vm_area_struct *vma;
740 struct map_info *curr = NULL;
741 struct map_info *prev = NULL;
742 struct map_info *info;
746 mutex_lock(&mapping->i_mmap_mutex);
747 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
748 if (!valid_vma(vma, is_register))
751 if (!prev && !more) {
753 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
754 * reclaim. This is optimistic, no harm done if it fails.
756 prev = kmalloc(sizeof(struct map_info),
757 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
766 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
774 info->mm = vma->vm_mm;
775 info->vaddr = offset_to_vaddr(vma, offset);
777 mutex_unlock(&mapping->i_mmap_mutex);
789 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
791 curr = ERR_PTR(-ENOMEM);
801 prev = free_map_info(prev);
805 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
807 struct map_info *info;
810 info = build_map_info(uprobe->inode->i_mapping,
811 uprobe->offset, is_register);
813 return PTR_ERR(info);
816 struct mm_struct *mm = info->mm;
817 struct vm_area_struct *vma;
822 down_write(&mm->mmap_sem);
823 vma = find_vma(mm, info->vaddr);
824 if (!vma || !valid_vma(vma, is_register) ||
825 vma->vm_file->f_mapping->host != uprobe->inode)
828 if (vma->vm_start > info->vaddr ||
829 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
833 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
835 remove_breakpoint(uprobe, mm, info->vaddr);
838 up_write(&mm->mmap_sem);
841 info = free_map_info(info);
847 static int __uprobe_register(struct uprobe *uprobe)
849 return register_for_each_vma(uprobe, true);
852 static void __uprobe_unregister(struct uprobe *uprobe)
854 if (!register_for_each_vma(uprobe, false))
855 delete_uprobe(uprobe);
857 /* TODO : cant unregister? schedule a worker thread */
861 * uprobe_register - register a probe
862 * @inode: the file in which the probe has to be placed.
863 * @offset: offset from the start of the file.
864 * @uc: information on howto handle the probe..
866 * Apart from the access refcount, uprobe_register() takes a creation
867 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
868 * inserted into the rbtree (i.e first consumer for a @inode:@offset
869 * tuple). Creation refcount stops uprobe_unregister from freeing the
870 * @uprobe even before the register operation is complete. Creation
871 * refcount is released when the last @uc for the @uprobe
874 * Return errno if it cannot successully install probes
875 * else return 0 (success)
877 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
879 struct uprobe *uprobe;
882 if (!inode || !uc || uc->next)
885 if (offset > i_size_read(inode))
889 mutex_lock(uprobes_hash(inode));
890 uprobe = alloc_uprobe(inode, offset);
892 if (uprobe && !consumer_add(uprobe, uc)) {
893 ret = __uprobe_register(uprobe);
895 uprobe->consumers = NULL;
896 __uprobe_unregister(uprobe);
898 uprobe->flags |= UPROBE_RUN_HANDLER;
902 mutex_unlock(uprobes_hash(inode));
910 * uprobe_unregister - unregister a already registered probe.
911 * @inode: the file in which the probe has to be removed.
912 * @offset: offset from the start of the file.
913 * @uc: identify which probe if multiple probes are colocated.
915 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
917 struct uprobe *uprobe;
922 uprobe = find_uprobe(inode, offset);
926 mutex_lock(uprobes_hash(inode));
928 if (consumer_del(uprobe, uc)) {
929 if (!uprobe->consumers) {
930 __uprobe_unregister(uprobe);
931 uprobe->flags &= ~UPROBE_RUN_HANDLER;
935 mutex_unlock(uprobes_hash(inode));
940 static struct rb_node *
941 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
943 struct rb_node *n = uprobes_tree.rb_node;
946 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
948 if (inode < u->inode) {
950 } else if (inode > u->inode) {
955 else if (min > u->offset)
966 * For a given range in vma, build a list of probes that need to be inserted.
968 static void build_probe_list(struct inode *inode,
969 struct vm_area_struct *vma,
970 unsigned long start, unsigned long end,
971 struct list_head *head)
974 struct rb_node *n, *t;
977 INIT_LIST_HEAD(head);
978 min = vaddr_to_offset(vma, start);
979 max = min + (end - start) - 1;
981 spin_lock(&uprobes_treelock);
982 n = find_node_in_range(inode, min, max);
984 for (t = n; t; t = rb_prev(t)) {
985 u = rb_entry(t, struct uprobe, rb_node);
986 if (u->inode != inode || u->offset < min)
988 list_add(&u->pending_list, head);
991 for (t = n; (t = rb_next(t)); ) {
992 u = rb_entry(t, struct uprobe, rb_node);
993 if (u->inode != inode || u->offset > max)
995 list_add(&u->pending_list, head);
999 spin_unlock(&uprobes_treelock);
1003 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1005 * Currently we ignore all errors and always return 0, the callers
1006 * can't handle the failure anyway.
1008 int uprobe_mmap(struct vm_area_struct *vma)
1010 struct list_head tmp_list;
1011 struct uprobe *uprobe, *u;
1012 struct inode *inode;
1014 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1017 inode = vma->vm_file->f_mapping->host;
1021 mutex_lock(uprobes_mmap_hash(inode));
1022 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1024 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1025 if (!fatal_signal_pending(current)) {
1026 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1027 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1031 mutex_unlock(uprobes_mmap_hash(inode));
1037 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1040 struct inode *inode;
1043 inode = vma->vm_file->f_mapping->host;
1045 min = vaddr_to_offset(vma, start);
1046 max = min + (end - start) - 1;
1048 spin_lock(&uprobes_treelock);
1049 n = find_node_in_range(inode, min, max);
1050 spin_unlock(&uprobes_treelock);
1056 * Called in context of a munmap of a vma.
1058 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1060 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1063 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1066 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1067 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1070 if (vma_has_uprobes(vma, start, end))
1071 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1074 /* Slot allocation for XOL */
1075 static int xol_add_vma(struct xol_area *area)
1077 struct mm_struct *mm;
1080 area->page = alloc_page(GFP_HIGHUSER);
1087 down_write(&mm->mmap_sem);
1088 if (mm->uprobes_state.xol_area)
1093 /* Try to map as high as possible, this is only a hint. */
1094 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1095 if (area->vaddr & ~PAGE_MASK) {
1100 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1101 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1105 smp_wmb(); /* pairs with get_xol_area() */
1106 mm->uprobes_state.xol_area = area;
1110 up_write(&mm->mmap_sem);
1112 __free_page(area->page);
1117 static struct xol_area *get_xol_area(struct mm_struct *mm)
1119 struct xol_area *area;
1121 area = mm->uprobes_state.xol_area;
1122 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1128 * xol_alloc_area - Allocate process's xol_area.
1129 * This area will be used for storing instructions for execution out of
1132 * Returns the allocated area or NULL.
1134 static struct xol_area *xol_alloc_area(void)
1136 struct xol_area *area;
1138 area = kzalloc(sizeof(*area), GFP_KERNEL);
1139 if (unlikely(!area))
1142 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1147 init_waitqueue_head(&area->wq);
1148 if (!xol_add_vma(area))
1152 kfree(area->bitmap);
1155 return get_xol_area(current->mm);
1159 * uprobe_clear_state - Free the area allocated for slots.
1161 void uprobe_clear_state(struct mm_struct *mm)
1163 struct xol_area *area = mm->uprobes_state.xol_area;
1168 put_page(area->page);
1169 kfree(area->bitmap);
1173 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1175 newmm->uprobes_state.xol_area = NULL;
1177 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1178 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1179 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1180 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1185 * - search for a free slot.
1187 static unsigned long xol_take_insn_slot(struct xol_area *area)
1189 unsigned long slot_addr;
1193 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1194 if (slot_nr < UINSNS_PER_PAGE) {
1195 if (!test_and_set_bit(slot_nr, area->bitmap))
1198 slot_nr = UINSNS_PER_PAGE;
1201 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1202 } while (slot_nr >= UINSNS_PER_PAGE);
1204 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1205 atomic_inc(&area->slot_count);
1211 * xol_get_insn_slot - If was not allocated a slot, then
1213 * Returns the allocated slot address or 0.
1215 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1217 struct xol_area *area;
1218 unsigned long offset;
1221 area = get_xol_area(current->mm);
1223 area = xol_alloc_area();
1227 current->utask->xol_vaddr = xol_take_insn_slot(area);
1230 * Initialize the slot if xol_vaddr points to valid
1233 if (unlikely(!current->utask->xol_vaddr))
1236 current->utask->vaddr = slot_addr;
1237 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1238 vaddr = kmap_atomic(area->page);
1239 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1240 kunmap_atomic(vaddr);
1242 return current->utask->xol_vaddr;
1246 * xol_free_insn_slot - If slot was earlier allocated by
1247 * @xol_get_insn_slot(), make the slot available for
1248 * subsequent requests.
1250 static void xol_free_insn_slot(struct task_struct *tsk)
1252 struct xol_area *area;
1253 unsigned long vma_end;
1254 unsigned long slot_addr;
1256 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1259 slot_addr = tsk->utask->xol_vaddr;
1261 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1264 area = tsk->mm->uprobes_state.xol_area;
1265 vma_end = area->vaddr + PAGE_SIZE;
1266 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1267 unsigned long offset;
1270 offset = slot_addr - area->vaddr;
1271 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1272 if (slot_nr >= UINSNS_PER_PAGE)
1275 clear_bit(slot_nr, area->bitmap);
1276 atomic_dec(&area->slot_count);
1277 if (waitqueue_active(&area->wq))
1280 tsk->utask->xol_vaddr = 0;
1285 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1286 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1288 * Return the address of the breakpoint instruction.
1290 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1292 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1296 * Called with no locks held.
1297 * Called in context of a exiting or a exec-ing thread.
1299 void uprobe_free_utask(struct task_struct *t)
1301 struct uprobe_task *utask = t->utask;
1306 if (utask->active_uprobe)
1307 put_uprobe(utask->active_uprobe);
1309 xol_free_insn_slot(t);
1315 * Called in context of a new clone/fork from copy_process.
1317 void uprobe_copy_process(struct task_struct *t)
1323 * Allocate a uprobe_task object for the task.
1324 * Called when the thread hits a breakpoint for the first time.
1327 * - pointer to new uprobe_task on success
1330 static struct uprobe_task *add_utask(void)
1332 struct uprobe_task *utask;
1334 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1335 if (unlikely(!utask))
1338 current->utask = utask;
1342 /* Prepare to single-step probed instruction out of line. */
1344 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1346 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1353 * If we are singlestepping, then ensure this thread is not connected to
1354 * non-fatal signals until completion of singlestep. When xol insn itself
1355 * triggers the signal, restart the original insn even if the task is
1356 * already SIGKILL'ed (since coredump should report the correct ip). This
1357 * is even more important if the task has a handler for SIGSEGV/etc, The
1358 * _same_ instruction should be repeated again after return from the signal
1359 * handler, and SSTEP can never finish in this case.
1361 bool uprobe_deny_signal(void)
1363 struct task_struct *t = current;
1364 struct uprobe_task *utask = t->utask;
1366 if (likely(!utask || !utask->active_uprobe))
1369 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1371 if (signal_pending(t)) {
1372 spin_lock_irq(&t->sighand->siglock);
1373 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1374 spin_unlock_irq(&t->sighand->siglock);
1376 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1377 utask->state = UTASK_SSTEP_TRAPPED;
1378 set_tsk_thread_flag(t, TIF_UPROBE);
1379 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1387 * Avoid singlestepping the original instruction if the original instruction
1388 * is a NOP or can be emulated.
1390 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1392 if (uprobe->flags & UPROBE_SKIP_SSTEP) {
1393 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1395 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1400 static void mmf_recalc_uprobes(struct mm_struct *mm)
1402 struct vm_area_struct *vma;
1404 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1405 if (!valid_vma(vma, false))
1408 * This is not strictly accurate, we can race with
1409 * uprobe_unregister() and see the already removed
1410 * uprobe if delete_uprobe() was not yet called.
1412 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1416 clear_bit(MMF_HAS_UPROBES, &mm->flags);
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);
1448 void __weak arch_uprobe_enable_step(struct arch_uprobe *arch)
1450 user_enable_single_step(current);
1453 void __weak arch_uprobe_disable_step(struct arch_uprobe *arch)
1455 user_disable_single_step(current);
1459 * Run handler and ask thread to singlestep.
1460 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1462 static void handle_swbp(struct pt_regs *regs)
1464 struct uprobe_task *utask;
1465 struct uprobe *uprobe;
1466 unsigned long bp_vaddr;
1467 int uninitialized_var(is_swbp);
1469 bp_vaddr = uprobe_get_swbp_addr(regs);
1470 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1474 /* No matching uprobe; signal SIGTRAP. */
1475 send_sig(SIGTRAP, current, 0);
1478 * Either we raced with uprobe_unregister() or we can't
1479 * access this memory. The latter is only possible if
1480 * another thread plays with our ->mm. In both cases
1481 * we can simply restart. If this vma was unmapped we
1482 * can pretend this insn was not executed yet and get
1483 * the (correct) SIGSEGV after restart.
1485 instruction_pointer_set(regs, bp_vaddr);
1490 utask = current->utask;
1492 utask = add_utask();
1493 /* Cannot allocate; re-execute the instruction. */
1498 handler_chain(uprobe, regs);
1499 if (can_skip_sstep(uprobe, regs))
1502 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1503 arch_uprobe_enable_step(&uprobe->arch);
1504 utask->active_uprobe = uprobe;
1505 utask->state = UTASK_SSTEP;
1511 * cannot singlestep; cannot skip instruction;
1512 * re-execute the instruction.
1514 instruction_pointer_set(regs, bp_vaddr);
1520 * Perform required fix-ups and disable singlestep.
1521 * Allow pending signals to take effect.
1523 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1525 struct uprobe *uprobe;
1527 uprobe = utask->active_uprobe;
1528 if (utask->state == UTASK_SSTEP_ACK)
1529 arch_uprobe_post_xol(&uprobe->arch, regs);
1530 else if (utask->state == UTASK_SSTEP_TRAPPED)
1531 arch_uprobe_abort_xol(&uprobe->arch, regs);
1535 arch_uprobe_disable_step(&uprobe->arch);
1537 utask->active_uprobe = NULL;
1538 utask->state = UTASK_RUNNING;
1539 xol_free_insn_slot(current);
1541 spin_lock_irq(¤t->sighand->siglock);
1542 recalc_sigpending(); /* see uprobe_deny_signal() */
1543 spin_unlock_irq(¤t->sighand->siglock);
1547 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1548 * allows the thread to return from interrupt. After that handle_swbp()
1549 * sets utask->active_uprobe.
1551 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1552 * and allows the thread to return from interrupt.
1554 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1555 * uprobe_notify_resume().
1557 void uprobe_notify_resume(struct pt_regs *regs)
1559 struct uprobe_task *utask;
1561 clear_thread_flag(TIF_UPROBE);
1563 utask = current->utask;
1564 if (utask && utask->active_uprobe)
1565 handle_singlestep(utask, regs);
1571 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1572 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1574 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1576 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1579 set_thread_flag(TIF_UPROBE);
1584 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1585 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1587 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1589 struct uprobe_task *utask = current->utask;
1591 if (!current->mm || !utask || !utask->active_uprobe)
1592 /* task is currently not uprobed */
1595 utask->state = UTASK_SSTEP_ACK;
1596 set_thread_flag(TIF_UPROBE);
1600 static struct notifier_block uprobe_exception_nb = {
1601 .notifier_call = arch_uprobe_exception_notify,
1602 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1605 static int __init init_uprobes(void)
1609 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1610 mutex_init(&uprobes_mutex[i]);
1611 mutex_init(&uprobes_mmap_mutex[i]);
1614 return register_die_notifier(&uprobe_exception_nb);
1616 module_init(init_uprobes);
1618 static void __exit exit_uprobes(void)
1621 module_exit(exit_uprobes);