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)
103 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
108 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
111 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
113 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
116 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
118 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
122 * __replace_page - replace page in vma by new page.
123 * based on replace_page in mm/ksm.c
125 * @vma: vma that holds the pte pointing to page
126 * @addr: address the old @page is mapped at
127 * @page: the cowed page we are replacing by kpage
128 * @kpage: the modified page we replace page by
130 * Returns 0 on success, -EFAULT on failure.
132 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
133 struct page *page, struct page *kpage)
135 struct mm_struct *mm = vma->vm_mm;
140 /* For try_to_free_swap() and munlock_vma_page() below */
144 ptep = page_check_address(page, mm, addr, &ptl, 0);
149 page_add_new_anon_rmap(kpage, vma, addr);
151 if (!PageAnon(page)) {
152 dec_mm_counter(mm, MM_FILEPAGES);
153 inc_mm_counter(mm, MM_ANONPAGES);
156 flush_cache_page(vma, addr, pte_pfn(*ptep));
157 ptep_clear_flush(vma, addr, ptep);
158 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
160 page_remove_rmap(page);
161 if (!page_mapped(page))
162 try_to_free_swap(page);
163 pte_unmap_unlock(ptep, ptl);
165 if (vma->vm_flags & VM_LOCKED)
166 munlock_vma_page(page);
176 * is_swbp_insn - check if instruction is breakpoint instruction.
177 * @insn: instruction to be checked.
178 * Default implementation of is_swbp_insn
179 * Returns true if @insn is a breakpoint instruction.
181 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
183 return *insn == UPROBE_SWBP_INSN;
186 static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode)
188 void *kaddr = kmap_atomic(page);
189 memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE);
190 kunmap_atomic(kaddr);
193 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
195 uprobe_opcode_t old_opcode;
198 copy_opcode(page, vaddr, &old_opcode);
199 is_swbp = is_swbp_insn(&old_opcode);
201 if (is_swbp_insn(new_opcode)) {
202 if (is_swbp) /* register: already installed? */
205 if (!is_swbp) /* unregister: was it changed by us? */
214 * Expect the breakpoint instruction to be the smallest size instruction for
215 * the architecture. If an arch has variable length instruction and the
216 * breakpoint instruction is not of the smallest length instruction
217 * supported by that architecture then we need to modify is_swbp_at_addr and
218 * write_opcode accordingly. This would never be a problem for archs that
219 * have fixed length instructions.
223 * write_opcode - write the opcode at a given virtual address.
224 * @mm: the probed process address space.
225 * @vaddr: the virtual address to store the opcode.
226 * @opcode: opcode to be written at @vaddr.
228 * Called with mm->mmap_sem held (for read and with a reference to
231 * For mm @mm, write the opcode at @vaddr.
232 * Return 0 (success) or a negative errno.
234 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
235 uprobe_opcode_t opcode)
237 struct page *old_page, *new_page;
238 void *vaddr_old, *vaddr_new;
239 struct vm_area_struct *vma;
243 /* Read the page with vaddr into memory */
244 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
248 ret = verify_opcode(old_page, vaddr, &opcode);
253 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
257 __SetPageUptodate(new_page);
259 /* copy the page now that we've got it stable */
260 vaddr_old = kmap_atomic(old_page);
261 vaddr_new = kmap_atomic(new_page);
263 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
264 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
266 kunmap_atomic(vaddr_new);
267 kunmap_atomic(vaddr_old);
269 ret = anon_vma_prepare(vma);
273 ret = __replace_page(vma, vaddr, old_page, new_page);
276 page_cache_release(new_page);
280 if (unlikely(ret == -EAGAIN))
286 * set_swbp - store breakpoint at a given address.
287 * @auprobe: arch specific probepoint information.
288 * @mm: the probed process address space.
289 * @vaddr: the virtual address to insert the opcode.
291 * For mm @mm, store the breakpoint instruction at @vaddr.
292 * Return 0 (success) or a negative errno.
294 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
296 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
300 * set_orig_insn - Restore the original instruction.
301 * @mm: the probed process address space.
302 * @auprobe: arch specific probepoint information.
303 * @vaddr: the virtual address to insert the opcode.
305 * For mm @mm, restore the original opcode (opcode) at @vaddr.
306 * Return 0 (success) or a negative errno.
309 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
311 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
314 static int match_uprobe(struct uprobe *l, struct uprobe *r)
316 if (l->inode < r->inode)
319 if (l->inode > r->inode)
322 if (l->offset < r->offset)
325 if (l->offset > r->offset)
331 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
333 struct uprobe u = { .inode = inode, .offset = offset };
334 struct rb_node *n = uprobes_tree.rb_node;
335 struct uprobe *uprobe;
339 uprobe = rb_entry(n, struct uprobe, rb_node);
340 match = match_uprobe(&u, uprobe);
342 atomic_inc(&uprobe->ref);
355 * Find a uprobe corresponding to a given inode:offset
356 * Acquires uprobes_treelock
358 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
360 struct uprobe *uprobe;
362 spin_lock(&uprobes_treelock);
363 uprobe = __find_uprobe(inode, offset);
364 spin_unlock(&uprobes_treelock);
369 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
371 struct rb_node **p = &uprobes_tree.rb_node;
372 struct rb_node *parent = NULL;
378 u = rb_entry(parent, struct uprobe, rb_node);
379 match = match_uprobe(uprobe, u);
386 p = &parent->rb_left;
388 p = &parent->rb_right;
393 rb_link_node(&uprobe->rb_node, parent, p);
394 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
395 /* get access + creation ref */
396 atomic_set(&uprobe->ref, 2);
402 * Acquire uprobes_treelock.
403 * Matching uprobe already exists in rbtree;
404 * increment (access refcount) and return the matching uprobe.
406 * No matching uprobe; insert the uprobe in rb_tree;
407 * get a double refcount (access + creation) and return NULL.
409 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
413 spin_lock(&uprobes_treelock);
414 u = __insert_uprobe(uprobe);
415 spin_unlock(&uprobes_treelock);
417 /* For now assume that the instruction need not be single-stepped */
418 uprobe->flags |= UPROBE_SKIP_SSTEP;
423 static void put_uprobe(struct uprobe *uprobe)
425 if (atomic_dec_and_test(&uprobe->ref))
429 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
431 struct uprobe *uprobe, *cur_uprobe;
433 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
437 uprobe->inode = igrab(inode);
438 uprobe->offset = offset;
439 init_rwsem(&uprobe->consumer_rwsem);
441 /* add to uprobes_tree, sorted on inode:offset */
442 cur_uprobe = insert_uprobe(uprobe);
444 /* a uprobe exists for this inode:offset combination */
450 atomic_inc(&uprobe_events);
456 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
458 struct uprobe_consumer *uc;
460 if (!(uprobe->flags & UPROBE_RUN_HANDLER))
463 down_read(&uprobe->consumer_rwsem);
464 for (uc = uprobe->consumers; uc; uc = uc->next) {
465 if (!uc->filter || uc->filter(uc, current))
466 uc->handler(uc, regs);
468 up_read(&uprobe->consumer_rwsem);
471 /* Returns the previous consumer */
472 static struct uprobe_consumer *
473 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
475 down_write(&uprobe->consumer_rwsem);
476 uc->next = uprobe->consumers;
477 uprobe->consumers = uc;
478 up_write(&uprobe->consumer_rwsem);
484 * For uprobe @uprobe, delete the consumer @uc.
485 * Return true if the @uc is deleted successfully
488 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
490 struct uprobe_consumer **con;
493 down_write(&uprobe->consumer_rwsem);
494 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
501 up_write(&uprobe->consumer_rwsem);
507 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
508 unsigned long nbytes, loff_t offset)
518 if (!mapping->a_ops->readpage)
521 idx = offset >> PAGE_CACHE_SHIFT;
522 off = offset & ~PAGE_MASK;
525 * Ensure that the page that has the original instruction is
526 * populated and in page-cache.
528 page = read_mapping_page(mapping, idx, filp);
530 return PTR_ERR(page);
532 vaddr = kmap_atomic(page);
533 memcpy(insn, vaddr + off, nbytes);
534 kunmap_atomic(vaddr);
535 page_cache_release(page);
540 static int copy_insn(struct uprobe *uprobe, struct file *filp)
542 struct address_space *mapping;
543 unsigned long nbytes;
546 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
547 mapping = uprobe->inode->i_mapping;
549 /* Instruction at end of binary; copy only available bytes */
550 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
551 bytes = uprobe->inode->i_size - uprobe->offset;
553 bytes = MAX_UINSN_BYTES;
555 /* Instruction at the page-boundary; copy bytes in second page */
556 if (nbytes < bytes) {
557 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
558 bytes - nbytes, uprobe->offset + nbytes);
563 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
567 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
568 struct vm_area_struct *vma, unsigned long vaddr)
574 * If probe is being deleted, unregister thread could be done with
575 * the vma-rmap-walk through. Adding a probe now can be fatal since
576 * nobody will be able to cleanup. Also we could be from fork or
577 * mremap path, where the probe might have already been inserted.
578 * Hence behave as if probe already existed.
580 if (!uprobe->consumers)
583 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
584 ret = copy_insn(uprobe, vma->vm_file);
588 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
591 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
595 /* write_opcode() assumes we don't cross page boundary */
596 BUG_ON((uprobe->offset & ~PAGE_MASK) +
597 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
599 uprobe->flags |= UPROBE_COPY_INSN;
603 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
604 * the task can hit this breakpoint right after __replace_page().
606 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
608 set_bit(MMF_HAS_UPROBES, &mm->flags);
610 ret = set_swbp(&uprobe->arch, mm, vaddr);
612 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
613 else if (first_uprobe)
614 clear_bit(MMF_HAS_UPROBES, &mm->flags);
620 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
622 /* can happen if uprobe_register() fails */
623 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
626 set_bit(MMF_RECALC_UPROBES, &mm->flags);
627 return set_orig_insn(&uprobe->arch, mm, vaddr);
631 * There could be threads that have already hit the breakpoint. They
632 * will recheck the current insn and restart if find_uprobe() fails.
633 * See find_active_uprobe().
635 static void delete_uprobe(struct uprobe *uprobe)
637 spin_lock(&uprobes_treelock);
638 rb_erase(&uprobe->rb_node, &uprobes_tree);
639 spin_unlock(&uprobes_treelock);
642 atomic_dec(&uprobe_events);
646 struct map_info *next;
647 struct mm_struct *mm;
651 static inline struct map_info *free_map_info(struct map_info *info)
653 struct map_info *next = info->next;
658 static struct map_info *
659 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
661 unsigned long pgoff = offset >> PAGE_SHIFT;
662 struct prio_tree_iter iter;
663 struct vm_area_struct *vma;
664 struct map_info *curr = NULL;
665 struct map_info *prev = NULL;
666 struct map_info *info;
670 mutex_lock(&mapping->i_mmap_mutex);
671 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
672 if (!valid_vma(vma, is_register))
675 if (!prev && !more) {
677 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
678 * reclaim. This is optimistic, no harm done if it fails.
680 prev = kmalloc(sizeof(struct map_info),
681 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
690 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
698 info->mm = vma->vm_mm;
699 info->vaddr = offset_to_vaddr(vma, offset);
701 mutex_unlock(&mapping->i_mmap_mutex);
713 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
715 curr = ERR_PTR(-ENOMEM);
725 prev = free_map_info(prev);
729 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
731 struct map_info *info;
734 info = build_map_info(uprobe->inode->i_mapping,
735 uprobe->offset, is_register);
737 return PTR_ERR(info);
740 struct mm_struct *mm = info->mm;
741 struct vm_area_struct *vma;
743 if (err && is_register)
746 down_write(&mm->mmap_sem);
747 vma = find_vma(mm, info->vaddr);
748 if (!vma || !valid_vma(vma, is_register) ||
749 vma->vm_file->f_mapping->host != uprobe->inode)
752 if (vma->vm_start > info->vaddr ||
753 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
757 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
759 err |= remove_breakpoint(uprobe, mm, info->vaddr);
762 up_write(&mm->mmap_sem);
765 info = free_map_info(info);
771 static int __uprobe_register(struct uprobe *uprobe)
773 return register_for_each_vma(uprobe, true);
776 static void __uprobe_unregister(struct uprobe *uprobe)
778 if (!register_for_each_vma(uprobe, false))
779 delete_uprobe(uprobe);
781 /* TODO : cant unregister? schedule a worker thread */
785 * uprobe_register - register a probe
786 * @inode: the file in which the probe has to be placed.
787 * @offset: offset from the start of the file.
788 * @uc: information on howto handle the probe..
790 * Apart from the access refcount, uprobe_register() takes a creation
791 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
792 * inserted into the rbtree (i.e first consumer for a @inode:@offset
793 * tuple). Creation refcount stops uprobe_unregister from freeing the
794 * @uprobe even before the register operation is complete. Creation
795 * refcount is released when the last @uc for the @uprobe
798 * Return errno if it cannot successully install probes
799 * else return 0 (success)
801 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
803 struct uprobe *uprobe;
806 if (!inode || !uc || uc->next)
809 if (offset > i_size_read(inode))
813 mutex_lock(uprobes_hash(inode));
814 uprobe = alloc_uprobe(inode, offset);
818 } else if (!consumer_add(uprobe, uc)) {
819 ret = __uprobe_register(uprobe);
821 uprobe->consumers = NULL;
822 __uprobe_unregister(uprobe);
824 uprobe->flags |= UPROBE_RUN_HANDLER;
828 mutex_unlock(uprobes_hash(inode));
836 * uprobe_unregister - unregister a already registered probe.
837 * @inode: the file in which the probe has to be removed.
838 * @offset: offset from the start of the file.
839 * @uc: identify which probe if multiple probes are colocated.
841 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
843 struct uprobe *uprobe;
848 uprobe = find_uprobe(inode, offset);
852 mutex_lock(uprobes_hash(inode));
854 if (consumer_del(uprobe, uc)) {
855 if (!uprobe->consumers) {
856 __uprobe_unregister(uprobe);
857 uprobe->flags &= ~UPROBE_RUN_HANDLER;
861 mutex_unlock(uprobes_hash(inode));
866 static struct rb_node *
867 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
869 struct rb_node *n = uprobes_tree.rb_node;
872 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
874 if (inode < u->inode) {
876 } else if (inode > u->inode) {
881 else if (min > u->offset)
892 * For a given range in vma, build a list of probes that need to be inserted.
894 static void build_probe_list(struct inode *inode,
895 struct vm_area_struct *vma,
896 unsigned long start, unsigned long end,
897 struct list_head *head)
900 struct rb_node *n, *t;
903 INIT_LIST_HEAD(head);
904 min = vaddr_to_offset(vma, start);
905 max = min + (end - start) - 1;
907 spin_lock(&uprobes_treelock);
908 n = find_node_in_range(inode, min, max);
910 for (t = n; t; t = rb_prev(t)) {
911 u = rb_entry(t, struct uprobe, rb_node);
912 if (u->inode != inode || u->offset < min)
914 list_add(&u->pending_list, head);
917 for (t = n; (t = rb_next(t)); ) {
918 u = rb_entry(t, struct uprobe, rb_node);
919 if (u->inode != inode || u->offset > max)
921 list_add(&u->pending_list, head);
925 spin_unlock(&uprobes_treelock);
929 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
931 * Currently we ignore all errors and always return 0, the callers
932 * can't handle the failure anyway.
934 int uprobe_mmap(struct vm_area_struct *vma)
936 struct list_head tmp_list;
937 struct uprobe *uprobe, *u;
940 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
943 inode = vma->vm_file->f_mapping->host;
947 mutex_lock(uprobes_mmap_hash(inode));
948 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
950 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
951 if (!fatal_signal_pending(current)) {
952 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
953 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
957 mutex_unlock(uprobes_mmap_hash(inode));
963 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
969 inode = vma->vm_file->f_mapping->host;
971 min = vaddr_to_offset(vma, start);
972 max = min + (end - start) - 1;
974 spin_lock(&uprobes_treelock);
975 n = find_node_in_range(inode, min, max);
976 spin_unlock(&uprobes_treelock);
982 * Called in context of a munmap of a vma.
984 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
986 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
989 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
992 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
993 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
996 if (vma_has_uprobes(vma, start, end))
997 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1000 /* Slot allocation for XOL */
1001 static int xol_add_vma(struct xol_area *area)
1003 struct mm_struct *mm;
1006 area->page = alloc_page(GFP_HIGHUSER);
1013 down_write(&mm->mmap_sem);
1014 if (mm->uprobes_state.xol_area)
1019 /* Try to map as high as possible, this is only a hint. */
1020 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1021 if (area->vaddr & ~PAGE_MASK) {
1026 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1027 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1031 smp_wmb(); /* pairs with get_xol_area() */
1032 mm->uprobes_state.xol_area = area;
1036 up_write(&mm->mmap_sem);
1038 __free_page(area->page);
1043 static struct xol_area *get_xol_area(struct mm_struct *mm)
1045 struct xol_area *area;
1047 area = mm->uprobes_state.xol_area;
1048 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1054 * xol_alloc_area - Allocate process's xol_area.
1055 * This area will be used for storing instructions for execution out of
1058 * Returns the allocated area or NULL.
1060 static struct xol_area *xol_alloc_area(void)
1062 struct xol_area *area;
1064 area = kzalloc(sizeof(*area), GFP_KERNEL);
1065 if (unlikely(!area))
1068 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1073 init_waitqueue_head(&area->wq);
1074 if (!xol_add_vma(area))
1078 kfree(area->bitmap);
1081 return get_xol_area(current->mm);
1085 * uprobe_clear_state - Free the area allocated for slots.
1087 void uprobe_clear_state(struct mm_struct *mm)
1089 struct xol_area *area = mm->uprobes_state.xol_area;
1094 put_page(area->page);
1095 kfree(area->bitmap);
1099 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1101 newmm->uprobes_state.xol_area = NULL;
1103 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1104 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1105 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1106 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1111 * - search for a free slot.
1113 static unsigned long xol_take_insn_slot(struct xol_area *area)
1115 unsigned long slot_addr;
1119 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1120 if (slot_nr < UINSNS_PER_PAGE) {
1121 if (!test_and_set_bit(slot_nr, area->bitmap))
1124 slot_nr = UINSNS_PER_PAGE;
1127 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1128 } while (slot_nr >= UINSNS_PER_PAGE);
1130 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1131 atomic_inc(&area->slot_count);
1137 * xol_get_insn_slot - If was not allocated a slot, then
1139 * Returns the allocated slot address or 0.
1141 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1143 struct xol_area *area;
1144 unsigned long offset;
1147 area = get_xol_area(current->mm);
1149 area = xol_alloc_area();
1153 current->utask->xol_vaddr = xol_take_insn_slot(area);
1156 * Initialize the slot if xol_vaddr points to valid
1159 if (unlikely(!current->utask->xol_vaddr))
1162 current->utask->vaddr = slot_addr;
1163 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1164 vaddr = kmap_atomic(area->page);
1165 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1166 kunmap_atomic(vaddr);
1168 return current->utask->xol_vaddr;
1172 * xol_free_insn_slot - If slot was earlier allocated by
1173 * @xol_get_insn_slot(), make the slot available for
1174 * subsequent requests.
1176 static void xol_free_insn_slot(struct task_struct *tsk)
1178 struct xol_area *area;
1179 unsigned long vma_end;
1180 unsigned long slot_addr;
1182 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1185 slot_addr = tsk->utask->xol_vaddr;
1187 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1190 area = tsk->mm->uprobes_state.xol_area;
1191 vma_end = area->vaddr + PAGE_SIZE;
1192 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1193 unsigned long offset;
1196 offset = slot_addr - area->vaddr;
1197 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1198 if (slot_nr >= UINSNS_PER_PAGE)
1201 clear_bit(slot_nr, area->bitmap);
1202 atomic_dec(&area->slot_count);
1203 if (waitqueue_active(&area->wq))
1206 tsk->utask->xol_vaddr = 0;
1211 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1212 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1214 * Return the address of the breakpoint instruction.
1216 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1218 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1222 * Called with no locks held.
1223 * Called in context of a exiting or a exec-ing thread.
1225 void uprobe_free_utask(struct task_struct *t)
1227 struct uprobe_task *utask = t->utask;
1232 if (utask->active_uprobe)
1233 put_uprobe(utask->active_uprobe);
1235 xol_free_insn_slot(t);
1241 * Called in context of a new clone/fork from copy_process.
1243 void uprobe_copy_process(struct task_struct *t)
1249 * Allocate a uprobe_task object for the task.
1250 * Called when the thread hits a breakpoint for the first time.
1253 * - pointer to new uprobe_task on success
1256 static struct uprobe_task *add_utask(void)
1258 struct uprobe_task *utask;
1260 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1261 if (unlikely(!utask))
1264 current->utask = utask;
1268 /* Prepare to single-step probed instruction out of line. */
1270 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1272 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1279 * If we are singlestepping, then ensure this thread is not connected to
1280 * non-fatal signals until completion of singlestep. When xol insn itself
1281 * triggers the signal, restart the original insn even if the task is
1282 * already SIGKILL'ed (since coredump should report the correct ip). This
1283 * is even more important if the task has a handler for SIGSEGV/etc, The
1284 * _same_ instruction should be repeated again after return from the signal
1285 * handler, and SSTEP can never finish in this case.
1287 bool uprobe_deny_signal(void)
1289 struct task_struct *t = current;
1290 struct uprobe_task *utask = t->utask;
1292 if (likely(!utask || !utask->active_uprobe))
1295 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1297 if (signal_pending(t)) {
1298 spin_lock_irq(&t->sighand->siglock);
1299 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1300 spin_unlock_irq(&t->sighand->siglock);
1302 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1303 utask->state = UTASK_SSTEP_TRAPPED;
1304 set_tsk_thread_flag(t, TIF_UPROBE);
1305 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1313 * Avoid singlestepping the original instruction if the original instruction
1314 * is a NOP or can be emulated.
1316 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1318 if (uprobe->flags & UPROBE_SKIP_SSTEP) {
1319 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1321 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1326 static void mmf_recalc_uprobes(struct mm_struct *mm)
1328 struct vm_area_struct *vma;
1330 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1331 if (!valid_vma(vma, false))
1334 * This is not strictly accurate, we can race with
1335 * uprobe_unregister() and see the already removed
1336 * uprobe if delete_uprobe() was not yet called.
1338 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1342 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1345 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1348 uprobe_opcode_t opcode;
1351 pagefault_disable();
1352 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1356 if (likely(result == 0))
1359 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1363 copy_opcode(page, vaddr, &opcode);
1366 return is_swbp_insn(&opcode);
1369 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1371 struct mm_struct *mm = current->mm;
1372 struct uprobe *uprobe = NULL;
1373 struct vm_area_struct *vma;
1375 down_read(&mm->mmap_sem);
1376 vma = find_vma(mm, bp_vaddr);
1377 if (vma && vma->vm_start <= bp_vaddr) {
1378 if (valid_vma(vma, false)) {
1379 struct inode *inode = vma->vm_file->f_mapping->host;
1380 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1382 uprobe = find_uprobe(inode, offset);
1386 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1391 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1392 mmf_recalc_uprobes(mm);
1393 up_read(&mm->mmap_sem);
1398 void __weak arch_uprobe_enable_step(struct arch_uprobe *arch)
1400 user_enable_single_step(current);
1403 void __weak arch_uprobe_disable_step(struct arch_uprobe *arch)
1405 user_disable_single_step(current);
1409 * Run handler and ask thread to singlestep.
1410 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1412 static void handle_swbp(struct pt_regs *regs)
1414 struct uprobe_task *utask;
1415 struct uprobe *uprobe;
1416 unsigned long bp_vaddr;
1417 int uninitialized_var(is_swbp);
1419 bp_vaddr = uprobe_get_swbp_addr(regs);
1420 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1424 /* No matching uprobe; signal SIGTRAP. */
1425 send_sig(SIGTRAP, current, 0);
1428 * Either we raced with uprobe_unregister() or we can't
1429 * access this memory. The latter is only possible if
1430 * another thread plays with our ->mm. In both cases
1431 * we can simply restart. If this vma was unmapped we
1432 * can pretend this insn was not executed yet and get
1433 * the (correct) SIGSEGV after restart.
1435 instruction_pointer_set(regs, bp_vaddr);
1440 utask = current->utask;
1442 utask = add_utask();
1443 /* Cannot allocate; re-execute the instruction. */
1448 handler_chain(uprobe, regs);
1449 if (can_skip_sstep(uprobe, regs))
1452 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1453 arch_uprobe_enable_step(&uprobe->arch);
1454 utask->active_uprobe = uprobe;
1455 utask->state = UTASK_SSTEP;
1461 * cannot singlestep; cannot skip instruction;
1462 * re-execute the instruction.
1464 instruction_pointer_set(regs, bp_vaddr);
1470 * Perform required fix-ups and disable singlestep.
1471 * Allow pending signals to take effect.
1473 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1475 struct uprobe *uprobe;
1477 uprobe = utask->active_uprobe;
1478 if (utask->state == UTASK_SSTEP_ACK)
1479 arch_uprobe_post_xol(&uprobe->arch, regs);
1480 else if (utask->state == UTASK_SSTEP_TRAPPED)
1481 arch_uprobe_abort_xol(&uprobe->arch, regs);
1485 arch_uprobe_disable_step(&uprobe->arch);
1487 utask->active_uprobe = NULL;
1488 utask->state = UTASK_RUNNING;
1489 xol_free_insn_slot(current);
1491 spin_lock_irq(¤t->sighand->siglock);
1492 recalc_sigpending(); /* see uprobe_deny_signal() */
1493 spin_unlock_irq(¤t->sighand->siglock);
1497 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1498 * allows the thread to return from interrupt. After that handle_swbp()
1499 * sets utask->active_uprobe.
1501 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1502 * and allows the thread to return from interrupt.
1504 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1505 * uprobe_notify_resume().
1507 void uprobe_notify_resume(struct pt_regs *regs)
1509 struct uprobe_task *utask;
1511 clear_thread_flag(TIF_UPROBE);
1513 utask = current->utask;
1514 if (utask && utask->active_uprobe)
1515 handle_singlestep(utask, regs);
1521 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1522 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1524 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1526 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1529 set_thread_flag(TIF_UPROBE);
1534 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1535 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1537 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1539 struct uprobe_task *utask = current->utask;
1541 if (!current->mm || !utask || !utask->active_uprobe)
1542 /* task is currently not uprobed */
1545 utask->state = UTASK_SSTEP_ACK;
1546 set_thread_flag(TIF_UPROBE);
1550 static struct notifier_block uprobe_exception_nb = {
1551 .notifier_call = arch_uprobe_exception_notify,
1552 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1555 static int __init init_uprobes(void)
1559 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1560 mutex_init(&uprobes_mutex[i]);
1561 mutex_init(&uprobes_mmap_mutex[i]);
1564 return register_die_notifier(&uprobe_exception_nb);
1566 module_init(init_uprobes);
1568 static void __exit exit_uprobes(void)
1571 module_exit(exit_uprobes);