2 * User-space Probes (UProbes)
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright (C) IBM Corporation, 2008-2012
22 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
25 #include <linux/kernel.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h> /* read_mapping_page */
28 #include <linux/slab.h>
29 #include <linux/sched.h>
30 #include <linux/rmap.h> /* anon_vma_prepare */
31 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
32 #include <linux/swap.h> /* try_to_free_swap */
33 #include <linux/ptrace.h> /* user_enable_single_step */
34 #include <linux/kdebug.h> /* notifier mechanism */
35 #include "../../mm/internal.h" /* munlock_vma_page */
36 #include <linux/percpu-rwsem.h>
38 #include <linux/uprobes.h>
40 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
41 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
43 static struct rb_root uprobes_tree = RB_ROOT;
45 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
47 #define UPROBES_HASH_SZ 13
50 * We need separate register/unregister and mmap/munmap lock hashes because
51 * of mmap_sem nesting.
53 * uprobe_register() needs to install probes on (potentially) all processes
54 * and thus needs to acquire multiple mmap_sems (consequtively, not
55 * concurrently), whereas uprobe_mmap() is called while holding mmap_sem
56 * for the particular process doing the mmap.
58 * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem
59 * because of lock order against i_mmap_mutex. This means there's a hole in
60 * the register vma iteration where a mmap() can happen.
62 * Thus uprobe_register() can race with uprobe_mmap() and we can try and
63 * install a probe where one is already installed.
66 /* serialize (un)register */
67 static struct mutex uprobes_mutex[UPROBES_HASH_SZ];
69 #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
71 /* serialize uprobe->pending_list */
72 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
73 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
75 static struct percpu_rw_semaphore dup_mmap_sem;
78 * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe
79 * events active at this time. Probably a fine grained per inode count is
82 static atomic_t uprobe_events = ATOMIC_INIT(0);
84 /* Have a copy of original instruction */
85 #define UPROBE_COPY_INSN 0
86 /* Dont run handlers when first register/ last unregister in progress*/
87 #define UPROBE_RUN_HANDLER 1
88 /* Can skip singlestep */
89 #define UPROBE_SKIP_SSTEP 2
92 struct rb_node rb_node; /* node in the rb tree */
94 struct rw_semaphore register_rwsem;
95 struct rw_semaphore consumer_rwsem;
96 struct mutex copy_mutex; /* TODO: kill me and UPROBE_COPY_INSN */
97 struct list_head pending_list;
98 struct uprobe_consumer *consumers;
99 struct inode *inode; /* Also hold a ref to inode */
102 struct arch_uprobe arch;
106 * valid_vma: Verify if the specified vma is an executable vma
107 * Relax restrictions while unregistering: vm_flags might have
108 * changed after breakpoint was inserted.
109 * - is_register: indicates if we are in register context.
110 * - Return 1 if the specified virtual address is in an
113 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
115 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
120 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
123 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
125 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
128 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
130 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
134 * __replace_page - replace page in vma by new page.
135 * based on replace_page in mm/ksm.c
137 * @vma: vma that holds the pte pointing to page
138 * @addr: address the old @page is mapped at
139 * @page: the cowed page we are replacing by kpage
140 * @kpage: the modified page we replace page by
142 * Returns 0 on success, -EFAULT on failure.
144 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
145 struct page *page, struct page *kpage)
147 struct mm_struct *mm = vma->vm_mm;
151 /* For mmu_notifiers */
152 const unsigned long mmun_start = addr;
153 const unsigned long mmun_end = addr + PAGE_SIZE;
155 /* For try_to_free_swap() and munlock_vma_page() below */
158 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
160 ptep = page_check_address(page, mm, addr, &ptl, 0);
165 page_add_new_anon_rmap(kpage, vma, addr);
167 if (!PageAnon(page)) {
168 dec_mm_counter(mm, MM_FILEPAGES);
169 inc_mm_counter(mm, MM_ANONPAGES);
172 flush_cache_page(vma, addr, pte_pfn(*ptep));
173 ptep_clear_flush(vma, addr, ptep);
174 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
176 page_remove_rmap(page);
177 if (!page_mapped(page))
178 try_to_free_swap(page);
179 pte_unmap_unlock(ptep, ptl);
181 if (vma->vm_flags & VM_LOCKED)
182 munlock_vma_page(page);
187 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
193 * is_swbp_insn - check if instruction is breakpoint instruction.
194 * @insn: instruction to be checked.
195 * Default implementation of is_swbp_insn
196 * Returns true if @insn is a breakpoint instruction.
198 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
200 return *insn == UPROBE_SWBP_INSN;
203 static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode)
205 void *kaddr = kmap_atomic(page);
206 memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE);
207 kunmap_atomic(kaddr);
210 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
212 uprobe_opcode_t old_opcode;
215 copy_opcode(page, vaddr, &old_opcode);
216 is_swbp = is_swbp_insn(&old_opcode);
218 if (is_swbp_insn(new_opcode)) {
219 if (is_swbp) /* register: already installed? */
222 if (!is_swbp) /* unregister: was it changed by us? */
231 * Expect the breakpoint instruction to be the smallest size instruction for
232 * the architecture. If an arch has variable length instruction and the
233 * breakpoint instruction is not of the smallest length instruction
234 * supported by that architecture then we need to modify is_swbp_at_addr and
235 * write_opcode accordingly. This would never be a problem for archs that
236 * have fixed length instructions.
240 * write_opcode - write the opcode at a given virtual address.
241 * @mm: the probed process address space.
242 * @vaddr: the virtual address to store the opcode.
243 * @opcode: opcode to be written at @vaddr.
245 * Called with mm->mmap_sem held (for read and with a reference to
248 * For mm @mm, write the opcode at @vaddr.
249 * Return 0 (success) or a negative errno.
251 static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
252 uprobe_opcode_t opcode)
254 struct page *old_page, *new_page;
255 void *vaddr_old, *vaddr_new;
256 struct vm_area_struct *vma;
260 /* Read the page with vaddr into memory */
261 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
265 ret = verify_opcode(old_page, vaddr, &opcode);
270 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
274 __SetPageUptodate(new_page);
276 /* copy the page now that we've got it stable */
277 vaddr_old = kmap_atomic(old_page);
278 vaddr_new = kmap_atomic(new_page);
280 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
281 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
283 kunmap_atomic(vaddr_new);
284 kunmap_atomic(vaddr_old);
286 ret = anon_vma_prepare(vma);
290 ret = __replace_page(vma, vaddr, old_page, new_page);
293 page_cache_release(new_page);
297 if (unlikely(ret == -EAGAIN))
303 * set_swbp - store breakpoint at a given address.
304 * @auprobe: arch specific probepoint information.
305 * @mm: the probed process address space.
306 * @vaddr: the virtual address to insert the opcode.
308 * For mm @mm, store the breakpoint instruction at @vaddr.
309 * Return 0 (success) or a negative errno.
311 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
313 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
317 * set_orig_insn - Restore the original instruction.
318 * @mm: the probed process address space.
319 * @auprobe: arch specific probepoint information.
320 * @vaddr: the virtual address to insert the opcode.
322 * For mm @mm, restore the original opcode (opcode) at @vaddr.
323 * Return 0 (success) or a negative errno.
326 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
328 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
331 static int match_uprobe(struct uprobe *l, struct uprobe *r)
333 if (l->inode < r->inode)
336 if (l->inode > r->inode)
339 if (l->offset < r->offset)
342 if (l->offset > r->offset)
348 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
350 struct uprobe u = { .inode = inode, .offset = offset };
351 struct rb_node *n = uprobes_tree.rb_node;
352 struct uprobe *uprobe;
356 uprobe = rb_entry(n, struct uprobe, rb_node);
357 match = match_uprobe(&u, uprobe);
359 atomic_inc(&uprobe->ref);
372 * Find a uprobe corresponding to a given inode:offset
373 * Acquires uprobes_treelock
375 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
377 struct uprobe *uprobe;
379 spin_lock(&uprobes_treelock);
380 uprobe = __find_uprobe(inode, offset);
381 spin_unlock(&uprobes_treelock);
386 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
388 struct rb_node **p = &uprobes_tree.rb_node;
389 struct rb_node *parent = NULL;
395 u = rb_entry(parent, struct uprobe, rb_node);
396 match = match_uprobe(uprobe, u);
403 p = &parent->rb_left;
405 p = &parent->rb_right;
410 rb_link_node(&uprobe->rb_node, parent, p);
411 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
412 /* get access + creation ref */
413 atomic_set(&uprobe->ref, 2);
419 * Acquire uprobes_treelock.
420 * Matching uprobe already exists in rbtree;
421 * increment (access refcount) and return the matching uprobe.
423 * No matching uprobe; insert the uprobe in rb_tree;
424 * get a double refcount (access + creation) and return NULL.
426 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
430 spin_lock(&uprobes_treelock);
431 u = __insert_uprobe(uprobe);
432 spin_unlock(&uprobes_treelock);
437 static void put_uprobe(struct uprobe *uprobe)
439 if (atomic_dec_and_test(&uprobe->ref))
443 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
445 struct uprobe *uprobe, *cur_uprobe;
447 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
451 uprobe->inode = igrab(inode);
452 uprobe->offset = offset;
453 init_rwsem(&uprobe->register_rwsem);
454 init_rwsem(&uprobe->consumer_rwsem);
455 mutex_init(&uprobe->copy_mutex);
456 /* For now assume that the instruction need not be single-stepped */
457 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
459 /* add to uprobes_tree, sorted on inode:offset */
460 cur_uprobe = insert_uprobe(uprobe);
462 /* a uprobe exists for this inode:offset combination */
468 atomic_inc(&uprobe_events);
474 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
476 struct uprobe_consumer *uc;
478 if (!test_bit(UPROBE_RUN_HANDLER, &uprobe->flags))
481 down_read(&uprobe->register_rwsem);
482 for (uc = uprobe->consumers; uc; uc = uc->next)
483 uc->handler(uc, regs);
484 up_read(&uprobe->register_rwsem);
487 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
489 down_write(&uprobe->consumer_rwsem);
490 uc->next = uprobe->consumers;
491 uprobe->consumers = uc;
492 up_write(&uprobe->consumer_rwsem);
496 * For uprobe @uprobe, delete the consumer @uc.
497 * Return true if the @uc is deleted successfully
500 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
502 struct uprobe_consumer **con;
505 down_write(&uprobe->consumer_rwsem);
506 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
513 up_write(&uprobe->consumer_rwsem);
519 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
520 unsigned long nbytes, loff_t offset)
530 if (!mapping->a_ops->readpage)
533 idx = offset >> PAGE_CACHE_SHIFT;
534 off = offset & ~PAGE_MASK;
537 * Ensure that the page that has the original instruction is
538 * populated and in page-cache.
540 page = read_mapping_page(mapping, idx, filp);
542 return PTR_ERR(page);
544 vaddr = kmap_atomic(page);
545 memcpy(insn, vaddr + off, nbytes);
546 kunmap_atomic(vaddr);
547 page_cache_release(page);
552 static int copy_insn(struct uprobe *uprobe, struct file *filp)
554 struct address_space *mapping;
555 unsigned long nbytes;
558 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
559 mapping = uprobe->inode->i_mapping;
561 /* Instruction at end of binary; copy only available bytes */
562 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
563 bytes = uprobe->inode->i_size - uprobe->offset;
565 bytes = MAX_UINSN_BYTES;
567 /* Instruction at the page-boundary; copy bytes in second page */
568 if (nbytes < bytes) {
569 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
570 bytes - nbytes, uprobe->offset + nbytes);
575 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
578 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
579 struct mm_struct *mm, unsigned long vaddr)
583 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
586 mutex_lock(&uprobe->copy_mutex);
587 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
590 ret = copy_insn(uprobe, file);
595 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
598 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
602 /* write_opcode() assumes we don't cross page boundary */
603 BUG_ON((uprobe->offset & ~PAGE_MASK) +
604 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
606 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
607 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
610 mutex_unlock(&uprobe->copy_mutex);
615 static bool filter_chain(struct uprobe *uprobe)
617 struct uprobe_consumer *uc;
620 down_read(&uprobe->consumer_rwsem);
621 for (uc = uprobe->consumers; uc; uc = uc->next) {
622 /* TODO: ret = uc->filter(...) */
627 up_read(&uprobe->consumer_rwsem);
633 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
634 struct vm_area_struct *vma, unsigned long vaddr)
640 * If probe is being deleted, unregister thread could be done with
641 * the vma-rmap-walk through. Adding a probe now can be fatal since
642 * nobody will be able to cleanup. But in this case filter_chain()
643 * must return false, all consumers have gone away.
645 if (!filter_chain(uprobe))
648 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
653 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
654 * the task can hit this breakpoint right after __replace_page().
656 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
658 set_bit(MMF_HAS_UPROBES, &mm->flags);
660 ret = set_swbp(&uprobe->arch, mm, vaddr);
662 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
663 else if (first_uprobe)
664 clear_bit(MMF_HAS_UPROBES, &mm->flags);
670 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
672 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
675 if (filter_chain(uprobe))
678 set_bit(MMF_RECALC_UPROBES, &mm->flags);
679 return set_orig_insn(&uprobe->arch, mm, vaddr);
683 * There could be threads that have already hit the breakpoint. They
684 * will recheck the current insn and restart if find_uprobe() fails.
685 * See find_active_uprobe().
687 static void delete_uprobe(struct uprobe *uprobe)
689 spin_lock(&uprobes_treelock);
690 rb_erase(&uprobe->rb_node, &uprobes_tree);
691 spin_unlock(&uprobes_treelock);
694 atomic_dec(&uprobe_events);
698 struct map_info *next;
699 struct mm_struct *mm;
703 static inline struct map_info *free_map_info(struct map_info *info)
705 struct map_info *next = info->next;
710 static struct map_info *
711 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
713 unsigned long pgoff = offset >> PAGE_SHIFT;
714 struct vm_area_struct *vma;
715 struct map_info *curr = NULL;
716 struct map_info *prev = NULL;
717 struct map_info *info;
721 mutex_lock(&mapping->i_mmap_mutex);
722 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
723 if (!valid_vma(vma, is_register))
726 if (!prev && !more) {
728 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
729 * reclaim. This is optimistic, no harm done if it fails.
731 prev = kmalloc(sizeof(struct map_info),
732 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
741 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
749 info->mm = vma->vm_mm;
750 info->vaddr = offset_to_vaddr(vma, offset);
752 mutex_unlock(&mapping->i_mmap_mutex);
764 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
766 curr = ERR_PTR(-ENOMEM);
776 prev = free_map_info(prev);
780 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
782 struct map_info *info;
785 percpu_down_write(&dup_mmap_sem);
786 info = build_map_info(uprobe->inode->i_mapping,
787 uprobe->offset, is_register);
794 struct mm_struct *mm = info->mm;
795 struct vm_area_struct *vma;
797 if (err && is_register)
800 down_write(&mm->mmap_sem);
801 vma = find_vma(mm, info->vaddr);
802 if (!vma || !valid_vma(vma, is_register) ||
803 vma->vm_file->f_mapping->host != uprobe->inode)
806 if (vma->vm_start > info->vaddr ||
807 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
811 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
813 err |= remove_breakpoint(uprobe, mm, info->vaddr);
816 up_write(&mm->mmap_sem);
819 info = free_map_info(info);
822 percpu_up_write(&dup_mmap_sem);
826 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
830 consumer_add(uprobe, uc);
831 err = register_for_each_vma(uprobe, true);
832 if (!err) /* TODO: pointless unless the first consumer */
833 set_bit(UPROBE_RUN_HANDLER, &uprobe->flags);
837 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
841 if (!consumer_del(uprobe, uc)) /* WARN? */
844 err = register_for_each_vma(uprobe, false);
845 if (!uprobe->consumers) {
846 clear_bit(UPROBE_RUN_HANDLER, &uprobe->flags);
847 /* TODO : cant unregister? schedule a worker thread */
849 delete_uprobe(uprobe);
854 * uprobe_register - register a probe
855 * @inode: the file in which the probe has to be placed.
856 * @offset: offset from the start of the file.
857 * @uc: information on howto handle the probe..
859 * Apart from the access refcount, uprobe_register() takes a creation
860 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
861 * inserted into the rbtree (i.e first consumer for a @inode:@offset
862 * tuple). Creation refcount stops uprobe_unregister from freeing the
863 * @uprobe even before the register operation is complete. Creation
864 * refcount is released when the last @uc for the @uprobe
867 * Return errno if it cannot successully install probes
868 * else return 0 (success)
870 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
872 struct uprobe *uprobe;
875 /* Racy, just to catch the obvious mistakes */
876 if (offset > i_size_read(inode))
880 mutex_lock(uprobes_hash(inode));
881 uprobe = alloc_uprobe(inode, offset);
883 down_write(&uprobe->register_rwsem);
884 ret = __uprobe_register(uprobe, uc);
886 __uprobe_unregister(uprobe, uc);
887 up_write(&uprobe->register_rwsem);
889 mutex_unlock(uprobes_hash(inode));
897 * uprobe_unregister - unregister a already registered probe.
898 * @inode: the file in which the probe has to be removed.
899 * @offset: offset from the start of the file.
900 * @uc: identify which probe if multiple probes are colocated.
902 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
904 struct uprobe *uprobe;
906 uprobe = find_uprobe(inode, offset);
910 mutex_lock(uprobes_hash(inode));
911 down_write(&uprobe->register_rwsem);
912 __uprobe_unregister(uprobe, uc);
913 up_write(&uprobe->register_rwsem);
914 mutex_unlock(uprobes_hash(inode));
918 static struct rb_node *
919 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
921 struct rb_node *n = uprobes_tree.rb_node;
924 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
926 if (inode < u->inode) {
928 } else if (inode > u->inode) {
933 else if (min > u->offset)
944 * For a given range in vma, build a list of probes that need to be inserted.
946 static void build_probe_list(struct inode *inode,
947 struct vm_area_struct *vma,
948 unsigned long start, unsigned long end,
949 struct list_head *head)
952 struct rb_node *n, *t;
955 INIT_LIST_HEAD(head);
956 min = vaddr_to_offset(vma, start);
957 max = min + (end - start) - 1;
959 spin_lock(&uprobes_treelock);
960 n = find_node_in_range(inode, min, max);
962 for (t = n; t; t = rb_prev(t)) {
963 u = rb_entry(t, struct uprobe, rb_node);
964 if (u->inode != inode || u->offset < min)
966 list_add(&u->pending_list, head);
969 for (t = n; (t = rb_next(t)); ) {
970 u = rb_entry(t, struct uprobe, rb_node);
971 if (u->inode != inode || u->offset > max)
973 list_add(&u->pending_list, head);
977 spin_unlock(&uprobes_treelock);
981 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
983 * Currently we ignore all errors and always return 0, the callers
984 * can't handle the failure anyway.
986 int uprobe_mmap(struct vm_area_struct *vma)
988 struct list_head tmp_list;
989 struct uprobe *uprobe, *u;
992 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
995 inode = vma->vm_file->f_mapping->host;
999 mutex_lock(uprobes_mmap_hash(inode));
1000 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1002 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1003 if (!fatal_signal_pending(current)) {
1004 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1005 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1009 mutex_unlock(uprobes_mmap_hash(inode));
1015 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1018 struct inode *inode;
1021 inode = vma->vm_file->f_mapping->host;
1023 min = vaddr_to_offset(vma, start);
1024 max = min + (end - start) - 1;
1026 spin_lock(&uprobes_treelock);
1027 n = find_node_in_range(inode, min, max);
1028 spin_unlock(&uprobes_treelock);
1034 * Called in context of a munmap of a vma.
1036 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1038 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1041 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1044 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1045 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1048 if (vma_has_uprobes(vma, start, end))
1049 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1052 /* Slot allocation for XOL */
1053 static int xol_add_vma(struct xol_area *area)
1055 struct mm_struct *mm;
1058 area->page = alloc_page(GFP_HIGHUSER);
1065 down_write(&mm->mmap_sem);
1066 if (mm->uprobes_state.xol_area)
1071 /* Try to map as high as possible, this is only a hint. */
1072 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1073 if (area->vaddr & ~PAGE_MASK) {
1078 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1079 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1083 smp_wmb(); /* pairs with get_xol_area() */
1084 mm->uprobes_state.xol_area = area;
1088 up_write(&mm->mmap_sem);
1090 __free_page(area->page);
1095 static struct xol_area *get_xol_area(struct mm_struct *mm)
1097 struct xol_area *area;
1099 area = mm->uprobes_state.xol_area;
1100 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1106 * xol_alloc_area - Allocate process's xol_area.
1107 * This area will be used for storing instructions for execution out of
1110 * Returns the allocated area or NULL.
1112 static struct xol_area *xol_alloc_area(void)
1114 struct xol_area *area;
1116 area = kzalloc(sizeof(*area), GFP_KERNEL);
1117 if (unlikely(!area))
1120 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1125 init_waitqueue_head(&area->wq);
1126 if (!xol_add_vma(area))
1130 kfree(area->bitmap);
1133 return get_xol_area(current->mm);
1137 * uprobe_clear_state - Free the area allocated for slots.
1139 void uprobe_clear_state(struct mm_struct *mm)
1141 struct xol_area *area = mm->uprobes_state.xol_area;
1146 put_page(area->page);
1147 kfree(area->bitmap);
1151 void uprobe_start_dup_mmap(void)
1153 percpu_down_read(&dup_mmap_sem);
1156 void uprobe_end_dup_mmap(void)
1158 percpu_up_read(&dup_mmap_sem);
1161 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1163 newmm->uprobes_state.xol_area = NULL;
1165 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1166 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1167 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1168 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1173 * - search for a free slot.
1175 static unsigned long xol_take_insn_slot(struct xol_area *area)
1177 unsigned long slot_addr;
1181 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1182 if (slot_nr < UINSNS_PER_PAGE) {
1183 if (!test_and_set_bit(slot_nr, area->bitmap))
1186 slot_nr = UINSNS_PER_PAGE;
1189 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1190 } while (slot_nr >= UINSNS_PER_PAGE);
1192 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1193 atomic_inc(&area->slot_count);
1199 * xol_get_insn_slot - If was not allocated a slot, then
1201 * Returns the allocated slot address or 0.
1203 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1205 struct xol_area *area;
1206 unsigned long offset;
1209 area = get_xol_area(current->mm);
1211 area = xol_alloc_area();
1215 current->utask->xol_vaddr = xol_take_insn_slot(area);
1218 * Initialize the slot if xol_vaddr points to valid
1221 if (unlikely(!current->utask->xol_vaddr))
1224 current->utask->vaddr = slot_addr;
1225 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1226 vaddr = kmap_atomic(area->page);
1227 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1228 kunmap_atomic(vaddr);
1230 * We probably need flush_icache_user_range() but it needs vma.
1231 * This should work on supported architectures too.
1233 flush_dcache_page(area->page);
1235 return current->utask->xol_vaddr;
1239 * xol_free_insn_slot - If slot was earlier allocated by
1240 * @xol_get_insn_slot(), make the slot available for
1241 * subsequent requests.
1243 static void xol_free_insn_slot(struct task_struct *tsk)
1245 struct xol_area *area;
1246 unsigned long vma_end;
1247 unsigned long slot_addr;
1249 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1252 slot_addr = tsk->utask->xol_vaddr;
1254 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1257 area = tsk->mm->uprobes_state.xol_area;
1258 vma_end = area->vaddr + PAGE_SIZE;
1259 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1260 unsigned long offset;
1263 offset = slot_addr - area->vaddr;
1264 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1265 if (slot_nr >= UINSNS_PER_PAGE)
1268 clear_bit(slot_nr, area->bitmap);
1269 atomic_dec(&area->slot_count);
1270 if (waitqueue_active(&area->wq))
1273 tsk->utask->xol_vaddr = 0;
1278 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1279 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1281 * Return the address of the breakpoint instruction.
1283 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1285 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1289 * Called with no locks held.
1290 * Called in context of a exiting or a exec-ing thread.
1292 void uprobe_free_utask(struct task_struct *t)
1294 struct uprobe_task *utask = t->utask;
1299 if (utask->active_uprobe)
1300 put_uprobe(utask->active_uprobe);
1302 xol_free_insn_slot(t);
1308 * Called in context of a new clone/fork from copy_process.
1310 void uprobe_copy_process(struct task_struct *t)
1316 * Allocate a uprobe_task object for the task.
1317 * Called when the thread hits a breakpoint for the first time.
1320 * - pointer to new uprobe_task on success
1323 static struct uprobe_task *add_utask(void)
1325 struct uprobe_task *utask;
1327 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1328 if (unlikely(!utask))
1331 current->utask = utask;
1335 /* Prepare to single-step probed instruction out of line. */
1337 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1339 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1346 * If we are singlestepping, then ensure this thread is not connected to
1347 * non-fatal signals until completion of singlestep. When xol insn itself
1348 * triggers the signal, restart the original insn even if the task is
1349 * already SIGKILL'ed (since coredump should report the correct ip). This
1350 * is even more important if the task has a handler for SIGSEGV/etc, The
1351 * _same_ instruction should be repeated again after return from the signal
1352 * handler, and SSTEP can never finish in this case.
1354 bool uprobe_deny_signal(void)
1356 struct task_struct *t = current;
1357 struct uprobe_task *utask = t->utask;
1359 if (likely(!utask || !utask->active_uprobe))
1362 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1364 if (signal_pending(t)) {
1365 spin_lock_irq(&t->sighand->siglock);
1366 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1367 spin_unlock_irq(&t->sighand->siglock);
1369 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1370 utask->state = UTASK_SSTEP_TRAPPED;
1371 set_tsk_thread_flag(t, TIF_UPROBE);
1372 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1380 * Avoid singlestepping the original instruction if the original instruction
1381 * is a NOP or can be emulated.
1383 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1385 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
1386 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1388 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
1393 static void mmf_recalc_uprobes(struct mm_struct *mm)
1395 struct vm_area_struct *vma;
1397 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1398 if (!valid_vma(vma, false))
1401 * This is not strictly accurate, we can race with
1402 * uprobe_unregister() and see the already removed
1403 * uprobe if delete_uprobe() was not yet called.
1404 * Or this uprobe can be filtered out.
1406 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1410 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1413 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1416 uprobe_opcode_t opcode;
1419 pagefault_disable();
1420 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1424 if (likely(result == 0))
1427 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1431 copy_opcode(page, vaddr, &opcode);
1434 return is_swbp_insn(&opcode);
1437 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1439 struct mm_struct *mm = current->mm;
1440 struct uprobe *uprobe = NULL;
1441 struct vm_area_struct *vma;
1443 down_read(&mm->mmap_sem);
1444 vma = find_vma(mm, bp_vaddr);
1445 if (vma && vma->vm_start <= bp_vaddr) {
1446 if (valid_vma(vma, false)) {
1447 struct inode *inode = vma->vm_file->f_mapping->host;
1448 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1450 uprobe = find_uprobe(inode, offset);
1454 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1459 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1460 mmf_recalc_uprobes(mm);
1461 up_read(&mm->mmap_sem);
1467 * Run handler and ask thread to singlestep.
1468 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1470 static void handle_swbp(struct pt_regs *regs)
1472 struct uprobe_task *utask;
1473 struct uprobe *uprobe;
1474 unsigned long bp_vaddr;
1475 int uninitialized_var(is_swbp);
1477 bp_vaddr = uprobe_get_swbp_addr(regs);
1478 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1482 /* No matching uprobe; signal SIGTRAP. */
1483 send_sig(SIGTRAP, current, 0);
1486 * Either we raced with uprobe_unregister() or we can't
1487 * access this memory. The latter is only possible if
1488 * another thread plays with our ->mm. In both cases
1489 * we can simply restart. If this vma was unmapped we
1490 * can pretend this insn was not executed yet and get
1491 * the (correct) SIGSEGV after restart.
1493 instruction_pointer_set(regs, bp_vaddr);
1498 * TODO: move copy_insn/etc into _register and remove this hack.
1499 * After we hit the bp, _unregister + _register can install the
1500 * new and not-yet-analyzed uprobe at the same address, restart.
1502 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1503 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1506 utask = current->utask;
1508 utask = add_utask();
1509 /* Cannot allocate; re-execute the instruction. */
1514 handler_chain(uprobe, regs);
1515 if (can_skip_sstep(uprobe, regs))
1518 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1519 utask->active_uprobe = uprobe;
1520 utask->state = UTASK_SSTEP;
1526 * cannot singlestep; cannot skip instruction;
1527 * re-execute the instruction.
1529 instruction_pointer_set(regs, bp_vaddr);
1535 * Perform required fix-ups and disable singlestep.
1536 * Allow pending signals to take effect.
1538 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1540 struct uprobe *uprobe;
1542 uprobe = utask->active_uprobe;
1543 if (utask->state == UTASK_SSTEP_ACK)
1544 arch_uprobe_post_xol(&uprobe->arch, regs);
1545 else if (utask->state == UTASK_SSTEP_TRAPPED)
1546 arch_uprobe_abort_xol(&uprobe->arch, regs);
1551 utask->active_uprobe = NULL;
1552 utask->state = UTASK_RUNNING;
1553 xol_free_insn_slot(current);
1555 spin_lock_irq(¤t->sighand->siglock);
1556 recalc_sigpending(); /* see uprobe_deny_signal() */
1557 spin_unlock_irq(¤t->sighand->siglock);
1561 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1562 * allows the thread to return from interrupt. After that handle_swbp()
1563 * sets utask->active_uprobe.
1565 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1566 * and allows the thread to return from interrupt.
1568 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1569 * uprobe_notify_resume().
1571 void uprobe_notify_resume(struct pt_regs *regs)
1573 struct uprobe_task *utask;
1575 clear_thread_flag(TIF_UPROBE);
1577 utask = current->utask;
1578 if (utask && utask->active_uprobe)
1579 handle_singlestep(utask, regs);
1585 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1586 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1588 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1590 if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags))
1593 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 if (percpu_init_rwsem(&dup_mmap_sem))
1631 return register_die_notifier(&uprobe_exception_nb);
1633 module_init(init_uprobes);
1635 static void __exit exit_uprobes(void)
1638 module_exit(exit_uprobes);