Linux 4.15.6
[linux/fpc-iii.git] / kernel / events / uprobes.c
blob267f6ef91d9709e2f53c35e053a792d3fcede64d
1 /*
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
19 * Authors:
20 * Srikar Dronamraju
21 * Jim Keniston
22 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra
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/sched/mm.h>
31 #include <linux/sched/coredump.h>
32 #include <linux/export.h>
33 #include <linux/rmap.h> /* anon_vma_prepare */
34 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
35 #include <linux/swap.h> /* try_to_free_swap */
36 #include <linux/ptrace.h> /* user_enable_single_step */
37 #include <linux/kdebug.h> /* notifier mechanism */
38 #include "../../mm/internal.h" /* munlock_vma_page */
39 #include <linux/percpu-rwsem.h>
40 #include <linux/task_work.h>
41 #include <linux/shmem_fs.h>
43 #include <linux/uprobes.h>
45 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
46 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
48 static struct rb_root uprobes_tree = RB_ROOT;
50 * allows us to skip the uprobe_mmap if there are no uprobe events active
51 * at this time. Probably a fine grained per inode count is better?
53 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
55 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
57 #define UPROBES_HASH_SZ 13
58 /* serialize uprobe->pending_list */
59 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
60 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
62 static struct percpu_rw_semaphore dup_mmap_sem;
64 /* Have a copy of original instruction */
65 #define UPROBE_COPY_INSN 0
67 struct uprobe {
68 struct rb_node rb_node; /* node in the rb tree */
69 atomic_t ref;
70 struct rw_semaphore register_rwsem;
71 struct rw_semaphore consumer_rwsem;
72 struct list_head pending_list;
73 struct uprobe_consumer *consumers;
74 struct inode *inode; /* Also hold a ref to inode */
75 loff_t offset;
76 unsigned long flags;
79 * The generic code assumes that it has two members of unknown type
80 * owned by the arch-specific code:
82 * insn - copy_insn() saves the original instruction here for
83 * arch_uprobe_analyze_insn().
85 * ixol - potentially modified instruction to execute out of
86 * line, copied to xol_area by xol_get_insn_slot().
88 struct arch_uprobe arch;
92 * Execute out of line area: anonymous executable mapping installed
93 * by the probed task to execute the copy of the original instruction
94 * mangled by set_swbp().
96 * On a breakpoint hit, thread contests for a slot. It frees the
97 * slot after singlestep. Currently a fixed number of slots are
98 * allocated.
100 struct xol_area {
101 wait_queue_head_t wq; /* if all slots are busy */
102 atomic_t slot_count; /* number of in-use slots */
103 unsigned long *bitmap; /* 0 = free slot */
105 struct vm_special_mapping xol_mapping;
106 struct page *pages[2];
108 * We keep the vma's vm_start rather than a pointer to the vma
109 * itself. The probed process or a naughty kernel module could make
110 * the vma go away, and we must handle that reasonably gracefully.
112 unsigned long vaddr; /* Page(s) of instruction slots */
116 * valid_vma: Verify if the specified vma is an executable vma
117 * Relax restrictions while unregistering: vm_flags might have
118 * changed after breakpoint was inserted.
119 * - is_register: indicates if we are in register context.
120 * - Return 1 if the specified virtual address is in an
121 * executable vma.
123 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
125 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
127 if (is_register)
128 flags |= VM_WRITE;
130 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
133 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
135 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
138 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
140 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
144 * __replace_page - replace page in vma by new page.
145 * based on replace_page in mm/ksm.c
147 * @vma: vma that holds the pte pointing to page
148 * @addr: address the old @page is mapped at
149 * @page: the cowed page we are replacing by kpage
150 * @kpage: the modified page we replace page by
152 * Returns 0 on success, -EFAULT on failure.
154 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
155 struct page *old_page, struct page *new_page)
157 struct mm_struct *mm = vma->vm_mm;
158 struct page_vma_mapped_walk pvmw = {
159 .page = old_page,
160 .vma = vma,
161 .address = addr,
163 int err;
164 /* For mmu_notifiers */
165 const unsigned long mmun_start = addr;
166 const unsigned long mmun_end = addr + PAGE_SIZE;
167 struct mem_cgroup *memcg;
169 VM_BUG_ON_PAGE(PageTransHuge(old_page), old_page);
171 err = mem_cgroup_try_charge(new_page, vma->vm_mm, GFP_KERNEL, &memcg,
172 false);
173 if (err)
174 return err;
176 /* For try_to_free_swap() and munlock_vma_page() below */
177 lock_page(old_page);
179 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
180 err = -EAGAIN;
181 if (!page_vma_mapped_walk(&pvmw)) {
182 mem_cgroup_cancel_charge(new_page, memcg, false);
183 goto unlock;
185 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
187 get_page(new_page);
188 page_add_new_anon_rmap(new_page, vma, addr, false);
189 mem_cgroup_commit_charge(new_page, memcg, false, false);
190 lru_cache_add_active_or_unevictable(new_page, vma);
192 if (!PageAnon(old_page)) {
193 dec_mm_counter(mm, mm_counter_file(old_page));
194 inc_mm_counter(mm, MM_ANONPAGES);
197 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
198 ptep_clear_flush_notify(vma, addr, pvmw.pte);
199 set_pte_at_notify(mm, addr, pvmw.pte,
200 mk_pte(new_page, vma->vm_page_prot));
202 page_remove_rmap(old_page, false);
203 if (!page_mapped(old_page))
204 try_to_free_swap(old_page);
205 page_vma_mapped_walk_done(&pvmw);
207 if (vma->vm_flags & VM_LOCKED)
208 munlock_vma_page(old_page);
209 put_page(old_page);
211 err = 0;
212 unlock:
213 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
214 unlock_page(old_page);
215 return err;
219 * is_swbp_insn - check if instruction is breakpoint instruction.
220 * @insn: instruction to be checked.
221 * Default implementation of is_swbp_insn
222 * Returns true if @insn is a breakpoint instruction.
224 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
226 return *insn == UPROBE_SWBP_INSN;
230 * is_trap_insn - check if instruction is breakpoint instruction.
231 * @insn: instruction to be checked.
232 * Default implementation of is_trap_insn
233 * Returns true if @insn is a breakpoint instruction.
235 * This function is needed for the case where an architecture has multiple
236 * trap instructions (like powerpc).
238 bool __weak is_trap_insn(uprobe_opcode_t *insn)
240 return is_swbp_insn(insn);
243 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
245 void *kaddr = kmap_atomic(page);
246 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
247 kunmap_atomic(kaddr);
250 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
252 void *kaddr = kmap_atomic(page);
253 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
254 kunmap_atomic(kaddr);
257 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
259 uprobe_opcode_t old_opcode;
260 bool is_swbp;
263 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
264 * We do not check if it is any other 'trap variant' which could
265 * be conditional trap instruction such as the one powerpc supports.
267 * The logic is that we do not care if the underlying instruction
268 * is a trap variant; uprobes always wins over any other (gdb)
269 * breakpoint.
271 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
272 is_swbp = is_swbp_insn(&old_opcode);
274 if (is_swbp_insn(new_opcode)) {
275 if (is_swbp) /* register: already installed? */
276 return 0;
277 } else {
278 if (!is_swbp) /* unregister: was it changed by us? */
279 return 0;
282 return 1;
286 * NOTE:
287 * Expect the breakpoint instruction to be the smallest size instruction for
288 * the architecture. If an arch has variable length instruction and the
289 * breakpoint instruction is not of the smallest length instruction
290 * supported by that architecture then we need to modify is_trap_at_addr and
291 * uprobe_write_opcode accordingly. This would never be a problem for archs
292 * that have fixed length instructions.
294 * uprobe_write_opcode - write the opcode at a given virtual address.
295 * @mm: the probed process address space.
296 * @vaddr: the virtual address to store the opcode.
297 * @opcode: opcode to be written at @vaddr.
299 * Called with mm->mmap_sem held for write.
300 * Return 0 (success) or a negative errno.
302 int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr,
303 uprobe_opcode_t opcode)
305 struct page *old_page, *new_page;
306 struct vm_area_struct *vma;
307 int ret;
309 retry:
310 /* Read the page with vaddr into memory */
311 ret = get_user_pages_remote(NULL, mm, vaddr, 1,
312 FOLL_FORCE | FOLL_SPLIT, &old_page, &vma, NULL);
313 if (ret <= 0)
314 return ret;
316 ret = verify_opcode(old_page, vaddr, &opcode);
317 if (ret <= 0)
318 goto put_old;
320 ret = anon_vma_prepare(vma);
321 if (ret)
322 goto put_old;
324 ret = -ENOMEM;
325 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
326 if (!new_page)
327 goto put_old;
329 __SetPageUptodate(new_page);
330 copy_highpage(new_page, old_page);
331 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
333 ret = __replace_page(vma, vaddr, old_page, new_page);
334 put_page(new_page);
335 put_old:
336 put_page(old_page);
338 if (unlikely(ret == -EAGAIN))
339 goto retry;
340 return ret;
344 * set_swbp - store breakpoint at a given address.
345 * @auprobe: arch specific probepoint information.
346 * @mm: the probed process address space.
347 * @vaddr: the virtual address to insert the opcode.
349 * For mm @mm, store the breakpoint instruction at @vaddr.
350 * Return 0 (success) or a negative errno.
352 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
354 return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
358 * set_orig_insn - Restore the original instruction.
359 * @mm: the probed process address space.
360 * @auprobe: arch specific probepoint information.
361 * @vaddr: the virtual address to insert the opcode.
363 * For mm @mm, restore the original opcode (opcode) at @vaddr.
364 * Return 0 (success) or a negative errno.
366 int __weak
367 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
369 return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn);
372 static struct uprobe *get_uprobe(struct uprobe *uprobe)
374 atomic_inc(&uprobe->ref);
375 return uprobe;
378 static void put_uprobe(struct uprobe *uprobe)
380 if (atomic_dec_and_test(&uprobe->ref))
381 kfree(uprobe);
384 static int match_uprobe(struct uprobe *l, struct uprobe *r)
386 if (l->inode < r->inode)
387 return -1;
389 if (l->inode > r->inode)
390 return 1;
392 if (l->offset < r->offset)
393 return -1;
395 if (l->offset > r->offset)
396 return 1;
398 return 0;
401 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
403 struct uprobe u = { .inode = inode, .offset = offset };
404 struct rb_node *n = uprobes_tree.rb_node;
405 struct uprobe *uprobe;
406 int match;
408 while (n) {
409 uprobe = rb_entry(n, struct uprobe, rb_node);
410 match = match_uprobe(&u, uprobe);
411 if (!match)
412 return get_uprobe(uprobe);
414 if (match < 0)
415 n = n->rb_left;
416 else
417 n = n->rb_right;
419 return NULL;
423 * Find a uprobe corresponding to a given inode:offset
424 * Acquires uprobes_treelock
426 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
428 struct uprobe *uprobe;
430 spin_lock(&uprobes_treelock);
431 uprobe = __find_uprobe(inode, offset);
432 spin_unlock(&uprobes_treelock);
434 return uprobe;
437 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
439 struct rb_node **p = &uprobes_tree.rb_node;
440 struct rb_node *parent = NULL;
441 struct uprobe *u;
442 int match;
444 while (*p) {
445 parent = *p;
446 u = rb_entry(parent, struct uprobe, rb_node);
447 match = match_uprobe(uprobe, u);
448 if (!match)
449 return get_uprobe(u);
451 if (match < 0)
452 p = &parent->rb_left;
453 else
454 p = &parent->rb_right;
458 u = NULL;
459 rb_link_node(&uprobe->rb_node, parent, p);
460 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
461 /* get access + creation ref */
462 atomic_set(&uprobe->ref, 2);
464 return u;
468 * Acquire uprobes_treelock.
469 * Matching uprobe already exists in rbtree;
470 * increment (access refcount) and return the matching uprobe.
472 * No matching uprobe; insert the uprobe in rb_tree;
473 * get a double refcount (access + creation) and return NULL.
475 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
477 struct uprobe *u;
479 spin_lock(&uprobes_treelock);
480 u = __insert_uprobe(uprobe);
481 spin_unlock(&uprobes_treelock);
483 return u;
486 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
488 struct uprobe *uprobe, *cur_uprobe;
490 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
491 if (!uprobe)
492 return NULL;
494 uprobe->inode = igrab(inode);
495 uprobe->offset = offset;
496 init_rwsem(&uprobe->register_rwsem);
497 init_rwsem(&uprobe->consumer_rwsem);
499 /* add to uprobes_tree, sorted on inode:offset */
500 cur_uprobe = insert_uprobe(uprobe);
501 /* a uprobe exists for this inode:offset combination */
502 if (cur_uprobe) {
503 kfree(uprobe);
504 uprobe = cur_uprobe;
505 iput(inode);
508 return uprobe;
511 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
513 down_write(&uprobe->consumer_rwsem);
514 uc->next = uprobe->consumers;
515 uprobe->consumers = uc;
516 up_write(&uprobe->consumer_rwsem);
520 * For uprobe @uprobe, delete the consumer @uc.
521 * Return true if the @uc is deleted successfully
522 * or return false.
524 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
526 struct uprobe_consumer **con;
527 bool ret = false;
529 down_write(&uprobe->consumer_rwsem);
530 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
531 if (*con == uc) {
532 *con = uc->next;
533 ret = true;
534 break;
537 up_write(&uprobe->consumer_rwsem);
539 return ret;
542 static int __copy_insn(struct address_space *mapping, struct file *filp,
543 void *insn, int nbytes, loff_t offset)
545 struct page *page;
547 * Ensure that the page that has the original instruction is populated
548 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
549 * see uprobe_register().
551 if (mapping->a_ops->readpage)
552 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
553 else
554 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
555 if (IS_ERR(page))
556 return PTR_ERR(page);
558 copy_from_page(page, offset, insn, nbytes);
559 put_page(page);
561 return 0;
564 static int copy_insn(struct uprobe *uprobe, struct file *filp)
566 struct address_space *mapping = uprobe->inode->i_mapping;
567 loff_t offs = uprobe->offset;
568 void *insn = &uprobe->arch.insn;
569 int size = sizeof(uprobe->arch.insn);
570 int len, err = -EIO;
572 /* Copy only available bytes, -EIO if nothing was read */
573 do {
574 if (offs >= i_size_read(uprobe->inode))
575 break;
577 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
578 err = __copy_insn(mapping, filp, insn, len, offs);
579 if (err)
580 break;
582 insn += len;
583 offs += len;
584 size -= len;
585 } while (size);
587 return err;
590 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
591 struct mm_struct *mm, unsigned long vaddr)
593 int ret = 0;
595 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
596 return ret;
598 /* TODO: move this into _register, until then we abuse this sem. */
599 down_write(&uprobe->consumer_rwsem);
600 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
601 goto out;
603 ret = copy_insn(uprobe, file);
604 if (ret)
605 goto out;
607 ret = -ENOTSUPP;
608 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
609 goto out;
611 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
612 if (ret)
613 goto out;
615 /* uprobe_write_opcode() assumes we don't cross page boundary */
616 BUG_ON((uprobe->offset & ~PAGE_MASK) +
617 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
619 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
620 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
622 out:
623 up_write(&uprobe->consumer_rwsem);
625 return ret;
628 static inline bool consumer_filter(struct uprobe_consumer *uc,
629 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
631 return !uc->filter || uc->filter(uc, ctx, mm);
634 static bool filter_chain(struct uprobe *uprobe,
635 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
637 struct uprobe_consumer *uc;
638 bool ret = false;
640 down_read(&uprobe->consumer_rwsem);
641 for (uc = uprobe->consumers; uc; uc = uc->next) {
642 ret = consumer_filter(uc, ctx, mm);
643 if (ret)
644 break;
646 up_read(&uprobe->consumer_rwsem);
648 return ret;
651 static int
652 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
653 struct vm_area_struct *vma, unsigned long vaddr)
655 bool first_uprobe;
656 int ret;
658 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
659 if (ret)
660 return ret;
663 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
664 * the task can hit this breakpoint right after __replace_page().
666 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
667 if (first_uprobe)
668 set_bit(MMF_HAS_UPROBES, &mm->flags);
670 ret = set_swbp(&uprobe->arch, mm, vaddr);
671 if (!ret)
672 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
673 else if (first_uprobe)
674 clear_bit(MMF_HAS_UPROBES, &mm->flags);
676 return ret;
679 static int
680 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
682 set_bit(MMF_RECALC_UPROBES, &mm->flags);
683 return set_orig_insn(&uprobe->arch, mm, vaddr);
686 static inline bool uprobe_is_active(struct uprobe *uprobe)
688 return !RB_EMPTY_NODE(&uprobe->rb_node);
691 * There could be threads that have already hit the breakpoint. They
692 * will recheck the current insn and restart if find_uprobe() fails.
693 * See find_active_uprobe().
695 static void delete_uprobe(struct uprobe *uprobe)
697 if (WARN_ON(!uprobe_is_active(uprobe)))
698 return;
700 spin_lock(&uprobes_treelock);
701 rb_erase(&uprobe->rb_node, &uprobes_tree);
702 spin_unlock(&uprobes_treelock);
703 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
704 iput(uprobe->inode);
705 put_uprobe(uprobe);
708 struct map_info {
709 struct map_info *next;
710 struct mm_struct *mm;
711 unsigned long vaddr;
714 static inline struct map_info *free_map_info(struct map_info *info)
716 struct map_info *next = info->next;
717 kfree(info);
718 return next;
721 static struct map_info *
722 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
724 unsigned long pgoff = offset >> PAGE_SHIFT;
725 struct vm_area_struct *vma;
726 struct map_info *curr = NULL;
727 struct map_info *prev = NULL;
728 struct map_info *info;
729 int more = 0;
731 again:
732 i_mmap_lock_read(mapping);
733 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
734 if (!valid_vma(vma, is_register))
735 continue;
737 if (!prev && !more) {
739 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
740 * reclaim. This is optimistic, no harm done if it fails.
742 prev = kmalloc(sizeof(struct map_info),
743 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
744 if (prev)
745 prev->next = NULL;
747 if (!prev) {
748 more++;
749 continue;
752 if (!mmget_not_zero(vma->vm_mm))
753 continue;
755 info = prev;
756 prev = prev->next;
757 info->next = curr;
758 curr = info;
760 info->mm = vma->vm_mm;
761 info->vaddr = offset_to_vaddr(vma, offset);
763 i_mmap_unlock_read(mapping);
765 if (!more)
766 goto out;
768 prev = curr;
769 while (curr) {
770 mmput(curr->mm);
771 curr = curr->next;
774 do {
775 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
776 if (!info) {
777 curr = ERR_PTR(-ENOMEM);
778 goto out;
780 info->next = prev;
781 prev = info;
782 } while (--more);
784 goto again;
785 out:
786 while (prev)
787 prev = free_map_info(prev);
788 return curr;
791 static int
792 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
794 bool is_register = !!new;
795 struct map_info *info;
796 int err = 0;
798 percpu_down_write(&dup_mmap_sem);
799 info = build_map_info(uprobe->inode->i_mapping,
800 uprobe->offset, is_register);
801 if (IS_ERR(info)) {
802 err = PTR_ERR(info);
803 goto out;
806 while (info) {
807 struct mm_struct *mm = info->mm;
808 struct vm_area_struct *vma;
810 if (err && is_register)
811 goto free;
813 down_write(&mm->mmap_sem);
814 vma = find_vma(mm, info->vaddr);
815 if (!vma || !valid_vma(vma, is_register) ||
816 file_inode(vma->vm_file) != uprobe->inode)
817 goto unlock;
819 if (vma->vm_start > info->vaddr ||
820 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
821 goto unlock;
823 if (is_register) {
824 /* consult only the "caller", new consumer. */
825 if (consumer_filter(new,
826 UPROBE_FILTER_REGISTER, mm))
827 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
828 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
829 if (!filter_chain(uprobe,
830 UPROBE_FILTER_UNREGISTER, mm))
831 err |= remove_breakpoint(uprobe, mm, info->vaddr);
834 unlock:
835 up_write(&mm->mmap_sem);
836 free:
837 mmput(mm);
838 info = free_map_info(info);
840 out:
841 percpu_up_write(&dup_mmap_sem);
842 return err;
845 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
847 consumer_add(uprobe, uc);
848 return register_for_each_vma(uprobe, uc);
851 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
853 int err;
855 if (WARN_ON(!consumer_del(uprobe, uc)))
856 return;
858 err = register_for_each_vma(uprobe, NULL);
859 /* TODO : cant unregister? schedule a worker thread */
860 if (!uprobe->consumers && !err)
861 delete_uprobe(uprobe);
865 * uprobe_register - register a probe
866 * @inode: the file in which the probe has to be placed.
867 * @offset: offset from the start of the file.
868 * @uc: information on howto handle the probe..
870 * Apart from the access refcount, uprobe_register() takes a creation
871 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
872 * inserted into the rbtree (i.e first consumer for a @inode:@offset
873 * tuple). Creation refcount stops uprobe_unregister from freeing the
874 * @uprobe even before the register operation is complete. Creation
875 * refcount is released when the last @uc for the @uprobe
876 * unregisters.
878 * Return errno if it cannot successully install probes
879 * else return 0 (success)
881 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
883 struct uprobe *uprobe;
884 int ret;
886 /* Uprobe must have at least one set consumer */
887 if (!uc->handler && !uc->ret_handler)
888 return -EINVAL;
890 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
891 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
892 return -EIO;
893 /* Racy, just to catch the obvious mistakes */
894 if (offset > i_size_read(inode))
895 return -EINVAL;
897 retry:
898 uprobe = alloc_uprobe(inode, offset);
899 if (!uprobe)
900 return -ENOMEM;
902 * We can race with uprobe_unregister()->delete_uprobe().
903 * Check uprobe_is_active() and retry if it is false.
905 down_write(&uprobe->register_rwsem);
906 ret = -EAGAIN;
907 if (likely(uprobe_is_active(uprobe))) {
908 ret = __uprobe_register(uprobe, uc);
909 if (ret)
910 __uprobe_unregister(uprobe, uc);
912 up_write(&uprobe->register_rwsem);
913 put_uprobe(uprobe);
915 if (unlikely(ret == -EAGAIN))
916 goto retry;
917 return ret;
919 EXPORT_SYMBOL_GPL(uprobe_register);
922 * uprobe_apply - unregister a already registered probe.
923 * @inode: the file in which the probe has to be removed.
924 * @offset: offset from the start of the file.
925 * @uc: consumer which wants to add more or remove some breakpoints
926 * @add: add or remove the breakpoints
928 int uprobe_apply(struct inode *inode, loff_t offset,
929 struct uprobe_consumer *uc, bool add)
931 struct uprobe *uprobe;
932 struct uprobe_consumer *con;
933 int ret = -ENOENT;
935 uprobe = find_uprobe(inode, offset);
936 if (WARN_ON(!uprobe))
937 return ret;
939 down_write(&uprobe->register_rwsem);
940 for (con = uprobe->consumers; con && con != uc ; con = con->next)
942 if (con)
943 ret = register_for_each_vma(uprobe, add ? uc : NULL);
944 up_write(&uprobe->register_rwsem);
945 put_uprobe(uprobe);
947 return ret;
951 * uprobe_unregister - unregister a already registered probe.
952 * @inode: the file in which the probe has to be removed.
953 * @offset: offset from the start of the file.
954 * @uc: identify which probe if multiple probes are colocated.
956 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
958 struct uprobe *uprobe;
960 uprobe = find_uprobe(inode, offset);
961 if (WARN_ON(!uprobe))
962 return;
964 down_write(&uprobe->register_rwsem);
965 __uprobe_unregister(uprobe, uc);
966 up_write(&uprobe->register_rwsem);
967 put_uprobe(uprobe);
969 EXPORT_SYMBOL_GPL(uprobe_unregister);
971 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
973 struct vm_area_struct *vma;
974 int err = 0;
976 down_read(&mm->mmap_sem);
977 for (vma = mm->mmap; vma; vma = vma->vm_next) {
978 unsigned long vaddr;
979 loff_t offset;
981 if (!valid_vma(vma, false) ||
982 file_inode(vma->vm_file) != uprobe->inode)
983 continue;
985 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
986 if (uprobe->offset < offset ||
987 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
988 continue;
990 vaddr = offset_to_vaddr(vma, uprobe->offset);
991 err |= remove_breakpoint(uprobe, mm, vaddr);
993 up_read(&mm->mmap_sem);
995 return err;
998 static struct rb_node *
999 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1001 struct rb_node *n = uprobes_tree.rb_node;
1003 while (n) {
1004 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1006 if (inode < u->inode) {
1007 n = n->rb_left;
1008 } else if (inode > u->inode) {
1009 n = n->rb_right;
1010 } else {
1011 if (max < u->offset)
1012 n = n->rb_left;
1013 else if (min > u->offset)
1014 n = n->rb_right;
1015 else
1016 break;
1020 return n;
1024 * For a given range in vma, build a list of probes that need to be inserted.
1026 static void build_probe_list(struct inode *inode,
1027 struct vm_area_struct *vma,
1028 unsigned long start, unsigned long end,
1029 struct list_head *head)
1031 loff_t min, max;
1032 struct rb_node *n, *t;
1033 struct uprobe *u;
1035 INIT_LIST_HEAD(head);
1036 min = vaddr_to_offset(vma, start);
1037 max = min + (end - start) - 1;
1039 spin_lock(&uprobes_treelock);
1040 n = find_node_in_range(inode, min, max);
1041 if (n) {
1042 for (t = n; t; t = rb_prev(t)) {
1043 u = rb_entry(t, struct uprobe, rb_node);
1044 if (u->inode != inode || u->offset < min)
1045 break;
1046 list_add(&u->pending_list, head);
1047 get_uprobe(u);
1049 for (t = n; (t = rb_next(t)); ) {
1050 u = rb_entry(t, struct uprobe, rb_node);
1051 if (u->inode != inode || u->offset > max)
1052 break;
1053 list_add(&u->pending_list, head);
1054 get_uprobe(u);
1057 spin_unlock(&uprobes_treelock);
1061 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1063 * Currently we ignore all errors and always return 0, the callers
1064 * can't handle the failure anyway.
1066 int uprobe_mmap(struct vm_area_struct *vma)
1068 struct list_head tmp_list;
1069 struct uprobe *uprobe, *u;
1070 struct inode *inode;
1072 if (no_uprobe_events() || !valid_vma(vma, true))
1073 return 0;
1075 inode = file_inode(vma->vm_file);
1076 if (!inode)
1077 return 0;
1079 mutex_lock(uprobes_mmap_hash(inode));
1080 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1082 * We can race with uprobe_unregister(), this uprobe can be already
1083 * removed. But in this case filter_chain() must return false, all
1084 * consumers have gone away.
1086 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1087 if (!fatal_signal_pending(current) &&
1088 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1089 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1090 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1092 put_uprobe(uprobe);
1094 mutex_unlock(uprobes_mmap_hash(inode));
1096 return 0;
1099 static bool
1100 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1102 loff_t min, max;
1103 struct inode *inode;
1104 struct rb_node *n;
1106 inode = file_inode(vma->vm_file);
1108 min = vaddr_to_offset(vma, start);
1109 max = min + (end - start) - 1;
1111 spin_lock(&uprobes_treelock);
1112 n = find_node_in_range(inode, min, max);
1113 spin_unlock(&uprobes_treelock);
1115 return !!n;
1119 * Called in context of a munmap of a vma.
1121 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1123 if (no_uprobe_events() || !valid_vma(vma, false))
1124 return;
1126 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1127 return;
1129 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1130 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1131 return;
1133 if (vma_has_uprobes(vma, start, end))
1134 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1137 /* Slot allocation for XOL */
1138 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1140 struct vm_area_struct *vma;
1141 int ret;
1143 if (down_write_killable(&mm->mmap_sem))
1144 return -EINTR;
1146 if (mm->uprobes_state.xol_area) {
1147 ret = -EALREADY;
1148 goto fail;
1151 if (!area->vaddr) {
1152 /* Try to map as high as possible, this is only a hint. */
1153 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1154 PAGE_SIZE, 0, 0);
1155 if (area->vaddr & ~PAGE_MASK) {
1156 ret = area->vaddr;
1157 goto fail;
1161 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1162 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1163 &area->xol_mapping);
1164 if (IS_ERR(vma)) {
1165 ret = PTR_ERR(vma);
1166 goto fail;
1169 ret = 0;
1170 smp_wmb(); /* pairs with get_xol_area() */
1171 mm->uprobes_state.xol_area = area;
1172 fail:
1173 up_write(&mm->mmap_sem);
1175 return ret;
1178 static struct xol_area *__create_xol_area(unsigned long vaddr)
1180 struct mm_struct *mm = current->mm;
1181 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1182 struct xol_area *area;
1184 area = kmalloc(sizeof(*area), GFP_KERNEL);
1185 if (unlikely(!area))
1186 goto out;
1188 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1189 if (!area->bitmap)
1190 goto free_area;
1192 area->xol_mapping.name = "[uprobes]";
1193 area->xol_mapping.fault = NULL;
1194 area->xol_mapping.pages = area->pages;
1195 area->pages[0] = alloc_page(GFP_HIGHUSER);
1196 if (!area->pages[0])
1197 goto free_bitmap;
1198 area->pages[1] = NULL;
1200 area->vaddr = vaddr;
1201 init_waitqueue_head(&area->wq);
1202 /* Reserve the 1st slot for get_trampoline_vaddr() */
1203 set_bit(0, area->bitmap);
1204 atomic_set(&area->slot_count, 1);
1205 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1207 if (!xol_add_vma(mm, area))
1208 return area;
1210 __free_page(area->pages[0]);
1211 free_bitmap:
1212 kfree(area->bitmap);
1213 free_area:
1214 kfree(area);
1215 out:
1216 return NULL;
1220 * get_xol_area - Allocate process's xol_area if necessary.
1221 * This area will be used for storing instructions for execution out of line.
1223 * Returns the allocated area or NULL.
1225 static struct xol_area *get_xol_area(void)
1227 struct mm_struct *mm = current->mm;
1228 struct xol_area *area;
1230 if (!mm->uprobes_state.xol_area)
1231 __create_xol_area(0);
1233 area = mm->uprobes_state.xol_area;
1234 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1235 return area;
1239 * uprobe_clear_state - Free the area allocated for slots.
1241 void uprobe_clear_state(struct mm_struct *mm)
1243 struct xol_area *area = mm->uprobes_state.xol_area;
1245 if (!area)
1246 return;
1248 put_page(area->pages[0]);
1249 kfree(area->bitmap);
1250 kfree(area);
1253 void uprobe_start_dup_mmap(void)
1255 percpu_down_read(&dup_mmap_sem);
1258 void uprobe_end_dup_mmap(void)
1260 percpu_up_read(&dup_mmap_sem);
1263 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1265 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1266 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1267 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1268 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1273 * - search for a free slot.
1275 static unsigned long xol_take_insn_slot(struct xol_area *area)
1277 unsigned long slot_addr;
1278 int slot_nr;
1280 do {
1281 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1282 if (slot_nr < UINSNS_PER_PAGE) {
1283 if (!test_and_set_bit(slot_nr, area->bitmap))
1284 break;
1286 slot_nr = UINSNS_PER_PAGE;
1287 continue;
1289 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1290 } while (slot_nr >= UINSNS_PER_PAGE);
1292 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1293 atomic_inc(&area->slot_count);
1295 return slot_addr;
1299 * xol_get_insn_slot - allocate a slot for xol.
1300 * Returns the allocated slot address or 0.
1302 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1304 struct xol_area *area;
1305 unsigned long xol_vaddr;
1307 area = get_xol_area();
1308 if (!area)
1309 return 0;
1311 xol_vaddr = xol_take_insn_slot(area);
1312 if (unlikely(!xol_vaddr))
1313 return 0;
1315 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1316 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1318 return xol_vaddr;
1322 * xol_free_insn_slot - If slot was earlier allocated by
1323 * @xol_get_insn_slot(), make the slot available for
1324 * subsequent requests.
1326 static void xol_free_insn_slot(struct task_struct *tsk)
1328 struct xol_area *area;
1329 unsigned long vma_end;
1330 unsigned long slot_addr;
1332 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1333 return;
1335 slot_addr = tsk->utask->xol_vaddr;
1336 if (unlikely(!slot_addr))
1337 return;
1339 area = tsk->mm->uprobes_state.xol_area;
1340 vma_end = area->vaddr + PAGE_SIZE;
1341 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1342 unsigned long offset;
1343 int slot_nr;
1345 offset = slot_addr - area->vaddr;
1346 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1347 if (slot_nr >= UINSNS_PER_PAGE)
1348 return;
1350 clear_bit(slot_nr, area->bitmap);
1351 atomic_dec(&area->slot_count);
1352 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1353 if (waitqueue_active(&area->wq))
1354 wake_up(&area->wq);
1356 tsk->utask->xol_vaddr = 0;
1360 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1361 void *src, unsigned long len)
1363 /* Initialize the slot */
1364 copy_to_page(page, vaddr, src, len);
1367 * We probably need flush_icache_user_range() but it needs vma.
1368 * This should work on most of architectures by default. If
1369 * architecture needs to do something different it can define
1370 * its own version of the function.
1372 flush_dcache_page(page);
1376 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1377 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1378 * instruction.
1379 * Return the address of the breakpoint instruction.
1381 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1383 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1386 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1388 struct uprobe_task *utask = current->utask;
1390 if (unlikely(utask && utask->active_uprobe))
1391 return utask->vaddr;
1393 return instruction_pointer(regs);
1396 static struct return_instance *free_ret_instance(struct return_instance *ri)
1398 struct return_instance *next = ri->next;
1399 put_uprobe(ri->uprobe);
1400 kfree(ri);
1401 return next;
1405 * Called with no locks held.
1406 * Called in context of a exiting or a exec-ing thread.
1408 void uprobe_free_utask(struct task_struct *t)
1410 struct uprobe_task *utask = t->utask;
1411 struct return_instance *ri;
1413 if (!utask)
1414 return;
1416 if (utask->active_uprobe)
1417 put_uprobe(utask->active_uprobe);
1419 ri = utask->return_instances;
1420 while (ri)
1421 ri = free_ret_instance(ri);
1423 xol_free_insn_slot(t);
1424 kfree(utask);
1425 t->utask = NULL;
1429 * Allocate a uprobe_task object for the task if if necessary.
1430 * Called when the thread hits a breakpoint.
1432 * Returns:
1433 * - pointer to new uprobe_task on success
1434 * - NULL otherwise
1436 static struct uprobe_task *get_utask(void)
1438 if (!current->utask)
1439 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1440 return current->utask;
1443 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1445 struct uprobe_task *n_utask;
1446 struct return_instance **p, *o, *n;
1448 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1449 if (!n_utask)
1450 return -ENOMEM;
1451 t->utask = n_utask;
1453 p = &n_utask->return_instances;
1454 for (o = o_utask->return_instances; o; o = o->next) {
1455 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1456 if (!n)
1457 return -ENOMEM;
1459 *n = *o;
1460 get_uprobe(n->uprobe);
1461 n->next = NULL;
1463 *p = n;
1464 p = &n->next;
1465 n_utask->depth++;
1468 return 0;
1471 static void uprobe_warn(struct task_struct *t, const char *msg)
1473 pr_warn("uprobe: %s:%d failed to %s\n",
1474 current->comm, current->pid, msg);
1477 static void dup_xol_work(struct callback_head *work)
1479 if (current->flags & PF_EXITING)
1480 return;
1482 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1483 !fatal_signal_pending(current))
1484 uprobe_warn(current, "dup xol area");
1488 * Called in context of a new clone/fork from copy_process.
1490 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1492 struct uprobe_task *utask = current->utask;
1493 struct mm_struct *mm = current->mm;
1494 struct xol_area *area;
1496 t->utask = NULL;
1498 if (!utask || !utask->return_instances)
1499 return;
1501 if (mm == t->mm && !(flags & CLONE_VFORK))
1502 return;
1504 if (dup_utask(t, utask))
1505 return uprobe_warn(t, "dup ret instances");
1507 /* The task can fork() after dup_xol_work() fails */
1508 area = mm->uprobes_state.xol_area;
1509 if (!area)
1510 return uprobe_warn(t, "dup xol area");
1512 if (mm == t->mm)
1513 return;
1515 t->utask->dup_xol_addr = area->vaddr;
1516 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1517 task_work_add(t, &t->utask->dup_xol_work, true);
1521 * Current area->vaddr notion assume the trampoline address is always
1522 * equal area->vaddr.
1524 * Returns -1 in case the xol_area is not allocated.
1526 static unsigned long get_trampoline_vaddr(void)
1528 struct xol_area *area;
1529 unsigned long trampoline_vaddr = -1;
1531 area = current->mm->uprobes_state.xol_area;
1532 smp_read_barrier_depends();
1533 if (area)
1534 trampoline_vaddr = area->vaddr;
1536 return trampoline_vaddr;
1539 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1540 struct pt_regs *regs)
1542 struct return_instance *ri = utask->return_instances;
1543 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1545 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1546 ri = free_ret_instance(ri);
1547 utask->depth--;
1549 utask->return_instances = ri;
1552 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1554 struct return_instance *ri;
1555 struct uprobe_task *utask;
1556 unsigned long orig_ret_vaddr, trampoline_vaddr;
1557 bool chained;
1559 if (!get_xol_area())
1560 return;
1562 utask = get_utask();
1563 if (!utask)
1564 return;
1566 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1567 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1568 " nestedness limit pid/tgid=%d/%d\n",
1569 current->pid, current->tgid);
1570 return;
1573 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1574 if (!ri)
1575 return;
1577 trampoline_vaddr = get_trampoline_vaddr();
1578 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1579 if (orig_ret_vaddr == -1)
1580 goto fail;
1582 /* drop the entries invalidated by longjmp() */
1583 chained = (orig_ret_vaddr == trampoline_vaddr);
1584 cleanup_return_instances(utask, chained, regs);
1587 * We don't want to keep trampoline address in stack, rather keep the
1588 * original return address of first caller thru all the consequent
1589 * instances. This also makes breakpoint unwrapping easier.
1591 if (chained) {
1592 if (!utask->return_instances) {
1594 * This situation is not possible. Likely we have an
1595 * attack from user-space.
1597 uprobe_warn(current, "handle tail call");
1598 goto fail;
1600 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1603 ri->uprobe = get_uprobe(uprobe);
1604 ri->func = instruction_pointer(regs);
1605 ri->stack = user_stack_pointer(regs);
1606 ri->orig_ret_vaddr = orig_ret_vaddr;
1607 ri->chained = chained;
1609 utask->depth++;
1610 ri->next = utask->return_instances;
1611 utask->return_instances = ri;
1613 return;
1614 fail:
1615 kfree(ri);
1618 /* Prepare to single-step probed instruction out of line. */
1619 static int
1620 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1622 struct uprobe_task *utask;
1623 unsigned long xol_vaddr;
1624 int err;
1626 utask = get_utask();
1627 if (!utask)
1628 return -ENOMEM;
1630 xol_vaddr = xol_get_insn_slot(uprobe);
1631 if (!xol_vaddr)
1632 return -ENOMEM;
1634 utask->xol_vaddr = xol_vaddr;
1635 utask->vaddr = bp_vaddr;
1637 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1638 if (unlikely(err)) {
1639 xol_free_insn_slot(current);
1640 return err;
1643 utask->active_uprobe = uprobe;
1644 utask->state = UTASK_SSTEP;
1645 return 0;
1649 * If we are singlestepping, then ensure this thread is not connected to
1650 * non-fatal signals until completion of singlestep. When xol insn itself
1651 * triggers the signal, restart the original insn even if the task is
1652 * already SIGKILL'ed (since coredump should report the correct ip). This
1653 * is even more important if the task has a handler for SIGSEGV/etc, The
1654 * _same_ instruction should be repeated again after return from the signal
1655 * handler, and SSTEP can never finish in this case.
1657 bool uprobe_deny_signal(void)
1659 struct task_struct *t = current;
1660 struct uprobe_task *utask = t->utask;
1662 if (likely(!utask || !utask->active_uprobe))
1663 return false;
1665 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1667 if (signal_pending(t)) {
1668 spin_lock_irq(&t->sighand->siglock);
1669 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1670 spin_unlock_irq(&t->sighand->siglock);
1672 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1673 utask->state = UTASK_SSTEP_TRAPPED;
1674 set_tsk_thread_flag(t, TIF_UPROBE);
1678 return true;
1681 static void mmf_recalc_uprobes(struct mm_struct *mm)
1683 struct vm_area_struct *vma;
1685 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1686 if (!valid_vma(vma, false))
1687 continue;
1689 * This is not strictly accurate, we can race with
1690 * uprobe_unregister() and see the already removed
1691 * uprobe if delete_uprobe() was not yet called.
1692 * Or this uprobe can be filtered out.
1694 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1695 return;
1698 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1701 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1703 struct page *page;
1704 uprobe_opcode_t opcode;
1705 int result;
1707 pagefault_disable();
1708 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
1709 pagefault_enable();
1711 if (likely(result == 0))
1712 goto out;
1715 * The NULL 'tsk' here ensures that any faults that occur here
1716 * will not be accounted to the task. 'mm' *is* current->mm,
1717 * but we treat this as a 'remote' access since it is
1718 * essentially a kernel access to the memory.
1720 result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
1721 NULL, NULL);
1722 if (result < 0)
1723 return result;
1725 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1726 put_page(page);
1727 out:
1728 /* This needs to return true for any variant of the trap insn */
1729 return is_trap_insn(&opcode);
1732 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1734 struct mm_struct *mm = current->mm;
1735 struct uprobe *uprobe = NULL;
1736 struct vm_area_struct *vma;
1738 down_read(&mm->mmap_sem);
1739 vma = find_vma(mm, bp_vaddr);
1740 if (vma && vma->vm_start <= bp_vaddr) {
1741 if (valid_vma(vma, false)) {
1742 struct inode *inode = file_inode(vma->vm_file);
1743 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1745 uprobe = find_uprobe(inode, offset);
1748 if (!uprobe)
1749 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1750 } else {
1751 *is_swbp = -EFAULT;
1754 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1755 mmf_recalc_uprobes(mm);
1756 up_read(&mm->mmap_sem);
1758 return uprobe;
1761 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1763 struct uprobe_consumer *uc;
1764 int remove = UPROBE_HANDLER_REMOVE;
1765 bool need_prep = false; /* prepare return uprobe, when needed */
1767 down_read(&uprobe->register_rwsem);
1768 for (uc = uprobe->consumers; uc; uc = uc->next) {
1769 int rc = 0;
1771 if (uc->handler) {
1772 rc = uc->handler(uc, regs);
1773 WARN(rc & ~UPROBE_HANDLER_MASK,
1774 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1777 if (uc->ret_handler)
1778 need_prep = true;
1780 remove &= rc;
1783 if (need_prep && !remove)
1784 prepare_uretprobe(uprobe, regs); /* put bp at return */
1786 if (remove && uprobe->consumers) {
1787 WARN_ON(!uprobe_is_active(uprobe));
1788 unapply_uprobe(uprobe, current->mm);
1790 up_read(&uprobe->register_rwsem);
1793 static void
1794 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1796 struct uprobe *uprobe = ri->uprobe;
1797 struct uprobe_consumer *uc;
1799 down_read(&uprobe->register_rwsem);
1800 for (uc = uprobe->consumers; uc; uc = uc->next) {
1801 if (uc->ret_handler)
1802 uc->ret_handler(uc, ri->func, regs);
1804 up_read(&uprobe->register_rwsem);
1807 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
1809 bool chained;
1811 do {
1812 chained = ri->chained;
1813 ri = ri->next; /* can't be NULL if chained */
1814 } while (chained);
1816 return ri;
1819 static void handle_trampoline(struct pt_regs *regs)
1821 struct uprobe_task *utask;
1822 struct return_instance *ri, *next;
1823 bool valid;
1825 utask = current->utask;
1826 if (!utask)
1827 goto sigill;
1829 ri = utask->return_instances;
1830 if (!ri)
1831 goto sigill;
1833 do {
1835 * We should throw out the frames invalidated by longjmp().
1836 * If this chain is valid, then the next one should be alive
1837 * or NULL; the latter case means that nobody but ri->func
1838 * could hit this trampoline on return. TODO: sigaltstack().
1840 next = find_next_ret_chain(ri);
1841 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
1843 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1844 do {
1845 if (valid)
1846 handle_uretprobe_chain(ri, regs);
1847 ri = free_ret_instance(ri);
1848 utask->depth--;
1849 } while (ri != next);
1850 } while (!valid);
1852 utask->return_instances = ri;
1853 return;
1855 sigill:
1856 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
1857 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1861 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
1863 return false;
1866 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
1867 struct pt_regs *regs)
1869 return true;
1873 * Run handler and ask thread to singlestep.
1874 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1876 static void handle_swbp(struct pt_regs *regs)
1878 struct uprobe *uprobe;
1879 unsigned long bp_vaddr;
1880 int uninitialized_var(is_swbp);
1882 bp_vaddr = uprobe_get_swbp_addr(regs);
1883 if (bp_vaddr == get_trampoline_vaddr())
1884 return handle_trampoline(regs);
1886 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1887 if (!uprobe) {
1888 if (is_swbp > 0) {
1889 /* No matching uprobe; signal SIGTRAP. */
1890 send_sig(SIGTRAP, current, 0);
1891 } else {
1893 * Either we raced with uprobe_unregister() or we can't
1894 * access this memory. The latter is only possible if
1895 * another thread plays with our ->mm. In both cases
1896 * we can simply restart. If this vma was unmapped we
1897 * can pretend this insn was not executed yet and get
1898 * the (correct) SIGSEGV after restart.
1900 instruction_pointer_set(regs, bp_vaddr);
1902 return;
1905 /* change it in advance for ->handler() and restart */
1906 instruction_pointer_set(regs, bp_vaddr);
1909 * TODO: move copy_insn/etc into _register and remove this hack.
1910 * After we hit the bp, _unregister + _register can install the
1911 * new and not-yet-analyzed uprobe at the same address, restart.
1913 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1914 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1915 goto out;
1917 /* Tracing handlers use ->utask to communicate with fetch methods */
1918 if (!get_utask())
1919 goto out;
1921 if (arch_uprobe_ignore(&uprobe->arch, regs))
1922 goto out;
1924 handler_chain(uprobe, regs);
1926 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1927 goto out;
1929 if (!pre_ssout(uprobe, regs, bp_vaddr))
1930 return;
1932 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1933 out:
1934 put_uprobe(uprobe);
1938 * Perform required fix-ups and disable singlestep.
1939 * Allow pending signals to take effect.
1941 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1943 struct uprobe *uprobe;
1944 int err = 0;
1946 uprobe = utask->active_uprobe;
1947 if (utask->state == UTASK_SSTEP_ACK)
1948 err = arch_uprobe_post_xol(&uprobe->arch, regs);
1949 else if (utask->state == UTASK_SSTEP_TRAPPED)
1950 arch_uprobe_abort_xol(&uprobe->arch, regs);
1951 else
1952 WARN_ON_ONCE(1);
1954 put_uprobe(uprobe);
1955 utask->active_uprobe = NULL;
1956 utask->state = UTASK_RUNNING;
1957 xol_free_insn_slot(current);
1959 spin_lock_irq(&current->sighand->siglock);
1960 recalc_sigpending(); /* see uprobe_deny_signal() */
1961 spin_unlock_irq(&current->sighand->siglock);
1963 if (unlikely(err)) {
1964 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
1965 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1970 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1971 * allows the thread to return from interrupt. After that handle_swbp()
1972 * sets utask->active_uprobe.
1974 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1975 * and allows the thread to return from interrupt.
1977 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1978 * uprobe_notify_resume().
1980 void uprobe_notify_resume(struct pt_regs *regs)
1982 struct uprobe_task *utask;
1984 clear_thread_flag(TIF_UPROBE);
1986 utask = current->utask;
1987 if (utask && utask->active_uprobe)
1988 handle_singlestep(utask, regs);
1989 else
1990 handle_swbp(regs);
1994 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1995 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1997 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1999 if (!current->mm)
2000 return 0;
2002 if (!test_bit(MMF_HAS_UPROBES, &current->mm->flags) &&
2003 (!current->utask || !current->utask->return_instances))
2004 return 0;
2006 set_thread_flag(TIF_UPROBE);
2007 return 1;
2011 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2012 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2014 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2016 struct uprobe_task *utask = current->utask;
2018 if (!current->mm || !utask || !utask->active_uprobe)
2019 /* task is currently not uprobed */
2020 return 0;
2022 utask->state = UTASK_SSTEP_ACK;
2023 set_thread_flag(TIF_UPROBE);
2024 return 1;
2027 static struct notifier_block uprobe_exception_nb = {
2028 .notifier_call = arch_uprobe_exception_notify,
2029 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2032 static int __init init_uprobes(void)
2034 int i;
2036 for (i = 0; i < UPROBES_HASH_SZ; i++)
2037 mutex_init(&uprobes_mmap_mutex[i]);
2039 if (percpu_init_rwsem(&dup_mmap_sem))
2040 return -ENOMEM;
2042 return register_die_notifier(&uprobe_exception_nb);
2044 __initcall(init_uprobes);