Loading...
Note: File does not exist in v3.1.
1/*
2 * User-space Probes (UProbes)
3 *
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.
8 *
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.
13 *
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.
17 *
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
23 */
24
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/export.h>
31#include <linux/rmap.h> /* anon_vma_prepare */
32#include <linux/mmu_notifier.h> /* set_pte_at_notify */
33#include <linux/swap.h> /* try_to_free_swap */
34#include <linux/ptrace.h> /* user_enable_single_step */
35#include <linux/kdebug.h> /* notifier mechanism */
36#include "../../mm/internal.h" /* munlock_vma_page */
37#include <linux/percpu-rwsem.h>
38#include <linux/task_work.h>
39#include <linux/shmem_fs.h>
40
41#include <linux/uprobes.h>
42
43#define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
44#define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
45
46static struct rb_root uprobes_tree = RB_ROOT;
47/*
48 * allows us to skip the uprobe_mmap if there are no uprobe events active
49 * at this time. Probably a fine grained per inode count is better?
50 */
51#define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
52
53static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
54
55#define UPROBES_HASH_SZ 13
56/* serialize uprobe->pending_list */
57static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
58#define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
59
60static struct percpu_rw_semaphore dup_mmap_sem;
61
62/* Have a copy of original instruction */
63#define UPROBE_COPY_INSN 0
64
65struct uprobe {
66 struct rb_node rb_node; /* node in the rb tree */
67 atomic_t ref;
68 struct rw_semaphore register_rwsem;
69 struct rw_semaphore consumer_rwsem;
70 struct list_head pending_list;
71 struct uprobe_consumer *consumers;
72 struct inode *inode; /* Also hold a ref to inode */
73 loff_t offset;
74 unsigned long flags;
75
76 /*
77 * The generic code assumes that it has two members of unknown type
78 * owned by the arch-specific code:
79 *
80 * insn - copy_insn() saves the original instruction here for
81 * arch_uprobe_analyze_insn().
82 *
83 * ixol - potentially modified instruction to execute out of
84 * line, copied to xol_area by xol_get_insn_slot().
85 */
86 struct arch_uprobe arch;
87};
88
89/*
90 * Execute out of line area: anonymous executable mapping installed
91 * by the probed task to execute the copy of the original instruction
92 * mangled by set_swbp().
93 *
94 * On a breakpoint hit, thread contests for a slot. It frees the
95 * slot after singlestep. Currently a fixed number of slots are
96 * allocated.
97 */
98struct xol_area {
99 wait_queue_head_t wq; /* if all slots are busy */
100 atomic_t slot_count; /* number of in-use slots */
101 unsigned long *bitmap; /* 0 = free slot */
102
103 struct vm_special_mapping xol_mapping;
104 struct page *pages[2];
105 /*
106 * We keep the vma's vm_start rather than a pointer to the vma
107 * itself. The probed process or a naughty kernel module could make
108 * the vma go away, and we must handle that reasonably gracefully.
109 */
110 unsigned long vaddr; /* Page(s) of instruction slots */
111};
112
113/*
114 * valid_vma: Verify if the specified vma is an executable vma
115 * Relax restrictions while unregistering: vm_flags might have
116 * changed after breakpoint was inserted.
117 * - is_register: indicates if we are in register context.
118 * - Return 1 if the specified virtual address is in an
119 * executable vma.
120 */
121static bool valid_vma(struct vm_area_struct *vma, bool is_register)
122{
123 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
124
125 if (is_register)
126 flags |= VM_WRITE;
127
128 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
129}
130
131static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
132{
133 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
134}
135
136static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
137{
138 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
139}
140
141/**
142 * __replace_page - replace page in vma by new page.
143 * based on replace_page in mm/ksm.c
144 *
145 * @vma: vma that holds the pte pointing to page
146 * @addr: address the old @page is mapped at
147 * @page: the cowed page we are replacing by kpage
148 * @kpage: the modified page we replace page by
149 *
150 * Returns 0 on success, -EFAULT on failure.
151 */
152static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
153 struct page *page, struct page *kpage)
154{
155 struct mm_struct *mm = vma->vm_mm;
156 spinlock_t *ptl;
157 pte_t *ptep;
158 int err;
159 /* For mmu_notifiers */
160 const unsigned long mmun_start = addr;
161 const unsigned long mmun_end = addr + PAGE_SIZE;
162 struct mem_cgroup *memcg;
163
164 err = mem_cgroup_try_charge(kpage, vma->vm_mm, GFP_KERNEL, &memcg,
165 false);
166 if (err)
167 return err;
168
169 /* For try_to_free_swap() and munlock_vma_page() below */
170 lock_page(page);
171
172 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
173 err = -EAGAIN;
174 ptep = page_check_address(page, mm, addr, &ptl, 0);
175 if (!ptep)
176 goto unlock;
177
178 get_page(kpage);
179 page_add_new_anon_rmap(kpage, vma, addr, false);
180 mem_cgroup_commit_charge(kpage, memcg, false, false);
181 lru_cache_add_active_or_unevictable(kpage, vma);
182
183 if (!PageAnon(page)) {
184 dec_mm_counter(mm, mm_counter_file(page));
185 inc_mm_counter(mm, MM_ANONPAGES);
186 }
187
188 flush_cache_page(vma, addr, pte_pfn(*ptep));
189 ptep_clear_flush_notify(vma, addr, ptep);
190 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
191
192 page_remove_rmap(page, false);
193 if (!page_mapped(page))
194 try_to_free_swap(page);
195 pte_unmap_unlock(ptep, ptl);
196
197 if (vma->vm_flags & VM_LOCKED)
198 munlock_vma_page(page);
199 put_page(page);
200
201 err = 0;
202 unlock:
203 mem_cgroup_cancel_charge(kpage, memcg, false);
204 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
205 unlock_page(page);
206 return err;
207}
208
209/**
210 * is_swbp_insn - check if instruction is breakpoint instruction.
211 * @insn: instruction to be checked.
212 * Default implementation of is_swbp_insn
213 * Returns true if @insn is a breakpoint instruction.
214 */
215bool __weak is_swbp_insn(uprobe_opcode_t *insn)
216{
217 return *insn == UPROBE_SWBP_INSN;
218}
219
220/**
221 * is_trap_insn - check if instruction is breakpoint instruction.
222 * @insn: instruction to be checked.
223 * Default implementation of is_trap_insn
224 * Returns true if @insn is a breakpoint instruction.
225 *
226 * This function is needed for the case where an architecture has multiple
227 * trap instructions (like powerpc).
228 */
229bool __weak is_trap_insn(uprobe_opcode_t *insn)
230{
231 return is_swbp_insn(insn);
232}
233
234static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
235{
236 void *kaddr = kmap_atomic(page);
237 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
238 kunmap_atomic(kaddr);
239}
240
241static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
242{
243 void *kaddr = kmap_atomic(page);
244 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
245 kunmap_atomic(kaddr);
246}
247
248static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
249{
250 uprobe_opcode_t old_opcode;
251 bool is_swbp;
252
253 /*
254 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
255 * We do not check if it is any other 'trap variant' which could
256 * be conditional trap instruction such as the one powerpc supports.
257 *
258 * The logic is that we do not care if the underlying instruction
259 * is a trap variant; uprobes always wins over any other (gdb)
260 * breakpoint.
261 */
262 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
263 is_swbp = is_swbp_insn(&old_opcode);
264
265 if (is_swbp_insn(new_opcode)) {
266 if (is_swbp) /* register: already installed? */
267 return 0;
268 } else {
269 if (!is_swbp) /* unregister: was it changed by us? */
270 return 0;
271 }
272
273 return 1;
274}
275
276/*
277 * NOTE:
278 * Expect the breakpoint instruction to be the smallest size instruction for
279 * the architecture. If an arch has variable length instruction and the
280 * breakpoint instruction is not of the smallest length instruction
281 * supported by that architecture then we need to modify is_trap_at_addr and
282 * uprobe_write_opcode accordingly. This would never be a problem for archs
283 * that have fixed length instructions.
284 *
285 * uprobe_write_opcode - write the opcode at a given virtual address.
286 * @mm: the probed process address space.
287 * @vaddr: the virtual address to store the opcode.
288 * @opcode: opcode to be written at @vaddr.
289 *
290 * Called with mm->mmap_sem held for write.
291 * Return 0 (success) or a negative errno.
292 */
293int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr,
294 uprobe_opcode_t opcode)
295{
296 struct page *old_page, *new_page;
297 struct vm_area_struct *vma;
298 int ret;
299
300retry:
301 /* Read the page with vaddr into memory */
302 ret = get_user_pages_remote(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
303 if (ret <= 0)
304 return ret;
305
306 ret = verify_opcode(old_page, vaddr, &opcode);
307 if (ret <= 0)
308 goto put_old;
309
310 ret = anon_vma_prepare(vma);
311 if (ret)
312 goto put_old;
313
314 ret = -ENOMEM;
315 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
316 if (!new_page)
317 goto put_old;
318
319 __SetPageUptodate(new_page);
320 copy_highpage(new_page, old_page);
321 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
322
323 ret = __replace_page(vma, vaddr, old_page, new_page);
324 put_page(new_page);
325put_old:
326 put_page(old_page);
327
328 if (unlikely(ret == -EAGAIN))
329 goto retry;
330 return ret;
331}
332
333/**
334 * set_swbp - store breakpoint at a given address.
335 * @auprobe: arch specific probepoint information.
336 * @mm: the probed process address space.
337 * @vaddr: the virtual address to insert the opcode.
338 *
339 * For mm @mm, store the breakpoint instruction at @vaddr.
340 * Return 0 (success) or a negative errno.
341 */
342int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
343{
344 return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
345}
346
347/**
348 * set_orig_insn - Restore the original instruction.
349 * @mm: the probed process address space.
350 * @auprobe: arch specific probepoint information.
351 * @vaddr: the virtual address to insert the opcode.
352 *
353 * For mm @mm, restore the original opcode (opcode) at @vaddr.
354 * Return 0 (success) or a negative errno.
355 */
356int __weak
357set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
358{
359 return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn);
360}
361
362static struct uprobe *get_uprobe(struct uprobe *uprobe)
363{
364 atomic_inc(&uprobe->ref);
365 return uprobe;
366}
367
368static void put_uprobe(struct uprobe *uprobe)
369{
370 if (atomic_dec_and_test(&uprobe->ref))
371 kfree(uprobe);
372}
373
374static int match_uprobe(struct uprobe *l, struct uprobe *r)
375{
376 if (l->inode < r->inode)
377 return -1;
378
379 if (l->inode > r->inode)
380 return 1;
381
382 if (l->offset < r->offset)
383 return -1;
384
385 if (l->offset > r->offset)
386 return 1;
387
388 return 0;
389}
390
391static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
392{
393 struct uprobe u = { .inode = inode, .offset = offset };
394 struct rb_node *n = uprobes_tree.rb_node;
395 struct uprobe *uprobe;
396 int match;
397
398 while (n) {
399 uprobe = rb_entry(n, struct uprobe, rb_node);
400 match = match_uprobe(&u, uprobe);
401 if (!match)
402 return get_uprobe(uprobe);
403
404 if (match < 0)
405 n = n->rb_left;
406 else
407 n = n->rb_right;
408 }
409 return NULL;
410}
411
412/*
413 * Find a uprobe corresponding to a given inode:offset
414 * Acquires uprobes_treelock
415 */
416static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
417{
418 struct uprobe *uprobe;
419
420 spin_lock(&uprobes_treelock);
421 uprobe = __find_uprobe(inode, offset);
422 spin_unlock(&uprobes_treelock);
423
424 return uprobe;
425}
426
427static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
428{
429 struct rb_node **p = &uprobes_tree.rb_node;
430 struct rb_node *parent = NULL;
431 struct uprobe *u;
432 int match;
433
434 while (*p) {
435 parent = *p;
436 u = rb_entry(parent, struct uprobe, rb_node);
437 match = match_uprobe(uprobe, u);
438 if (!match)
439 return get_uprobe(u);
440
441 if (match < 0)
442 p = &parent->rb_left;
443 else
444 p = &parent->rb_right;
445
446 }
447
448 u = NULL;
449 rb_link_node(&uprobe->rb_node, parent, p);
450 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
451 /* get access + creation ref */
452 atomic_set(&uprobe->ref, 2);
453
454 return u;
455}
456
457/*
458 * Acquire uprobes_treelock.
459 * Matching uprobe already exists in rbtree;
460 * increment (access refcount) and return the matching uprobe.
461 *
462 * No matching uprobe; insert the uprobe in rb_tree;
463 * get a double refcount (access + creation) and return NULL.
464 */
465static struct uprobe *insert_uprobe(struct uprobe *uprobe)
466{
467 struct uprobe *u;
468
469 spin_lock(&uprobes_treelock);
470 u = __insert_uprobe(uprobe);
471 spin_unlock(&uprobes_treelock);
472
473 return u;
474}
475
476static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
477{
478 struct uprobe *uprobe, *cur_uprobe;
479
480 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
481 if (!uprobe)
482 return NULL;
483
484 uprobe->inode = igrab(inode);
485 uprobe->offset = offset;
486 init_rwsem(&uprobe->register_rwsem);
487 init_rwsem(&uprobe->consumer_rwsem);
488
489 /* add to uprobes_tree, sorted on inode:offset */
490 cur_uprobe = insert_uprobe(uprobe);
491 /* a uprobe exists for this inode:offset combination */
492 if (cur_uprobe) {
493 kfree(uprobe);
494 uprobe = cur_uprobe;
495 iput(inode);
496 }
497
498 return uprobe;
499}
500
501static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
502{
503 down_write(&uprobe->consumer_rwsem);
504 uc->next = uprobe->consumers;
505 uprobe->consumers = uc;
506 up_write(&uprobe->consumer_rwsem);
507}
508
509/*
510 * For uprobe @uprobe, delete the consumer @uc.
511 * Return true if the @uc is deleted successfully
512 * or return false.
513 */
514static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
515{
516 struct uprobe_consumer **con;
517 bool ret = false;
518
519 down_write(&uprobe->consumer_rwsem);
520 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
521 if (*con == uc) {
522 *con = uc->next;
523 ret = true;
524 break;
525 }
526 }
527 up_write(&uprobe->consumer_rwsem);
528
529 return ret;
530}
531
532static int __copy_insn(struct address_space *mapping, struct file *filp,
533 void *insn, int nbytes, loff_t offset)
534{
535 struct page *page;
536 /*
537 * Ensure that the page that has the original instruction is populated
538 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
539 * see uprobe_register().
540 */
541 if (mapping->a_ops->readpage)
542 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
543 else
544 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
545 if (IS_ERR(page))
546 return PTR_ERR(page);
547
548 copy_from_page(page, offset, insn, nbytes);
549 put_page(page);
550
551 return 0;
552}
553
554static int copy_insn(struct uprobe *uprobe, struct file *filp)
555{
556 struct address_space *mapping = uprobe->inode->i_mapping;
557 loff_t offs = uprobe->offset;
558 void *insn = &uprobe->arch.insn;
559 int size = sizeof(uprobe->arch.insn);
560 int len, err = -EIO;
561
562 /* Copy only available bytes, -EIO if nothing was read */
563 do {
564 if (offs >= i_size_read(uprobe->inode))
565 break;
566
567 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
568 err = __copy_insn(mapping, filp, insn, len, offs);
569 if (err)
570 break;
571
572 insn += len;
573 offs += len;
574 size -= len;
575 } while (size);
576
577 return err;
578}
579
580static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
581 struct mm_struct *mm, unsigned long vaddr)
582{
583 int ret = 0;
584
585 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
586 return ret;
587
588 /* TODO: move this into _register, until then we abuse this sem. */
589 down_write(&uprobe->consumer_rwsem);
590 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
591 goto out;
592
593 ret = copy_insn(uprobe, file);
594 if (ret)
595 goto out;
596
597 ret = -ENOTSUPP;
598 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
599 goto out;
600
601 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
602 if (ret)
603 goto out;
604
605 /* uprobe_write_opcode() assumes we don't cross page boundary */
606 BUG_ON((uprobe->offset & ~PAGE_MASK) +
607 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
608
609 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
610 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
611
612 out:
613 up_write(&uprobe->consumer_rwsem);
614
615 return ret;
616}
617
618static inline bool consumer_filter(struct uprobe_consumer *uc,
619 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
620{
621 return !uc->filter || uc->filter(uc, ctx, mm);
622}
623
624static bool filter_chain(struct uprobe *uprobe,
625 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
626{
627 struct uprobe_consumer *uc;
628 bool ret = false;
629
630 down_read(&uprobe->consumer_rwsem);
631 for (uc = uprobe->consumers; uc; uc = uc->next) {
632 ret = consumer_filter(uc, ctx, mm);
633 if (ret)
634 break;
635 }
636 up_read(&uprobe->consumer_rwsem);
637
638 return ret;
639}
640
641static int
642install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
643 struct vm_area_struct *vma, unsigned long vaddr)
644{
645 bool first_uprobe;
646 int ret;
647
648 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
649 if (ret)
650 return ret;
651
652 /*
653 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
654 * the task can hit this breakpoint right after __replace_page().
655 */
656 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
657 if (first_uprobe)
658 set_bit(MMF_HAS_UPROBES, &mm->flags);
659
660 ret = set_swbp(&uprobe->arch, mm, vaddr);
661 if (!ret)
662 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
663 else if (first_uprobe)
664 clear_bit(MMF_HAS_UPROBES, &mm->flags);
665
666 return ret;
667}
668
669static int
670remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
671{
672 set_bit(MMF_RECALC_UPROBES, &mm->flags);
673 return set_orig_insn(&uprobe->arch, mm, vaddr);
674}
675
676static inline bool uprobe_is_active(struct uprobe *uprobe)
677{
678 return !RB_EMPTY_NODE(&uprobe->rb_node);
679}
680/*
681 * There could be threads that have already hit the breakpoint. They
682 * will recheck the current insn and restart if find_uprobe() fails.
683 * See find_active_uprobe().
684 */
685static void delete_uprobe(struct uprobe *uprobe)
686{
687 if (WARN_ON(!uprobe_is_active(uprobe)))
688 return;
689
690 spin_lock(&uprobes_treelock);
691 rb_erase(&uprobe->rb_node, &uprobes_tree);
692 spin_unlock(&uprobes_treelock);
693 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
694 iput(uprobe->inode);
695 put_uprobe(uprobe);
696}
697
698struct map_info {
699 struct map_info *next;
700 struct mm_struct *mm;
701 unsigned long vaddr;
702};
703
704static inline struct map_info *free_map_info(struct map_info *info)
705{
706 struct map_info *next = info->next;
707 kfree(info);
708 return next;
709}
710
711static struct map_info *
712build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
713{
714 unsigned long pgoff = offset >> PAGE_SHIFT;
715 struct vm_area_struct *vma;
716 struct map_info *curr = NULL;
717 struct map_info *prev = NULL;
718 struct map_info *info;
719 int more = 0;
720
721 again:
722 i_mmap_lock_read(mapping);
723 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
724 if (!valid_vma(vma, is_register))
725 continue;
726
727 if (!prev && !more) {
728 /*
729 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
730 * reclaim. This is optimistic, no harm done if it fails.
731 */
732 prev = kmalloc(sizeof(struct map_info),
733 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
734 if (prev)
735 prev->next = NULL;
736 }
737 if (!prev) {
738 more++;
739 continue;
740 }
741
742 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
743 continue;
744
745 info = prev;
746 prev = prev->next;
747 info->next = curr;
748 curr = info;
749
750 info->mm = vma->vm_mm;
751 info->vaddr = offset_to_vaddr(vma, offset);
752 }
753 i_mmap_unlock_read(mapping);
754
755 if (!more)
756 goto out;
757
758 prev = curr;
759 while (curr) {
760 mmput(curr->mm);
761 curr = curr->next;
762 }
763
764 do {
765 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
766 if (!info) {
767 curr = ERR_PTR(-ENOMEM);
768 goto out;
769 }
770 info->next = prev;
771 prev = info;
772 } while (--more);
773
774 goto again;
775 out:
776 while (prev)
777 prev = free_map_info(prev);
778 return curr;
779}
780
781static int
782register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
783{
784 bool is_register = !!new;
785 struct map_info *info;
786 int err = 0;
787
788 percpu_down_write(&dup_mmap_sem);
789 info = build_map_info(uprobe->inode->i_mapping,
790 uprobe->offset, is_register);
791 if (IS_ERR(info)) {
792 err = PTR_ERR(info);
793 goto out;
794 }
795
796 while (info) {
797 struct mm_struct *mm = info->mm;
798 struct vm_area_struct *vma;
799
800 if (err && is_register)
801 goto free;
802
803 down_write(&mm->mmap_sem);
804 vma = find_vma(mm, info->vaddr);
805 if (!vma || !valid_vma(vma, is_register) ||
806 file_inode(vma->vm_file) != uprobe->inode)
807 goto unlock;
808
809 if (vma->vm_start > info->vaddr ||
810 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
811 goto unlock;
812
813 if (is_register) {
814 /* consult only the "caller", new consumer. */
815 if (consumer_filter(new,
816 UPROBE_FILTER_REGISTER, mm))
817 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
818 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
819 if (!filter_chain(uprobe,
820 UPROBE_FILTER_UNREGISTER, mm))
821 err |= remove_breakpoint(uprobe, mm, info->vaddr);
822 }
823
824 unlock:
825 up_write(&mm->mmap_sem);
826 free:
827 mmput(mm);
828 info = free_map_info(info);
829 }
830 out:
831 percpu_up_write(&dup_mmap_sem);
832 return err;
833}
834
835static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
836{
837 consumer_add(uprobe, uc);
838 return register_for_each_vma(uprobe, uc);
839}
840
841static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
842{
843 int err;
844
845 if (WARN_ON(!consumer_del(uprobe, uc)))
846 return;
847
848 err = register_for_each_vma(uprobe, NULL);
849 /* TODO : cant unregister? schedule a worker thread */
850 if (!uprobe->consumers && !err)
851 delete_uprobe(uprobe);
852}
853
854/*
855 * uprobe_register - register a probe
856 * @inode: the file in which the probe has to be placed.
857 * @offset: offset from the start of the file.
858 * @uc: information on howto handle the probe..
859 *
860 * Apart from the access refcount, uprobe_register() takes a creation
861 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
862 * inserted into the rbtree (i.e first consumer for a @inode:@offset
863 * tuple). Creation refcount stops uprobe_unregister from freeing the
864 * @uprobe even before the register operation is complete. Creation
865 * refcount is released when the last @uc for the @uprobe
866 * unregisters.
867 *
868 * Return errno if it cannot successully install probes
869 * else return 0 (success)
870 */
871int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
872{
873 struct uprobe *uprobe;
874 int ret;
875
876 /* Uprobe must have at least one set consumer */
877 if (!uc->handler && !uc->ret_handler)
878 return -EINVAL;
879
880 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
881 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
882 return -EIO;
883 /* Racy, just to catch the obvious mistakes */
884 if (offset > i_size_read(inode))
885 return -EINVAL;
886
887 retry:
888 uprobe = alloc_uprobe(inode, offset);
889 if (!uprobe)
890 return -ENOMEM;
891 /*
892 * We can race with uprobe_unregister()->delete_uprobe().
893 * Check uprobe_is_active() and retry if it is false.
894 */
895 down_write(&uprobe->register_rwsem);
896 ret = -EAGAIN;
897 if (likely(uprobe_is_active(uprobe))) {
898 ret = __uprobe_register(uprobe, uc);
899 if (ret)
900 __uprobe_unregister(uprobe, uc);
901 }
902 up_write(&uprobe->register_rwsem);
903 put_uprobe(uprobe);
904
905 if (unlikely(ret == -EAGAIN))
906 goto retry;
907 return ret;
908}
909EXPORT_SYMBOL_GPL(uprobe_register);
910
911/*
912 * uprobe_apply - unregister a already registered probe.
913 * @inode: the file in which the probe has to be removed.
914 * @offset: offset from the start of the file.
915 * @uc: consumer which wants to add more or remove some breakpoints
916 * @add: add or remove the breakpoints
917 */
918int uprobe_apply(struct inode *inode, loff_t offset,
919 struct uprobe_consumer *uc, bool add)
920{
921 struct uprobe *uprobe;
922 struct uprobe_consumer *con;
923 int ret = -ENOENT;
924
925 uprobe = find_uprobe(inode, offset);
926 if (WARN_ON(!uprobe))
927 return ret;
928
929 down_write(&uprobe->register_rwsem);
930 for (con = uprobe->consumers; con && con != uc ; con = con->next)
931 ;
932 if (con)
933 ret = register_for_each_vma(uprobe, add ? uc : NULL);
934 up_write(&uprobe->register_rwsem);
935 put_uprobe(uprobe);
936
937 return ret;
938}
939
940/*
941 * uprobe_unregister - unregister a already registered probe.
942 * @inode: the file in which the probe has to be removed.
943 * @offset: offset from the start of the file.
944 * @uc: identify which probe if multiple probes are colocated.
945 */
946void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
947{
948 struct uprobe *uprobe;
949
950 uprobe = find_uprobe(inode, offset);
951 if (WARN_ON(!uprobe))
952 return;
953
954 down_write(&uprobe->register_rwsem);
955 __uprobe_unregister(uprobe, uc);
956 up_write(&uprobe->register_rwsem);
957 put_uprobe(uprobe);
958}
959EXPORT_SYMBOL_GPL(uprobe_unregister);
960
961static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
962{
963 struct vm_area_struct *vma;
964 int err = 0;
965
966 down_read(&mm->mmap_sem);
967 for (vma = mm->mmap; vma; vma = vma->vm_next) {
968 unsigned long vaddr;
969 loff_t offset;
970
971 if (!valid_vma(vma, false) ||
972 file_inode(vma->vm_file) != uprobe->inode)
973 continue;
974
975 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
976 if (uprobe->offset < offset ||
977 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
978 continue;
979
980 vaddr = offset_to_vaddr(vma, uprobe->offset);
981 err |= remove_breakpoint(uprobe, mm, vaddr);
982 }
983 up_read(&mm->mmap_sem);
984
985 return err;
986}
987
988static struct rb_node *
989find_node_in_range(struct inode *inode, loff_t min, loff_t max)
990{
991 struct rb_node *n = uprobes_tree.rb_node;
992
993 while (n) {
994 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
995
996 if (inode < u->inode) {
997 n = n->rb_left;
998 } else if (inode > u->inode) {
999 n = n->rb_right;
1000 } else {
1001 if (max < u->offset)
1002 n = n->rb_left;
1003 else if (min > u->offset)
1004 n = n->rb_right;
1005 else
1006 break;
1007 }
1008 }
1009
1010 return n;
1011}
1012
1013/*
1014 * For a given range in vma, build a list of probes that need to be inserted.
1015 */
1016static void build_probe_list(struct inode *inode,
1017 struct vm_area_struct *vma,
1018 unsigned long start, unsigned long end,
1019 struct list_head *head)
1020{
1021 loff_t min, max;
1022 struct rb_node *n, *t;
1023 struct uprobe *u;
1024
1025 INIT_LIST_HEAD(head);
1026 min = vaddr_to_offset(vma, start);
1027 max = min + (end - start) - 1;
1028
1029 spin_lock(&uprobes_treelock);
1030 n = find_node_in_range(inode, min, max);
1031 if (n) {
1032 for (t = n; t; t = rb_prev(t)) {
1033 u = rb_entry(t, struct uprobe, rb_node);
1034 if (u->inode != inode || u->offset < min)
1035 break;
1036 list_add(&u->pending_list, head);
1037 get_uprobe(u);
1038 }
1039 for (t = n; (t = rb_next(t)); ) {
1040 u = rb_entry(t, struct uprobe, rb_node);
1041 if (u->inode != inode || u->offset > max)
1042 break;
1043 list_add(&u->pending_list, head);
1044 get_uprobe(u);
1045 }
1046 }
1047 spin_unlock(&uprobes_treelock);
1048}
1049
1050/*
1051 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1052 *
1053 * Currently we ignore all errors and always return 0, the callers
1054 * can't handle the failure anyway.
1055 */
1056int uprobe_mmap(struct vm_area_struct *vma)
1057{
1058 struct list_head tmp_list;
1059 struct uprobe *uprobe, *u;
1060 struct inode *inode;
1061
1062 if (no_uprobe_events() || !valid_vma(vma, true))
1063 return 0;
1064
1065 inode = file_inode(vma->vm_file);
1066 if (!inode)
1067 return 0;
1068
1069 mutex_lock(uprobes_mmap_hash(inode));
1070 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1071 /*
1072 * We can race with uprobe_unregister(), this uprobe can be already
1073 * removed. But in this case filter_chain() must return false, all
1074 * consumers have gone away.
1075 */
1076 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1077 if (!fatal_signal_pending(current) &&
1078 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1079 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1080 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1081 }
1082 put_uprobe(uprobe);
1083 }
1084 mutex_unlock(uprobes_mmap_hash(inode));
1085
1086 return 0;
1087}
1088
1089static bool
1090vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1091{
1092 loff_t min, max;
1093 struct inode *inode;
1094 struct rb_node *n;
1095
1096 inode = file_inode(vma->vm_file);
1097
1098 min = vaddr_to_offset(vma, start);
1099 max = min + (end - start) - 1;
1100
1101 spin_lock(&uprobes_treelock);
1102 n = find_node_in_range(inode, min, max);
1103 spin_unlock(&uprobes_treelock);
1104
1105 return !!n;
1106}
1107
1108/*
1109 * Called in context of a munmap of a vma.
1110 */
1111void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1112{
1113 if (no_uprobe_events() || !valid_vma(vma, false))
1114 return;
1115
1116 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1117 return;
1118
1119 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1120 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1121 return;
1122
1123 if (vma_has_uprobes(vma, start, end))
1124 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1125}
1126
1127/* Slot allocation for XOL */
1128static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1129{
1130 struct vm_area_struct *vma;
1131 int ret;
1132
1133 down_write(&mm->mmap_sem);
1134 if (mm->uprobes_state.xol_area) {
1135 ret = -EALREADY;
1136 goto fail;
1137 }
1138
1139 if (!area->vaddr) {
1140 /* Try to map as high as possible, this is only a hint. */
1141 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1142 PAGE_SIZE, 0, 0);
1143 if (area->vaddr & ~PAGE_MASK) {
1144 ret = area->vaddr;
1145 goto fail;
1146 }
1147 }
1148
1149 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1150 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1151 &area->xol_mapping);
1152 if (IS_ERR(vma)) {
1153 ret = PTR_ERR(vma);
1154 goto fail;
1155 }
1156
1157 ret = 0;
1158 smp_wmb(); /* pairs with get_xol_area() */
1159 mm->uprobes_state.xol_area = area;
1160 fail:
1161 up_write(&mm->mmap_sem);
1162
1163 return ret;
1164}
1165
1166static struct xol_area *__create_xol_area(unsigned long vaddr)
1167{
1168 struct mm_struct *mm = current->mm;
1169 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1170 struct xol_area *area;
1171
1172 area = kmalloc(sizeof(*area), GFP_KERNEL);
1173 if (unlikely(!area))
1174 goto out;
1175
1176 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1177 if (!area->bitmap)
1178 goto free_area;
1179
1180 area->xol_mapping.name = "[uprobes]";
1181 area->xol_mapping.fault = NULL;
1182 area->xol_mapping.pages = area->pages;
1183 area->pages[0] = alloc_page(GFP_HIGHUSER);
1184 if (!area->pages[0])
1185 goto free_bitmap;
1186 area->pages[1] = NULL;
1187
1188 area->vaddr = vaddr;
1189 init_waitqueue_head(&area->wq);
1190 /* Reserve the 1st slot for get_trampoline_vaddr() */
1191 set_bit(0, area->bitmap);
1192 atomic_set(&area->slot_count, 1);
1193 copy_to_page(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1194
1195 if (!xol_add_vma(mm, area))
1196 return area;
1197
1198 __free_page(area->pages[0]);
1199 free_bitmap:
1200 kfree(area->bitmap);
1201 free_area:
1202 kfree(area);
1203 out:
1204 return NULL;
1205}
1206
1207/*
1208 * get_xol_area - Allocate process's xol_area if necessary.
1209 * This area will be used for storing instructions for execution out of line.
1210 *
1211 * Returns the allocated area or NULL.
1212 */
1213static struct xol_area *get_xol_area(void)
1214{
1215 struct mm_struct *mm = current->mm;
1216 struct xol_area *area;
1217
1218 if (!mm->uprobes_state.xol_area)
1219 __create_xol_area(0);
1220
1221 area = mm->uprobes_state.xol_area;
1222 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1223 return area;
1224}
1225
1226/*
1227 * uprobe_clear_state - Free the area allocated for slots.
1228 */
1229void uprobe_clear_state(struct mm_struct *mm)
1230{
1231 struct xol_area *area = mm->uprobes_state.xol_area;
1232
1233 if (!area)
1234 return;
1235
1236 put_page(area->pages[0]);
1237 kfree(area->bitmap);
1238 kfree(area);
1239}
1240
1241void uprobe_start_dup_mmap(void)
1242{
1243 percpu_down_read(&dup_mmap_sem);
1244}
1245
1246void uprobe_end_dup_mmap(void)
1247{
1248 percpu_up_read(&dup_mmap_sem);
1249}
1250
1251void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1252{
1253 newmm->uprobes_state.xol_area = NULL;
1254
1255 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1256 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1257 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1258 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1259 }
1260}
1261
1262/*
1263 * - search for a free slot.
1264 */
1265static unsigned long xol_take_insn_slot(struct xol_area *area)
1266{
1267 unsigned long slot_addr;
1268 int slot_nr;
1269
1270 do {
1271 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1272 if (slot_nr < UINSNS_PER_PAGE) {
1273 if (!test_and_set_bit(slot_nr, area->bitmap))
1274 break;
1275
1276 slot_nr = UINSNS_PER_PAGE;
1277 continue;
1278 }
1279 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1280 } while (slot_nr >= UINSNS_PER_PAGE);
1281
1282 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1283 atomic_inc(&area->slot_count);
1284
1285 return slot_addr;
1286}
1287
1288/*
1289 * xol_get_insn_slot - allocate a slot for xol.
1290 * Returns the allocated slot address or 0.
1291 */
1292static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1293{
1294 struct xol_area *area;
1295 unsigned long xol_vaddr;
1296
1297 area = get_xol_area();
1298 if (!area)
1299 return 0;
1300
1301 xol_vaddr = xol_take_insn_slot(area);
1302 if (unlikely(!xol_vaddr))
1303 return 0;
1304
1305 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1306 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1307
1308 return xol_vaddr;
1309}
1310
1311/*
1312 * xol_free_insn_slot - If slot was earlier allocated by
1313 * @xol_get_insn_slot(), make the slot available for
1314 * subsequent requests.
1315 */
1316static void xol_free_insn_slot(struct task_struct *tsk)
1317{
1318 struct xol_area *area;
1319 unsigned long vma_end;
1320 unsigned long slot_addr;
1321
1322 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1323 return;
1324
1325 slot_addr = tsk->utask->xol_vaddr;
1326 if (unlikely(!slot_addr))
1327 return;
1328
1329 area = tsk->mm->uprobes_state.xol_area;
1330 vma_end = area->vaddr + PAGE_SIZE;
1331 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1332 unsigned long offset;
1333 int slot_nr;
1334
1335 offset = slot_addr - area->vaddr;
1336 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1337 if (slot_nr >= UINSNS_PER_PAGE)
1338 return;
1339
1340 clear_bit(slot_nr, area->bitmap);
1341 atomic_dec(&area->slot_count);
1342 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1343 if (waitqueue_active(&area->wq))
1344 wake_up(&area->wq);
1345
1346 tsk->utask->xol_vaddr = 0;
1347 }
1348}
1349
1350void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1351 void *src, unsigned long len)
1352{
1353 /* Initialize the slot */
1354 copy_to_page(page, vaddr, src, len);
1355
1356 /*
1357 * We probably need flush_icache_user_range() but it needs vma.
1358 * This should work on most of architectures by default. If
1359 * architecture needs to do something different it can define
1360 * its own version of the function.
1361 */
1362 flush_dcache_page(page);
1363}
1364
1365/**
1366 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1367 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1368 * instruction.
1369 * Return the address of the breakpoint instruction.
1370 */
1371unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1372{
1373 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1374}
1375
1376unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1377{
1378 struct uprobe_task *utask = current->utask;
1379
1380 if (unlikely(utask && utask->active_uprobe))
1381 return utask->vaddr;
1382
1383 return instruction_pointer(regs);
1384}
1385
1386static struct return_instance *free_ret_instance(struct return_instance *ri)
1387{
1388 struct return_instance *next = ri->next;
1389 put_uprobe(ri->uprobe);
1390 kfree(ri);
1391 return next;
1392}
1393
1394/*
1395 * Called with no locks held.
1396 * Called in context of a exiting or a exec-ing thread.
1397 */
1398void uprobe_free_utask(struct task_struct *t)
1399{
1400 struct uprobe_task *utask = t->utask;
1401 struct return_instance *ri;
1402
1403 if (!utask)
1404 return;
1405
1406 if (utask->active_uprobe)
1407 put_uprobe(utask->active_uprobe);
1408
1409 ri = utask->return_instances;
1410 while (ri)
1411 ri = free_ret_instance(ri);
1412
1413 xol_free_insn_slot(t);
1414 kfree(utask);
1415 t->utask = NULL;
1416}
1417
1418/*
1419 * Allocate a uprobe_task object for the task if if necessary.
1420 * Called when the thread hits a breakpoint.
1421 *
1422 * Returns:
1423 * - pointer to new uprobe_task on success
1424 * - NULL otherwise
1425 */
1426static struct uprobe_task *get_utask(void)
1427{
1428 if (!current->utask)
1429 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1430 return current->utask;
1431}
1432
1433static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1434{
1435 struct uprobe_task *n_utask;
1436 struct return_instance **p, *o, *n;
1437
1438 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1439 if (!n_utask)
1440 return -ENOMEM;
1441 t->utask = n_utask;
1442
1443 p = &n_utask->return_instances;
1444 for (o = o_utask->return_instances; o; o = o->next) {
1445 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1446 if (!n)
1447 return -ENOMEM;
1448
1449 *n = *o;
1450 get_uprobe(n->uprobe);
1451 n->next = NULL;
1452
1453 *p = n;
1454 p = &n->next;
1455 n_utask->depth++;
1456 }
1457
1458 return 0;
1459}
1460
1461static void uprobe_warn(struct task_struct *t, const char *msg)
1462{
1463 pr_warn("uprobe: %s:%d failed to %s\n",
1464 current->comm, current->pid, msg);
1465}
1466
1467static void dup_xol_work(struct callback_head *work)
1468{
1469 if (current->flags & PF_EXITING)
1470 return;
1471
1472 if (!__create_xol_area(current->utask->dup_xol_addr))
1473 uprobe_warn(current, "dup xol area");
1474}
1475
1476/*
1477 * Called in context of a new clone/fork from copy_process.
1478 */
1479void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1480{
1481 struct uprobe_task *utask = current->utask;
1482 struct mm_struct *mm = current->mm;
1483 struct xol_area *area;
1484
1485 t->utask = NULL;
1486
1487 if (!utask || !utask->return_instances)
1488 return;
1489
1490 if (mm == t->mm && !(flags & CLONE_VFORK))
1491 return;
1492
1493 if (dup_utask(t, utask))
1494 return uprobe_warn(t, "dup ret instances");
1495
1496 /* The task can fork() after dup_xol_work() fails */
1497 area = mm->uprobes_state.xol_area;
1498 if (!area)
1499 return uprobe_warn(t, "dup xol area");
1500
1501 if (mm == t->mm)
1502 return;
1503
1504 t->utask->dup_xol_addr = area->vaddr;
1505 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1506 task_work_add(t, &t->utask->dup_xol_work, true);
1507}
1508
1509/*
1510 * Current area->vaddr notion assume the trampoline address is always
1511 * equal area->vaddr.
1512 *
1513 * Returns -1 in case the xol_area is not allocated.
1514 */
1515static unsigned long get_trampoline_vaddr(void)
1516{
1517 struct xol_area *area;
1518 unsigned long trampoline_vaddr = -1;
1519
1520 area = current->mm->uprobes_state.xol_area;
1521 smp_read_barrier_depends();
1522 if (area)
1523 trampoline_vaddr = area->vaddr;
1524
1525 return trampoline_vaddr;
1526}
1527
1528static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1529 struct pt_regs *regs)
1530{
1531 struct return_instance *ri = utask->return_instances;
1532 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1533
1534 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1535 ri = free_ret_instance(ri);
1536 utask->depth--;
1537 }
1538 utask->return_instances = ri;
1539}
1540
1541static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1542{
1543 struct return_instance *ri;
1544 struct uprobe_task *utask;
1545 unsigned long orig_ret_vaddr, trampoline_vaddr;
1546 bool chained;
1547
1548 if (!get_xol_area())
1549 return;
1550
1551 utask = get_utask();
1552 if (!utask)
1553 return;
1554
1555 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1556 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1557 " nestedness limit pid/tgid=%d/%d\n",
1558 current->pid, current->tgid);
1559 return;
1560 }
1561
1562 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1563 if (!ri)
1564 return;
1565
1566 trampoline_vaddr = get_trampoline_vaddr();
1567 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1568 if (orig_ret_vaddr == -1)
1569 goto fail;
1570
1571 /* drop the entries invalidated by longjmp() */
1572 chained = (orig_ret_vaddr == trampoline_vaddr);
1573 cleanup_return_instances(utask, chained, regs);
1574
1575 /*
1576 * We don't want to keep trampoline address in stack, rather keep the
1577 * original return address of first caller thru all the consequent
1578 * instances. This also makes breakpoint unwrapping easier.
1579 */
1580 if (chained) {
1581 if (!utask->return_instances) {
1582 /*
1583 * This situation is not possible. Likely we have an
1584 * attack from user-space.
1585 */
1586 uprobe_warn(current, "handle tail call");
1587 goto fail;
1588 }
1589 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1590 }
1591
1592 ri->uprobe = get_uprobe(uprobe);
1593 ri->func = instruction_pointer(regs);
1594 ri->stack = user_stack_pointer(regs);
1595 ri->orig_ret_vaddr = orig_ret_vaddr;
1596 ri->chained = chained;
1597
1598 utask->depth++;
1599 ri->next = utask->return_instances;
1600 utask->return_instances = ri;
1601
1602 return;
1603 fail:
1604 kfree(ri);
1605}
1606
1607/* Prepare to single-step probed instruction out of line. */
1608static int
1609pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1610{
1611 struct uprobe_task *utask;
1612 unsigned long xol_vaddr;
1613 int err;
1614
1615 utask = get_utask();
1616 if (!utask)
1617 return -ENOMEM;
1618
1619 xol_vaddr = xol_get_insn_slot(uprobe);
1620 if (!xol_vaddr)
1621 return -ENOMEM;
1622
1623 utask->xol_vaddr = xol_vaddr;
1624 utask->vaddr = bp_vaddr;
1625
1626 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1627 if (unlikely(err)) {
1628 xol_free_insn_slot(current);
1629 return err;
1630 }
1631
1632 utask->active_uprobe = uprobe;
1633 utask->state = UTASK_SSTEP;
1634 return 0;
1635}
1636
1637/*
1638 * If we are singlestepping, then ensure this thread is not connected to
1639 * non-fatal signals until completion of singlestep. When xol insn itself
1640 * triggers the signal, restart the original insn even if the task is
1641 * already SIGKILL'ed (since coredump should report the correct ip). This
1642 * is even more important if the task has a handler for SIGSEGV/etc, The
1643 * _same_ instruction should be repeated again after return from the signal
1644 * handler, and SSTEP can never finish in this case.
1645 */
1646bool uprobe_deny_signal(void)
1647{
1648 struct task_struct *t = current;
1649 struct uprobe_task *utask = t->utask;
1650
1651 if (likely(!utask || !utask->active_uprobe))
1652 return false;
1653
1654 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1655
1656 if (signal_pending(t)) {
1657 spin_lock_irq(&t->sighand->siglock);
1658 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1659 spin_unlock_irq(&t->sighand->siglock);
1660
1661 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1662 utask->state = UTASK_SSTEP_TRAPPED;
1663 set_tsk_thread_flag(t, TIF_UPROBE);
1664 }
1665 }
1666
1667 return true;
1668}
1669
1670static void mmf_recalc_uprobes(struct mm_struct *mm)
1671{
1672 struct vm_area_struct *vma;
1673
1674 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1675 if (!valid_vma(vma, false))
1676 continue;
1677 /*
1678 * This is not strictly accurate, we can race with
1679 * uprobe_unregister() and see the already removed
1680 * uprobe if delete_uprobe() was not yet called.
1681 * Or this uprobe can be filtered out.
1682 */
1683 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1684 return;
1685 }
1686
1687 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1688}
1689
1690static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1691{
1692 struct page *page;
1693 uprobe_opcode_t opcode;
1694 int result;
1695
1696 pagefault_disable();
1697 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1698 sizeof(opcode));
1699 pagefault_enable();
1700
1701 if (likely(result == 0))
1702 goto out;
1703
1704 /*
1705 * The NULL 'tsk' here ensures that any faults that occur here
1706 * will not be accounted to the task. 'mm' *is* current->mm,
1707 * but we treat this as a 'remote' access since it is
1708 * essentially a kernel access to the memory.
1709 */
1710 result = get_user_pages_remote(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1711 if (result < 0)
1712 return result;
1713
1714 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1715 put_page(page);
1716 out:
1717 /* This needs to return true for any variant of the trap insn */
1718 return is_trap_insn(&opcode);
1719}
1720
1721static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1722{
1723 struct mm_struct *mm = current->mm;
1724 struct uprobe *uprobe = NULL;
1725 struct vm_area_struct *vma;
1726
1727 down_read(&mm->mmap_sem);
1728 vma = find_vma(mm, bp_vaddr);
1729 if (vma && vma->vm_start <= bp_vaddr) {
1730 if (valid_vma(vma, false)) {
1731 struct inode *inode = file_inode(vma->vm_file);
1732 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1733
1734 uprobe = find_uprobe(inode, offset);
1735 }
1736
1737 if (!uprobe)
1738 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1739 } else {
1740 *is_swbp = -EFAULT;
1741 }
1742
1743 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1744 mmf_recalc_uprobes(mm);
1745 up_read(&mm->mmap_sem);
1746
1747 return uprobe;
1748}
1749
1750static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1751{
1752 struct uprobe_consumer *uc;
1753 int remove = UPROBE_HANDLER_REMOVE;
1754 bool need_prep = false; /* prepare return uprobe, when needed */
1755
1756 down_read(&uprobe->register_rwsem);
1757 for (uc = uprobe->consumers; uc; uc = uc->next) {
1758 int rc = 0;
1759
1760 if (uc->handler) {
1761 rc = uc->handler(uc, regs);
1762 WARN(rc & ~UPROBE_HANDLER_MASK,
1763 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1764 }
1765
1766 if (uc->ret_handler)
1767 need_prep = true;
1768
1769 remove &= rc;
1770 }
1771
1772 if (need_prep && !remove)
1773 prepare_uretprobe(uprobe, regs); /* put bp at return */
1774
1775 if (remove && uprobe->consumers) {
1776 WARN_ON(!uprobe_is_active(uprobe));
1777 unapply_uprobe(uprobe, current->mm);
1778 }
1779 up_read(&uprobe->register_rwsem);
1780}
1781
1782static void
1783handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1784{
1785 struct uprobe *uprobe = ri->uprobe;
1786 struct uprobe_consumer *uc;
1787
1788 down_read(&uprobe->register_rwsem);
1789 for (uc = uprobe->consumers; uc; uc = uc->next) {
1790 if (uc->ret_handler)
1791 uc->ret_handler(uc, ri->func, regs);
1792 }
1793 up_read(&uprobe->register_rwsem);
1794}
1795
1796static struct return_instance *find_next_ret_chain(struct return_instance *ri)
1797{
1798 bool chained;
1799
1800 do {
1801 chained = ri->chained;
1802 ri = ri->next; /* can't be NULL if chained */
1803 } while (chained);
1804
1805 return ri;
1806}
1807
1808static void handle_trampoline(struct pt_regs *regs)
1809{
1810 struct uprobe_task *utask;
1811 struct return_instance *ri, *next;
1812 bool valid;
1813
1814 utask = current->utask;
1815 if (!utask)
1816 goto sigill;
1817
1818 ri = utask->return_instances;
1819 if (!ri)
1820 goto sigill;
1821
1822 do {
1823 /*
1824 * We should throw out the frames invalidated by longjmp().
1825 * If this chain is valid, then the next one should be alive
1826 * or NULL; the latter case means that nobody but ri->func
1827 * could hit this trampoline on return. TODO: sigaltstack().
1828 */
1829 next = find_next_ret_chain(ri);
1830 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
1831
1832 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1833 do {
1834 if (valid)
1835 handle_uretprobe_chain(ri, regs);
1836 ri = free_ret_instance(ri);
1837 utask->depth--;
1838 } while (ri != next);
1839 } while (!valid);
1840
1841 utask->return_instances = ri;
1842 return;
1843
1844 sigill:
1845 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
1846 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1847
1848}
1849
1850bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
1851{
1852 return false;
1853}
1854
1855bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
1856 struct pt_regs *regs)
1857{
1858 return true;
1859}
1860
1861/*
1862 * Run handler and ask thread to singlestep.
1863 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1864 */
1865static void handle_swbp(struct pt_regs *regs)
1866{
1867 struct uprobe *uprobe;
1868 unsigned long bp_vaddr;
1869 int uninitialized_var(is_swbp);
1870
1871 bp_vaddr = uprobe_get_swbp_addr(regs);
1872 if (bp_vaddr == get_trampoline_vaddr())
1873 return handle_trampoline(regs);
1874
1875 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1876 if (!uprobe) {
1877 if (is_swbp > 0) {
1878 /* No matching uprobe; signal SIGTRAP. */
1879 send_sig(SIGTRAP, current, 0);
1880 } else {
1881 /*
1882 * Either we raced with uprobe_unregister() or we can't
1883 * access this memory. The latter is only possible if
1884 * another thread plays with our ->mm. In both cases
1885 * we can simply restart. If this vma was unmapped we
1886 * can pretend this insn was not executed yet and get
1887 * the (correct) SIGSEGV after restart.
1888 */
1889 instruction_pointer_set(regs, bp_vaddr);
1890 }
1891 return;
1892 }
1893
1894 /* change it in advance for ->handler() and restart */
1895 instruction_pointer_set(regs, bp_vaddr);
1896
1897 /*
1898 * TODO: move copy_insn/etc into _register and remove this hack.
1899 * After we hit the bp, _unregister + _register can install the
1900 * new and not-yet-analyzed uprobe at the same address, restart.
1901 */
1902 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1903 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1904 goto out;
1905
1906 /* Tracing handlers use ->utask to communicate with fetch methods */
1907 if (!get_utask())
1908 goto out;
1909
1910 if (arch_uprobe_ignore(&uprobe->arch, regs))
1911 goto out;
1912
1913 handler_chain(uprobe, regs);
1914
1915 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1916 goto out;
1917
1918 if (!pre_ssout(uprobe, regs, bp_vaddr))
1919 return;
1920
1921 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1922out:
1923 put_uprobe(uprobe);
1924}
1925
1926/*
1927 * Perform required fix-ups and disable singlestep.
1928 * Allow pending signals to take effect.
1929 */
1930static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1931{
1932 struct uprobe *uprobe;
1933 int err = 0;
1934
1935 uprobe = utask->active_uprobe;
1936 if (utask->state == UTASK_SSTEP_ACK)
1937 err = arch_uprobe_post_xol(&uprobe->arch, regs);
1938 else if (utask->state == UTASK_SSTEP_TRAPPED)
1939 arch_uprobe_abort_xol(&uprobe->arch, regs);
1940 else
1941 WARN_ON_ONCE(1);
1942
1943 put_uprobe(uprobe);
1944 utask->active_uprobe = NULL;
1945 utask->state = UTASK_RUNNING;
1946 xol_free_insn_slot(current);
1947
1948 spin_lock_irq(¤t->sighand->siglock);
1949 recalc_sigpending(); /* see uprobe_deny_signal() */
1950 spin_unlock_irq(¤t->sighand->siglock);
1951
1952 if (unlikely(err)) {
1953 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
1954 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1955 }
1956}
1957
1958/*
1959 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1960 * allows the thread to return from interrupt. After that handle_swbp()
1961 * sets utask->active_uprobe.
1962 *
1963 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1964 * and allows the thread to return from interrupt.
1965 *
1966 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1967 * uprobe_notify_resume().
1968 */
1969void uprobe_notify_resume(struct pt_regs *regs)
1970{
1971 struct uprobe_task *utask;
1972
1973 clear_thread_flag(TIF_UPROBE);
1974
1975 utask = current->utask;
1976 if (utask && utask->active_uprobe)
1977 handle_singlestep(utask, regs);
1978 else
1979 handle_swbp(regs);
1980}
1981
1982/*
1983 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1984 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1985 */
1986int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1987{
1988 if (!current->mm)
1989 return 0;
1990
1991 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
1992 (!current->utask || !current->utask->return_instances))
1993 return 0;
1994
1995 set_thread_flag(TIF_UPROBE);
1996 return 1;
1997}
1998
1999/*
2000 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2001 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2002 */
2003int uprobe_post_sstep_notifier(struct pt_regs *regs)
2004{
2005 struct uprobe_task *utask = current->utask;
2006
2007 if (!current->mm || !utask || !utask->active_uprobe)
2008 /* task is currently not uprobed */
2009 return 0;
2010
2011 utask->state = UTASK_SSTEP_ACK;
2012 set_thread_flag(TIF_UPROBE);
2013 return 1;
2014}
2015
2016static struct notifier_block uprobe_exception_nb = {
2017 .notifier_call = arch_uprobe_exception_notify,
2018 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2019};
2020
2021static int __init init_uprobes(void)
2022{
2023 int i;
2024
2025 for (i = 0; i < UPROBES_HASH_SZ; i++)
2026 mutex_init(&uprobes_mmap_mutex[i]);
2027
2028 if (percpu_init_rwsem(&dup_mmap_sem))
2029 return -ENOMEM;
2030
2031 return register_die_notifier(&uprobe_exception_nb);
2032}
2033__initcall(init_uprobes);