uprobes: Kill uprobes_mutex[], separate alloc_uprobe() and __uprobe_register()
[deliverable/linux.git] / kernel / events / uprobes.c
CommitLineData
2b144498 1/*
7b2d81d4 2 * User-space Probes (UProbes)
2b144498
SD
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 *
35aa621b 18 * Copyright (C) IBM Corporation, 2008-2012
2b144498
SD
19 * Authors:
20 * Srikar Dronamraju
21 * Jim Keniston
35aa621b 22 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
2b144498
SD
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/rmap.h> /* anon_vma_prepare */
31#include <linux/mmu_notifier.h> /* set_pte_at_notify */
32#include <linux/swap.h> /* try_to_free_swap */
0326f5a9
SD
33#include <linux/ptrace.h> /* user_enable_single_step */
34#include <linux/kdebug.h> /* notifier mechanism */
194f8dcb 35#include "../../mm/internal.h" /* munlock_vma_page */
32cdba1e 36#include <linux/percpu-rwsem.h>
7b2d81d4 37
2b144498
SD
38#include <linux/uprobes.h>
39
d4b3b638
SD
40#define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
41#define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
42
2b144498 43static struct rb_root uprobes_tree = RB_ROOT;
441f1eb7
ON
44/*
45 * allows us to skip the uprobe_mmap if there are no uprobe events active
46 * at this time. Probably a fine grained per inode count is better?
47 */
48#define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
7b2d81d4 49
2b144498
SD
50static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
51
52#define UPROBES_HASH_SZ 13
2b144498
SD
53/* serialize uprobe->pending_list */
54static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
7b2d81d4 55#define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
2b144498 56
32cdba1e
ON
57static struct percpu_rw_semaphore dup_mmap_sem;
58
cb9a19fe 59/* Have a copy of original instruction */
71434f2f 60#define UPROBE_COPY_INSN 0
cb9a19fe 61/* Can skip singlestep */
bb929284 62#define UPROBE_SKIP_SSTEP 1
cb9a19fe 63
3ff54efd
SD
64struct uprobe {
65 struct rb_node rb_node; /* node in the rb tree */
66 atomic_t ref;
e591c8d7 67 struct rw_semaphore register_rwsem;
3ff54efd
SD
68 struct rw_semaphore consumer_rwsem;
69 struct list_head pending_list;
70 struct uprobe_consumer *consumers;
71 struct inode *inode; /* Also hold a ref to inode */
72 loff_t offset;
71434f2f 73 unsigned long flags;
3ff54efd
SD
74 struct arch_uprobe arch;
75};
76
2b144498
SD
77/*
78 * valid_vma: Verify if the specified vma is an executable vma
79 * Relax restrictions while unregistering: vm_flags might have
80 * changed after breakpoint was inserted.
81 * - is_register: indicates if we are in register context.
82 * - Return 1 if the specified virtual address is in an
83 * executable vma.
84 */
85static bool valid_vma(struct vm_area_struct *vma, bool is_register)
86{
e40cfce6 87 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED;
2b144498 88
e40cfce6
ON
89 if (is_register)
90 flags |= VM_WRITE;
2b144498 91
e40cfce6 92 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
2b144498
SD
93}
94
57683f72 95static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
2b144498 96{
57683f72 97 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
2b144498
SD
98}
99
cb113b47
ON
100static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
101{
102 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
103}
104
2b144498
SD
105/**
106 * __replace_page - replace page in vma by new page.
107 * based on replace_page in mm/ksm.c
108 *
109 * @vma: vma that holds the pte pointing to page
c517ee74 110 * @addr: address the old @page is mapped at
2b144498
SD
111 * @page: the cowed page we are replacing by kpage
112 * @kpage: the modified page we replace page by
113 *
114 * Returns 0 on success, -EFAULT on failure.
115 */
c517ee74
ON
116static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
117 struct page *page, struct page *kpage)
2b144498
SD
118{
119 struct mm_struct *mm = vma->vm_mm;
5323ce71
ON
120 spinlock_t *ptl;
121 pte_t *ptep;
9f92448c 122 int err;
6bdb913f
HE
123 /* For mmu_notifiers */
124 const unsigned long mmun_start = addr;
125 const unsigned long mmun_end = addr + PAGE_SIZE;
2b144498 126
194f8dcb 127 /* For try_to_free_swap() and munlock_vma_page() below */
9f92448c
ON
128 lock_page(page);
129
6bdb913f 130 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
9f92448c 131 err = -EAGAIN;
5323ce71 132 ptep = page_check_address(page, mm, addr, &ptl, 0);
2b144498 133 if (!ptep)
9f92448c 134 goto unlock;
2b144498
SD
135
136 get_page(kpage);
137 page_add_new_anon_rmap(kpage, vma, addr);
138
7396fa81
SD
139 if (!PageAnon(page)) {
140 dec_mm_counter(mm, MM_FILEPAGES);
141 inc_mm_counter(mm, MM_ANONPAGES);
142 }
143
2b144498
SD
144 flush_cache_page(vma, addr, pte_pfn(*ptep));
145 ptep_clear_flush(vma, addr, ptep);
146 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
147
148 page_remove_rmap(page);
149 if (!page_mapped(page))
150 try_to_free_swap(page);
2b144498 151 pte_unmap_unlock(ptep, ptl);
2b144498 152
194f8dcb
ON
153 if (vma->vm_flags & VM_LOCKED)
154 munlock_vma_page(page);
155 put_page(page);
156
9f92448c
ON
157 err = 0;
158 unlock:
6bdb913f 159 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
9f92448c
ON
160 unlock_page(page);
161 return err;
2b144498
SD
162}
163
164/**
5cb4ac3a 165 * is_swbp_insn - check if instruction is breakpoint instruction.
2b144498 166 * @insn: instruction to be checked.
5cb4ac3a 167 * Default implementation of is_swbp_insn
2b144498
SD
168 * Returns true if @insn is a breakpoint instruction.
169 */
5cb4ac3a 170bool __weak is_swbp_insn(uprobe_opcode_t *insn)
2b144498 171{
5cb4ac3a 172 return *insn == UPROBE_SWBP_INSN;
2b144498
SD
173}
174
cceb55aa
ON
175static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode)
176{
177 void *kaddr = kmap_atomic(page);
178 memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE);
179 kunmap_atomic(kaddr);
180}
181
ed6f6a50
ON
182static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
183{
184 uprobe_opcode_t old_opcode;
185 bool is_swbp;
186
187 copy_opcode(page, vaddr, &old_opcode);
188 is_swbp = is_swbp_insn(&old_opcode);
189
190 if (is_swbp_insn(new_opcode)) {
191 if (is_swbp) /* register: already installed? */
192 return 0;
193 } else {
194 if (!is_swbp) /* unregister: was it changed by us? */
076a365b 195 return 0;
ed6f6a50
ON
196 }
197
198 return 1;
199}
200
2b144498
SD
201/*
202 * NOTE:
203 * Expect the breakpoint instruction to be the smallest size instruction for
204 * the architecture. If an arch has variable length instruction and the
205 * breakpoint instruction is not of the smallest length instruction
cceb55aa 206 * supported by that architecture then we need to modify is_swbp_at_addr and
2b144498
SD
207 * write_opcode accordingly. This would never be a problem for archs that
208 * have fixed length instructions.
209 */
210
211/*
212 * write_opcode - write the opcode at a given virtual address.
213 * @mm: the probed process address space.
2b144498
SD
214 * @vaddr: the virtual address to store the opcode.
215 * @opcode: opcode to be written at @vaddr.
216 *
217 * Called with mm->mmap_sem held (for read and with a reference to
218 * mm).
219 *
220 * For mm @mm, write the opcode at @vaddr.
221 * Return 0 (success) or a negative errno.
222 */
cceb55aa
ON
223static int write_opcode(struct mm_struct *mm, unsigned long vaddr,
224 uprobe_opcode_t opcode)
2b144498
SD
225{
226 struct page *old_page, *new_page;
2b144498
SD
227 void *vaddr_old, *vaddr_new;
228 struct vm_area_struct *vma;
2b144498 229 int ret;
f403072c 230
5323ce71 231retry:
2b144498 232 /* Read the page with vaddr into memory */
75ed82ea 233 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma);
2b144498
SD
234 if (ret <= 0)
235 return ret;
7b2d81d4 236
ed6f6a50
ON
237 ret = verify_opcode(old_page, vaddr, &opcode);
238 if (ret <= 0)
239 goto put_old;
240
2b144498
SD
241 ret = -ENOMEM;
242 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
243 if (!new_page)
9f92448c 244 goto put_old;
2b144498
SD
245
246 __SetPageUptodate(new_page);
247
2b144498
SD
248 /* copy the page now that we've got it stable */
249 vaddr_old = kmap_atomic(old_page);
250 vaddr_new = kmap_atomic(new_page);
251
252 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
d9c4a30e 253 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
2b144498
SD
254
255 kunmap_atomic(vaddr_new);
256 kunmap_atomic(vaddr_old);
257
258 ret = anon_vma_prepare(vma);
259 if (ret)
9f92448c 260 goto put_new;
2b144498 261
c517ee74 262 ret = __replace_page(vma, vaddr, old_page, new_page);
2b144498 263
9f92448c 264put_new:
2b144498 265 page_cache_release(new_page);
9f92448c 266put_old:
7b2d81d4
IM
267 put_page(old_page);
268
5323ce71
ON
269 if (unlikely(ret == -EAGAIN))
270 goto retry;
2b144498
SD
271 return ret;
272}
273
2b144498 274/**
5cb4ac3a 275 * set_swbp - store breakpoint at a given address.
e3343e6a 276 * @auprobe: arch specific probepoint information.
2b144498 277 * @mm: the probed process address space.
2b144498
SD
278 * @vaddr: the virtual address to insert the opcode.
279 *
280 * For mm @mm, store the breakpoint instruction at @vaddr.
281 * Return 0 (success) or a negative errno.
282 */
5cb4ac3a 283int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
2b144498 284{
cceb55aa 285 return write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
2b144498
SD
286}
287
288/**
289 * set_orig_insn - Restore the original instruction.
290 * @mm: the probed process address space.
e3343e6a 291 * @auprobe: arch specific probepoint information.
2b144498 292 * @vaddr: the virtual address to insert the opcode.
2b144498
SD
293 *
294 * For mm @mm, restore the original opcode (opcode) at @vaddr.
295 * Return 0 (success) or a negative errno.
296 */
7b2d81d4 297int __weak
ded86e7c 298set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
2b144498 299{
cceb55aa 300 return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
2b144498
SD
301}
302
303static int match_uprobe(struct uprobe *l, struct uprobe *r)
304{
305 if (l->inode < r->inode)
306 return -1;
7b2d81d4 307
2b144498
SD
308 if (l->inode > r->inode)
309 return 1;
2b144498 310
7b2d81d4
IM
311 if (l->offset < r->offset)
312 return -1;
313
314 if (l->offset > r->offset)
315 return 1;
2b144498
SD
316
317 return 0;
318}
319
320static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
321{
322 struct uprobe u = { .inode = inode, .offset = offset };
323 struct rb_node *n = uprobes_tree.rb_node;
324 struct uprobe *uprobe;
325 int match;
326
327 while (n) {
328 uprobe = rb_entry(n, struct uprobe, rb_node);
329 match = match_uprobe(&u, uprobe);
330 if (!match) {
331 atomic_inc(&uprobe->ref);
332 return uprobe;
333 }
7b2d81d4 334
2b144498
SD
335 if (match < 0)
336 n = n->rb_left;
337 else
338 n = n->rb_right;
339 }
340 return NULL;
341}
342
343/*
344 * Find a uprobe corresponding to a given inode:offset
345 * Acquires uprobes_treelock
346 */
347static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
348{
349 struct uprobe *uprobe;
2b144498 350
6f47caa0 351 spin_lock(&uprobes_treelock);
2b144498 352 uprobe = __find_uprobe(inode, offset);
6f47caa0 353 spin_unlock(&uprobes_treelock);
7b2d81d4 354
2b144498
SD
355 return uprobe;
356}
357
358static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
359{
360 struct rb_node **p = &uprobes_tree.rb_node;
361 struct rb_node *parent = NULL;
362 struct uprobe *u;
363 int match;
364
365 while (*p) {
366 parent = *p;
367 u = rb_entry(parent, struct uprobe, rb_node);
368 match = match_uprobe(uprobe, u);
369 if (!match) {
370 atomic_inc(&u->ref);
371 return u;
372 }
373
374 if (match < 0)
375 p = &parent->rb_left;
376 else
377 p = &parent->rb_right;
378
379 }
7b2d81d4 380
2b144498
SD
381 u = NULL;
382 rb_link_node(&uprobe->rb_node, parent, p);
383 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
384 /* get access + creation ref */
385 atomic_set(&uprobe->ref, 2);
7b2d81d4 386
2b144498
SD
387 return u;
388}
389
390/*
7b2d81d4 391 * Acquire uprobes_treelock.
2b144498
SD
392 * Matching uprobe already exists in rbtree;
393 * increment (access refcount) and return the matching uprobe.
394 *
395 * No matching uprobe; insert the uprobe in rb_tree;
396 * get a double refcount (access + creation) and return NULL.
397 */
398static struct uprobe *insert_uprobe(struct uprobe *uprobe)
399{
2b144498
SD
400 struct uprobe *u;
401
6f47caa0 402 spin_lock(&uprobes_treelock);
2b144498 403 u = __insert_uprobe(uprobe);
6f47caa0 404 spin_unlock(&uprobes_treelock);
7b2d81d4 405
2b144498
SD
406 return u;
407}
408
409static void put_uprobe(struct uprobe *uprobe)
410{
411 if (atomic_dec_and_test(&uprobe->ref))
412 kfree(uprobe);
413}
414
415static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
416{
417 struct uprobe *uprobe, *cur_uprobe;
418
419 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
420 if (!uprobe)
421 return NULL;
422
423 uprobe->inode = igrab(inode);
424 uprobe->offset = offset;
e591c8d7 425 init_rwsem(&uprobe->register_rwsem);
2b144498 426 init_rwsem(&uprobe->consumer_rwsem);
bbc33d05
ON
427 /* For now assume that the instruction need not be single-stepped */
428 __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
2b144498
SD
429
430 /* add to uprobes_tree, sorted on inode:offset */
431 cur_uprobe = insert_uprobe(uprobe);
432
433 /* a uprobe exists for this inode:offset combination */
434 if (cur_uprobe) {
435 kfree(uprobe);
436 uprobe = cur_uprobe;
437 iput(inode);
7b2d81d4
IM
438 }
439
2b144498
SD
440 return uprobe;
441}
442
0326f5a9
SD
443static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
444{
445 struct uprobe_consumer *uc;
446
e591c8d7 447 down_read(&uprobe->register_rwsem);
fe20d71f
ON
448 for (uc = uprobe->consumers; uc; uc = uc->next)
449 uc->handler(uc, regs);
e591c8d7 450 up_read(&uprobe->register_rwsem);
0326f5a9
SD
451}
452
9a98e03c 453static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
2b144498
SD
454{
455 down_write(&uprobe->consumer_rwsem);
e3343e6a
SD
456 uc->next = uprobe->consumers;
457 uprobe->consumers = uc;
2b144498 458 up_write(&uprobe->consumer_rwsem);
2b144498
SD
459}
460
461/*
e3343e6a
SD
462 * For uprobe @uprobe, delete the consumer @uc.
463 * Return true if the @uc is deleted successfully
2b144498
SD
464 * or return false.
465 */
e3343e6a 466static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
2b144498
SD
467{
468 struct uprobe_consumer **con;
469 bool ret = false;
470
471 down_write(&uprobe->consumer_rwsem);
472 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
e3343e6a
SD
473 if (*con == uc) {
474 *con = uc->next;
2b144498
SD
475 ret = true;
476 break;
477 }
478 }
479 up_write(&uprobe->consumer_rwsem);
7b2d81d4 480
2b144498
SD
481 return ret;
482}
483
e3343e6a 484static int
d436615e 485__copy_insn(struct address_space *mapping, struct file *filp, char *insn,
593609a5 486 unsigned long nbytes, loff_t offset)
2b144498 487{
2b144498
SD
488 struct page *page;
489 void *vaddr;
593609a5
ON
490 unsigned long off;
491 pgoff_t idx;
2b144498
SD
492
493 if (!filp)
494 return -EINVAL;
495
cc359d18
ON
496 if (!mapping->a_ops->readpage)
497 return -EIO;
498
593609a5
ON
499 idx = offset >> PAGE_CACHE_SHIFT;
500 off = offset & ~PAGE_MASK;
2b144498
SD
501
502 /*
503 * Ensure that the page that has the original instruction is
504 * populated and in page-cache.
505 */
506 page = read_mapping_page(mapping, idx, filp);
507 if (IS_ERR(page))
508 return PTR_ERR(page);
509
510 vaddr = kmap_atomic(page);
593609a5 511 memcpy(insn, vaddr + off, nbytes);
2b144498
SD
512 kunmap_atomic(vaddr);
513 page_cache_release(page);
7b2d81d4 514
2b144498
SD
515 return 0;
516}
517
d436615e 518static int copy_insn(struct uprobe *uprobe, struct file *filp)
2b144498
SD
519{
520 struct address_space *mapping;
2b144498 521 unsigned long nbytes;
7b2d81d4 522 int bytes;
2b144498 523
d436615e 524 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
2b144498
SD
525 mapping = uprobe->inode->i_mapping;
526
527 /* Instruction at end of binary; copy only available bytes */
528 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
529 bytes = uprobe->inode->i_size - uprobe->offset;
530 else
531 bytes = MAX_UINSN_BYTES;
532
533 /* Instruction at the page-boundary; copy bytes in second page */
534 if (nbytes < bytes) {
fc36f595
ON
535 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
536 bytes - nbytes, uprobe->offset + nbytes);
537 if (err)
538 return err;
2b144498
SD
539 bytes = nbytes;
540 }
d436615e 541 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
2b144498
SD
542}
543
cb9a19fe
ON
544static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
545 struct mm_struct *mm, unsigned long vaddr)
546{
547 int ret = 0;
548
71434f2f 549 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
cb9a19fe
ON
550 return ret;
551
d4d3ccc6
ON
552 /* TODO: move this into _register, until then we abuse this sem. */
553 down_write(&uprobe->consumer_rwsem);
71434f2f 554 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
4710f05f
ON
555 goto out;
556
cb9a19fe
ON
557 ret = copy_insn(uprobe, file);
558 if (ret)
559 goto out;
560
561 ret = -ENOTSUPP;
562 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
563 goto out;
564
565 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
566 if (ret)
567 goto out;
568
569 /* write_opcode() assumes we don't cross page boundary */
570 BUG_ON((uprobe->offset & ~PAGE_MASK) +
571 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
572
573 smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
71434f2f 574 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
cb9a19fe
ON
575
576 out:
d4d3ccc6 577 up_write(&uprobe->consumer_rwsem);
4710f05f 578
cb9a19fe
ON
579 return ret;
580}
581
63633cbf
ON
582static bool filter_chain(struct uprobe *uprobe)
583{
1ff6fee5
ON
584 struct uprobe_consumer *uc;
585 bool ret = false;
586
587 down_read(&uprobe->consumer_rwsem);
588 for (uc = uprobe->consumers; uc; uc = uc->next) {
589 /* TODO: ret = uc->filter(...) */
590 ret = true;
591 if (ret)
592 break;
593 }
594 up_read(&uprobe->consumer_rwsem);
595
596 return ret;
63633cbf
ON
597}
598
e3343e6a
SD
599static int
600install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
816c03fb 601 struct vm_area_struct *vma, unsigned long vaddr)
2b144498 602{
f8ac4ec9 603 bool first_uprobe;
2b144498
SD
604 int ret;
605
606 /*
607 * If probe is being deleted, unregister thread could be done with
608 * the vma-rmap-walk through. Adding a probe now can be fatal since
63633cbf
ON
609 * nobody will be able to cleanup. But in this case filter_chain()
610 * must return false, all consumers have gone away.
2b144498 611 */
63633cbf 612 if (!filter_chain(uprobe))
78f74116 613 return 0;
2b144498 614
cb9a19fe
ON
615 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
616 if (ret)
617 return ret;
682968e0 618
f8ac4ec9
ON
619 /*
620 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
621 * the task can hit this breakpoint right after __replace_page().
622 */
623 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
624 if (first_uprobe)
625 set_bit(MMF_HAS_UPROBES, &mm->flags);
626
816c03fb 627 ret = set_swbp(&uprobe->arch, mm, vaddr);
9f68f672
ON
628 if (!ret)
629 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
630 else if (first_uprobe)
f8ac4ec9 631 clear_bit(MMF_HAS_UPROBES, &mm->flags);
2b144498
SD
632
633 return ret;
634}
635
076a365b 636static int
816c03fb 637remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
2b144498 638{
9f68f672 639 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
076a365b 640 return 0;
9f68f672 641
63633cbf
ON
642 if (filter_chain(uprobe))
643 return 0;
644
9f68f672 645 set_bit(MMF_RECALC_UPROBES, &mm->flags);
076a365b 646 return set_orig_insn(&uprobe->arch, mm, vaddr);
2b144498
SD
647}
648
06b7bcd8
ON
649static inline bool uprobe_is_active(struct uprobe *uprobe)
650{
651 return !RB_EMPTY_NODE(&uprobe->rb_node);
652}
0326f5a9 653/*
778b032d
ON
654 * There could be threads that have already hit the breakpoint. They
655 * will recheck the current insn and restart if find_uprobe() fails.
656 * See find_active_uprobe().
0326f5a9 657 */
2b144498
SD
658static void delete_uprobe(struct uprobe *uprobe)
659{
06b7bcd8
ON
660 if (WARN_ON(!uprobe_is_active(uprobe)))
661 return;
662
6f47caa0 663 spin_lock(&uprobes_treelock);
2b144498 664 rb_erase(&uprobe->rb_node, &uprobes_tree);
6f47caa0 665 spin_unlock(&uprobes_treelock);
06b7bcd8 666 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
2b144498
SD
667 iput(uprobe->inode);
668 put_uprobe(uprobe);
2b144498
SD
669}
670
26872090
ON
671struct map_info {
672 struct map_info *next;
673 struct mm_struct *mm;
816c03fb 674 unsigned long vaddr;
26872090
ON
675};
676
677static inline struct map_info *free_map_info(struct map_info *info)
2b144498 678{
26872090
ON
679 struct map_info *next = info->next;
680 kfree(info);
681 return next;
682}
683
684static struct map_info *
685build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
686{
687 unsigned long pgoff = offset >> PAGE_SHIFT;
2b144498 688 struct vm_area_struct *vma;
26872090
ON
689 struct map_info *curr = NULL;
690 struct map_info *prev = NULL;
691 struct map_info *info;
692 int more = 0;
2b144498 693
26872090
ON
694 again:
695 mutex_lock(&mapping->i_mmap_mutex);
6b2dbba8 696 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
2b144498
SD
697 if (!valid_vma(vma, is_register))
698 continue;
699
7a5bfb66
ON
700 if (!prev && !more) {
701 /*
702 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
703 * reclaim. This is optimistic, no harm done if it fails.
704 */
705 prev = kmalloc(sizeof(struct map_info),
706 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
707 if (prev)
708 prev->next = NULL;
709 }
26872090
ON
710 if (!prev) {
711 more++;
712 continue;
2b144498 713 }
2b144498 714
26872090
ON
715 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
716 continue;
7b2d81d4 717
26872090
ON
718 info = prev;
719 prev = prev->next;
720 info->next = curr;
721 curr = info;
2b144498 722
26872090 723 info->mm = vma->vm_mm;
57683f72 724 info->vaddr = offset_to_vaddr(vma, offset);
26872090 725 }
2b144498
SD
726 mutex_unlock(&mapping->i_mmap_mutex);
727
26872090
ON
728 if (!more)
729 goto out;
730
731 prev = curr;
732 while (curr) {
733 mmput(curr->mm);
734 curr = curr->next;
735 }
7b2d81d4 736
26872090
ON
737 do {
738 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
739 if (!info) {
740 curr = ERR_PTR(-ENOMEM);
741 goto out;
742 }
743 info->next = prev;
744 prev = info;
745 } while (--more);
746
747 goto again;
748 out:
749 while (prev)
750 prev = free_map_info(prev);
751 return curr;
2b144498
SD
752}
753
754static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
755{
26872090
ON
756 struct map_info *info;
757 int err = 0;
2b144498 758
32cdba1e 759 percpu_down_write(&dup_mmap_sem);
26872090
ON
760 info = build_map_info(uprobe->inode->i_mapping,
761 uprobe->offset, is_register);
32cdba1e
ON
762 if (IS_ERR(info)) {
763 err = PTR_ERR(info);
764 goto out;
765 }
7b2d81d4 766
26872090
ON
767 while (info) {
768 struct mm_struct *mm = info->mm;
769 struct vm_area_struct *vma;
7b2d81d4 770
076a365b 771 if (err && is_register)
26872090 772 goto free;
7b2d81d4 773
77fc4af1 774 down_write(&mm->mmap_sem);
f4d6dfe5
ON
775 vma = find_vma(mm, info->vaddr);
776 if (!vma || !valid_vma(vma, is_register) ||
777 vma->vm_file->f_mapping->host != uprobe->inode)
26872090
ON
778 goto unlock;
779
f4d6dfe5
ON
780 if (vma->vm_start > info->vaddr ||
781 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
26872090 782 goto unlock;
2b144498 783
78f74116 784 if (is_register)
26872090 785 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
78f74116 786 else
076a365b 787 err |= remove_breakpoint(uprobe, mm, info->vaddr);
78f74116 788
26872090
ON
789 unlock:
790 up_write(&mm->mmap_sem);
791 free:
792 mmput(mm);
793 info = free_map_info(info);
2b144498 794 }
32cdba1e
ON
795 out:
796 percpu_up_write(&dup_mmap_sem);
26872090 797 return err;
2b144498
SD
798}
799
9a98e03c 800static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
2b144498 801{
9a98e03c 802 consumer_add(uprobe, uc);
bb929284 803 return register_for_each_vma(uprobe, true);
2b144498
SD
804}
805
04aab9b2 806static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
2b144498 807{
04aab9b2
ON
808 int err;
809
810 if (!consumer_del(uprobe, uc)) /* WARN? */
811 return;
2b144498 812
04aab9b2 813 err = register_for_each_vma(uprobe, false);
bb929284
ON
814 /* TODO : cant unregister? schedule a worker thread */
815 if (!uprobe->consumers && !err)
816 delete_uprobe(uprobe);
2b144498
SD
817}
818
819/*
7b2d81d4 820 * uprobe_register - register a probe
2b144498
SD
821 * @inode: the file in which the probe has to be placed.
822 * @offset: offset from the start of the file.
e3343e6a 823 * @uc: information on howto handle the probe..
2b144498 824 *
7b2d81d4 825 * Apart from the access refcount, uprobe_register() takes a creation
2b144498
SD
826 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
827 * inserted into the rbtree (i.e first consumer for a @inode:@offset
7b2d81d4 828 * tuple). Creation refcount stops uprobe_unregister from freeing the
2b144498 829 * @uprobe even before the register operation is complete. Creation
e3343e6a 830 * refcount is released when the last @uc for the @uprobe
2b144498
SD
831 * unregisters.
832 *
833 * Return errno if it cannot successully install probes
834 * else return 0 (success)
835 */
e3343e6a 836int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
2b144498
SD
837{
838 struct uprobe *uprobe;
7b2d81d4 839 int ret;
2b144498 840
f0744af7 841 /* Racy, just to catch the obvious mistakes */
2b144498 842 if (offset > i_size_read(inode))
7b2d81d4 843 return -EINVAL;
2b144498 844
66d06dff 845 retry:
2b144498 846 uprobe = alloc_uprobe(inode, offset);
66d06dff
ON
847 if (!uprobe)
848 return -ENOMEM;
849 /*
850 * We can race with uprobe_unregister()->delete_uprobe().
851 * Check uprobe_is_active() and retry if it is false.
852 */
853 down_write(&uprobe->register_rwsem);
854 ret = -EAGAIN;
855 if (likely(uprobe_is_active(uprobe))) {
9a98e03c
ON
856 ret = __uprobe_register(uprobe, uc);
857 if (ret)
04aab9b2 858 __uprobe_unregister(uprobe, uc);
2b144498 859 }
66d06dff
ON
860 up_write(&uprobe->register_rwsem);
861 put_uprobe(uprobe);
2b144498 862
66d06dff
ON
863 if (unlikely(ret == -EAGAIN))
864 goto retry;
2b144498
SD
865 return ret;
866}
867
868/*
7b2d81d4 869 * uprobe_unregister - unregister a already registered probe.
2b144498
SD
870 * @inode: the file in which the probe has to be removed.
871 * @offset: offset from the start of the file.
e3343e6a 872 * @uc: identify which probe if multiple probes are colocated.
2b144498 873 */
e3343e6a 874void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
2b144498 875{
7b2d81d4 876 struct uprobe *uprobe;
2b144498 877
2b144498
SD
878 uprobe = find_uprobe(inode, offset);
879 if (!uprobe)
880 return;
881
e591c8d7 882 down_write(&uprobe->register_rwsem);
04aab9b2 883 __uprobe_unregister(uprobe, uc);
e591c8d7 884 up_write(&uprobe->register_rwsem);
c91368c4 885 put_uprobe(uprobe);
2b144498
SD
886}
887
891c3970
ON
888static struct rb_node *
889find_node_in_range(struct inode *inode, loff_t min, loff_t max)
2b144498 890{
2b144498 891 struct rb_node *n = uprobes_tree.rb_node;
2b144498
SD
892
893 while (n) {
891c3970 894 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
2b144498 895
891c3970 896 if (inode < u->inode) {
2b144498 897 n = n->rb_left;
891c3970 898 } else if (inode > u->inode) {
2b144498 899 n = n->rb_right;
891c3970
ON
900 } else {
901 if (max < u->offset)
902 n = n->rb_left;
903 else if (min > u->offset)
904 n = n->rb_right;
905 else
906 break;
907 }
2b144498 908 }
7b2d81d4 909
891c3970 910 return n;
2b144498
SD
911}
912
913/*
891c3970 914 * For a given range in vma, build a list of probes that need to be inserted.
2b144498 915 */
891c3970
ON
916static void build_probe_list(struct inode *inode,
917 struct vm_area_struct *vma,
918 unsigned long start, unsigned long end,
919 struct list_head *head)
2b144498 920{
891c3970 921 loff_t min, max;
891c3970
ON
922 struct rb_node *n, *t;
923 struct uprobe *u;
7b2d81d4 924
891c3970 925 INIT_LIST_HEAD(head);
cb113b47 926 min = vaddr_to_offset(vma, start);
891c3970 927 max = min + (end - start) - 1;
2b144498 928
6f47caa0 929 spin_lock(&uprobes_treelock);
891c3970
ON
930 n = find_node_in_range(inode, min, max);
931 if (n) {
932 for (t = n; t; t = rb_prev(t)) {
933 u = rb_entry(t, struct uprobe, rb_node);
934 if (u->inode != inode || u->offset < min)
935 break;
936 list_add(&u->pending_list, head);
937 atomic_inc(&u->ref);
938 }
939 for (t = n; (t = rb_next(t)); ) {
940 u = rb_entry(t, struct uprobe, rb_node);
941 if (u->inode != inode || u->offset > max)
942 break;
943 list_add(&u->pending_list, head);
944 atomic_inc(&u->ref);
945 }
2b144498 946 }
6f47caa0 947 spin_unlock(&uprobes_treelock);
2b144498
SD
948}
949
950/*
5e5be71a 951 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
2b144498 952 *
5e5be71a
ON
953 * Currently we ignore all errors and always return 0, the callers
954 * can't handle the failure anyway.
2b144498 955 */
7b2d81d4 956int uprobe_mmap(struct vm_area_struct *vma)
2b144498
SD
957{
958 struct list_head tmp_list;
665605a2 959 struct uprobe *uprobe, *u;
2b144498 960 struct inode *inode;
2b144498 961
441f1eb7 962 if (no_uprobe_events() || !valid_vma(vma, true))
7b2d81d4 963 return 0;
2b144498
SD
964
965 inode = vma->vm_file->f_mapping->host;
966 if (!inode)
7b2d81d4 967 return 0;
2b144498 968
2b144498 969 mutex_lock(uprobes_mmap_hash(inode));
891c3970 970 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
7b2d81d4 971
665605a2 972 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
5e5be71a 973 if (!fatal_signal_pending(current)) {
57683f72 974 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
5e5be71a 975 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
2b144498
SD
976 }
977 put_uprobe(uprobe);
978 }
2b144498
SD
979 mutex_unlock(uprobes_mmap_hash(inode));
980
5e5be71a 981 return 0;
2b144498
SD
982}
983
9f68f672
ON
984static bool
985vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
986{
987 loff_t min, max;
988 struct inode *inode;
989 struct rb_node *n;
990
991 inode = vma->vm_file->f_mapping->host;
992
993 min = vaddr_to_offset(vma, start);
994 max = min + (end - start) - 1;
995
996 spin_lock(&uprobes_treelock);
997 n = find_node_in_range(inode, min, max);
998 spin_unlock(&uprobes_treelock);
999
1000 return !!n;
1001}
1002
682968e0
SD
1003/*
1004 * Called in context of a munmap of a vma.
1005 */
cbc91f71 1006void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
682968e0 1007{
441f1eb7 1008 if (no_uprobe_events() || !valid_vma(vma, false))
682968e0
SD
1009 return;
1010
2fd611a9
ON
1011 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1012 return;
1013
9f68f672
ON
1014 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1015 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
f8ac4ec9
ON
1016 return;
1017
9f68f672
ON
1018 if (vma_has_uprobes(vma, start, end))
1019 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
682968e0
SD
1020}
1021
d4b3b638
SD
1022/* Slot allocation for XOL */
1023static int xol_add_vma(struct xol_area *area)
1024{
1025 struct mm_struct *mm;
1026 int ret;
1027
1028 area->page = alloc_page(GFP_HIGHUSER);
1029 if (!area->page)
1030 return -ENOMEM;
1031
1032 ret = -EALREADY;
1033 mm = current->mm;
1034
1035 down_write(&mm->mmap_sem);
1036 if (mm->uprobes_state.xol_area)
1037 goto fail;
1038
1039 ret = -ENOMEM;
1040
1041 /* Try to map as high as possible, this is only a hint. */
1042 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1043 if (area->vaddr & ~PAGE_MASK) {
1044 ret = area->vaddr;
1045 goto fail;
1046 }
1047
1048 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1049 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1050 if (ret)
1051 goto fail;
1052
1053 smp_wmb(); /* pairs with get_xol_area() */
1054 mm->uprobes_state.xol_area = area;
1055 ret = 0;
1056
1057fail:
1058 up_write(&mm->mmap_sem);
1059 if (ret)
1060 __free_page(area->page);
1061
1062 return ret;
1063}
1064
1065static struct xol_area *get_xol_area(struct mm_struct *mm)
1066{
1067 struct xol_area *area;
1068
1069 area = mm->uprobes_state.xol_area;
1070 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1071
1072 return area;
1073}
1074
1075/*
1076 * xol_alloc_area - Allocate process's xol_area.
1077 * This area will be used for storing instructions for execution out of
1078 * line.
1079 *
1080 * Returns the allocated area or NULL.
1081 */
1082static struct xol_area *xol_alloc_area(void)
1083{
1084 struct xol_area *area;
1085
1086 area = kzalloc(sizeof(*area), GFP_KERNEL);
1087 if (unlikely(!area))
1088 return NULL;
1089
1090 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1091
1092 if (!area->bitmap)
1093 goto fail;
1094
1095 init_waitqueue_head(&area->wq);
1096 if (!xol_add_vma(area))
1097 return area;
1098
1099fail:
1100 kfree(area->bitmap);
1101 kfree(area);
1102
1103 return get_xol_area(current->mm);
1104}
1105
1106/*
1107 * uprobe_clear_state - Free the area allocated for slots.
1108 */
1109void uprobe_clear_state(struct mm_struct *mm)
1110{
1111 struct xol_area *area = mm->uprobes_state.xol_area;
1112
1113 if (!area)
1114 return;
1115
1116 put_page(area->page);
1117 kfree(area->bitmap);
1118 kfree(area);
1119}
1120
32cdba1e
ON
1121void uprobe_start_dup_mmap(void)
1122{
1123 percpu_down_read(&dup_mmap_sem);
1124}
1125
1126void uprobe_end_dup_mmap(void)
1127{
1128 percpu_up_read(&dup_mmap_sem);
1129}
1130
f8ac4ec9
ON
1131void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1132{
61559a81
ON
1133 newmm->uprobes_state.xol_area = NULL;
1134
9f68f672 1135 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
f8ac4ec9 1136 set_bit(MMF_HAS_UPROBES, &newmm->flags);
9f68f672
ON
1137 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1138 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1139 }
f8ac4ec9
ON
1140}
1141
d4b3b638
SD
1142/*
1143 * - search for a free slot.
1144 */
1145static unsigned long xol_take_insn_slot(struct xol_area *area)
1146{
1147 unsigned long slot_addr;
1148 int slot_nr;
1149
1150 do {
1151 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1152 if (slot_nr < UINSNS_PER_PAGE) {
1153 if (!test_and_set_bit(slot_nr, area->bitmap))
1154 break;
1155
1156 slot_nr = UINSNS_PER_PAGE;
1157 continue;
1158 }
1159 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1160 } while (slot_nr >= UINSNS_PER_PAGE);
1161
1162 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1163 atomic_inc(&area->slot_count);
1164
1165 return slot_addr;
1166}
1167
1168/*
1169 * xol_get_insn_slot - If was not allocated a slot, then
1170 * allocate a slot.
1171 * Returns the allocated slot address or 0.
1172 */
1173static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1174{
1175 struct xol_area *area;
1176 unsigned long offset;
1177 void *vaddr;
1178
1179 area = get_xol_area(current->mm);
1180 if (!area) {
1181 area = xol_alloc_area();
1182 if (!area)
1183 return 0;
1184 }
1185 current->utask->xol_vaddr = xol_take_insn_slot(area);
1186
1187 /*
1188 * Initialize the slot if xol_vaddr points to valid
1189 * instruction slot.
1190 */
1191 if (unlikely(!current->utask->xol_vaddr))
1192 return 0;
1193
1194 current->utask->vaddr = slot_addr;
1195 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1196 vaddr = kmap_atomic(area->page);
1197 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1198 kunmap_atomic(vaddr);
65b6ecc0
RV
1199 /*
1200 * We probably need flush_icache_user_range() but it needs vma.
1201 * This should work on supported architectures too.
1202 */
1203 flush_dcache_page(area->page);
d4b3b638
SD
1204
1205 return current->utask->xol_vaddr;
1206}
1207
1208/*
1209 * xol_free_insn_slot - If slot was earlier allocated by
1210 * @xol_get_insn_slot(), make the slot available for
1211 * subsequent requests.
1212 */
1213static void xol_free_insn_slot(struct task_struct *tsk)
1214{
1215 struct xol_area *area;
1216 unsigned long vma_end;
1217 unsigned long slot_addr;
1218
1219 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1220 return;
1221
1222 slot_addr = tsk->utask->xol_vaddr;
1223
1224 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1225 return;
1226
1227 area = tsk->mm->uprobes_state.xol_area;
1228 vma_end = area->vaddr + PAGE_SIZE;
1229 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1230 unsigned long offset;
1231 int slot_nr;
1232
1233 offset = slot_addr - area->vaddr;
1234 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1235 if (slot_nr >= UINSNS_PER_PAGE)
1236 return;
1237
1238 clear_bit(slot_nr, area->bitmap);
1239 atomic_dec(&area->slot_count);
1240 if (waitqueue_active(&area->wq))
1241 wake_up(&area->wq);
1242
1243 tsk->utask->xol_vaddr = 0;
1244 }
1245}
1246
0326f5a9
SD
1247/**
1248 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1249 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1250 * instruction.
1251 * Return the address of the breakpoint instruction.
1252 */
1253unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1254{
1255 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1256}
1257
1258/*
1259 * Called with no locks held.
1260 * Called in context of a exiting or a exec-ing thread.
1261 */
1262void uprobe_free_utask(struct task_struct *t)
1263{
1264 struct uprobe_task *utask = t->utask;
1265
0326f5a9
SD
1266 if (!utask)
1267 return;
1268
1269 if (utask->active_uprobe)
1270 put_uprobe(utask->active_uprobe);
1271
d4b3b638 1272 xol_free_insn_slot(t);
0326f5a9
SD
1273 kfree(utask);
1274 t->utask = NULL;
1275}
1276
1277/*
1278 * Called in context of a new clone/fork from copy_process.
1279 */
1280void uprobe_copy_process(struct task_struct *t)
1281{
1282 t->utask = NULL;
0326f5a9
SD
1283}
1284
1285/*
1286 * Allocate a uprobe_task object for the task.
1287 * Called when the thread hits a breakpoint for the first time.
1288 *
1289 * Returns:
1290 * - pointer to new uprobe_task on success
1291 * - NULL otherwise
1292 */
1293static struct uprobe_task *add_utask(void)
1294{
1295 struct uprobe_task *utask;
1296
1297 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1298 if (unlikely(!utask))
1299 return NULL;
1300
0326f5a9
SD
1301 current->utask = utask;
1302 return utask;
1303}
1304
1305/* Prepare to single-step probed instruction out of line. */
1306static int
1307pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1308{
d4b3b638
SD
1309 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1310 return 0;
1311
0326f5a9
SD
1312 return -EFAULT;
1313}
1314
1315/*
1316 * If we are singlestepping, then ensure this thread is not connected to
1317 * non-fatal signals until completion of singlestep. When xol insn itself
1318 * triggers the signal, restart the original insn even if the task is
1319 * already SIGKILL'ed (since coredump should report the correct ip). This
1320 * is even more important if the task has a handler for SIGSEGV/etc, The
1321 * _same_ instruction should be repeated again after return from the signal
1322 * handler, and SSTEP can never finish in this case.
1323 */
1324bool uprobe_deny_signal(void)
1325{
1326 struct task_struct *t = current;
1327 struct uprobe_task *utask = t->utask;
1328
1329 if (likely(!utask || !utask->active_uprobe))
1330 return false;
1331
1332 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1333
1334 if (signal_pending(t)) {
1335 spin_lock_irq(&t->sighand->siglock);
1336 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1337 spin_unlock_irq(&t->sighand->siglock);
1338
1339 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1340 utask->state = UTASK_SSTEP_TRAPPED;
1341 set_tsk_thread_flag(t, TIF_UPROBE);
1342 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1343 }
1344 }
1345
1346 return true;
1347}
1348
1349/*
1350 * Avoid singlestepping the original instruction if the original instruction
1351 * is a NOP or can be emulated.
1352 */
1353static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1354{
71434f2f 1355 if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) {
0578a970
ON
1356 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1357 return true;
71434f2f 1358 clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags);
0578a970 1359 }
0326f5a9
SD
1360 return false;
1361}
1362
499a4f3e
ON
1363static void mmf_recalc_uprobes(struct mm_struct *mm)
1364{
1365 struct vm_area_struct *vma;
1366
1367 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1368 if (!valid_vma(vma, false))
1369 continue;
1370 /*
1371 * This is not strictly accurate, we can race with
1372 * uprobe_unregister() and see the already removed
1373 * uprobe if delete_uprobe() was not yet called.
63633cbf 1374 * Or this uprobe can be filtered out.
499a4f3e
ON
1375 */
1376 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1377 return;
1378 }
1379
1380 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1381}
1382
ec75fba9
ON
1383static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
1384{
1385 struct page *page;
1386 uprobe_opcode_t opcode;
1387 int result;
1388
1389 pagefault_disable();
1390 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
1391 sizeof(opcode));
1392 pagefault_enable();
1393
1394 if (likely(result == 0))
1395 goto out;
1396
1397 result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
1398 if (result < 0)
1399 return result;
1400
1401 copy_opcode(page, vaddr, &opcode);
1402 put_page(page);
1403 out:
1404 return is_swbp_insn(&opcode);
1405}
1406
d790d346 1407static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
0326f5a9 1408{
3a9ea052
ON
1409 struct mm_struct *mm = current->mm;
1410 struct uprobe *uprobe = NULL;
0326f5a9 1411 struct vm_area_struct *vma;
0326f5a9 1412
0326f5a9
SD
1413 down_read(&mm->mmap_sem);
1414 vma = find_vma(mm, bp_vaddr);
3a9ea052
ON
1415 if (vma && vma->vm_start <= bp_vaddr) {
1416 if (valid_vma(vma, false)) {
cb113b47
ON
1417 struct inode *inode = vma->vm_file->f_mapping->host;
1418 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
0326f5a9 1419
3a9ea052
ON
1420 uprobe = find_uprobe(inode, offset);
1421 }
d790d346
ON
1422
1423 if (!uprobe)
1424 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1425 } else {
1426 *is_swbp = -EFAULT;
0326f5a9 1427 }
499a4f3e
ON
1428
1429 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1430 mmf_recalc_uprobes(mm);
0326f5a9
SD
1431 up_read(&mm->mmap_sem);
1432
3a9ea052
ON
1433 return uprobe;
1434}
1435
1436/*
1437 * Run handler and ask thread to singlestep.
1438 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1439 */
1440static void handle_swbp(struct pt_regs *regs)
1441{
1442 struct uprobe_task *utask;
1443 struct uprobe *uprobe;
1444 unsigned long bp_vaddr;
56bb4cf6 1445 int uninitialized_var(is_swbp);
3a9ea052
ON
1446
1447 bp_vaddr = uprobe_get_swbp_addr(regs);
d790d346 1448 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
3a9ea052 1449
0326f5a9 1450 if (!uprobe) {
56bb4cf6
ON
1451 if (is_swbp > 0) {
1452 /* No matching uprobe; signal SIGTRAP. */
1453 send_sig(SIGTRAP, current, 0);
1454 } else {
1455 /*
1456 * Either we raced with uprobe_unregister() or we can't
1457 * access this memory. The latter is only possible if
1458 * another thread plays with our ->mm. In both cases
1459 * we can simply restart. If this vma was unmapped we
1460 * can pretend this insn was not executed yet and get
1461 * the (correct) SIGSEGV after restart.
1462 */
1463 instruction_pointer_set(regs, bp_vaddr);
1464 }
0326f5a9
SD
1465 return;
1466 }
142b18dd
ON
1467 /*
1468 * TODO: move copy_insn/etc into _register and remove this hack.
1469 * After we hit the bp, _unregister + _register can install the
1470 * new and not-yet-analyzed uprobe at the same address, restart.
1471 */
1472 smp_rmb(); /* pairs with wmb() in install_breakpoint() */
71434f2f 1473 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
142b18dd 1474 goto restart;
0326f5a9
SD
1475
1476 utask = current->utask;
1477 if (!utask) {
1478 utask = add_utask();
1479 /* Cannot allocate; re-execute the instruction. */
1480 if (!utask)
0578a970 1481 goto restart;
0326f5a9 1482 }
746a9e6b 1483
0326f5a9 1484 handler_chain(uprobe, regs);
0578a970
ON
1485 if (can_skip_sstep(uprobe, regs))
1486 goto out;
0326f5a9 1487
0326f5a9 1488 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
746a9e6b
ON
1489 utask->active_uprobe = uprobe;
1490 utask->state = UTASK_SSTEP;
0326f5a9
SD
1491 return;
1492 }
1493
0578a970
ON
1494restart:
1495 /*
1496 * cannot singlestep; cannot skip instruction;
1497 * re-execute the instruction.
1498 */
1499 instruction_pointer_set(regs, bp_vaddr);
1500out:
8bd87445 1501 put_uprobe(uprobe);
0326f5a9
SD
1502}
1503
1504/*
1505 * Perform required fix-ups and disable singlestep.
1506 * Allow pending signals to take effect.
1507 */
1508static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1509{
1510 struct uprobe *uprobe;
1511
1512 uprobe = utask->active_uprobe;
1513 if (utask->state == UTASK_SSTEP_ACK)
1514 arch_uprobe_post_xol(&uprobe->arch, regs);
1515 else if (utask->state == UTASK_SSTEP_TRAPPED)
1516 arch_uprobe_abort_xol(&uprobe->arch, regs);
1517 else
1518 WARN_ON_ONCE(1);
1519
1520 put_uprobe(uprobe);
1521 utask->active_uprobe = NULL;
1522 utask->state = UTASK_RUNNING;
d4b3b638 1523 xol_free_insn_slot(current);
0326f5a9
SD
1524
1525 spin_lock_irq(&current->sighand->siglock);
1526 recalc_sigpending(); /* see uprobe_deny_signal() */
1527 spin_unlock_irq(&current->sighand->siglock);
1528}
1529
1530/*
1b08e907
ON
1531 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1532 * allows the thread to return from interrupt. After that handle_swbp()
1533 * sets utask->active_uprobe.
0326f5a9 1534 *
1b08e907
ON
1535 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1536 * and allows the thread to return from interrupt.
0326f5a9
SD
1537 *
1538 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1539 * uprobe_notify_resume().
1540 */
1541void uprobe_notify_resume(struct pt_regs *regs)
1542{
1543 struct uprobe_task *utask;
1544
db023ea5
ON
1545 clear_thread_flag(TIF_UPROBE);
1546
0326f5a9 1547 utask = current->utask;
1b08e907 1548 if (utask && utask->active_uprobe)
0326f5a9 1549 handle_singlestep(utask, regs);
1b08e907
ON
1550 else
1551 handle_swbp(regs);
0326f5a9
SD
1552}
1553
1554/*
1555 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1556 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1557 */
1558int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1559{
f8ac4ec9 1560 if (!current->mm || !test_bit(MMF_HAS_UPROBES, &current->mm->flags))
0326f5a9
SD
1561 return 0;
1562
0326f5a9 1563 set_thread_flag(TIF_UPROBE);
0326f5a9
SD
1564 return 1;
1565}
1566
1567/*
1568 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1569 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1570 */
1571int uprobe_post_sstep_notifier(struct pt_regs *regs)
1572{
1573 struct uprobe_task *utask = current->utask;
1574
1575 if (!current->mm || !utask || !utask->active_uprobe)
1576 /* task is currently not uprobed */
1577 return 0;
1578
1579 utask->state = UTASK_SSTEP_ACK;
1580 set_thread_flag(TIF_UPROBE);
1581 return 1;
1582}
1583
1584static struct notifier_block uprobe_exception_nb = {
1585 .notifier_call = arch_uprobe_exception_notify,
1586 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1587};
1588
2b144498
SD
1589static int __init init_uprobes(void)
1590{
1591 int i;
1592
66d06dff 1593 for (i = 0; i < UPROBES_HASH_SZ; i++)
2b144498 1594 mutex_init(&uprobes_mmap_mutex[i]);
0326f5a9 1595
32cdba1e
ON
1596 if (percpu_init_rwsem(&dup_mmap_sem))
1597 return -ENOMEM;
1598
0326f5a9 1599 return register_die_notifier(&uprobe_exception_nb);
2b144498 1600}
0326f5a9 1601module_init(init_uprobes);
2b144498
SD
1602
1603static void __exit exit_uprobes(void)
1604{
1605}
2b144498 1606module_exit(exit_uprobes);
This page took 0.159294 seconds and 5 git commands to generate.