~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/events/uprobes.c

Version: ~ [ linux-5.4.2 ] ~ [ linux-5.3.15 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.88 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.158 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.206 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.206 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.78 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp