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

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

Version: ~ [ linux-6.0-rc6 ] ~ [ linux-5.19.10 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.69 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.144 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.214 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.259 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.294 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.329 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ 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
 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/sched/mm.h>
 31 #include <linux/sched/coredump.h>
 32 #include <linux/export.h>
 33 #include <linux/rmap.h>         /* anon_vma_prepare */
 34 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
 35 #include <linux/swap.h>         /* try_to_free_swap */
 36 #include <linux/ptrace.h>       /* user_enable_single_step */
 37 #include <linux/kdebug.h>       /* notifier mechanism */
 38 #include "../../mm/internal.h"  /* munlock_vma_page */
 39 #include <linux/percpu-rwsem.h>
 40 #include <linux/task_work.h>
 41 #include <linux/shmem_fs.h>
 42 
 43 #include <linux/uprobes.h>
 44 
 45 #define UINSNS_PER_PAGE                 (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
 46 #define MAX_UPROBE_XOL_SLOTS            UINSNS_PER_PAGE
 47 
 48 static struct rb_root uprobes_tree = RB_ROOT;
 49 /*
 50  * allows us to skip the uprobe_mmap if there are no uprobe events active
 51  * at this time.  Probably a fine grained per inode count is better?
 52  */
 53 #define no_uprobe_events()      RB_EMPTY_ROOT(&uprobes_tree)
 54 
 55 static DEFINE_SPINLOCK(uprobes_treelock);       /* serialize rbtree access */
 56 
 57 #define UPROBES_HASH_SZ 13
 58 /* serialize uprobe->pending_list */
 59 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
 60 #define uprobes_mmap_hash(v)    (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
 61 
 62 static struct percpu_rw_semaphore dup_mmap_sem;
 63 
 64 /* Have a copy of original instruction */
 65 #define UPROBE_COPY_INSN        0
 66 
 67 struct uprobe {
 68         struct rb_node          rb_node;        /* node in the rb tree */
 69         atomic_t                ref;
 70         struct rw_semaphore     register_rwsem;
 71         struct rw_semaphore     consumer_rwsem;
 72         struct list_head        pending_list;
 73         struct uprobe_consumer  *consumers;
 74         struct inode            *inode;         /* Also hold a ref to inode */
 75         loff_t                  offset;
 76         unsigned long           flags;
 77 
 78         /*
 79          * The generic code assumes that it has two members of unknown type
 80          * owned by the arch-specific code:
 81          *
 82          *      insn -  copy_insn() saves the original instruction here for
 83          *              arch_uprobe_analyze_insn().
 84          *
 85          *      ixol -  potentially modified instruction to execute out of
 86          *              line, copied to xol_area by xol_get_insn_slot().
 87          */
 88         struct arch_uprobe      arch;
 89 };
 90 
 91 /*
 92  * Execute out of line area: anonymous executable mapping installed
 93  * by the probed task to execute the copy of the original instruction
 94  * mangled by set_swbp().
 95  *
 96  * On a breakpoint hit, thread contests for a slot.  It frees the
 97  * slot after singlestep. Currently a fixed number of slots are
 98  * allocated.
 99  */
100 struct xol_area {
101         wait_queue_head_t               wq;             /* if all slots are busy */
102         atomic_t                        slot_count;     /* number of in-use slots */
103         unsigned long                   *bitmap;        /* 0 = free slot */
104 
105         struct vm_special_mapping       xol_mapping;
106         struct page                     *pages[2];
107         /*
108          * We keep the vma's vm_start rather than a pointer to the vma
109          * itself.  The probed process or a naughty kernel module could make
110          * the vma go away, and we must handle that reasonably gracefully.
111          */
112         unsigned long                   vaddr;          /* Page(s) of instruction slots */
113 };
114 
115 /*
116  * valid_vma: Verify if the specified vma is an executable vma
117  * Relax restrictions while unregistering: vm_flags might have
118  * changed after breakpoint was inserted.
119  *      - is_register: indicates if we are in register context.
120  *      - Return 1 if the specified virtual address is in an
121  *        executable vma.
122  */
123 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
124 {
125         vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
126 
127         if (is_register)
128                 flags |= VM_WRITE;
129 
130         return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
131 }
132 
133 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
134 {
135         return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
136 }
137 
138 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
139 {
140         return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
141 }
142 
143 /**
144  * __replace_page - replace page in vma by new page.
145  * based on replace_page in mm/ksm.c
146  *
147  * @vma:      vma that holds the pte pointing to page
148  * @addr:     address the old @page is mapped at
149  * @page:     the cowed page we are replacing by kpage
150  * @kpage:    the modified page we replace page by
151  *
152  * Returns 0 on success, -EFAULT on failure.
153  */
154 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
155                                 struct page *old_page, struct page *new_page)
156 {
157         struct mm_struct *mm = vma->vm_mm;
158         struct page_vma_mapped_walk pvmw = {
159                 .page = old_page,
160                 .vma = vma,
161                 .address = addr,
162         };
163         int err;
164         /* For mmu_notifiers */
165         const unsigned long mmun_start = addr;
166         const unsigned long mmun_end   = addr + PAGE_SIZE;
167         struct mem_cgroup *memcg;
168 
169         VM_BUG_ON_PAGE(PageTransHuge(old_page), old_page);
170 
171         err = mem_cgroup_try_charge(new_page, vma->vm_mm, GFP_KERNEL, &memcg,
172                         false);
173         if (err)
174                 return err;
175 
176         /* For try_to_free_swap() and munlock_vma_page() below */
177         lock_page(old_page);
178 
179         mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
180         err = -EAGAIN;
181         if (!page_vma_mapped_walk(&pvmw)) {
182                 mem_cgroup_cancel_charge(new_page, memcg, false);
183                 goto unlock;
184         }
185         VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
186 
187         get_page(new_page);
188         page_add_new_anon_rmap(new_page, vma, addr, false);
189         mem_cgroup_commit_charge(new_page, memcg, false, false);
190         lru_cache_add_active_or_unevictable(new_page, vma);
191 
192         if (!PageAnon(old_page)) {
193                 dec_mm_counter(mm, mm_counter_file(old_page));
194                 inc_mm_counter(mm, MM_ANONPAGES);
195         }
196 
197         flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
198         ptep_clear_flush_notify(vma, addr, pvmw.pte);
199         set_pte_at_notify(mm, addr, pvmw.pte,
200                         mk_pte(new_page, vma->vm_page_prot));
201 
202         page_remove_rmap(old_page, false);
203         if (!page_mapped(old_page))
204                 try_to_free_swap(old_page);
205         page_vma_mapped_walk_done(&pvmw);
206 
207         if (vma->vm_flags & VM_LOCKED)
208                 munlock_vma_page(old_page);
209         put_page(old_page);
210 
211         err = 0;
212  unlock:
213         mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
214         unlock_page(old_page);
215         return err;
216 }
217 
218 /**
219  * is_swbp_insn - check if instruction is breakpoint instruction.
220  * @insn: instruction to be checked.
221  * Default implementation of is_swbp_insn
222  * Returns true if @insn is a breakpoint instruction.
223  */
224 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
225 {
226         return *insn == UPROBE_SWBP_INSN;
227 }
228 
229 /**
230  * is_trap_insn - check if instruction is breakpoint instruction.
231  * @insn: instruction to be checked.
232  * Default implementation of is_trap_insn
233  * Returns true if @insn is a breakpoint instruction.
234  *
235  * This function is needed for the case where an architecture has multiple
236  * trap instructions (like powerpc).
237  */
238 bool __weak is_trap_insn(uprobe_opcode_t *insn)
239 {
240         return is_swbp_insn(insn);
241 }
242 
243 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
244 {
245         void *kaddr = kmap_atomic(page);
246         memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
247         kunmap_atomic(kaddr);
248 }
249 
250 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
251 {
252         void *kaddr = kmap_atomic(page);
253         memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
254         kunmap_atomic(kaddr);
255 }
256 
257 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
258 {
259         uprobe_opcode_t old_opcode;
260         bool is_swbp;
261 
262         /*
263          * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
264          * We do not check if it is any other 'trap variant' which could
265          * be conditional trap instruction such as the one powerpc supports.
266          *
267          * The logic is that we do not care if the underlying instruction
268          * is a trap variant; uprobes always wins over any other (gdb)
269          * breakpoint.
270          */
271         copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
272         is_swbp = is_swbp_insn(&old_opcode);
273 
274         if (is_swbp_insn(new_opcode)) {
275                 if (is_swbp)            /* register: already installed? */
276                         return 0;
277         } else {
278                 if (!is_swbp)           /* unregister: was it changed by us? */
279                         return 0;
280         }
281 
282         return 1;
283 }
284 
285 /*
286  * NOTE:
287  * Expect the breakpoint instruction to be the smallest size instruction for
288  * the architecture. If an arch has variable length instruction and the
289  * breakpoint instruction is not of the smallest length instruction
290  * supported by that architecture then we need to modify is_trap_at_addr and
291  * uprobe_write_opcode accordingly. This would never be a problem for archs
292  * that have fixed length instructions.
293  *
294  * uprobe_write_opcode - write the opcode at a given virtual address.
295  * @mm: the probed process address space.
296  * @vaddr: the virtual address to store the opcode.
297  * @opcode: opcode to be written at @vaddr.
298  *
299  * Called with mm->mmap_sem held for write.
300  * Return 0 (success) or a negative errno.
301  */
302 int uprobe_write_opcode(struct mm_struct *mm, unsigned long vaddr,
303                         uprobe_opcode_t opcode)
304 {
305         struct page *old_page, *new_page;
306         struct vm_area_struct *vma;
307         int ret;
308 
309 retry:
310         /* Read the page with vaddr into memory */
311         ret = get_user_pages_remote(NULL, mm, vaddr, 1,
312                         FOLL_FORCE | FOLL_SPLIT, &old_page, &vma, NULL);
313         if (ret <= 0)
314                 return ret;
315 
316         ret = verify_opcode(old_page, vaddr, &opcode);
317         if (ret <= 0)
318                 goto put_old;
319 
320         ret = anon_vma_prepare(vma);
321         if (ret)
322                 goto put_old;
323 
324         ret = -ENOMEM;
325         new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
326         if (!new_page)
327                 goto put_old;
328 
329         __SetPageUptodate(new_page);
330         copy_highpage(new_page, old_page);
331         copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
332 
333         ret = __replace_page(vma, vaddr, old_page, new_page);
334         put_page(new_page);
335 put_old:
336         put_page(old_page);
337 
338         if (unlikely(ret == -EAGAIN))
339                 goto retry;
340         return ret;
341 }
342 
343 /**
344  * set_swbp - store breakpoint at a given address.
345  * @auprobe: arch specific probepoint information.
346  * @mm: the probed process address space.
347  * @vaddr: the virtual address to insert the opcode.
348  *
349  * For mm @mm, store the breakpoint instruction at @vaddr.
350  * Return 0 (success) or a negative errno.
351  */
352 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
353 {
354         return uprobe_write_opcode(mm, vaddr, UPROBE_SWBP_INSN);
355 }
356 
357 /**
358  * set_orig_insn - Restore the original instruction.
359  * @mm: the probed process address space.
360  * @auprobe: arch specific probepoint information.
361  * @vaddr: the virtual address to insert the opcode.
362  *
363  * For mm @mm, restore the original opcode (opcode) at @vaddr.
364  * Return 0 (success) or a negative errno.
365  */
366 int __weak
367 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
368 {
369         return uprobe_write_opcode(mm, vaddr, *(uprobe_opcode_t *)&auprobe->insn);
370 }
371 
372 static struct uprobe *get_uprobe(struct uprobe *uprobe)
373 {
374         atomic_inc(&uprobe->ref);
375         return uprobe;
376 }
377 
378 static void put_uprobe(struct uprobe *uprobe)
379 {
380         if (atomic_dec_and_test(&uprobe->ref))
381                 kfree(uprobe);
382 }
383 
384 static int match_uprobe(struct uprobe *l, struct uprobe *r)
385 {
386         if (l->inode < r->inode)
387                 return -1;
388 
389         if (l->inode > r->inode)
390                 return 1;
391 
392         if (l->offset < r->offset)
393                 return -1;
394 
395         if (l->offset > r->offset)
396                 return 1;
397 
398         return 0;
399 }
400 
401 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
402 {
403         struct uprobe u = { .inode = inode, .offset = offset };
404         struct rb_node *n = uprobes_tree.rb_node;
405         struct uprobe *uprobe;
406         int match;
407 
408         while (n) {
409                 uprobe = rb_entry(n, struct uprobe, rb_node);
410                 match = match_uprobe(&u, uprobe);
411                 if (!match)
412                         return get_uprobe(uprobe);
413 
414                 if (match < 0)
415                         n = n->rb_left;
416                 else
417                         n = n->rb_right;
418         }
419         return NULL;
420 }
421 
422 /*
423  * Find a uprobe corresponding to a given inode:offset
424  * Acquires uprobes_treelock
425  */
426 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
427 {
428         struct uprobe *uprobe;
429 
430         spin_lock(&uprobes_treelock);
431         uprobe = __find_uprobe(inode, offset);
432         spin_unlock(&uprobes_treelock);
433 
434         return uprobe;
435 }
436 
437 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
438 {
439         struct rb_node **p = &uprobes_tree.rb_node;
440         struct rb_node *parent = NULL;
441         struct uprobe *u;
442         int match;
443 
444         while (*p) {
445                 parent = *p;
446                 u = rb_entry(parent, struct uprobe, rb_node);
447                 match = match_uprobe(uprobe, u);
448                 if (!match)
449                         return get_uprobe(u);
450 
451                 if (match < 0)
452                         p = &parent->rb_left;
453                 else
454                         p = &parent->rb_right;
455 
456         }
457 
458         u = NULL;
459         rb_link_node(&uprobe->rb_node, parent, p);
460         rb_insert_color(&uprobe->rb_node, &uprobes_tree);
461         /* get access + creation ref */
462         atomic_set(&uprobe->ref, 2);
463 
464         return u;
465 }
466 
467 /*
468  * Acquire uprobes_treelock.
469  * Matching uprobe already exists in rbtree;
470  *      increment (access refcount) and return the matching uprobe.
471  *
472  * No matching uprobe; insert the uprobe in rb_tree;
473  *      get a double refcount (access + creation) and return NULL.
474  */
475 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
476 {
477         struct uprobe *u;
478 
479         spin_lock(&uprobes_treelock);
480         u = __insert_uprobe(uprobe);
481         spin_unlock(&uprobes_treelock);
482 
483         return u;
484 }
485 
486 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
487 {
488         struct uprobe *uprobe, *cur_uprobe;
489 
490         uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
491         if (!uprobe)
492                 return NULL;
493 
494         uprobe->inode = inode;
495         uprobe->offset = offset;
496         init_rwsem(&uprobe->register_rwsem);
497         init_rwsem(&uprobe->consumer_rwsem);
498 
499         /* add to uprobes_tree, sorted on inode:offset */
500         cur_uprobe = insert_uprobe(uprobe);
501         /* a uprobe exists for this inode:offset combination */
502         if (cur_uprobe) {
503                 kfree(uprobe);
504                 uprobe = cur_uprobe;
505         }
506 
507         return uprobe;
508 }
509 
510 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
511 {
512         down_write(&uprobe->consumer_rwsem);
513         uc->next = uprobe->consumers;
514         uprobe->consumers = uc;
515         up_write(&uprobe->consumer_rwsem);
516 }
517 
518 /*
519  * For uprobe @uprobe, delete the consumer @uc.
520  * Return true if the @uc is deleted successfully
521  * or return false.
522  */
523 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
524 {
525         struct uprobe_consumer **con;
526         bool ret = false;
527 
528         down_write(&uprobe->consumer_rwsem);
529         for (con = &uprobe->consumers; *con; con = &(*con)->next) {
530                 if (*con == uc) {
531                         *con = uc->next;
532                         ret = true;
533                         break;
534                 }
535         }
536         up_write(&uprobe->consumer_rwsem);
537 
538         return ret;
539 }
540 
541 static int __copy_insn(struct address_space *mapping, struct file *filp,
542                         void *insn, int nbytes, loff_t offset)
543 {
544         struct page *page;
545         /*
546          * Ensure that the page that has the original instruction is populated
547          * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
548          * see uprobe_register().
549          */
550         if (mapping->a_ops->readpage)
551                 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
552         else
553                 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
554         if (IS_ERR(page))
555                 return PTR_ERR(page);
556 
557         copy_from_page(page, offset, insn, nbytes);
558         put_page(page);
559 
560         return 0;
561 }
562 
563 static int copy_insn(struct uprobe *uprobe, struct file *filp)
564 {
565         struct address_space *mapping = uprobe->inode->i_mapping;
566         loff_t offs = uprobe->offset;
567         void *insn = &uprobe->arch.insn;
568         int size = sizeof(uprobe->arch.insn);
569         int len, err = -EIO;
570 
571         /* Copy only available bytes, -EIO if nothing was read */
572         do {
573                 if (offs >= i_size_read(uprobe->inode))
574                         break;
575 
576                 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
577                 err = __copy_insn(mapping, filp, insn, len, offs);
578                 if (err)
579                         break;
580 
581                 insn += len;
582                 offs += len;
583                 size -= len;
584         } while (size);
585 
586         return err;
587 }
588 
589 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
590                                 struct mm_struct *mm, unsigned long vaddr)
591 {
592         int ret = 0;
593 
594         if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
595                 return ret;
596 
597         /* TODO: move this into _register, until then we abuse this sem. */
598         down_write(&uprobe->consumer_rwsem);
599         if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
600                 goto out;
601 
602         ret = copy_insn(uprobe, file);
603         if (ret)
604                 goto out;
605 
606         ret = -ENOTSUPP;
607         if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
608                 goto out;
609 
610         ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
611         if (ret)
612                 goto out;
613 
614         /* uprobe_write_opcode() assumes we don't cross page boundary */
615         BUG_ON((uprobe->offset & ~PAGE_MASK) +
616                         UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
617 
618         smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
619         set_bit(UPROBE_COPY_INSN, &uprobe->flags);
620 
621  out:
622         up_write(&uprobe->consumer_rwsem);
623 
624         return ret;
625 }
626 
627 static inline bool consumer_filter(struct uprobe_consumer *uc,
628                                    enum uprobe_filter_ctx ctx, struct mm_struct *mm)
629 {
630         return !uc->filter || uc->filter(uc, ctx, mm);
631 }
632 
633 static bool filter_chain(struct uprobe *uprobe,
634                          enum uprobe_filter_ctx ctx, struct mm_struct *mm)
635 {
636         struct uprobe_consumer *uc;
637         bool ret = false;
638 
639         down_read(&uprobe->consumer_rwsem);
640         for (uc = uprobe->consumers; uc; uc = uc->next) {
641                 ret = consumer_filter(uc, ctx, mm);
642                 if (ret)
643                         break;
644         }
645         up_read(&uprobe->consumer_rwsem);
646 
647         return ret;
648 }
649 
650 static int
651 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
652                         struct vm_area_struct *vma, unsigned long vaddr)
653 {
654         bool first_uprobe;
655         int ret;
656 
657         ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
658         if (ret)
659                 return ret;
660 
661         /*
662          * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
663          * the task can hit this breakpoint right after __replace_page().
664          */
665         first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
666         if (first_uprobe)
667                 set_bit(MMF_HAS_UPROBES, &mm->flags);
668 
669         ret = set_swbp(&uprobe->arch, mm, vaddr);
670         if (!ret)
671                 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
672         else if (first_uprobe)
673                 clear_bit(MMF_HAS_UPROBES, &mm->flags);
674 
675         return ret;
676 }
677 
678 static int
679 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
680 {
681         set_bit(MMF_RECALC_UPROBES, &mm->flags);
682         return set_orig_insn(&uprobe->arch, mm, vaddr);
683 }
684 
685 static inline bool uprobe_is_active(struct uprobe *uprobe)
686 {
687         return !RB_EMPTY_NODE(&uprobe->rb_node);
688 }
689 /*
690  * There could be threads that have already hit the breakpoint. They
691  * will recheck the current insn and restart if find_uprobe() fails.
692  * See find_active_uprobe().
693  */
694 static void delete_uprobe(struct uprobe *uprobe)
695 {
696         if (WARN_ON(!uprobe_is_active(uprobe)))
697                 return;
698 
699         spin_lock(&uprobes_treelock);
700         rb_erase(&uprobe->rb_node, &uprobes_tree);
701         spin_unlock(&uprobes_treelock);
702         RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
703         put_uprobe(uprobe);
704 }
705 
706 struct map_info {
707         struct map_info *next;
708         struct mm_struct *mm;
709         unsigned long vaddr;
710 };
711 
712 static inline struct map_info *free_map_info(struct map_info *info)
713 {
714         struct map_info *next = info->next;
715         kfree(info);
716         return next;
717 }
718 
719 static struct map_info *
720 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
721 {
722         unsigned long pgoff = offset >> PAGE_SHIFT;
723         struct vm_area_struct *vma;
724         struct map_info *curr = NULL;
725         struct map_info *prev = NULL;
726         struct map_info *info;
727         int more = 0;
728 
729  again:
730         i_mmap_lock_read(mapping);
731         vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
732                 if (!valid_vma(vma, is_register))
733                         continue;
734 
735                 if (!prev && !more) {
736                         /*
737                          * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
738                          * reclaim. This is optimistic, no harm done if it fails.
739                          */
740                         prev = kmalloc(sizeof(struct map_info),
741                                         GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
742                         if (prev)
743                                 prev->next = NULL;
744                 }
745                 if (!prev) {
746                         more++;
747                         continue;
748                 }
749 
750                 if (!mmget_not_zero(vma->vm_mm))
751                         continue;
752 
753                 info = prev;
754                 prev = prev->next;
755                 info->next = curr;
756                 curr = info;
757 
758                 info->mm = vma->vm_mm;
759                 info->vaddr = offset_to_vaddr(vma, offset);
760         }
761         i_mmap_unlock_read(mapping);
762 
763         if (!more)
764                 goto out;
765 
766         prev = curr;
767         while (curr) {
768                 mmput(curr->mm);
769                 curr = curr->next;
770         }
771 
772         do {
773                 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
774                 if (!info) {
775                         curr = ERR_PTR(-ENOMEM);
776                         goto out;
777                 }
778                 info->next = prev;
779                 prev = info;
780         } while (--more);
781 
782         goto again;
783  out:
784         while (prev)
785                 prev = free_map_info(prev);
786         return curr;
787 }
788 
789 static int
790 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
791 {
792         bool is_register = !!new;
793         struct map_info *info;
794         int err = 0;
795 
796         percpu_down_write(&dup_mmap_sem);
797         info = build_map_info(uprobe->inode->i_mapping,
798                                         uprobe->offset, is_register);
799         if (IS_ERR(info)) {
800                 err = PTR_ERR(info);
801                 goto out;
802         }
803 
804         while (info) {
805                 struct mm_struct *mm = info->mm;
806                 struct vm_area_struct *vma;
807 
808                 if (err && is_register)
809                         goto free;
810 
811                 down_write(&mm->mmap_sem);
812                 vma = find_vma(mm, info->vaddr);
813                 if (!vma || !valid_vma(vma, is_register) ||
814                     file_inode(vma->vm_file) != uprobe->inode)
815                         goto unlock;
816 
817                 if (vma->vm_start > info->vaddr ||
818                     vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
819                         goto unlock;
820 
821                 if (is_register) {
822                         /* consult only the "caller", new consumer. */
823                         if (consumer_filter(new,
824                                         UPROBE_FILTER_REGISTER, mm))
825                                 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
826                 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
827                         if (!filter_chain(uprobe,
828                                         UPROBE_FILTER_UNREGISTER, mm))
829                                 err |= remove_breakpoint(uprobe, mm, info->vaddr);
830                 }
831 
832  unlock:
833                 up_write(&mm->mmap_sem);
834  free:
835                 mmput(mm);
836                 info = free_map_info(info);
837         }
838  out:
839         percpu_up_write(&dup_mmap_sem);
840         return err;
841 }
842 
843 static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc)
844 {
845         consumer_add(uprobe, uc);
846         return register_for_each_vma(uprobe, uc);
847 }
848 
849 static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
850 {
851         int err;
852 
853         if (WARN_ON(!consumer_del(uprobe, uc)))
854                 return;
855 
856         err = register_for_each_vma(uprobe, NULL);
857         /* TODO : cant unregister? schedule a worker thread */
858         if (!uprobe->consumers && !err)
859                 delete_uprobe(uprobe);
860 }
861 
862 /*
863  * uprobe_register - register a probe
864  * @inode: the file in which the probe has to be placed.
865  * @offset: offset from the start of the file.
866  * @uc: information on howto handle the probe..
867  *
868  * Apart from the access refcount, uprobe_register() takes a creation
869  * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
870  * inserted into the rbtree (i.e first consumer for a @inode:@offset
871  * tuple).  Creation refcount stops uprobe_unregister from freeing the
872  * @uprobe even before the register operation is complete. Creation
873  * refcount is released when the last @uc for the @uprobe
874  * unregisters. Caller of uprobe_register() is required to keep @inode
875  * (and the containing mount) referenced.
876  *
877  * Return errno if it cannot successully install probes
878  * else return 0 (success)
879  */
880 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
881 {
882         struct uprobe *uprobe;
883         int ret;
884 
885         /* Uprobe must have at least one set consumer */
886         if (!uc->handler && !uc->ret_handler)
887                 return -EINVAL;
888 
889         /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
890         if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
891                 return -EIO;
892         /* Racy, just to catch the obvious mistakes */
893         if (offset > i_size_read(inode))
894                 return -EINVAL;
895 
896  retry:
897         uprobe = alloc_uprobe(inode, offset);
898         if (!uprobe)
899                 return -ENOMEM;
900         /*
901          * We can race with uprobe_unregister()->delete_uprobe().
902          * Check uprobe_is_active() and retry if it is false.
903          */
904         down_write(&uprobe->register_rwsem);
905         ret = -EAGAIN;
906         if (likely(uprobe_is_active(uprobe))) {
907                 ret = __uprobe_register(uprobe, uc);
908                 if (ret)
909                         __uprobe_unregister(uprobe, uc);
910         }
911         up_write(&uprobe->register_rwsem);
912         put_uprobe(uprobe);
913 
914         if (unlikely(ret == -EAGAIN))
915                 goto retry;
916         return ret;
917 }
918 EXPORT_SYMBOL_GPL(uprobe_register);
919 
920 /*
921  * uprobe_apply - unregister a already registered probe.
922  * @inode: the file in which the probe has to be removed.
923  * @offset: offset from the start of the file.
924  * @uc: consumer which wants to add more or remove some breakpoints
925  * @add: add or remove the breakpoints
926  */
927 int uprobe_apply(struct inode *inode, loff_t offset,
928                         struct uprobe_consumer *uc, bool add)
929 {
930         struct uprobe *uprobe;
931         struct uprobe_consumer *con;
932         int ret = -ENOENT;
933 
934         uprobe = find_uprobe(inode, offset);
935         if (WARN_ON(!uprobe))
936                 return ret;
937 
938         down_write(&uprobe->register_rwsem);
939         for (con = uprobe->consumers; con && con != uc ; con = con->next)
940                 ;
941         if (con)
942                 ret = register_for_each_vma(uprobe, add ? uc : NULL);
943         up_write(&uprobe->register_rwsem);
944         put_uprobe(uprobe);
945 
946         return ret;
947 }
948 
949 /*
950  * uprobe_unregister - unregister a already registered probe.
951  * @inode: the file in which the probe has to be removed.
952  * @offset: offset from the start of the file.
953  * @uc: identify which probe if multiple probes are colocated.
954  */
955 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
956 {
957         struct uprobe *uprobe;
958 
959         uprobe = find_uprobe(inode, offset);
960         if (WARN_ON(!uprobe))
961                 return;
962 
963         down_write(&uprobe->register_rwsem);
964         __uprobe_unregister(uprobe, uc);
965         up_write(&uprobe->register_rwsem);
966         put_uprobe(uprobe);
967 }
968 EXPORT_SYMBOL_GPL(uprobe_unregister);
969 
970 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
971 {
972         struct vm_area_struct *vma;
973         int err = 0;
974 
975         down_read(&mm->mmap_sem);
976         for (vma = mm->mmap; vma; vma = vma->vm_next) {
977                 unsigned long vaddr;
978                 loff_t offset;
979 
980                 if (!valid_vma(vma, false) ||
981                     file_inode(vma->vm_file) != uprobe->inode)
982                         continue;
983 
984                 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
985                 if (uprobe->offset <  offset ||
986                     uprobe->offset >= offset + vma->vm_end - vma->vm_start)
987                         continue;
988 
989                 vaddr = offset_to_vaddr(vma, uprobe->offset);
990                 err |= remove_breakpoint(uprobe, mm, vaddr);
991         }
992         up_read(&mm->mmap_sem);
993 
994         return err;
995 }
996 
997 static struct rb_node *
998 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
999 {
1000         struct rb_node *n = uprobes_tree.rb_node;
1001 
1002         while (n) {
1003                 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1004 
1005                 if (inode < u->inode) {
1006                         n = n->rb_left;
1007                 } else if (inode > u->inode) {
1008                         n = n->rb_right;
1009                 } else {
1010                         if (max < u->offset)
1011                                 n = n->rb_left;
1012                         else if (min > u->offset)
1013                                 n = n->rb_right;
1014                         else
1015                                 break;
1016                 }
1017         }
1018 
1019         return n;
1020 }
1021 
1022 /*
1023  * For a given range in vma, build a list of probes that need to be inserted.
1024  */
1025 static void build_probe_list(struct inode *inode,
1026                                 struct vm_area_struct *vma,
1027                                 unsigned long start, unsigned long end,
1028                                 struct list_head *head)
1029 {
1030         loff_t min, max;
1031         struct rb_node *n, *t;
1032         struct uprobe *u;
1033 
1034         INIT_LIST_HEAD(head);
1035         min = vaddr_to_offset(vma, start);
1036         max = min + (end - start) - 1;
1037 
1038         spin_lock(&uprobes_treelock);
1039         n = find_node_in_range(inode, min, max);
1040         if (n) {
1041                 for (t = n; t; t = rb_prev(t)) {
1042                         u = rb_entry(t, struct uprobe, rb_node);
1043                         if (u->inode != inode || u->offset < min)
1044                                 break;
1045                         list_add(&u->pending_list, head);
1046                         get_uprobe(u);
1047                 }
1048                 for (t = n; (t = rb_next(t)); ) {
1049                         u = rb_entry(t, struct uprobe, rb_node);
1050                         if (u->inode != inode || u->offset > max)
1051                                 break;
1052                         list_add(&u->pending_list, head);
1053                         get_uprobe(u);
1054                 }
1055         }
1056         spin_unlock(&uprobes_treelock);
1057 }
1058 
1059 /*
1060  * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1061  *
1062  * Currently we ignore all errors and always return 0, the callers
1063  * can't handle the failure anyway.
1064  */
1065 int uprobe_mmap(struct vm_area_struct *vma)
1066 {
1067         struct list_head tmp_list;
1068         struct uprobe *uprobe, *u;
1069         struct inode *inode;
1070 
1071         if (no_uprobe_events() || !valid_vma(vma, true))
1072                 return 0;
1073 
1074         inode = file_inode(vma->vm_file);
1075         if (!inode)
1076                 return 0;
1077 
1078         mutex_lock(uprobes_mmap_hash(inode));
1079         build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1080         /*
1081          * We can race with uprobe_unregister(), this uprobe can be already
1082          * removed. But in this case filter_chain() must return false, all
1083          * consumers have gone away.
1084          */
1085         list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1086                 if (!fatal_signal_pending(current) &&
1087                     filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1088                         unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1089                         install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1090                 }
1091                 put_uprobe(uprobe);
1092         }
1093         mutex_unlock(uprobes_mmap_hash(inode));
1094 
1095         return 0;
1096 }
1097 
1098 static bool
1099 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1100 {
1101         loff_t min, max;
1102         struct inode *inode;
1103         struct rb_node *n;
1104 
1105         inode = file_inode(vma->vm_file);
1106 
1107         min = vaddr_to_offset(vma, start);
1108         max = min + (end - start) - 1;
1109 
1110         spin_lock(&uprobes_treelock);
1111         n = find_node_in_range(inode, min, max);
1112         spin_unlock(&uprobes_treelock);
1113 
1114         return !!n;
1115 }
1116 
1117 /*
1118  * Called in context of a munmap of a vma.
1119  */
1120 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1121 {
1122         if (no_uprobe_events() || !valid_vma(vma, false))
1123                 return;
1124 
1125         if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1126                 return;
1127 
1128         if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1129              test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1130                 return;
1131 
1132         if (vma_has_uprobes(vma, start, end))
1133                 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1134 }
1135 
1136 /* Slot allocation for XOL */
1137 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1138 {
1139         struct vm_area_struct *vma;
1140         int ret;
1141 
1142         if (down_write_killable(&mm->mmap_sem))
1143                 return -EINTR;
1144 
1145         if (mm->uprobes_state.xol_area) {
1146                 ret = -EALREADY;
1147                 goto fail;
1148         }
1149 
1150         if (!area->vaddr) {
1151                 /* Try to map as high as possible, this is only a hint. */
1152                 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1153                                                 PAGE_SIZE, 0, 0);
1154                 if (area->vaddr & ~PAGE_MASK) {
1155                         ret = area->vaddr;
1156                         goto fail;
1157                 }
1158         }
1159 
1160         vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1161                                 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1162                                 &area->xol_mapping);
1163         if (IS_ERR(vma)) {
1164                 ret = PTR_ERR(vma);
1165                 goto fail;
1166         }
1167 
1168         ret = 0;
1169         /* pairs with get_xol_area() */
1170         smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1171  fail:
1172         up_write(&mm->mmap_sem);
1173 
1174         return ret;
1175 }
1176 
1177 static struct xol_area *__create_xol_area(unsigned long vaddr)
1178 {
1179         struct mm_struct *mm = current->mm;
1180         uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1181         struct xol_area *area;
1182 
1183         area = kmalloc(sizeof(*area), GFP_KERNEL);
1184         if (unlikely(!area))
1185                 goto out;
1186 
1187         area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1188         if (!area->bitmap)
1189                 goto free_area;
1190 
1191         area->xol_mapping.name = "[uprobes]";
1192         area->xol_mapping.fault = NULL;
1193         area->xol_mapping.pages = area->pages;
1194         area->pages[0] = alloc_page(GFP_HIGHUSER);
1195         if (!area->pages[0])
1196                 goto free_bitmap;
1197         area->pages[1] = NULL;
1198 
1199         area->vaddr = vaddr;
1200         init_waitqueue_head(&area->wq);
1201         /* Reserve the 1st slot for get_trampoline_vaddr() */
1202         set_bit(0, area->bitmap);
1203         atomic_set(&area->slot_count, 1);
1204         arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1205 
1206         if (!xol_add_vma(mm, area))
1207                 return area;
1208 
1209         __free_page(area->pages[0]);
1210  free_bitmap:
1211         kfree(area->bitmap);
1212  free_area:
1213         kfree(area);
1214  out:
1215         return NULL;
1216 }
1217 
1218 /*
1219  * get_xol_area - Allocate process's xol_area if necessary.
1220  * This area will be used for storing instructions for execution out of line.
1221  *
1222  * Returns the allocated area or NULL.
1223  */
1224 static struct xol_area *get_xol_area(void)
1225 {
1226         struct mm_struct *mm = current->mm;
1227         struct xol_area *area;
1228 
1229         if (!mm->uprobes_state.xol_area)
1230                 __create_xol_area(0);
1231 
1232         /* Pairs with xol_add_vma() smp_store_release() */
1233         area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1234         return area;
1235 }
1236 
1237 /*
1238  * uprobe_clear_state - Free the area allocated for slots.
1239  */
1240 void uprobe_clear_state(struct mm_struct *mm)
1241 {
1242         struct xol_area *area = mm->uprobes_state.xol_area;
1243 
1244         if (!area)
1245                 return;
1246 
1247         put_page(area->pages[0]);
1248         kfree(area->bitmap);
1249         kfree(area);
1250 }
1251 
1252 void uprobe_start_dup_mmap(void)
1253 {
1254         percpu_down_read(&dup_mmap_sem);
1255 }
1256 
1257 void uprobe_end_dup_mmap(void)
1258 {
1259         percpu_up_read(&dup_mmap_sem);
1260 }
1261 
1262 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1263 {
1264         if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1265                 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1266                 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1267                 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1268         }
1269 }
1270 
1271 /*
1272  *  - search for a free slot.
1273  */
1274 static unsigned long xol_take_insn_slot(struct xol_area *area)
1275 {
1276         unsigned long slot_addr;
1277         int slot_nr;
1278 
1279         do {
1280                 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1281                 if (slot_nr < UINSNS_PER_PAGE) {
1282                         if (!test_and_set_bit(slot_nr, area->bitmap))
1283                                 break;
1284 
1285                         slot_nr = UINSNS_PER_PAGE;
1286                         continue;
1287                 }
1288                 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1289         } while (slot_nr >= UINSNS_PER_PAGE);
1290 
1291         slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1292         atomic_inc(&area->slot_count);
1293 
1294         return slot_addr;
1295 }
1296 
1297 /*
1298  * xol_get_insn_slot - allocate a slot for xol.
1299  * Returns the allocated slot address or 0.
1300  */
1301 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1302 {
1303         struct xol_area *area;
1304         unsigned long xol_vaddr;
1305 
1306         area = get_xol_area();
1307         if (!area)
1308                 return 0;
1309 
1310         xol_vaddr = xol_take_insn_slot(area);
1311         if (unlikely(!xol_vaddr))
1312                 return 0;
1313 
1314         arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1315                               &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1316 
1317         return xol_vaddr;
1318 }
1319 
1320 /*
1321  * xol_free_insn_slot - If slot was earlier allocated by
1322  * @xol_get_insn_slot(), make the slot available for
1323  * subsequent requests.
1324  */
1325 static void xol_free_insn_slot(struct task_struct *tsk)
1326 {
1327         struct xol_area *area;
1328         unsigned long vma_end;
1329         unsigned long slot_addr;
1330 
1331         if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1332                 return;
1333 
1334         slot_addr = tsk->utask->xol_vaddr;
1335         if (unlikely(!slot_addr))
1336                 return;
1337 
1338         area = tsk->mm->uprobes_state.xol_area;
1339         vma_end = area->vaddr + PAGE_SIZE;
1340         if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1341                 unsigned long offset;
1342                 int slot_nr;
1343 
1344                 offset = slot_addr - area->vaddr;
1345                 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1346                 if (slot_nr >= UINSNS_PER_PAGE)
1347                         return;
1348 
1349                 clear_bit(slot_nr, area->bitmap);
1350                 atomic_dec(&area->slot_count);
1351                 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1352                 if (waitqueue_active(&area->wq))
1353                         wake_up(&area->wq);
1354 
1355                 tsk->utask->xol_vaddr = 0;
1356         }
1357 }
1358 
1359 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1360                                   void *src, unsigned long len)
1361 {
1362         /* Initialize the slot */
1363         copy_to_page(page, vaddr, src, len);
1364 
1365         /*
1366          * We probably need flush_icache_user_range() but it needs vma.
1367          * This should work on most of architectures by default. If
1368          * architecture needs to do something different it can define
1369          * its own version of the function.
1370          */
1371         flush_dcache_page(page);
1372 }
1373 
1374 /**
1375  * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1376  * @regs: Reflects the saved state of the task after it has hit a breakpoint
1377  * instruction.
1378  * Return the address of the breakpoint instruction.
1379  */
1380 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1381 {
1382         return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1383 }
1384 
1385 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1386 {
1387         struct uprobe_task *utask = current->utask;
1388 
1389         if (unlikely(utask && utask->active_uprobe))
1390                 return utask->vaddr;
1391 
1392         return instruction_pointer(regs);
1393 }
1394 
1395 static struct return_instance *free_ret_instance(struct return_instance *ri)
1396 {
1397         struct return_instance *next = ri->next;
1398         put_uprobe(ri->uprobe);
1399         kfree(ri);
1400         return next;
1401 }
1402 
1403 /*
1404  * Called with no locks held.
1405  * Called in context of a exiting or a exec-ing thread.
1406  */
1407 void uprobe_free_utask(struct task_struct *t)
1408 {
1409         struct uprobe_task *utask = t->utask;
1410         struct return_instance *ri;
1411 
1412         if (!utask)
1413                 return;
1414 
1415         if (utask->active_uprobe)
1416                 put_uprobe(utask->active_uprobe);
1417 
1418         ri = utask->return_instances;
1419         while (ri)
1420                 ri = free_ret_instance(ri);
1421 
1422         xol_free_insn_slot(t);
1423         kfree(utask);
1424         t->utask = NULL;
1425 }
1426 
1427 /*
1428  * Allocate a uprobe_task object for the task if if necessary.
1429  * Called when the thread hits a breakpoint.
1430  *
1431  * Returns:
1432  * - pointer to new uprobe_task on success
1433  * - NULL otherwise
1434  */
1435 static struct uprobe_task *get_utask(void)
1436 {
1437         if (!current->utask)
1438                 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1439         return current->utask;
1440 }
1441 
1442 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1443 {
1444         struct uprobe_task *n_utask;
1445         struct return_instance **p, *o, *n;
1446 
1447         n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1448         if (!n_utask)
1449                 return -ENOMEM;
1450         t->utask = n_utask;
1451 
1452         p = &n_utask->return_instances;
1453         for (o = o_utask->return_instances; o; o = o->next) {
1454                 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1455                 if (!n)
1456                         return -ENOMEM;
1457 
1458                 *n = *o;
1459                 get_uprobe(n->uprobe);
1460                 n->next = NULL;
1461 
1462                 *p = n;
1463                 p = &n->next;
1464                 n_utask->depth++;
1465         }
1466 
1467         return 0;
1468 }
1469 
1470 static void uprobe_warn(struct task_struct *t, const char *msg)
1471 {
1472         pr_warn("uprobe: %s:%d failed to %s\n",
1473                         current->comm, current->pid, msg);
1474 }
1475 
1476 static void dup_xol_work(struct callback_head *work)
1477 {
1478         if (current->flags & PF_EXITING)
1479                 return;
1480 
1481         if (!__create_xol_area(current->utask->dup_xol_addr) &&
1482                         !fatal_signal_pending(current))
1483                 uprobe_warn(current, "dup xol area");
1484 }
1485 
1486 /*
1487  * Called in context of a new clone/fork from copy_process.
1488  */
1489 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1490 {
1491         struct uprobe_task *utask = current->utask;
1492         struct mm_struct *mm = current->mm;
1493         struct xol_area *area;
1494 
1495         t->utask = NULL;
1496 
1497         if (!utask || !utask->return_instances)
1498                 return;
1499 
1500         if (mm == t->mm && !(flags & CLONE_VFORK))
1501                 return;
1502 
1503         if (dup_utask(t, utask))
1504                 return uprobe_warn(t, "dup ret instances");
1505 
1506         /* The task can fork() after dup_xol_work() fails */
1507         area = mm->uprobes_state.xol_area;
1508         if (!area)
1509                 return uprobe_warn(t, "dup xol area");
1510 
1511         if (mm == t->mm)
1512                 return;
1513 
1514         t->utask->dup_xol_addr = area->vaddr;
1515         init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1516         task_work_add(t, &t->utask->dup_xol_work, true);
1517 }
1518 
1519 /*
1520  * Current area->vaddr notion assume the trampoline address is always
1521  * equal area->vaddr.
1522  *
1523  * Returns -1 in case the xol_area is not allocated.
1524  */
1525 static unsigned long get_trampoline_vaddr(void)
1526 {
1527         struct xol_area *area;
1528         unsigned long trampoline_vaddr = -1;
1529 
1530         /* Pairs with xol_add_vma() smp_store_release() */
1531         area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1532         if (area)
1533                 trampoline_vaddr = area->vaddr;
1534 
1535         return trampoline_vaddr;
1536 }
1537 
1538 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1539                                         struct pt_regs *regs)
1540 {
1541         struct return_instance *ri = utask->return_instances;
1542         enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1543 
1544         while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1545                 ri = free_ret_instance(ri);
1546                 utask->depth--;
1547         }
1548         utask->return_instances = ri;
1549 }
1550 
1551 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1552 {
1553         struct return_instance *ri;
1554         struct uprobe_task *utask;
1555         unsigned long orig_ret_vaddr, trampoline_vaddr;
1556         bool chained;
1557 
1558         if (!get_xol_area())
1559                 return;
1560 
1561         utask = get_utask();
1562         if (!utask)
1563                 return;
1564 
1565         if (utask->depth >= MAX_URETPROBE_DEPTH) {
1566                 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1567                                 " nestedness limit pid/tgid=%d/%d\n",
1568                                 current->pid, current->tgid);
1569                 return;
1570         }
1571 
1572         ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1573         if (!ri)
1574                 return;
1575 
1576         trampoline_vaddr = get_trampoline_vaddr();
1577         orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1578         if (orig_ret_vaddr == -1)
1579                 goto fail;
1580 
1581         /* drop the entries invalidated by longjmp() */
1582         chained = (orig_ret_vaddr == trampoline_vaddr);
1583         cleanup_return_instances(utask, chained, regs);
1584 
1585         /*
1586          * We don't want to keep trampoline address in stack, rather keep the
1587          * original return address of first caller thru all the consequent
1588          * instances. This also makes breakpoint unwrapping easier.
1589          */
1590         if (chained) {
1591                 if (!utask->return_instances) {
1592                         /*
1593                          * This situation is not possible. Likely we have an
1594                          * attack from user-space.
1595                          */
1596                         uprobe_warn(current, "handle tail call");
1597                         goto fail;
1598                 }
1599                 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1600         }
1601 
1602         ri->uprobe = get_uprobe(uprobe);
1603         ri->func = instruction_pointer(regs);
1604         ri->stack = user_stack_pointer(regs);
1605         ri->orig_ret_vaddr = orig_ret_vaddr;
1606         ri->chained = chained;
1607 
1608         utask->depth++;
1609         ri->next = utask->return_instances;
1610         utask->return_instances = ri;
1611 
1612         return;
1613  fail:
1614         kfree(ri);
1615 }
1616 
1617 /* Prepare to single-step probed instruction out of line. */
1618 static int
1619 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1620 {
1621         struct uprobe_task *utask;
1622         unsigned long xol_vaddr;
1623         int err;
1624 
1625         utask = get_utask();
1626         if (!utask)
1627                 return -ENOMEM;
1628 
1629         xol_vaddr = xol_get_insn_slot(uprobe);
1630         if (!xol_vaddr)
1631                 return -ENOMEM;
1632 
1633         utask->xol_vaddr = xol_vaddr;
1634         utask->vaddr = bp_vaddr;
1635 
1636         err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1637         if (unlikely(err)) {
1638                 xol_free_insn_slot(current);
1639                 return err;
1640         }
1641 
1642         utask->active_uprobe = uprobe;
1643         utask->state = UTASK_SSTEP;
1644         return 0;
1645 }
1646 
1647 /*
1648  * If we are singlestepping, then ensure this thread is not connected to
1649  * non-fatal signals until completion of singlestep.  When xol insn itself
1650  * triggers the signal,  restart the original insn even if the task is
1651  * already SIGKILL'ed (since coredump should report the correct ip).  This
1652  * is even more important if the task has a handler for SIGSEGV/etc, The
1653  * _same_ instruction should be repeated again after return from the signal
1654  * handler, and SSTEP can never finish in this case.
1655  */
1656 bool uprobe_deny_signal(void)
1657 {
1658         struct task_struct *t = current;
1659         struct uprobe_task *utask = t->utask;
1660 
1661         if (likely(!utask || !utask->active_uprobe))
1662                 return false;
1663 
1664         WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1665 
1666         if (signal_pending(t)) {
1667                 spin_lock_irq(&t->sighand->siglock);
1668                 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1669                 spin_unlock_irq(&t->sighand->siglock);
1670 
1671                 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1672                         utask->state = UTASK_SSTEP_TRAPPED;
1673                         set_tsk_thread_flag(t, TIF_UPROBE);
1674                 }
1675         }
1676 
1677         return true;
1678 }
1679 
1680 static void mmf_recalc_uprobes(struct mm_struct *mm)
1681 {
1682         struct vm_area_struct *vma;
1683 
1684         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1685                 if (!valid_vma(vma, false))
1686                         continue;
1687                 /*
1688                  * This is not strictly accurate, we can race with
1689                  * uprobe_unregister() and see the already removed
1690                  * uprobe if delete_uprobe() was not yet called.
1691                  * Or this uprobe can be filtered out.
1692                  */
1693                 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1694                         return;
1695         }
1696 
1697         clear_bit(MMF_HAS_UPROBES, &mm->flags);
1698 }
1699 
1700 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
1701 {
1702         struct page *page;
1703         uprobe_opcode_t opcode;
1704         int result;
1705 
1706         pagefault_disable();
1707         result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
1708         pagefault_enable();
1709 
1710         if (likely(result == 0))
1711                 goto out;
1712 
1713         /*
1714          * The NULL 'tsk' here ensures that any faults that occur here
1715          * will not be accounted to the task.  'mm' *is* current->mm,
1716          * but we treat this as a 'remote' access since it is
1717          * essentially a kernel access to the memory.
1718          */
1719         result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page,
1720                         NULL, NULL);
1721         if (result < 0)
1722                 return result;
1723 
1724         copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
1725         put_page(page);
1726  out:
1727         /* This needs to return true for any variant of the trap insn */
1728         return is_trap_insn(&opcode);
1729 }
1730 
1731 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1732 {
1733         struct mm_struct *mm = current->mm;
1734         struct uprobe *uprobe = NULL;
1735         struct vm_area_struct *vma;
1736 
1737         down_read(&mm->mmap_sem);
1738         vma = find_vma(mm, bp_vaddr);
1739         if (vma && vma->vm_start <= bp_vaddr) {
1740                 if (valid_vma(vma, false)) {
1741                         struct inode *inode = file_inode(vma->vm_file);
1742                         loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1743 
1744                         uprobe = find_uprobe(inode, offset);
1745                 }
1746 
1747                 if (!uprobe)
1748                         *is_swbp = is_trap_at_addr(mm, bp_vaddr);
1749         } else {
1750                 *is_swbp = -EFAULT;
1751         }
1752 
1753         if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1754                 mmf_recalc_uprobes(mm);
1755         up_read(&mm->mmap_sem);
1756 
1757         return uprobe;
1758 }
1759 
1760 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
1761 {
1762         struct uprobe_consumer *uc;
1763         int remove = UPROBE_HANDLER_REMOVE;
1764         bool need_prep = false; /* prepare return uprobe, when needed */
1765 
1766         down_read(&uprobe->register_rwsem);
1767         for (uc = uprobe->consumers; uc; uc = uc->next) {
1768                 int rc = 0;
1769 
1770                 if (uc->handler) {
1771                         rc = uc->handler(uc, regs);
1772                         WARN(rc & ~UPROBE_HANDLER_MASK,
1773                                 "bad rc=0x%x from %pf()\n", rc, uc->handler);
1774                 }
1775 
1776                 if (uc->ret_handler)
1777                         need_prep = true;
1778 
1779                 remove &= rc;
1780         }
1781 
1782         if (need_prep && !remove)
1783                 prepare_uretprobe(uprobe, regs); /* put bp at return */
1784 
1785         if (remove && uprobe->consumers) {
1786                 WARN_ON(!uprobe_is_active(uprobe));
1787                 unapply_uprobe(uprobe, current->mm);
1788         }
1789         up_read(&uprobe->register_rwsem);
1790 }
1791 
1792 static void
1793 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
1794 {
1795         struct uprobe *uprobe = ri->uprobe;
1796         struct uprobe_consumer *uc;
1797 
1798         down_read(&uprobe->register_rwsem);
1799         for (uc = uprobe->consumers; uc; uc = uc->next) {
1800                 if (uc->ret_handler)
1801                         uc->ret_handler(uc, ri->func, regs);
1802         }
1803         up_read(&uprobe->register_rwsem);
1804 }
1805 
1806 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
1807 {
1808         bool chained;
1809 
1810         do {
1811                 chained = ri->chained;
1812                 ri = ri->next;  /* can't be NULL if chained */
1813         } while (chained);
1814 
1815         return ri;
1816 }
1817 
1818 static void handle_trampoline(struct pt_regs *regs)
1819 {
1820         struct uprobe_task *utask;
1821         struct return_instance *ri, *next;
1822         bool valid;
1823 
1824         utask = current->utask;
1825         if (!utask)
1826                 goto sigill;
1827 
1828         ri = utask->return_instances;
1829         if (!ri)
1830                 goto sigill;
1831 
1832         do {
1833                 /*
1834                  * We should throw out the frames invalidated by longjmp().
1835                  * If this chain is valid, then the next one should be alive
1836                  * or NULL; the latter case means that nobody but ri->func
1837                  * could hit this trampoline on return. TODO: sigaltstack().
1838                  */
1839                 next = find_next_ret_chain(ri);
1840                 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
1841 
1842                 instruction_pointer_set(regs, ri->orig_ret_vaddr);
1843                 do {
1844                         if (valid)
1845                                 handle_uretprobe_chain(ri, regs);
1846                         ri = free_ret_instance(ri);
1847                         utask->depth--;
1848                 } while (ri != next);
1849         } while (!valid);
1850 
1851         utask->return_instances = ri;
1852         return;
1853 
1854  sigill:
1855         uprobe_warn(current, "handle uretprobe, sending SIGILL.");
1856         force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1857 
1858 }
1859 
1860 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
1861 {
1862         return false;
1863 }
1864 
1865 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
1866                                         struct pt_regs *regs)
1867 {
1868         return true;
1869 }
1870 
1871 /*
1872  * Run handler and ask thread to singlestep.
1873  * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1874  */
1875 static void handle_swbp(struct pt_regs *regs)
1876 {
1877         struct uprobe *uprobe;
1878         unsigned long bp_vaddr;
1879         int uninitialized_var(is_swbp);
1880 
1881         bp_vaddr = uprobe_get_swbp_addr(regs);
1882         if (bp_vaddr == get_trampoline_vaddr())
1883                 return handle_trampoline(regs);
1884 
1885         uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1886         if (!uprobe) {
1887                 if (is_swbp > 0) {
1888                         /* No matching uprobe; signal SIGTRAP. */
1889                         send_sig(SIGTRAP, current, 0);
1890                 } else {
1891                         /*
1892                          * Either we raced with uprobe_unregister() or we can't
1893                          * access this memory. The latter is only possible if
1894                          * another thread plays with our ->mm. In both cases
1895                          * we can simply restart. If this vma was unmapped we
1896                          * can pretend this insn was not executed yet and get
1897                          * the (correct) SIGSEGV after restart.
1898                          */
1899                         instruction_pointer_set(regs, bp_vaddr);
1900                 }
1901                 return;
1902         }
1903 
1904         /* change it in advance for ->handler() and restart */
1905         instruction_pointer_set(regs, bp_vaddr);
1906 
1907         /*
1908          * TODO: move copy_insn/etc into _register and remove this hack.
1909          * After we hit the bp, _unregister + _register can install the
1910          * new and not-yet-analyzed uprobe at the same address, restart.
1911          */
1912         smp_rmb(); /* pairs with wmb() in install_breakpoint() */
1913         if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
1914                 goto out;
1915 
1916         /* Tracing handlers use ->utask to communicate with fetch methods */
1917         if (!get_utask())
1918                 goto out;
1919 
1920         if (arch_uprobe_ignore(&uprobe->arch, regs))
1921                 goto out;
1922 
1923         handler_chain(uprobe, regs);
1924 
1925         if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1926                 goto out;
1927 
1928         if (!pre_ssout(uprobe, regs, bp_vaddr))
1929                 return;
1930 
1931         /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
1932 out:
1933         put_uprobe(uprobe);
1934 }
1935 
1936 /*
1937  * Perform required fix-ups and disable singlestep.
1938  * Allow pending signals to take effect.
1939  */
1940 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1941 {
1942         struct uprobe *uprobe;
1943         int err = 0;
1944 
1945         uprobe = utask->active_uprobe;
1946         if (utask->state == UTASK_SSTEP_ACK)
1947                 err = arch_uprobe_post_xol(&uprobe->arch, regs);
1948         else if (utask->state == UTASK_SSTEP_TRAPPED)
1949                 arch_uprobe_abort_xol(&uprobe->arch, regs);
1950         else
1951                 WARN_ON_ONCE(1);
1952 
1953         put_uprobe(uprobe);
1954         utask->active_uprobe = NULL;
1955         utask->state = UTASK_RUNNING;
1956         xol_free_insn_slot(current);
1957 
1958         spin_lock_irq(&current->sighand->siglock);
1959         recalc_sigpending(); /* see uprobe_deny_signal() */
1960         spin_unlock_irq(&current->sighand->siglock);
1961 
1962         if (unlikely(err)) {
1963                 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
1964                 force_sig_info(SIGILL, SEND_SIG_FORCED, current);
1965         }
1966 }
1967 
1968 /*
1969  * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
1970  * allows the thread to return from interrupt. After that handle_swbp()
1971  * sets utask->active_uprobe.
1972  *
1973  * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
1974  * and allows the thread to return from interrupt.
1975  *
1976  * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1977  * uprobe_notify_resume().
1978  */
1979 void uprobe_notify_resume(struct pt_regs *regs)
1980 {
1981         struct uprobe_task *utask;
1982 
1983         clear_thread_flag(TIF_UPROBE);
1984 
1985         utask = current->utask;
1986         if (utask && utask->active_uprobe)
1987                 handle_singlestep(utask, regs);
1988         else
1989                 handle_swbp(regs);
1990 }
1991 
1992 /*
1993  * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1994  * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1995  */
1996 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1997 {
1998         if (!current->mm)
1999                 return 0;
2000 
2001         if (!test_bit(MMF_HAS_UPROBES, &current->mm->flags) &&
2002             (!current->utask || !current->utask->return_instances))
2003                 return 0;
2004 
2005         set_thread_flag(TIF_UPROBE);
2006         return 1;
2007 }
2008 
2009 /*
2010  * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2011  * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2012  */
2013 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2014 {
2015         struct uprobe_task *utask = current->utask;
2016 
2017         if (!current->mm || !utask || !utask->active_uprobe)
2018                 /* task is currently not uprobed */
2019                 return 0;
2020 
2021         utask->state = UTASK_SSTEP_ACK;
2022         set_thread_flag(TIF_UPROBE);
2023         return 1;
2024 }
2025 
2026 static struct notifier_block uprobe_exception_nb = {
2027         .notifier_call          = arch_uprobe_exception_notify,
2028         .priority               = INT_MAX-1,    /* notified after kprobes, kgdb */
2029 };
2030 
2031 static int __init init_uprobes(void)
2032 {
2033         int i;
2034 
2035         for (i = 0; i < UPROBES_HASH_SZ; i++)
2036                 mutex_init(&uprobes_mmap_mutex[i]);
2037 
2038         if (percpu_init_rwsem(&dup_mmap_sem))
2039                 return -ENOMEM;
2040 
2041         return register_die_notifier(&uprobe_exception_nb);
2042 }
2043 __initcall(init_uprobes);
2044 

~ [ 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