2 * User-space Probes (UProbes)
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright (C) IBM Corporation, 2008-2012
22 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
25 #include <linux/kernel.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h> /* read_mapping_page */
28 #include <linux/slab.h>
29 #include <linux/sched.h>
30 #include <linux/rmap.h> /* anon_vma_prepare */
31 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
32 #include <linux/swap.h> /* try_to_free_swap */
33 #include <linux/ptrace.h> /* user_enable_single_step */
34 #include <linux/kdebug.h> /* notifier mechanism */
36 #include <linux/uprobes.h>
38 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
39 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
41 static struct rb_root uprobes_tree = RB_ROOT;
43 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
45 #define UPROBES_HASH_SZ 13
47 /* serialize (un)register */
48 static struct mutex uprobes_mutex[UPROBES_HASH_SZ];
50 #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
52 /* serialize uprobe->pending_list */
53 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
54 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
57 * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe
58 * events active at this time. Probably a fine grained per inode count is
61 static atomic_t uprobe_events = ATOMIC_INIT(0);
64 * Maintain a temporary per vma info that can be used to search if a vma
65 * has already been handled. This structure is introduced since extending
66 * vm_area_struct wasnt recommended.
69 struct list_head probe_list;
75 struct rb_node rb_node; /* node in the rb tree */
77 struct rw_semaphore consumer_rwsem;
78 struct list_head pending_list;
79 struct uprobe_consumer *consumers;
80 struct inode *inode; /* Also hold a ref to inode */
83 struct arch_uprobe arch;
87 * valid_vma: Verify if the specified vma is an executable vma
88 * Relax restrictions while unregistering: vm_flags might have
89 * changed after breakpoint was inserted.
90 * - is_register: indicates if we are in register context.
91 * - Return 1 if the specified virtual address is in an
94 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
102 if ((vma->vm_flags & (VM_HUGETLB|VM_READ|VM_WRITE|VM_EXEC|VM_SHARED))
103 == (VM_READ|VM_EXEC))
109 static loff_t vma_address(struct vm_area_struct *vma, loff_t offset)
113 vaddr = vma->vm_start + offset;
114 vaddr -= vma->vm_pgoff << PAGE_SHIFT;
120 * __replace_page - replace page in vma by new page.
121 * based on replace_page in mm/ksm.c
123 * @vma: vma that holds the pte pointing to page
124 * @page: the cowed page we are replacing by kpage
125 * @kpage: the modified page we replace page by
127 * Returns 0 on success, -EFAULT on failure.
129 static int __replace_page(struct vm_area_struct *vma, struct page *page, struct page *kpage)
131 struct mm_struct *mm = vma->vm_mm;
140 addr = page_address_in_vma(page, vma);
144 pgd = pgd_offset(mm, addr);
145 if (!pgd_present(*pgd))
148 pud = pud_offset(pgd, addr);
149 if (!pud_present(*pud))
152 pmd = pmd_offset(pud, addr);
153 if (!pmd_present(*pmd))
156 ptep = pte_offset_map_lock(mm, pmd, addr, &ptl);
161 page_add_new_anon_rmap(kpage, vma, addr);
163 if (!PageAnon(page)) {
164 dec_mm_counter(mm, MM_FILEPAGES);
165 inc_mm_counter(mm, MM_ANONPAGES);
168 flush_cache_page(vma, addr, pte_pfn(*ptep));
169 ptep_clear_flush(vma, addr, ptep);
170 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
172 page_remove_rmap(page);
173 if (!page_mapped(page))
174 try_to_free_swap(page);
176 pte_unmap_unlock(ptep, ptl);
184 * is_swbp_insn - check if instruction is breakpoint instruction.
185 * @insn: instruction to be checked.
186 * Default implementation of is_swbp_insn
187 * Returns true if @insn is a breakpoint instruction.
189 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
191 return *insn == UPROBE_SWBP_INSN;
196 * Expect the breakpoint instruction to be the smallest size instruction for
197 * the architecture. If an arch has variable length instruction and the
198 * breakpoint instruction is not of the smallest length instruction
199 * supported by that architecture then we need to modify read_opcode /
200 * write_opcode accordingly. This would never be a problem for archs that
201 * have fixed length instructions.
205 * write_opcode - write the opcode at a given virtual address.
206 * @auprobe: arch breakpointing information.
207 * @mm: the probed process address space.
208 * @vaddr: the virtual address to store the opcode.
209 * @opcode: opcode to be written at @vaddr.
211 * Called with mm->mmap_sem held (for read and with a reference to
214 * For mm @mm, write the opcode at @vaddr.
215 * Return 0 (success) or a negative errno.
217 static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
218 unsigned long vaddr, uprobe_opcode_t opcode)
220 struct page *old_page, *new_page;
221 struct address_space *mapping;
222 void *vaddr_old, *vaddr_new;
223 struct vm_area_struct *vma;
224 struct uprobe *uprobe;
228 /* Read the page with vaddr into memory */
229 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
236 * We are interested in text pages only. Our pages of interest
237 * should be mapped for read and execute only. We desist from
238 * adding probes in write mapped pages since the breakpoints
239 * might end up in the file copy.
241 if (!valid_vma(vma, is_swbp_insn(&opcode)))
244 uprobe = container_of(auprobe, struct uprobe, arch);
245 mapping = uprobe->inode->i_mapping;
246 if (mapping != vma->vm_file->f_mapping)
249 addr = vma_address(vma, uprobe->offset);
250 if (vaddr != (unsigned long)addr)
254 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
258 __SetPageUptodate(new_page);
261 * lock page will serialize against do_wp_page()'s
262 * PageAnon() handling
265 /* copy the page now that we've got it stable */
266 vaddr_old = kmap_atomic(old_page);
267 vaddr_new = kmap_atomic(new_page);
269 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
271 /* poke the new insn in, ASSUMES we don't cross page boundary */
273 BUG_ON(vaddr + UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
274 memcpy(vaddr_new + vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
276 kunmap_atomic(vaddr_new);
277 kunmap_atomic(vaddr_old);
279 ret = anon_vma_prepare(vma);
284 ret = __replace_page(vma, old_page, new_page);
285 unlock_page(new_page);
288 unlock_page(old_page);
289 page_cache_release(new_page);
298 * read_opcode - read the opcode at a given virtual address.
299 * @mm: the probed process address space.
300 * @vaddr: the virtual address to read the opcode.
301 * @opcode: location to store the read opcode.
303 * Called with mm->mmap_sem held (for read and with a reference to
306 * For mm @mm, read the opcode at @vaddr and store it in @opcode.
307 * Return 0 (success) or a negative errno.
309 static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode)
315 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
320 vaddr_new = kmap_atomic(page);
322 memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE);
323 kunmap_atomic(vaddr_new);
331 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
333 uprobe_opcode_t opcode;
336 if (current->mm == mm) {
338 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
342 if (likely(result == 0))
346 result = read_opcode(mm, vaddr, &opcode);
350 if (is_swbp_insn(&opcode))
357 * set_swbp - store breakpoint at a given address.
358 * @auprobe: arch specific probepoint information.
359 * @mm: the probed process address space.
360 * @vaddr: the virtual address to insert the opcode.
362 * For mm @mm, store the breakpoint instruction at @vaddr.
363 * Return 0 (success) or a negative errno.
365 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
369 result = is_swbp_at_addr(mm, vaddr);
376 return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
380 * set_orig_insn - Restore the original instruction.
381 * @mm: the probed process address space.
382 * @auprobe: arch specific probepoint information.
383 * @vaddr: the virtual address to insert the opcode.
384 * @verify: if true, verify existance of breakpoint instruction.
386 * For mm @mm, restore the original opcode (opcode) at @vaddr.
387 * Return 0 (success) or a negative errno.
390 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify)
395 result = is_swbp_at_addr(mm, vaddr);
402 return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
405 static int match_uprobe(struct uprobe *l, struct uprobe *r)
407 if (l->inode < r->inode)
410 if (l->inode > r->inode)
413 if (l->offset < r->offset)
416 if (l->offset > r->offset)
422 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
424 struct uprobe u = { .inode = inode, .offset = offset };
425 struct rb_node *n = uprobes_tree.rb_node;
426 struct uprobe *uprobe;
430 uprobe = rb_entry(n, struct uprobe, rb_node);
431 match = match_uprobe(&u, uprobe);
433 atomic_inc(&uprobe->ref);
446 * Find a uprobe corresponding to a given inode:offset
447 * Acquires uprobes_treelock
449 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
451 struct uprobe *uprobe;
454 spin_lock_irqsave(&uprobes_treelock, flags);
455 uprobe = __find_uprobe(inode, offset);
456 spin_unlock_irqrestore(&uprobes_treelock, flags);
461 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
463 struct rb_node **p = &uprobes_tree.rb_node;
464 struct rb_node *parent = NULL;
470 u = rb_entry(parent, struct uprobe, rb_node);
471 match = match_uprobe(uprobe, u);
478 p = &parent->rb_left;
480 p = &parent->rb_right;
485 rb_link_node(&uprobe->rb_node, parent, p);
486 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
487 /* get access + creation ref */
488 atomic_set(&uprobe->ref, 2);
494 * Acquire uprobes_treelock.
495 * Matching uprobe already exists in rbtree;
496 * increment (access refcount) and return the matching uprobe.
498 * No matching uprobe; insert the uprobe in rb_tree;
499 * get a double refcount (access + creation) and return NULL.
501 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
506 spin_lock_irqsave(&uprobes_treelock, flags);
507 u = __insert_uprobe(uprobe);
508 spin_unlock_irqrestore(&uprobes_treelock, flags);
510 /* For now assume that the instruction need not be single-stepped */
511 uprobe->flags |= UPROBE_SKIP_SSTEP;
516 static void put_uprobe(struct uprobe *uprobe)
518 if (atomic_dec_and_test(&uprobe->ref))
522 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
524 struct uprobe *uprobe, *cur_uprobe;
526 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
530 uprobe->inode = igrab(inode);
531 uprobe->offset = offset;
532 init_rwsem(&uprobe->consumer_rwsem);
533 INIT_LIST_HEAD(&uprobe->pending_list);
535 /* add to uprobes_tree, sorted on inode:offset */
536 cur_uprobe = insert_uprobe(uprobe);
538 /* a uprobe exists for this inode:offset combination */
544 atomic_inc(&uprobe_events);
550 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
552 struct uprobe_consumer *uc;
554 if (!(uprobe->flags & UPROBE_RUN_HANDLER))
557 down_read(&uprobe->consumer_rwsem);
558 for (uc = uprobe->consumers; uc; uc = uc->next) {
559 if (!uc->filter || uc->filter(uc, current))
560 uc->handler(uc, regs);
562 up_read(&uprobe->consumer_rwsem);
565 /* Returns the previous consumer */
566 static struct uprobe_consumer *
567 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
569 down_write(&uprobe->consumer_rwsem);
570 uc->next = uprobe->consumers;
571 uprobe->consumers = uc;
572 up_write(&uprobe->consumer_rwsem);
578 * For uprobe @uprobe, delete the consumer @uc.
579 * Return true if the @uc is deleted successfully
582 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
584 struct uprobe_consumer **con;
587 down_write(&uprobe->consumer_rwsem);
588 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
595 up_write(&uprobe->consumer_rwsem);
601 __copy_insn(struct address_space *mapping, struct vm_area_struct *vma, char *insn,
602 unsigned long nbytes, unsigned long offset)
604 struct file *filp = vma->vm_file;
613 if (!mapping->a_ops->readpage)
616 idx = (unsigned long)(offset >> PAGE_CACHE_SHIFT);
617 off1 = offset &= ~PAGE_MASK;
620 * Ensure that the page that has the original instruction is
621 * populated and in page-cache.
623 page = read_mapping_page(mapping, idx, filp);
625 return PTR_ERR(page);
627 vaddr = kmap_atomic(page);
628 memcpy(insn, vaddr + off1, nbytes);
629 kunmap_atomic(vaddr);
630 page_cache_release(page);
636 copy_insn(struct uprobe *uprobe, struct vm_area_struct *vma, unsigned long addr)
638 struct address_space *mapping;
639 unsigned long nbytes;
643 nbytes = PAGE_SIZE - addr;
644 mapping = uprobe->inode->i_mapping;
646 /* Instruction at end of binary; copy only available bytes */
647 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
648 bytes = uprobe->inode->i_size - uprobe->offset;
650 bytes = MAX_UINSN_BYTES;
652 /* Instruction at the page-boundary; copy bytes in second page */
653 if (nbytes < bytes) {
654 if (__copy_insn(mapping, vma, uprobe->arch.insn + nbytes,
655 bytes - nbytes, uprobe->offset + nbytes))
660 return __copy_insn(mapping, vma, uprobe->arch.insn, bytes, uprobe->offset);
664 * How mm->uprobes_state.count gets updated
665 * uprobe_mmap() increments the count if
666 * - it successfully adds a breakpoint.
667 * - it cannot add a breakpoint, but sees that there is a underlying
668 * breakpoint (via a is_swbp_at_addr()).
670 * uprobe_munmap() decrements the count if
671 * - it sees a underlying breakpoint, (via is_swbp_at_addr)
672 * (Subsequent uprobe_unregister wouldnt find the breakpoint
673 * unless a uprobe_mmap kicks in, since the old vma would be
674 * dropped just after uprobe_munmap.)
676 * uprobe_register increments the count if:
677 * - it successfully adds a breakpoint.
679 * uprobe_unregister decrements the count if:
680 * - it sees a underlying breakpoint and removes successfully.
681 * (via is_swbp_at_addr)
682 * (Subsequent uprobe_munmap wouldnt find the breakpoint
683 * since there is no underlying breakpoint after the
684 * breakpoint removal.)
687 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
688 struct vm_area_struct *vma, loff_t vaddr)
694 * If probe is being deleted, unregister thread could be done with
695 * the vma-rmap-walk through. Adding a probe now can be fatal since
696 * nobody will be able to cleanup. Also we could be from fork or
697 * mremap path, where the probe might have already been inserted.
698 * Hence behave as if probe already existed.
700 if (!uprobe->consumers)
703 addr = (unsigned long)vaddr;
705 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
706 ret = copy_insn(uprobe, vma, addr);
710 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
713 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, addr);
717 uprobe->flags |= UPROBE_COPY_INSN;
721 * Ideally, should be updating the probe count after the breakpoint
722 * has been successfully inserted. However a thread could hit the
723 * breakpoint we just inserted even before the probe count is
724 * incremented. If this is the first breakpoint placed, breakpoint
725 * notifier might ignore uprobes and pass the trap to the thread.
726 * Hence increment before and decrement on failure.
728 atomic_inc(&mm->uprobes_state.count);
729 ret = set_swbp(&uprobe->arch, mm, addr);
731 atomic_dec(&mm->uprobes_state.count);
737 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, loff_t vaddr)
739 if (!set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true))
740 atomic_dec(&mm->uprobes_state.count);
744 * There could be threads that have already hit the breakpoint. They
745 * will recheck the current insn and restart if find_uprobe() fails.
746 * See find_active_uprobe().
748 static void delete_uprobe(struct uprobe *uprobe)
752 spin_lock_irqsave(&uprobes_treelock, flags);
753 rb_erase(&uprobe->rb_node, &uprobes_tree);
754 spin_unlock_irqrestore(&uprobes_treelock, flags);
757 atomic_dec(&uprobe_events);
760 static struct vma_info *
761 __find_next_vma_info(struct address_space *mapping, struct list_head *head,
762 struct vma_info *vi, loff_t offset, bool is_register)
764 struct prio_tree_iter iter;
765 struct vm_area_struct *vma;
766 struct vma_info *tmpvi;
771 pgoff = offset >> PAGE_SHIFT;
773 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
774 if (!valid_vma(vma, is_register))
778 vaddr = vma_address(vma, offset);
780 list_for_each_entry(tmpvi, head, probe_list) {
781 if (tmpvi->mm == vma->vm_mm && tmpvi->vaddr == vaddr) {
788 * Another vma needs a probe to be installed. However skip
789 * installing the probe if the vma is about to be unlinked.
791 if (!existing_vma && atomic_inc_not_zero(&vma->vm_mm->mm_users)) {
794 list_add(&vi->probe_list, head);
804 * Iterate in the rmap prio tree and find a vma where a probe has not
807 static struct vma_info *
808 find_next_vma_info(struct address_space *mapping, struct list_head *head,
809 loff_t offset, bool is_register)
811 struct vma_info *vi, *retvi;
813 vi = kzalloc(sizeof(struct vma_info), GFP_KERNEL);
815 return ERR_PTR(-ENOMEM);
817 mutex_lock(&mapping->i_mmap_mutex);
818 retvi = __find_next_vma_info(mapping, head, vi, offset, is_register);
819 mutex_unlock(&mapping->i_mmap_mutex);
827 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
829 struct list_head try_list;
830 struct vm_area_struct *vma;
831 struct address_space *mapping;
832 struct vma_info *vi, *tmpvi;
833 struct mm_struct *mm;
837 mapping = uprobe->inode->i_mapping;
838 INIT_LIST_HEAD(&try_list);
843 vi = find_next_vma_info(mapping, &try_list, uprobe->offset, is_register);
853 down_write(&mm->mmap_sem);
854 vma = find_vma(mm, (unsigned long)vi->vaddr);
855 if (!vma || !valid_vma(vma, is_register)) {
856 list_del(&vi->probe_list);
858 up_write(&mm->mmap_sem);
862 vaddr = vma_address(vma, uprobe->offset);
863 if (vma->vm_file->f_mapping->host != uprobe->inode ||
864 vaddr != vi->vaddr) {
865 list_del(&vi->probe_list);
867 up_write(&mm->mmap_sem);
873 ret = install_breakpoint(uprobe, mm, vma, vi->vaddr);
875 remove_breakpoint(uprobe, mm, vi->vaddr);
877 up_write(&mm->mmap_sem);
880 if (ret && ret == -EEXIST)
887 list_for_each_entry_safe(vi, tmpvi, &try_list, probe_list) {
888 list_del(&vi->probe_list);
895 static int __uprobe_register(struct uprobe *uprobe)
897 return register_for_each_vma(uprobe, true);
900 static void __uprobe_unregister(struct uprobe *uprobe)
902 if (!register_for_each_vma(uprobe, false))
903 delete_uprobe(uprobe);
905 /* TODO : cant unregister? schedule a worker thread */
909 * uprobe_register - register a probe
910 * @inode: the file in which the probe has to be placed.
911 * @offset: offset from the start of the file.
912 * @uc: information on howto handle the probe..
914 * Apart from the access refcount, uprobe_register() takes a creation
915 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
916 * inserted into the rbtree (i.e first consumer for a @inode:@offset
917 * tuple). Creation refcount stops uprobe_unregister from freeing the
918 * @uprobe even before the register operation is complete. Creation
919 * refcount is released when the last @uc for the @uprobe
922 * Return errno if it cannot successully install probes
923 * else return 0 (success)
925 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
927 struct uprobe *uprobe;
930 if (!inode || !uc || uc->next)
933 if (offset > i_size_read(inode))
937 mutex_lock(uprobes_hash(inode));
938 uprobe = alloc_uprobe(inode, offset);
940 if (uprobe && !consumer_add(uprobe, uc)) {
941 ret = __uprobe_register(uprobe);
943 uprobe->consumers = NULL;
944 __uprobe_unregister(uprobe);
946 uprobe->flags |= UPROBE_RUN_HANDLER;
950 mutex_unlock(uprobes_hash(inode));
957 * uprobe_unregister - unregister a already registered probe.
958 * @inode: the file in which the probe has to be removed.
959 * @offset: offset from the start of the file.
960 * @uc: identify which probe if multiple probes are colocated.
962 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
964 struct uprobe *uprobe;
969 uprobe = find_uprobe(inode, offset);
973 mutex_lock(uprobes_hash(inode));
975 if (consumer_del(uprobe, uc)) {
976 if (!uprobe->consumers) {
977 __uprobe_unregister(uprobe);
978 uprobe->flags &= ~UPROBE_RUN_HANDLER;
982 mutex_unlock(uprobes_hash(inode));
988 * Of all the nodes that correspond to the given inode, return the node
989 * with the least offset.
991 static struct rb_node *find_least_offset_node(struct inode *inode)
993 struct uprobe u = { .inode = inode, .offset = 0};
994 struct rb_node *n = uprobes_tree.rb_node;
995 struct rb_node *close_node = NULL;
996 struct uprobe *uprobe;
1000 uprobe = rb_entry(n, struct uprobe, rb_node);
1001 match = match_uprobe(&u, uprobe);
1003 if (uprobe->inode == inode)
1019 * For a given inode, build a list of probes that need to be inserted.
1021 static void build_probe_list(struct inode *inode, struct list_head *head)
1023 struct uprobe *uprobe;
1024 unsigned long flags;
1027 spin_lock_irqsave(&uprobes_treelock, flags);
1029 n = find_least_offset_node(inode);
1031 for (; n; n = rb_next(n)) {
1032 uprobe = rb_entry(n, struct uprobe, rb_node);
1033 if (uprobe->inode != inode)
1036 list_add(&uprobe->pending_list, head);
1037 atomic_inc(&uprobe->ref);
1040 spin_unlock_irqrestore(&uprobes_treelock, flags);
1044 * Called from mmap_region.
1045 * called with mm->mmap_sem acquired.
1047 * Return -ve no if we fail to insert probes and we cannot
1049 * Return 0 otherwise. i.e:
1051 * - successful insertion of probes
1052 * - (or) no possible probes to be inserted.
1053 * - (or) insertion of probes failed but we can bail-out.
1055 int uprobe_mmap(struct vm_area_struct *vma)
1057 struct list_head tmp_list;
1058 struct uprobe *uprobe, *u;
1059 struct inode *inode;
1062 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1065 inode = vma->vm_file->f_mapping->host;
1069 INIT_LIST_HEAD(&tmp_list);
1070 mutex_lock(uprobes_mmap_hash(inode));
1071 build_probe_list(inode, &tmp_list);
1076 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1079 list_del(&uprobe->pending_list);
1081 vaddr = vma_address(vma, uprobe->offset);
1083 if (vaddr < vma->vm_start || vaddr >= vma->vm_end) {
1088 ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1090 /* Ignore double add: */
1091 if (ret == -EEXIST) {
1094 if (!is_swbp_at_addr(vma->vm_mm, vaddr))
1098 * Unable to insert a breakpoint, but
1099 * breakpoint lies underneath. Increment the
1102 atomic_inc(&vma->vm_mm->uprobes_state.count);
1111 mutex_unlock(uprobes_mmap_hash(inode));
1114 atomic_sub(count, &vma->vm_mm->uprobes_state.count);
1120 * Called in context of a munmap of a vma.
1122 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1124 struct list_head tmp_list;
1125 struct uprobe *uprobe, *u;
1126 struct inode *inode;
1128 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1131 if (!atomic_read(&vma->vm_mm->uprobes_state.count))
1134 inode = vma->vm_file->f_mapping->host;
1138 INIT_LIST_HEAD(&tmp_list);
1139 mutex_lock(uprobes_mmap_hash(inode));
1140 build_probe_list(inode, &tmp_list);
1142 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1145 list_del(&uprobe->pending_list);
1146 vaddr = vma_address(vma, uprobe->offset);
1148 if (vaddr >= start && vaddr < end) {
1150 * An unregister could have removed the probe before
1151 * unmap. So check before we decrement the count.
1153 if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1)
1154 atomic_dec(&vma->vm_mm->uprobes_state.count);
1158 mutex_unlock(uprobes_mmap_hash(inode));
1161 /* Slot allocation for XOL */
1162 static int xol_add_vma(struct xol_area *area)
1164 struct mm_struct *mm;
1167 area->page = alloc_page(GFP_HIGHUSER);
1174 down_write(&mm->mmap_sem);
1175 if (mm->uprobes_state.xol_area)
1180 /* Try to map as high as possible, this is only a hint. */
1181 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1182 if (area->vaddr & ~PAGE_MASK) {
1187 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1188 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1192 smp_wmb(); /* pairs with get_xol_area() */
1193 mm->uprobes_state.xol_area = area;
1197 up_write(&mm->mmap_sem);
1199 __free_page(area->page);
1204 static struct xol_area *get_xol_area(struct mm_struct *mm)
1206 struct xol_area *area;
1208 area = mm->uprobes_state.xol_area;
1209 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1215 * xol_alloc_area - Allocate process's xol_area.
1216 * This area will be used for storing instructions for execution out of
1219 * Returns the allocated area or NULL.
1221 static struct xol_area *xol_alloc_area(void)
1223 struct xol_area *area;
1225 area = kzalloc(sizeof(*area), GFP_KERNEL);
1226 if (unlikely(!area))
1229 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1234 init_waitqueue_head(&area->wq);
1235 if (!xol_add_vma(area))
1239 kfree(area->bitmap);
1242 return get_xol_area(current->mm);
1246 * uprobe_clear_state - Free the area allocated for slots.
1248 void uprobe_clear_state(struct mm_struct *mm)
1250 struct xol_area *area = mm->uprobes_state.xol_area;
1255 put_page(area->page);
1256 kfree(area->bitmap);
1261 * uprobe_reset_state - Free the area allocated for slots.
1263 void uprobe_reset_state(struct mm_struct *mm)
1265 mm->uprobes_state.xol_area = NULL;
1266 atomic_set(&mm->uprobes_state.count, 0);
1270 * - search for a free slot.
1272 static unsigned long xol_take_insn_slot(struct xol_area *area)
1274 unsigned long slot_addr;
1278 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1279 if (slot_nr < UINSNS_PER_PAGE) {
1280 if (!test_and_set_bit(slot_nr, area->bitmap))
1283 slot_nr = UINSNS_PER_PAGE;
1286 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1287 } while (slot_nr >= UINSNS_PER_PAGE);
1289 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1290 atomic_inc(&area->slot_count);
1296 * xol_get_insn_slot - If was not allocated a slot, then
1298 * Returns the allocated slot address or 0.
1300 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1302 struct xol_area *area;
1303 unsigned long offset;
1306 area = get_xol_area(current->mm);
1308 area = xol_alloc_area();
1312 current->utask->xol_vaddr = xol_take_insn_slot(area);
1315 * Initialize the slot if xol_vaddr points to valid
1318 if (unlikely(!current->utask->xol_vaddr))
1321 current->utask->vaddr = slot_addr;
1322 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1323 vaddr = kmap_atomic(area->page);
1324 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1325 kunmap_atomic(vaddr);
1327 return current->utask->xol_vaddr;
1331 * xol_free_insn_slot - If slot was earlier allocated by
1332 * @xol_get_insn_slot(), make the slot available for
1333 * subsequent requests.
1335 static void xol_free_insn_slot(struct task_struct *tsk)
1337 struct xol_area *area;
1338 unsigned long vma_end;
1339 unsigned long slot_addr;
1341 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1344 slot_addr = tsk->utask->xol_vaddr;
1346 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1349 area = tsk->mm->uprobes_state.xol_area;
1350 vma_end = area->vaddr + PAGE_SIZE;
1351 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1352 unsigned long offset;
1355 offset = slot_addr - area->vaddr;
1356 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1357 if (slot_nr >= UINSNS_PER_PAGE)
1360 clear_bit(slot_nr, area->bitmap);
1361 atomic_dec(&area->slot_count);
1362 if (waitqueue_active(&area->wq))
1365 tsk->utask->xol_vaddr = 0;
1370 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1371 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1373 * Return the address of the breakpoint instruction.
1375 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1377 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1381 * Called with no locks held.
1382 * Called in context of a exiting or a exec-ing thread.
1384 void uprobe_free_utask(struct task_struct *t)
1386 struct uprobe_task *utask = t->utask;
1391 if (utask->active_uprobe)
1392 put_uprobe(utask->active_uprobe);
1394 xol_free_insn_slot(t);
1400 * Called in context of a new clone/fork from copy_process.
1402 void uprobe_copy_process(struct task_struct *t)
1408 * Allocate a uprobe_task object for the task.
1409 * Called when the thread hits a breakpoint for the first time.
1412 * - pointer to new uprobe_task on success
1415 static struct uprobe_task *add_utask(void)
1417 struct uprobe_task *utask;
1419 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1420 if (unlikely(!utask))
1423 utask->active_uprobe = NULL;
1424 current->utask = utask;
1428 /* Prepare to single-step probed instruction out of line. */
1430 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1432 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1439 * If we are singlestepping, then ensure this thread is not connected to
1440 * non-fatal signals until completion of singlestep. When xol insn itself
1441 * triggers the signal, restart the original insn even if the task is
1442 * already SIGKILL'ed (since coredump should report the correct ip). This
1443 * is even more important if the task has a handler for SIGSEGV/etc, The
1444 * _same_ instruction should be repeated again after return from the signal
1445 * handler, and SSTEP can never finish in this case.
1447 bool uprobe_deny_signal(void)
1449 struct task_struct *t = current;
1450 struct uprobe_task *utask = t->utask;
1452 if (likely(!utask || !utask->active_uprobe))
1455 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1457 if (signal_pending(t)) {
1458 spin_lock_irq(&t->sighand->siglock);
1459 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1460 spin_unlock_irq(&t->sighand->siglock);
1462 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1463 utask->state = UTASK_SSTEP_TRAPPED;
1464 set_tsk_thread_flag(t, TIF_UPROBE);
1465 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1473 * Avoid singlestepping the original instruction if the original instruction
1474 * is a NOP or can be emulated.
1476 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1478 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1481 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1485 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1487 struct mm_struct *mm = current->mm;
1488 struct uprobe *uprobe = NULL;
1489 struct vm_area_struct *vma;
1491 down_read(&mm->mmap_sem);
1492 vma = find_vma(mm, bp_vaddr);
1493 if (vma && vma->vm_start <= bp_vaddr) {
1494 if (valid_vma(vma, false)) {
1495 struct inode *inode;
1498 inode = vma->vm_file->f_mapping->host;
1499 offset = bp_vaddr - vma->vm_start;
1500 offset += (vma->vm_pgoff << PAGE_SHIFT);
1501 uprobe = find_uprobe(inode, offset);
1505 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1509 up_read(&mm->mmap_sem);
1515 * Run handler and ask thread to singlestep.
1516 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1518 static void handle_swbp(struct pt_regs *regs)
1520 struct uprobe_task *utask;
1521 struct uprobe *uprobe;
1522 unsigned long bp_vaddr;
1523 int uninitialized_var(is_swbp);
1525 bp_vaddr = uprobe_get_swbp_addr(regs);
1526 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1530 /* No matching uprobe; signal SIGTRAP. */
1531 send_sig(SIGTRAP, current, 0);
1534 * Either we raced with uprobe_unregister() or we can't
1535 * access this memory. The latter is only possible if
1536 * another thread plays with our ->mm. In both cases
1537 * we can simply restart. If this vma was unmapped we
1538 * can pretend this insn was not executed yet and get
1539 * the (correct) SIGSEGV after restart.
1541 instruction_pointer_set(regs, bp_vaddr);
1546 utask = current->utask;
1548 utask = add_utask();
1549 /* Cannot allocate; re-execute the instruction. */
1553 utask->active_uprobe = uprobe;
1554 handler_chain(uprobe, regs);
1555 if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs))
1558 utask->state = UTASK_SSTEP;
1559 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1560 user_enable_single_step(current);
1566 utask->active_uprobe = NULL;
1567 utask->state = UTASK_RUNNING;
1570 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1573 * cannot singlestep; cannot skip instruction;
1574 * re-execute the instruction.
1576 instruction_pointer_set(regs, bp_vaddr);
1583 * Perform required fix-ups and disable singlestep.
1584 * Allow pending signals to take effect.
1586 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1588 struct uprobe *uprobe;
1590 uprobe = utask->active_uprobe;
1591 if (utask->state == UTASK_SSTEP_ACK)
1592 arch_uprobe_post_xol(&uprobe->arch, regs);
1593 else if (utask->state == UTASK_SSTEP_TRAPPED)
1594 arch_uprobe_abort_xol(&uprobe->arch, regs);
1599 utask->active_uprobe = NULL;
1600 utask->state = UTASK_RUNNING;
1601 user_disable_single_step(current);
1602 xol_free_insn_slot(current);
1604 spin_lock_irq(¤t->sighand->siglock);
1605 recalc_sigpending(); /* see uprobe_deny_signal() */
1606 spin_unlock_irq(¤t->sighand->siglock);
1610 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on
1611 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1612 * allows the thread to return from interrupt.
1614 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1615 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1618 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1619 * uprobe_notify_resume().
1621 void uprobe_notify_resume(struct pt_regs *regs)
1623 struct uprobe_task *utask;
1625 utask = current->utask;
1626 if (!utask || utask->state == UTASK_BP_HIT)
1629 handle_singlestep(utask, regs);
1633 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1634 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1636 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1638 struct uprobe_task *utask;
1640 if (!current->mm || !atomic_read(¤t->mm->uprobes_state.count))
1641 /* task is currently not uprobed */
1644 utask = current->utask;
1646 utask->state = UTASK_BP_HIT;
1648 set_thread_flag(TIF_UPROBE);
1654 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1655 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1657 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1659 struct uprobe_task *utask = current->utask;
1661 if (!current->mm || !utask || !utask->active_uprobe)
1662 /* task is currently not uprobed */
1665 utask->state = UTASK_SSTEP_ACK;
1666 set_thread_flag(TIF_UPROBE);
1670 static struct notifier_block uprobe_exception_nb = {
1671 .notifier_call = arch_uprobe_exception_notify,
1672 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1675 static int __init init_uprobes(void)
1679 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1680 mutex_init(&uprobes_mutex[i]);
1681 mutex_init(&uprobes_mmap_mutex[i]);
1684 return register_die_notifier(&uprobe_exception_nb);
1686 module_init(init_uprobes);
1688 static void __exit exit_uprobes(void)
1691 module_exit(exit_uprobes);