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 idx = (unsigned long)(offset >> PAGE_CACHE_SHIFT);
614 off1 = offset &= ~PAGE_MASK;
617 * Ensure that the page that has the original instruction is
618 * populated and in page-cache.
620 page = read_mapping_page(mapping, idx, filp);
622 return PTR_ERR(page);
624 vaddr = kmap_atomic(page);
625 memcpy(insn, vaddr + off1, nbytes);
626 kunmap_atomic(vaddr);
627 page_cache_release(page);
633 copy_insn(struct uprobe *uprobe, struct vm_area_struct *vma, unsigned long addr)
635 struct address_space *mapping;
636 unsigned long nbytes;
640 nbytes = PAGE_SIZE - addr;
641 mapping = uprobe->inode->i_mapping;
643 /* Instruction at end of binary; copy only available bytes */
644 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
645 bytes = uprobe->inode->i_size - uprobe->offset;
647 bytes = MAX_UINSN_BYTES;
649 /* Instruction at the page-boundary; copy bytes in second page */
650 if (nbytes < bytes) {
651 if (__copy_insn(mapping, vma, uprobe->arch.insn + nbytes,
652 bytes - nbytes, uprobe->offset + nbytes))
657 return __copy_insn(mapping, vma, uprobe->arch.insn, bytes, uprobe->offset);
661 * How mm->uprobes_state.count gets updated
662 * uprobe_mmap() increments the count if
663 * - it successfully adds a breakpoint.
664 * - it cannot add a breakpoint, but sees that there is a underlying
665 * breakpoint (via a is_swbp_at_addr()).
667 * uprobe_munmap() decrements the count if
668 * - it sees a underlying breakpoint, (via is_swbp_at_addr)
669 * (Subsequent uprobe_unregister wouldnt find the breakpoint
670 * unless a uprobe_mmap kicks in, since the old vma would be
671 * dropped just after uprobe_munmap.)
673 * uprobe_register increments the count if:
674 * - it successfully adds a breakpoint.
676 * uprobe_unregister decrements the count if:
677 * - it sees a underlying breakpoint and removes successfully.
678 * (via is_swbp_at_addr)
679 * (Subsequent uprobe_munmap wouldnt find the breakpoint
680 * since there is no underlying breakpoint after the
681 * breakpoint removal.)
684 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
685 struct vm_area_struct *vma, loff_t vaddr)
691 * If probe is being deleted, unregister thread could be done with
692 * the vma-rmap-walk through. Adding a probe now can be fatal since
693 * nobody will be able to cleanup. Also we could be from fork or
694 * mremap path, where the probe might have already been inserted.
695 * Hence behave as if probe already existed.
697 if (!uprobe->consumers)
700 addr = (unsigned long)vaddr;
702 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
703 ret = copy_insn(uprobe, vma, addr);
707 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
710 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, addr);
714 uprobe->flags |= UPROBE_COPY_INSN;
718 * Ideally, should be updating the probe count after the breakpoint
719 * has been successfully inserted. However a thread could hit the
720 * breakpoint we just inserted even before the probe count is
721 * incremented. If this is the first breakpoint placed, breakpoint
722 * notifier might ignore uprobes and pass the trap to the thread.
723 * Hence increment before and decrement on failure.
725 atomic_inc(&mm->uprobes_state.count);
726 ret = set_swbp(&uprobe->arch, mm, addr);
728 atomic_dec(&mm->uprobes_state.count);
734 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, loff_t vaddr)
736 if (!set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true))
737 atomic_dec(&mm->uprobes_state.count);
741 * There could be threads that have already hit the breakpoint. They
742 * will recheck the current insn and restart if find_uprobe() fails.
743 * See find_active_uprobe().
745 static void delete_uprobe(struct uprobe *uprobe)
749 spin_lock_irqsave(&uprobes_treelock, flags);
750 rb_erase(&uprobe->rb_node, &uprobes_tree);
751 spin_unlock_irqrestore(&uprobes_treelock, flags);
754 atomic_dec(&uprobe_events);
757 static struct vma_info *
758 __find_next_vma_info(struct address_space *mapping, struct list_head *head,
759 struct vma_info *vi, loff_t offset, bool is_register)
761 struct prio_tree_iter iter;
762 struct vm_area_struct *vma;
763 struct vma_info *tmpvi;
768 pgoff = offset >> PAGE_SHIFT;
770 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
771 if (!valid_vma(vma, is_register))
775 vaddr = vma_address(vma, offset);
777 list_for_each_entry(tmpvi, head, probe_list) {
778 if (tmpvi->mm == vma->vm_mm && tmpvi->vaddr == vaddr) {
785 * Another vma needs a probe to be installed. However skip
786 * installing the probe if the vma is about to be unlinked.
788 if (!existing_vma && atomic_inc_not_zero(&vma->vm_mm->mm_users)) {
791 list_add(&vi->probe_list, head);
801 * Iterate in the rmap prio tree and find a vma where a probe has not
804 static struct vma_info *
805 find_next_vma_info(struct address_space *mapping, struct list_head *head,
806 loff_t offset, bool is_register)
808 struct vma_info *vi, *retvi;
810 vi = kzalloc(sizeof(struct vma_info), GFP_KERNEL);
812 return ERR_PTR(-ENOMEM);
814 mutex_lock(&mapping->i_mmap_mutex);
815 retvi = __find_next_vma_info(mapping, head, vi, offset, is_register);
816 mutex_unlock(&mapping->i_mmap_mutex);
824 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
826 struct list_head try_list;
827 struct vm_area_struct *vma;
828 struct address_space *mapping;
829 struct vma_info *vi, *tmpvi;
830 struct mm_struct *mm;
834 mapping = uprobe->inode->i_mapping;
835 INIT_LIST_HEAD(&try_list);
840 vi = find_next_vma_info(mapping, &try_list, uprobe->offset, is_register);
850 down_write(&mm->mmap_sem);
851 vma = find_vma(mm, (unsigned long)vi->vaddr);
852 if (!vma || !valid_vma(vma, is_register)) {
853 list_del(&vi->probe_list);
855 up_write(&mm->mmap_sem);
859 vaddr = vma_address(vma, uprobe->offset);
860 if (vma->vm_file->f_mapping->host != uprobe->inode ||
861 vaddr != vi->vaddr) {
862 list_del(&vi->probe_list);
864 up_write(&mm->mmap_sem);
870 ret = install_breakpoint(uprobe, mm, vma, vi->vaddr);
872 remove_breakpoint(uprobe, mm, vi->vaddr);
874 up_write(&mm->mmap_sem);
877 if (ret && ret == -EEXIST)
884 list_for_each_entry_safe(vi, tmpvi, &try_list, probe_list) {
885 list_del(&vi->probe_list);
892 static int __uprobe_register(struct uprobe *uprobe)
894 return register_for_each_vma(uprobe, true);
897 static void __uprobe_unregister(struct uprobe *uprobe)
899 if (!register_for_each_vma(uprobe, false))
900 delete_uprobe(uprobe);
902 /* TODO : cant unregister? schedule a worker thread */
906 * uprobe_register - register a probe
907 * @inode: the file in which the probe has to be placed.
908 * @offset: offset from the start of the file.
909 * @uc: information on howto handle the probe..
911 * Apart from the access refcount, uprobe_register() takes a creation
912 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
913 * inserted into the rbtree (i.e first consumer for a @inode:@offset
914 * tuple). Creation refcount stops uprobe_unregister from freeing the
915 * @uprobe even before the register operation is complete. Creation
916 * refcount is released when the last @uc for the @uprobe
919 * Return errno if it cannot successully install probes
920 * else return 0 (success)
922 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
924 struct uprobe *uprobe;
927 if (!inode || !uc || uc->next)
930 if (offset > i_size_read(inode))
934 mutex_lock(uprobes_hash(inode));
935 uprobe = alloc_uprobe(inode, offset);
937 if (uprobe && !consumer_add(uprobe, uc)) {
938 ret = __uprobe_register(uprobe);
940 uprobe->consumers = NULL;
941 __uprobe_unregister(uprobe);
943 uprobe->flags |= UPROBE_RUN_HANDLER;
947 mutex_unlock(uprobes_hash(inode));
954 * uprobe_unregister - unregister a already registered probe.
955 * @inode: the file in which the probe has to be removed.
956 * @offset: offset from the start of the file.
957 * @uc: identify which probe if multiple probes are colocated.
959 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
961 struct uprobe *uprobe;
966 uprobe = find_uprobe(inode, offset);
970 mutex_lock(uprobes_hash(inode));
972 if (consumer_del(uprobe, uc)) {
973 if (!uprobe->consumers) {
974 __uprobe_unregister(uprobe);
975 uprobe->flags &= ~UPROBE_RUN_HANDLER;
979 mutex_unlock(uprobes_hash(inode));
985 * Of all the nodes that correspond to the given inode, return the node
986 * with the least offset.
988 static struct rb_node *find_least_offset_node(struct inode *inode)
990 struct uprobe u = { .inode = inode, .offset = 0};
991 struct rb_node *n = uprobes_tree.rb_node;
992 struct rb_node *close_node = NULL;
993 struct uprobe *uprobe;
997 uprobe = rb_entry(n, struct uprobe, rb_node);
998 match = match_uprobe(&u, uprobe);
1000 if (uprobe->inode == inode)
1016 * For a given inode, build a list of probes that need to be inserted.
1018 static void build_probe_list(struct inode *inode, struct list_head *head)
1020 struct uprobe *uprobe;
1021 unsigned long flags;
1024 spin_lock_irqsave(&uprobes_treelock, flags);
1026 n = find_least_offset_node(inode);
1028 for (; n; n = rb_next(n)) {
1029 uprobe = rb_entry(n, struct uprobe, rb_node);
1030 if (uprobe->inode != inode)
1033 list_add(&uprobe->pending_list, head);
1034 atomic_inc(&uprobe->ref);
1037 spin_unlock_irqrestore(&uprobes_treelock, flags);
1041 * Called from mmap_region.
1042 * called with mm->mmap_sem acquired.
1044 * Return -ve no if we fail to insert probes and we cannot
1046 * Return 0 otherwise. i.e:
1048 * - successful insertion of probes
1049 * - (or) no possible probes to be inserted.
1050 * - (or) insertion of probes failed but we can bail-out.
1052 int uprobe_mmap(struct vm_area_struct *vma)
1054 struct list_head tmp_list;
1055 struct uprobe *uprobe, *u;
1056 struct inode *inode;
1059 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1062 inode = vma->vm_file->f_mapping->host;
1066 INIT_LIST_HEAD(&tmp_list);
1067 mutex_lock(uprobes_mmap_hash(inode));
1068 build_probe_list(inode, &tmp_list);
1073 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1076 list_del(&uprobe->pending_list);
1078 vaddr = vma_address(vma, uprobe->offset);
1080 if (vaddr < vma->vm_start || vaddr >= vma->vm_end) {
1085 ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1087 /* Ignore double add: */
1088 if (ret == -EEXIST) {
1091 if (!is_swbp_at_addr(vma->vm_mm, vaddr))
1095 * Unable to insert a breakpoint, but
1096 * breakpoint lies underneath. Increment the
1099 atomic_inc(&vma->vm_mm->uprobes_state.count);
1108 mutex_unlock(uprobes_mmap_hash(inode));
1111 atomic_sub(count, &vma->vm_mm->uprobes_state.count);
1117 * Called in context of a munmap of a vma.
1119 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1121 struct list_head tmp_list;
1122 struct uprobe *uprobe, *u;
1123 struct inode *inode;
1125 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1128 if (!atomic_read(&vma->vm_mm->uprobes_state.count))
1131 inode = vma->vm_file->f_mapping->host;
1135 INIT_LIST_HEAD(&tmp_list);
1136 mutex_lock(uprobes_mmap_hash(inode));
1137 build_probe_list(inode, &tmp_list);
1139 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1142 list_del(&uprobe->pending_list);
1143 vaddr = vma_address(vma, uprobe->offset);
1145 if (vaddr >= start && vaddr < end) {
1147 * An unregister could have removed the probe before
1148 * unmap. So check before we decrement the count.
1150 if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1)
1151 atomic_dec(&vma->vm_mm->uprobes_state.count);
1155 mutex_unlock(uprobes_mmap_hash(inode));
1158 /* Slot allocation for XOL */
1159 static int xol_add_vma(struct xol_area *area)
1161 struct mm_struct *mm;
1164 area->page = alloc_page(GFP_HIGHUSER);
1171 down_write(&mm->mmap_sem);
1172 if (mm->uprobes_state.xol_area)
1177 /* Try to map as high as possible, this is only a hint. */
1178 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1179 if (area->vaddr & ~PAGE_MASK) {
1184 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1185 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1189 smp_wmb(); /* pairs with get_xol_area() */
1190 mm->uprobes_state.xol_area = area;
1194 up_write(&mm->mmap_sem);
1196 __free_page(area->page);
1201 static struct xol_area *get_xol_area(struct mm_struct *mm)
1203 struct xol_area *area;
1205 area = mm->uprobes_state.xol_area;
1206 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1212 * xol_alloc_area - Allocate process's xol_area.
1213 * This area will be used for storing instructions for execution out of
1216 * Returns the allocated area or NULL.
1218 static struct xol_area *xol_alloc_area(void)
1220 struct xol_area *area;
1222 area = kzalloc(sizeof(*area), GFP_KERNEL);
1223 if (unlikely(!area))
1226 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1231 init_waitqueue_head(&area->wq);
1232 if (!xol_add_vma(area))
1236 kfree(area->bitmap);
1239 return get_xol_area(current->mm);
1243 * uprobe_clear_state - Free the area allocated for slots.
1245 void uprobe_clear_state(struct mm_struct *mm)
1247 struct xol_area *area = mm->uprobes_state.xol_area;
1252 put_page(area->page);
1253 kfree(area->bitmap);
1258 * uprobe_reset_state - Free the area allocated for slots.
1260 void uprobe_reset_state(struct mm_struct *mm)
1262 mm->uprobes_state.xol_area = NULL;
1263 atomic_set(&mm->uprobes_state.count, 0);
1267 * - search for a free slot.
1269 static unsigned long xol_take_insn_slot(struct xol_area *area)
1271 unsigned long slot_addr;
1275 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1276 if (slot_nr < UINSNS_PER_PAGE) {
1277 if (!test_and_set_bit(slot_nr, area->bitmap))
1280 slot_nr = UINSNS_PER_PAGE;
1283 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1284 } while (slot_nr >= UINSNS_PER_PAGE);
1286 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1287 atomic_inc(&area->slot_count);
1293 * xol_get_insn_slot - If was not allocated a slot, then
1295 * Returns the allocated slot address or 0.
1297 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1299 struct xol_area *area;
1300 unsigned long offset;
1303 area = get_xol_area(current->mm);
1305 area = xol_alloc_area();
1309 current->utask->xol_vaddr = xol_take_insn_slot(area);
1312 * Initialize the slot if xol_vaddr points to valid
1315 if (unlikely(!current->utask->xol_vaddr))
1318 current->utask->vaddr = slot_addr;
1319 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1320 vaddr = kmap_atomic(area->page);
1321 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1322 kunmap_atomic(vaddr);
1324 return current->utask->xol_vaddr;
1328 * xol_free_insn_slot - If slot was earlier allocated by
1329 * @xol_get_insn_slot(), make the slot available for
1330 * subsequent requests.
1332 static void xol_free_insn_slot(struct task_struct *tsk)
1334 struct xol_area *area;
1335 unsigned long vma_end;
1336 unsigned long slot_addr;
1338 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1341 slot_addr = tsk->utask->xol_vaddr;
1343 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1346 area = tsk->mm->uprobes_state.xol_area;
1347 vma_end = area->vaddr + PAGE_SIZE;
1348 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1349 unsigned long offset;
1352 offset = slot_addr - area->vaddr;
1353 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1354 if (slot_nr >= UINSNS_PER_PAGE)
1357 clear_bit(slot_nr, area->bitmap);
1358 atomic_dec(&area->slot_count);
1359 if (waitqueue_active(&area->wq))
1362 tsk->utask->xol_vaddr = 0;
1367 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1368 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1370 * Return the address of the breakpoint instruction.
1372 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1374 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1378 * Called with no locks held.
1379 * Called in context of a exiting or a exec-ing thread.
1381 void uprobe_free_utask(struct task_struct *t)
1383 struct uprobe_task *utask = t->utask;
1388 if (utask->active_uprobe)
1389 put_uprobe(utask->active_uprobe);
1391 xol_free_insn_slot(t);
1397 * Called in context of a new clone/fork from copy_process.
1399 void uprobe_copy_process(struct task_struct *t)
1405 * Allocate a uprobe_task object for the task.
1406 * Called when the thread hits a breakpoint for the first time.
1409 * - pointer to new uprobe_task on success
1412 static struct uprobe_task *add_utask(void)
1414 struct uprobe_task *utask;
1416 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1417 if (unlikely(!utask))
1420 utask->active_uprobe = NULL;
1421 current->utask = utask;
1425 /* Prepare to single-step probed instruction out of line. */
1427 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1429 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1436 * If we are singlestepping, then ensure this thread is not connected to
1437 * non-fatal signals until completion of singlestep. When xol insn itself
1438 * triggers the signal, restart the original insn even if the task is
1439 * already SIGKILL'ed (since coredump should report the correct ip). This
1440 * is even more important if the task has a handler for SIGSEGV/etc, The
1441 * _same_ instruction should be repeated again after return from the signal
1442 * handler, and SSTEP can never finish in this case.
1444 bool uprobe_deny_signal(void)
1446 struct task_struct *t = current;
1447 struct uprobe_task *utask = t->utask;
1449 if (likely(!utask || !utask->active_uprobe))
1452 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1454 if (signal_pending(t)) {
1455 spin_lock_irq(&t->sighand->siglock);
1456 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1457 spin_unlock_irq(&t->sighand->siglock);
1459 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1460 utask->state = UTASK_SSTEP_TRAPPED;
1461 set_tsk_thread_flag(t, TIF_UPROBE);
1462 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1470 * Avoid singlestepping the original instruction if the original instruction
1471 * is a NOP or can be emulated.
1473 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1475 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1478 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1482 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1484 struct mm_struct *mm = current->mm;
1485 struct uprobe *uprobe = NULL;
1486 struct vm_area_struct *vma;
1488 down_read(&mm->mmap_sem);
1489 vma = find_vma(mm, bp_vaddr);
1490 if (vma && vma->vm_start <= bp_vaddr) {
1491 if (valid_vma(vma, false)) {
1492 struct inode *inode;
1495 inode = vma->vm_file->f_mapping->host;
1496 offset = bp_vaddr - vma->vm_start;
1497 offset += (vma->vm_pgoff << PAGE_SHIFT);
1498 uprobe = find_uprobe(inode, offset);
1502 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1506 up_read(&mm->mmap_sem);
1512 * Run handler and ask thread to singlestep.
1513 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1515 static void handle_swbp(struct pt_regs *regs)
1517 struct uprobe_task *utask;
1518 struct uprobe *uprobe;
1519 unsigned long bp_vaddr;
1520 int uninitialized_var(is_swbp);
1522 bp_vaddr = uprobe_get_swbp_addr(regs);
1523 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1527 /* No matching uprobe; signal SIGTRAP. */
1528 send_sig(SIGTRAP, current, 0);
1531 * Either we raced with uprobe_unregister() or we can't
1532 * access this memory. The latter is only possible if
1533 * another thread plays with our ->mm. In both cases
1534 * we can simply restart. If this vma was unmapped we
1535 * can pretend this insn was not executed yet and get
1536 * the (correct) SIGSEGV after restart.
1538 instruction_pointer_set(regs, bp_vaddr);
1543 utask = current->utask;
1545 utask = add_utask();
1546 /* Cannot allocate; re-execute the instruction. */
1550 utask->active_uprobe = uprobe;
1551 handler_chain(uprobe, regs);
1552 if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs))
1555 utask->state = UTASK_SSTEP;
1556 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1557 user_enable_single_step(current);
1563 utask->active_uprobe = NULL;
1564 utask->state = UTASK_RUNNING;
1567 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1570 * cannot singlestep; cannot skip instruction;
1571 * re-execute the instruction.
1573 instruction_pointer_set(regs, bp_vaddr);
1580 * Perform required fix-ups and disable singlestep.
1581 * Allow pending signals to take effect.
1583 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1585 struct uprobe *uprobe;
1587 uprobe = utask->active_uprobe;
1588 if (utask->state == UTASK_SSTEP_ACK)
1589 arch_uprobe_post_xol(&uprobe->arch, regs);
1590 else if (utask->state == UTASK_SSTEP_TRAPPED)
1591 arch_uprobe_abort_xol(&uprobe->arch, regs);
1596 utask->active_uprobe = NULL;
1597 utask->state = UTASK_RUNNING;
1598 user_disable_single_step(current);
1599 xol_free_insn_slot(current);
1601 spin_lock_irq(¤t->sighand->siglock);
1602 recalc_sigpending(); /* see uprobe_deny_signal() */
1603 spin_unlock_irq(¤t->sighand->siglock);
1607 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on
1608 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1609 * allows the thread to return from interrupt.
1611 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1612 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1615 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1616 * uprobe_notify_resume().
1618 void uprobe_notify_resume(struct pt_regs *regs)
1620 struct uprobe_task *utask;
1622 utask = current->utask;
1623 if (!utask || utask->state == UTASK_BP_HIT)
1626 handle_singlestep(utask, regs);
1630 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1631 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1633 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1635 struct uprobe_task *utask;
1637 if (!current->mm || !atomic_read(¤t->mm->uprobes_state.count))
1638 /* task is currently not uprobed */
1641 utask = current->utask;
1643 utask->state = UTASK_BP_HIT;
1645 set_thread_flag(TIF_UPROBE);
1651 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1652 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1654 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1656 struct uprobe_task *utask = current->utask;
1658 if (!current->mm || !utask || !utask->active_uprobe)
1659 /* task is currently not uprobed */
1662 utask->state = UTASK_SSTEP_ACK;
1663 set_thread_flag(TIF_UPROBE);
1667 static struct notifier_block uprobe_exception_nb = {
1668 .notifier_call = arch_uprobe_exception_notify,
1669 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1672 static int __init init_uprobes(void)
1676 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1677 mutex_init(&uprobes_mutex[i]);
1678 mutex_init(&uprobes_mmap_mutex[i]);
1681 return register_die_notifier(&uprobe_exception_nb);
1683 module_init(init_uprobes);
1685 static void __exit exit_uprobes(void)
1688 module_exit(exit_uprobes);