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
48 * We need separate register/unregister and mmap/munmap lock hashes because
49 * of mmap_sem nesting.
51 * uprobe_register() needs to install probes on (potentially) all processes
52 * and thus needs to acquire multiple mmap_sems (consequtively, not
53 * concurrently), whereas uprobe_mmap() is called while holding mmap_sem
54 * for the particular process doing the mmap.
56 * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem
57 * because of lock order against i_mmap_mutex. This means there's a hole in
58 * the register vma iteration where a mmap() can happen.
60 * Thus uprobe_register() can race with uprobe_mmap() and we can try and
61 * install a probe where one is already installed.
64 /* serialize (un)register */
65 static struct mutex uprobes_mutex[UPROBES_HASH_SZ];
67 #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
69 /* serialize uprobe->pending_list */
70 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
71 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
74 * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe
75 * events active at this time. Probably a fine grained per inode count is
78 static atomic_t uprobe_events = ATOMIC_INIT(0);
81 struct rb_node rb_node; /* node in the rb tree */
83 struct rw_semaphore consumer_rwsem;
84 struct list_head pending_list;
85 struct uprobe_consumer *consumers;
86 struct inode *inode; /* Also hold a ref to inode */
89 struct arch_uprobe arch;
93 * valid_vma: Verify if the specified vma is an executable vma
94 * Relax restrictions while unregistering: vm_flags might have
95 * changed after breakpoint was inserted.
96 * - is_register: indicates if we are in register context.
97 * - Return 1 if the specified virtual address is in an
100 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
108 if ((vma->vm_flags & (VM_HUGETLB|VM_READ|VM_WRITE|VM_EXEC|VM_SHARED))
109 == (VM_READ|VM_EXEC))
115 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
117 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
120 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
122 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
126 * __replace_page - replace page in vma by new page.
127 * based on replace_page in mm/ksm.c
129 * @vma: vma that holds the pte pointing to page
130 * @addr: address the old @page is mapped at
131 * @page: the cowed page we are replacing by kpage
132 * @kpage: the modified page we replace page by
134 * Returns 0 on success, -EFAULT on failure.
136 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
137 struct page *page, struct page *kpage)
139 struct mm_struct *mm = vma->vm_mm;
144 /* freeze PageSwapCache() for try_to_free_swap() below */
148 ptep = page_check_address(page, mm, addr, &ptl, 0);
153 page_add_new_anon_rmap(kpage, vma, addr);
155 if (!PageAnon(page)) {
156 dec_mm_counter(mm, MM_FILEPAGES);
157 inc_mm_counter(mm, MM_ANONPAGES);
160 flush_cache_page(vma, addr, pte_pfn(*ptep));
161 ptep_clear_flush(vma, addr, ptep);
162 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
164 page_remove_rmap(page);
165 if (!page_mapped(page))
166 try_to_free_swap(page);
168 pte_unmap_unlock(ptep, ptl);
177 * is_swbp_insn - check if instruction is breakpoint instruction.
178 * @insn: instruction to be checked.
179 * Default implementation of is_swbp_insn
180 * Returns true if @insn is a breakpoint instruction.
182 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
184 return *insn == UPROBE_SWBP_INSN;
189 * Expect the breakpoint instruction to be the smallest size instruction for
190 * the architecture. If an arch has variable length instruction and the
191 * breakpoint instruction is not of the smallest length instruction
192 * supported by that architecture then we need to modify read_opcode /
193 * write_opcode accordingly. This would never be a problem for archs that
194 * have fixed length instructions.
198 * write_opcode - write the opcode at a given virtual address.
199 * @auprobe: arch breakpointing information.
200 * @mm: the probed process address space.
201 * @vaddr: the virtual address to store the opcode.
202 * @opcode: opcode to be written at @vaddr.
204 * Called with mm->mmap_sem held (for read and with a reference to
207 * For mm @mm, write the opcode at @vaddr.
208 * Return 0 (success) or a negative errno.
210 static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
211 unsigned long vaddr, uprobe_opcode_t opcode)
213 struct page *old_page, *new_page;
214 void *vaddr_old, *vaddr_new;
215 struct vm_area_struct *vma;
219 /* Read the page with vaddr into memory */
220 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
225 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
229 __SetPageUptodate(new_page);
231 /* copy the page now that we've got it stable */
232 vaddr_old = kmap_atomic(old_page);
233 vaddr_new = kmap_atomic(new_page);
235 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
236 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
238 kunmap_atomic(vaddr_new);
239 kunmap_atomic(vaddr_old);
241 ret = anon_vma_prepare(vma);
245 ret = __replace_page(vma, vaddr, old_page, new_page);
248 page_cache_release(new_page);
252 if (unlikely(ret == -EAGAIN))
258 * read_opcode - read the opcode at a given virtual address.
259 * @mm: the probed process address space.
260 * @vaddr: the virtual address to read the opcode.
261 * @opcode: location to store the read opcode.
263 * Called with mm->mmap_sem held (for read and with a reference to
266 * For mm @mm, read the opcode at @vaddr and store it in @opcode.
267 * Return 0 (success) or a negative errno.
269 static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode)
275 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
280 vaddr_new = kmap_atomic(page);
282 memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE);
283 kunmap_atomic(vaddr_new);
291 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
293 uprobe_opcode_t opcode;
296 if (current->mm == mm) {
298 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
302 if (likely(result == 0))
306 result = read_opcode(mm, vaddr, &opcode);
310 if (is_swbp_insn(&opcode))
317 * set_swbp - store breakpoint at a given address.
318 * @auprobe: arch specific probepoint information.
319 * @mm: the probed process address space.
320 * @vaddr: the virtual address to insert the opcode.
322 * For mm @mm, store the breakpoint instruction at @vaddr.
323 * Return 0 (success) or a negative errno.
325 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
329 * See the comment near uprobes_hash().
331 result = is_swbp_at_addr(mm, vaddr);
338 return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
342 * set_orig_insn - Restore the original instruction.
343 * @mm: the probed process address space.
344 * @auprobe: arch specific probepoint information.
345 * @vaddr: the virtual address to insert the opcode.
346 * @verify: if true, verify existance of breakpoint instruction.
348 * For mm @mm, restore the original opcode (opcode) at @vaddr.
349 * Return 0 (success) or a negative errno.
352 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify)
357 result = is_swbp_at_addr(mm, vaddr);
364 return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
367 static int match_uprobe(struct uprobe *l, struct uprobe *r)
369 if (l->inode < r->inode)
372 if (l->inode > r->inode)
375 if (l->offset < r->offset)
378 if (l->offset > r->offset)
384 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
386 struct uprobe u = { .inode = inode, .offset = offset };
387 struct rb_node *n = uprobes_tree.rb_node;
388 struct uprobe *uprobe;
392 uprobe = rb_entry(n, struct uprobe, rb_node);
393 match = match_uprobe(&u, uprobe);
395 atomic_inc(&uprobe->ref);
408 * Find a uprobe corresponding to a given inode:offset
409 * Acquires uprobes_treelock
411 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
413 struct uprobe *uprobe;
416 spin_lock_irqsave(&uprobes_treelock, flags);
417 uprobe = __find_uprobe(inode, offset);
418 spin_unlock_irqrestore(&uprobes_treelock, flags);
423 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
425 struct rb_node **p = &uprobes_tree.rb_node;
426 struct rb_node *parent = NULL;
432 u = rb_entry(parent, struct uprobe, rb_node);
433 match = match_uprobe(uprobe, u);
440 p = &parent->rb_left;
442 p = &parent->rb_right;
447 rb_link_node(&uprobe->rb_node, parent, p);
448 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
449 /* get access + creation ref */
450 atomic_set(&uprobe->ref, 2);
456 * Acquire uprobes_treelock.
457 * Matching uprobe already exists in rbtree;
458 * increment (access refcount) and return the matching uprobe.
460 * No matching uprobe; insert the uprobe in rb_tree;
461 * get a double refcount (access + creation) and return NULL.
463 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
468 spin_lock_irqsave(&uprobes_treelock, flags);
469 u = __insert_uprobe(uprobe);
470 spin_unlock_irqrestore(&uprobes_treelock, flags);
472 /* For now assume that the instruction need not be single-stepped */
473 uprobe->flags |= UPROBE_SKIP_SSTEP;
478 static void put_uprobe(struct uprobe *uprobe)
480 if (atomic_dec_and_test(&uprobe->ref))
484 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
486 struct uprobe *uprobe, *cur_uprobe;
488 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
492 uprobe->inode = igrab(inode);
493 uprobe->offset = offset;
494 init_rwsem(&uprobe->consumer_rwsem);
496 /* add to uprobes_tree, sorted on inode:offset */
497 cur_uprobe = insert_uprobe(uprobe);
499 /* a uprobe exists for this inode:offset combination */
505 atomic_inc(&uprobe_events);
511 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
513 struct uprobe_consumer *uc;
515 if (!(uprobe->flags & UPROBE_RUN_HANDLER))
518 down_read(&uprobe->consumer_rwsem);
519 for (uc = uprobe->consumers; uc; uc = uc->next) {
520 if (!uc->filter || uc->filter(uc, current))
521 uc->handler(uc, regs);
523 up_read(&uprobe->consumer_rwsem);
526 /* Returns the previous consumer */
527 static struct uprobe_consumer *
528 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
530 down_write(&uprobe->consumer_rwsem);
531 uc->next = uprobe->consumers;
532 uprobe->consumers = uc;
533 up_write(&uprobe->consumer_rwsem);
539 * For uprobe @uprobe, delete the consumer @uc.
540 * Return true if the @uc is deleted successfully
543 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
545 struct uprobe_consumer **con;
548 down_write(&uprobe->consumer_rwsem);
549 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
556 up_write(&uprobe->consumer_rwsem);
562 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
563 unsigned long nbytes, loff_t offset)
573 if (!mapping->a_ops->readpage)
576 idx = offset >> PAGE_CACHE_SHIFT;
577 off = offset & ~PAGE_MASK;
580 * Ensure that the page that has the original instruction is
581 * populated and in page-cache.
583 page = read_mapping_page(mapping, idx, filp);
585 return PTR_ERR(page);
587 vaddr = kmap_atomic(page);
588 memcpy(insn, vaddr + off, nbytes);
589 kunmap_atomic(vaddr);
590 page_cache_release(page);
595 static int copy_insn(struct uprobe *uprobe, struct file *filp)
597 struct address_space *mapping;
598 unsigned long nbytes;
601 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
602 mapping = uprobe->inode->i_mapping;
604 /* Instruction at end of binary; copy only available bytes */
605 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
606 bytes = uprobe->inode->i_size - uprobe->offset;
608 bytes = MAX_UINSN_BYTES;
610 /* Instruction at the page-boundary; copy bytes in second page */
611 if (nbytes < bytes) {
612 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
613 bytes - nbytes, uprobe->offset + nbytes);
618 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
622 * How mm->uprobes_state.count gets updated
623 * uprobe_mmap() increments the count if
624 * - it successfully adds a breakpoint.
625 * - it cannot add a breakpoint, but sees that there is a underlying
626 * breakpoint (via a is_swbp_at_addr()).
628 * uprobe_munmap() decrements the count if
629 * - it sees a underlying breakpoint, (via is_swbp_at_addr)
630 * (Subsequent uprobe_unregister wouldnt find the breakpoint
631 * unless a uprobe_mmap kicks in, since the old vma would be
632 * dropped just after uprobe_munmap.)
634 * uprobe_register increments the count if:
635 * - it successfully adds a breakpoint.
637 * uprobe_unregister decrements the count if:
638 * - it sees a underlying breakpoint and removes successfully.
639 * (via is_swbp_at_addr)
640 * (Subsequent uprobe_munmap wouldnt find the breakpoint
641 * since there is no underlying breakpoint after the
642 * breakpoint removal.)
645 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
646 struct vm_area_struct *vma, unsigned long vaddr)
651 * If probe is being deleted, unregister thread could be done with
652 * the vma-rmap-walk through. Adding a probe now can be fatal since
653 * nobody will be able to cleanup. Also we could be from fork or
654 * mremap path, where the probe might have already been inserted.
655 * Hence behave as if probe already existed.
657 if (!uprobe->consumers)
660 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
661 ret = copy_insn(uprobe, vma->vm_file);
665 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
668 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
672 /* write_opcode() assumes we don't cross page boundary */
673 BUG_ON((uprobe->offset & ~PAGE_MASK) +
674 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
676 uprobe->flags |= UPROBE_COPY_INSN;
680 * Ideally, should be updating the probe count after the breakpoint
681 * has been successfully inserted. However a thread could hit the
682 * breakpoint we just inserted even before the probe count is
683 * incremented. If this is the first breakpoint placed, breakpoint
684 * notifier might ignore uprobes and pass the trap to the thread.
685 * Hence increment before and decrement on failure.
687 atomic_inc(&mm->uprobes_state.count);
688 ret = set_swbp(&uprobe->arch, mm, vaddr);
690 atomic_dec(&mm->uprobes_state.count);
696 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
698 if (!set_orig_insn(&uprobe->arch, mm, vaddr, true))
699 atomic_dec(&mm->uprobes_state.count);
703 * There could be threads that have already hit the breakpoint. They
704 * will recheck the current insn and restart if find_uprobe() fails.
705 * See find_active_uprobe().
707 static void delete_uprobe(struct uprobe *uprobe)
711 spin_lock_irqsave(&uprobes_treelock, flags);
712 rb_erase(&uprobe->rb_node, &uprobes_tree);
713 spin_unlock_irqrestore(&uprobes_treelock, flags);
716 atomic_dec(&uprobe_events);
720 struct map_info *next;
721 struct mm_struct *mm;
725 static inline struct map_info *free_map_info(struct map_info *info)
727 struct map_info *next = info->next;
732 static struct map_info *
733 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
735 unsigned long pgoff = offset >> PAGE_SHIFT;
736 struct prio_tree_iter iter;
737 struct vm_area_struct *vma;
738 struct map_info *curr = NULL;
739 struct map_info *prev = NULL;
740 struct map_info *info;
744 mutex_lock(&mapping->i_mmap_mutex);
745 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
746 if (!valid_vma(vma, is_register))
749 if (!prev && !more) {
751 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
752 * reclaim. This is optimistic, no harm done if it fails.
754 prev = kmalloc(sizeof(struct map_info),
755 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
764 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
772 info->mm = vma->vm_mm;
773 info->vaddr = offset_to_vaddr(vma, offset);
775 mutex_unlock(&mapping->i_mmap_mutex);
787 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
789 curr = ERR_PTR(-ENOMEM);
799 prev = free_map_info(prev);
803 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
805 struct map_info *info;
808 info = build_map_info(uprobe->inode->i_mapping,
809 uprobe->offset, is_register);
811 return PTR_ERR(info);
814 struct mm_struct *mm = info->mm;
815 struct vm_area_struct *vma;
820 down_write(&mm->mmap_sem);
821 vma = find_vma(mm, info->vaddr);
822 if (!vma || !valid_vma(vma, is_register) ||
823 vma->vm_file->f_mapping->host != uprobe->inode)
826 if (vma->vm_start > info->vaddr ||
827 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
831 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
833 * We can race against uprobe_mmap(), see the
834 * comment near uprobe_hash().
839 remove_breakpoint(uprobe, mm, info->vaddr);
842 up_write(&mm->mmap_sem);
845 info = free_map_info(info);
851 static int __uprobe_register(struct uprobe *uprobe)
853 return register_for_each_vma(uprobe, true);
856 static void __uprobe_unregister(struct uprobe *uprobe)
858 if (!register_for_each_vma(uprobe, false))
859 delete_uprobe(uprobe);
861 /* TODO : cant unregister? schedule a worker thread */
865 * uprobe_register - register a probe
866 * @inode: the file in which the probe has to be placed.
867 * @offset: offset from the start of the file.
868 * @uc: information on howto handle the probe..
870 * Apart from the access refcount, uprobe_register() takes a creation
871 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
872 * inserted into the rbtree (i.e first consumer for a @inode:@offset
873 * tuple). Creation refcount stops uprobe_unregister from freeing the
874 * @uprobe even before the register operation is complete. Creation
875 * refcount is released when the last @uc for the @uprobe
878 * Return errno if it cannot successully install probes
879 * else return 0 (success)
881 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
883 struct uprobe *uprobe;
886 if (!inode || !uc || uc->next)
889 if (offset > i_size_read(inode))
893 mutex_lock(uprobes_hash(inode));
894 uprobe = alloc_uprobe(inode, offset);
896 if (uprobe && !consumer_add(uprobe, uc)) {
897 ret = __uprobe_register(uprobe);
899 uprobe->consumers = NULL;
900 __uprobe_unregister(uprobe);
902 uprobe->flags |= UPROBE_RUN_HANDLER;
906 mutex_unlock(uprobes_hash(inode));
913 * uprobe_unregister - unregister a already registered probe.
914 * @inode: the file in which the probe has to be removed.
915 * @offset: offset from the start of the file.
916 * @uc: identify which probe if multiple probes are colocated.
918 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
920 struct uprobe *uprobe;
925 uprobe = find_uprobe(inode, offset);
929 mutex_lock(uprobes_hash(inode));
931 if (consumer_del(uprobe, uc)) {
932 if (!uprobe->consumers) {
933 __uprobe_unregister(uprobe);
934 uprobe->flags &= ~UPROBE_RUN_HANDLER;
938 mutex_unlock(uprobes_hash(inode));
943 static struct rb_node *
944 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
946 struct rb_node *n = uprobes_tree.rb_node;
949 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
951 if (inode < u->inode) {
953 } else if (inode > u->inode) {
958 else if (min > u->offset)
969 * For a given range in vma, build a list of probes that need to be inserted.
971 static void build_probe_list(struct inode *inode,
972 struct vm_area_struct *vma,
973 unsigned long start, unsigned long end,
974 struct list_head *head)
978 struct rb_node *n, *t;
981 INIT_LIST_HEAD(head);
982 min = vaddr_to_offset(vma, start);
983 max = min + (end - start) - 1;
985 spin_lock_irqsave(&uprobes_treelock, flags);
986 n = find_node_in_range(inode, min, max);
988 for (t = n; t; t = rb_prev(t)) {
989 u = rb_entry(t, struct uprobe, rb_node);
990 if (u->inode != inode || u->offset < min)
992 list_add(&u->pending_list, head);
995 for (t = n; (t = rb_next(t)); ) {
996 u = rb_entry(t, struct uprobe, rb_node);
997 if (u->inode != inode || u->offset > max)
999 list_add(&u->pending_list, head);
1000 atomic_inc(&u->ref);
1003 spin_unlock_irqrestore(&uprobes_treelock, flags);
1007 * Called from mmap_region.
1008 * called with mm->mmap_sem acquired.
1010 * Return -ve no if we fail to insert probes and we cannot
1012 * Return 0 otherwise. i.e:
1014 * - successful insertion of probes
1015 * - (or) no possible probes to be inserted.
1016 * - (or) insertion of probes failed but we can bail-out.
1018 int uprobe_mmap(struct vm_area_struct *vma)
1020 struct list_head tmp_list;
1021 struct uprobe *uprobe, *u;
1022 struct inode *inode;
1025 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1028 inode = vma->vm_file->f_mapping->host;
1032 mutex_lock(uprobes_mmap_hash(inode));
1033 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1038 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1040 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1042 ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1044 * We can race against uprobe_register(), see the
1045 * comment near uprobe_hash().
1047 if (ret == -EEXIST) {
1050 if (!is_swbp_at_addr(vma->vm_mm, vaddr))
1054 * Unable to insert a breakpoint, but
1055 * breakpoint lies underneath. Increment the
1058 atomic_inc(&vma->vm_mm->uprobes_state.count);
1067 mutex_unlock(uprobes_mmap_hash(inode));
1070 atomic_sub(count, &vma->vm_mm->uprobes_state.count);
1076 * Called in context of a munmap of a vma.
1078 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1080 struct list_head tmp_list;
1081 struct uprobe *uprobe, *u;
1082 struct inode *inode;
1084 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1087 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1090 if (!atomic_read(&vma->vm_mm->uprobes_state.count))
1093 inode = vma->vm_file->f_mapping->host;
1097 mutex_lock(uprobes_mmap_hash(inode));
1098 build_probe_list(inode, vma, start, end, &tmp_list);
1100 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1101 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1103 * An unregister could have removed the probe before
1104 * unmap. So check before we decrement the count.
1106 if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1)
1107 atomic_dec(&vma->vm_mm->uprobes_state.count);
1110 mutex_unlock(uprobes_mmap_hash(inode));
1113 /* Slot allocation for XOL */
1114 static int xol_add_vma(struct xol_area *area)
1116 struct mm_struct *mm;
1119 area->page = alloc_page(GFP_HIGHUSER);
1126 down_write(&mm->mmap_sem);
1127 if (mm->uprobes_state.xol_area)
1132 /* Try to map as high as possible, this is only a hint. */
1133 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1134 if (area->vaddr & ~PAGE_MASK) {
1139 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1140 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1144 smp_wmb(); /* pairs with get_xol_area() */
1145 mm->uprobes_state.xol_area = area;
1149 up_write(&mm->mmap_sem);
1151 __free_page(area->page);
1156 static struct xol_area *get_xol_area(struct mm_struct *mm)
1158 struct xol_area *area;
1160 area = mm->uprobes_state.xol_area;
1161 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1167 * xol_alloc_area - Allocate process's xol_area.
1168 * This area will be used for storing instructions for execution out of
1171 * Returns the allocated area or NULL.
1173 static struct xol_area *xol_alloc_area(void)
1175 struct xol_area *area;
1177 area = kzalloc(sizeof(*area), GFP_KERNEL);
1178 if (unlikely(!area))
1181 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1186 init_waitqueue_head(&area->wq);
1187 if (!xol_add_vma(area))
1191 kfree(area->bitmap);
1194 return get_xol_area(current->mm);
1198 * uprobe_clear_state - Free the area allocated for slots.
1200 void uprobe_clear_state(struct mm_struct *mm)
1202 struct xol_area *area = mm->uprobes_state.xol_area;
1207 put_page(area->page);
1208 kfree(area->bitmap);
1213 * uprobe_reset_state - Free the area allocated for slots.
1215 void uprobe_reset_state(struct mm_struct *mm)
1217 mm->uprobes_state.xol_area = NULL;
1218 atomic_set(&mm->uprobes_state.count, 0);
1222 * - search for a free slot.
1224 static unsigned long xol_take_insn_slot(struct xol_area *area)
1226 unsigned long slot_addr;
1230 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1231 if (slot_nr < UINSNS_PER_PAGE) {
1232 if (!test_and_set_bit(slot_nr, area->bitmap))
1235 slot_nr = UINSNS_PER_PAGE;
1238 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1239 } while (slot_nr >= UINSNS_PER_PAGE);
1241 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1242 atomic_inc(&area->slot_count);
1248 * xol_get_insn_slot - If was not allocated a slot, then
1250 * Returns the allocated slot address or 0.
1252 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1254 struct xol_area *area;
1255 unsigned long offset;
1258 area = get_xol_area(current->mm);
1260 area = xol_alloc_area();
1264 current->utask->xol_vaddr = xol_take_insn_slot(area);
1267 * Initialize the slot if xol_vaddr points to valid
1270 if (unlikely(!current->utask->xol_vaddr))
1273 current->utask->vaddr = slot_addr;
1274 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1275 vaddr = kmap_atomic(area->page);
1276 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1277 kunmap_atomic(vaddr);
1279 return current->utask->xol_vaddr;
1283 * xol_free_insn_slot - If slot was earlier allocated by
1284 * @xol_get_insn_slot(), make the slot available for
1285 * subsequent requests.
1287 static void xol_free_insn_slot(struct task_struct *tsk)
1289 struct xol_area *area;
1290 unsigned long vma_end;
1291 unsigned long slot_addr;
1293 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1296 slot_addr = tsk->utask->xol_vaddr;
1298 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1301 area = tsk->mm->uprobes_state.xol_area;
1302 vma_end = area->vaddr + PAGE_SIZE;
1303 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1304 unsigned long offset;
1307 offset = slot_addr - area->vaddr;
1308 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1309 if (slot_nr >= UINSNS_PER_PAGE)
1312 clear_bit(slot_nr, area->bitmap);
1313 atomic_dec(&area->slot_count);
1314 if (waitqueue_active(&area->wq))
1317 tsk->utask->xol_vaddr = 0;
1322 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1323 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1325 * Return the address of the breakpoint instruction.
1327 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1329 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1333 * Called with no locks held.
1334 * Called in context of a exiting or a exec-ing thread.
1336 void uprobe_free_utask(struct task_struct *t)
1338 struct uprobe_task *utask = t->utask;
1343 if (utask->active_uprobe)
1344 put_uprobe(utask->active_uprobe);
1346 xol_free_insn_slot(t);
1352 * Called in context of a new clone/fork from copy_process.
1354 void uprobe_copy_process(struct task_struct *t)
1360 * Allocate a uprobe_task object for the task.
1361 * Called when the thread hits a breakpoint for the first time.
1364 * - pointer to new uprobe_task on success
1367 static struct uprobe_task *add_utask(void)
1369 struct uprobe_task *utask;
1371 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1372 if (unlikely(!utask))
1375 current->utask = utask;
1379 /* Prepare to single-step probed instruction out of line. */
1381 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1383 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1390 * If we are singlestepping, then ensure this thread is not connected to
1391 * non-fatal signals until completion of singlestep. When xol insn itself
1392 * triggers the signal, restart the original insn even if the task is
1393 * already SIGKILL'ed (since coredump should report the correct ip). This
1394 * is even more important if the task has a handler for SIGSEGV/etc, The
1395 * _same_ instruction should be repeated again after return from the signal
1396 * handler, and SSTEP can never finish in this case.
1398 bool uprobe_deny_signal(void)
1400 struct task_struct *t = current;
1401 struct uprobe_task *utask = t->utask;
1403 if (likely(!utask || !utask->active_uprobe))
1406 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1408 if (signal_pending(t)) {
1409 spin_lock_irq(&t->sighand->siglock);
1410 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1411 spin_unlock_irq(&t->sighand->siglock);
1413 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1414 utask->state = UTASK_SSTEP_TRAPPED;
1415 set_tsk_thread_flag(t, TIF_UPROBE);
1416 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1424 * Avoid singlestepping the original instruction if the original instruction
1425 * is a NOP or can be emulated.
1427 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1429 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1432 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1436 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1438 struct mm_struct *mm = current->mm;
1439 struct uprobe *uprobe = NULL;
1440 struct vm_area_struct *vma;
1442 down_read(&mm->mmap_sem);
1443 vma = find_vma(mm, bp_vaddr);
1444 if (vma && vma->vm_start <= bp_vaddr) {
1445 if (valid_vma(vma, false)) {
1446 struct inode *inode = vma->vm_file->f_mapping->host;
1447 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1449 uprobe = find_uprobe(inode, offset);
1453 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1457 up_read(&mm->mmap_sem);
1463 * Run handler and ask thread to singlestep.
1464 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1466 static void handle_swbp(struct pt_regs *regs)
1468 struct uprobe_task *utask;
1469 struct uprobe *uprobe;
1470 unsigned long bp_vaddr;
1471 int uninitialized_var(is_swbp);
1473 bp_vaddr = uprobe_get_swbp_addr(regs);
1474 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1478 /* No matching uprobe; signal SIGTRAP. */
1479 send_sig(SIGTRAP, current, 0);
1482 * Either we raced with uprobe_unregister() or we can't
1483 * access this memory. The latter is only possible if
1484 * another thread plays with our ->mm. In both cases
1485 * we can simply restart. If this vma was unmapped we
1486 * can pretend this insn was not executed yet and get
1487 * the (correct) SIGSEGV after restart.
1489 instruction_pointer_set(regs, bp_vaddr);
1494 utask = current->utask;
1496 utask = add_utask();
1497 /* Cannot allocate; re-execute the instruction. */
1501 utask->active_uprobe = uprobe;
1502 handler_chain(uprobe, regs);
1503 if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs))
1506 utask->state = UTASK_SSTEP;
1507 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1508 user_enable_single_step(current);
1514 utask->active_uprobe = NULL;
1515 utask->state = UTASK_RUNNING;
1518 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1521 * cannot singlestep; cannot skip instruction;
1522 * re-execute the instruction.
1524 instruction_pointer_set(regs, bp_vaddr);
1531 * Perform required fix-ups and disable singlestep.
1532 * Allow pending signals to take effect.
1534 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1536 struct uprobe *uprobe;
1538 uprobe = utask->active_uprobe;
1539 if (utask->state == UTASK_SSTEP_ACK)
1540 arch_uprobe_post_xol(&uprobe->arch, regs);
1541 else if (utask->state == UTASK_SSTEP_TRAPPED)
1542 arch_uprobe_abort_xol(&uprobe->arch, regs);
1547 utask->active_uprobe = NULL;
1548 utask->state = UTASK_RUNNING;
1549 user_disable_single_step(current);
1550 xol_free_insn_slot(current);
1552 spin_lock_irq(¤t->sighand->siglock);
1553 recalc_sigpending(); /* see uprobe_deny_signal() */
1554 spin_unlock_irq(¤t->sighand->siglock);
1558 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on
1559 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1560 * allows the thread to return from interrupt.
1562 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1563 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1566 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1567 * uprobe_notify_resume().
1569 void uprobe_notify_resume(struct pt_regs *regs)
1571 struct uprobe_task *utask;
1573 utask = current->utask;
1574 if (!utask || utask->state == UTASK_BP_HIT)
1577 handle_singlestep(utask, regs);
1581 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1582 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1584 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1586 struct uprobe_task *utask;
1588 if (!current->mm || !atomic_read(¤t->mm->uprobes_state.count))
1589 /* task is currently not uprobed */
1592 utask = current->utask;
1594 utask->state = UTASK_BP_HIT;
1596 set_thread_flag(TIF_UPROBE);
1602 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1603 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1605 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1607 struct uprobe_task *utask = current->utask;
1609 if (!current->mm || !utask || !utask->active_uprobe)
1610 /* task is currently not uprobed */
1613 utask->state = UTASK_SSTEP_ACK;
1614 set_thread_flag(TIF_UPROBE);
1618 static struct notifier_block uprobe_exception_nb = {
1619 .notifier_call = arch_uprobe_exception_notify,
1620 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1623 static int __init init_uprobes(void)
1627 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1628 mutex_init(&uprobes_mutex[i]);
1629 mutex_init(&uprobes_mmap_mutex[i]);
1632 return register_die_notifier(&uprobe_exception_nb);
1634 module_init(init_uprobes);
1636 static void __exit exit_uprobes(void)
1639 module_exit(exit_uprobes);