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 loff_t vma_address(struct vm_area_struct *vma, loff_t offset)
119 vaddr = vma->vm_start + offset;
120 vaddr -= vma->vm_pgoff << PAGE_SHIFT;
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 * @page: the cowed page we are replacing by kpage
131 * @kpage: the modified page we replace page by
133 * Returns 0 on success, -EFAULT on failure.
135 static int __replace_page(struct vm_area_struct *vma, struct page *page, struct page *kpage)
137 struct mm_struct *mm = vma->vm_mm;
142 addr = page_address_in_vma(page, vma);
146 ptep = page_check_address(page, mm, addr, &ptl, 0);
151 page_add_new_anon_rmap(kpage, vma, addr);
153 if (!PageAnon(page)) {
154 dec_mm_counter(mm, MM_FILEPAGES);
155 inc_mm_counter(mm, MM_ANONPAGES);
158 flush_cache_page(vma, addr, pte_pfn(*ptep));
159 ptep_clear_flush(vma, addr, ptep);
160 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
162 page_remove_rmap(page);
163 if (!page_mapped(page))
164 try_to_free_swap(page);
166 pte_unmap_unlock(ptep, ptl);
172 * is_swbp_insn - check if instruction is breakpoint instruction.
173 * @insn: instruction to be checked.
174 * Default implementation of is_swbp_insn
175 * Returns true if @insn is a breakpoint instruction.
177 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
179 return *insn == UPROBE_SWBP_INSN;
184 * Expect the breakpoint instruction to be the smallest size instruction for
185 * the architecture. If an arch has variable length instruction and the
186 * breakpoint instruction is not of the smallest length instruction
187 * supported by that architecture then we need to modify read_opcode /
188 * write_opcode accordingly. This would never be a problem for archs that
189 * have fixed length instructions.
193 * write_opcode - write the opcode at a given virtual address.
194 * @auprobe: arch breakpointing information.
195 * @mm: the probed process address space.
196 * @vaddr: the virtual address to store the opcode.
197 * @opcode: opcode to be written at @vaddr.
199 * Called with mm->mmap_sem held (for read and with a reference to
202 * For mm @mm, write the opcode at @vaddr.
203 * Return 0 (success) or a negative errno.
205 static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
206 unsigned long vaddr, uprobe_opcode_t opcode)
208 struct page *old_page, *new_page;
209 struct address_space *mapping;
210 void *vaddr_old, *vaddr_new;
211 struct vm_area_struct *vma;
212 struct uprobe *uprobe;
215 /* Read the page with vaddr into memory */
216 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
223 * We are interested in text pages only. Our pages of interest
224 * should be mapped for read and execute only. We desist from
225 * adding probes in write mapped pages since the breakpoints
226 * might end up in the file copy.
228 if (!valid_vma(vma, is_swbp_insn(&opcode)))
231 uprobe = container_of(auprobe, struct uprobe, arch);
232 mapping = uprobe->inode->i_mapping;
233 if (mapping != vma->vm_file->f_mapping)
237 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
241 __SetPageUptodate(new_page);
244 * lock page will serialize against do_wp_page()'s
245 * PageAnon() handling
248 /* copy the page now that we've got it stable */
249 vaddr_old = kmap_atomic(old_page);
250 vaddr_new = kmap_atomic(new_page);
252 memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
253 memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
255 kunmap_atomic(vaddr_new);
256 kunmap_atomic(vaddr_old);
258 ret = anon_vma_prepare(vma);
263 ret = __replace_page(vma, old_page, new_page);
264 unlock_page(new_page);
267 unlock_page(old_page);
268 page_cache_release(new_page);
273 if (unlikely(ret == -EAGAIN))
279 * read_opcode - read the opcode at a given virtual address.
280 * @mm: the probed process address space.
281 * @vaddr: the virtual address to read the opcode.
282 * @opcode: location to store the read opcode.
284 * Called with mm->mmap_sem held (for read and with a reference to
287 * For mm @mm, read the opcode at @vaddr and store it in @opcode.
288 * Return 0 (success) or a negative errno.
290 static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode)
296 ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
301 vaddr_new = kmap_atomic(page);
303 memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE);
304 kunmap_atomic(vaddr_new);
312 static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
314 uprobe_opcode_t opcode;
317 if (current->mm == mm) {
319 result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
323 if (likely(result == 0))
327 result = read_opcode(mm, vaddr, &opcode);
331 if (is_swbp_insn(&opcode))
338 * set_swbp - store breakpoint at a given address.
339 * @auprobe: arch specific probepoint information.
340 * @mm: the probed process address space.
341 * @vaddr: the virtual address to insert the opcode.
343 * For mm @mm, store the breakpoint instruction at @vaddr.
344 * Return 0 (success) or a negative errno.
346 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
350 * See the comment near uprobes_hash().
352 result = is_swbp_at_addr(mm, vaddr);
359 return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
363 * set_orig_insn - Restore the original instruction.
364 * @mm: the probed process address space.
365 * @auprobe: arch specific probepoint information.
366 * @vaddr: the virtual address to insert the opcode.
367 * @verify: if true, verify existance of breakpoint instruction.
369 * For mm @mm, restore the original opcode (opcode) at @vaddr.
370 * Return 0 (success) or a negative errno.
373 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify)
378 result = is_swbp_at_addr(mm, vaddr);
385 return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
388 static int match_uprobe(struct uprobe *l, struct uprobe *r)
390 if (l->inode < r->inode)
393 if (l->inode > r->inode)
396 if (l->offset < r->offset)
399 if (l->offset > r->offset)
405 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
407 struct uprobe u = { .inode = inode, .offset = offset };
408 struct rb_node *n = uprobes_tree.rb_node;
409 struct uprobe *uprobe;
413 uprobe = rb_entry(n, struct uprobe, rb_node);
414 match = match_uprobe(&u, uprobe);
416 atomic_inc(&uprobe->ref);
429 * Find a uprobe corresponding to a given inode:offset
430 * Acquires uprobes_treelock
432 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
434 struct uprobe *uprobe;
437 spin_lock_irqsave(&uprobes_treelock, flags);
438 uprobe = __find_uprobe(inode, offset);
439 spin_unlock_irqrestore(&uprobes_treelock, flags);
444 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
446 struct rb_node **p = &uprobes_tree.rb_node;
447 struct rb_node *parent = NULL;
453 u = rb_entry(parent, struct uprobe, rb_node);
454 match = match_uprobe(uprobe, u);
461 p = &parent->rb_left;
463 p = &parent->rb_right;
468 rb_link_node(&uprobe->rb_node, parent, p);
469 rb_insert_color(&uprobe->rb_node, &uprobes_tree);
470 /* get access + creation ref */
471 atomic_set(&uprobe->ref, 2);
477 * Acquire uprobes_treelock.
478 * Matching uprobe already exists in rbtree;
479 * increment (access refcount) and return the matching uprobe.
481 * No matching uprobe; insert the uprobe in rb_tree;
482 * get a double refcount (access + creation) and return NULL.
484 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
489 spin_lock_irqsave(&uprobes_treelock, flags);
490 u = __insert_uprobe(uprobe);
491 spin_unlock_irqrestore(&uprobes_treelock, flags);
493 /* For now assume that the instruction need not be single-stepped */
494 uprobe->flags |= UPROBE_SKIP_SSTEP;
499 static void put_uprobe(struct uprobe *uprobe)
501 if (atomic_dec_and_test(&uprobe->ref))
505 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
507 struct uprobe *uprobe, *cur_uprobe;
509 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
513 uprobe->inode = igrab(inode);
514 uprobe->offset = offset;
515 init_rwsem(&uprobe->consumer_rwsem);
516 INIT_LIST_HEAD(&uprobe->pending_list);
518 /* add to uprobes_tree, sorted on inode:offset */
519 cur_uprobe = insert_uprobe(uprobe);
521 /* a uprobe exists for this inode:offset combination */
527 atomic_inc(&uprobe_events);
533 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
535 struct uprobe_consumer *uc;
537 if (!(uprobe->flags & UPROBE_RUN_HANDLER))
540 down_read(&uprobe->consumer_rwsem);
541 for (uc = uprobe->consumers; uc; uc = uc->next) {
542 if (!uc->filter || uc->filter(uc, current))
543 uc->handler(uc, regs);
545 up_read(&uprobe->consumer_rwsem);
548 /* Returns the previous consumer */
549 static struct uprobe_consumer *
550 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
552 down_write(&uprobe->consumer_rwsem);
553 uc->next = uprobe->consumers;
554 uprobe->consumers = uc;
555 up_write(&uprobe->consumer_rwsem);
561 * For uprobe @uprobe, delete the consumer @uc.
562 * Return true if the @uc is deleted successfully
565 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
567 struct uprobe_consumer **con;
570 down_write(&uprobe->consumer_rwsem);
571 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
578 up_write(&uprobe->consumer_rwsem);
584 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
585 unsigned long nbytes, unsigned long offset)
595 if (!mapping->a_ops->readpage)
598 idx = (unsigned long)(offset >> PAGE_CACHE_SHIFT);
599 off1 = offset &= ~PAGE_MASK;
602 * Ensure that the page that has the original instruction is
603 * populated and in page-cache.
605 page = read_mapping_page(mapping, idx, filp);
607 return PTR_ERR(page);
609 vaddr = kmap_atomic(page);
610 memcpy(insn, vaddr + off1, nbytes);
611 kunmap_atomic(vaddr);
612 page_cache_release(page);
617 static int copy_insn(struct uprobe *uprobe, struct file *filp)
619 struct address_space *mapping;
620 unsigned long nbytes;
623 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
624 mapping = uprobe->inode->i_mapping;
626 /* Instruction at end of binary; copy only available bytes */
627 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
628 bytes = uprobe->inode->i_size - uprobe->offset;
630 bytes = MAX_UINSN_BYTES;
632 /* Instruction at the page-boundary; copy bytes in second page */
633 if (nbytes < bytes) {
634 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
635 bytes - nbytes, uprobe->offset + nbytes);
640 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
644 * How mm->uprobes_state.count gets updated
645 * uprobe_mmap() increments the count if
646 * - it successfully adds a breakpoint.
647 * - it cannot add a breakpoint, but sees that there is a underlying
648 * breakpoint (via a is_swbp_at_addr()).
650 * uprobe_munmap() decrements the count if
651 * - it sees a underlying breakpoint, (via is_swbp_at_addr)
652 * (Subsequent uprobe_unregister wouldnt find the breakpoint
653 * unless a uprobe_mmap kicks in, since the old vma would be
654 * dropped just after uprobe_munmap.)
656 * uprobe_register increments the count if:
657 * - it successfully adds a breakpoint.
659 * uprobe_unregister decrements the count if:
660 * - it sees a underlying breakpoint and removes successfully.
661 * (via is_swbp_at_addr)
662 * (Subsequent uprobe_munmap wouldnt find the breakpoint
663 * since there is no underlying breakpoint after the
664 * breakpoint removal.)
667 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
668 struct vm_area_struct *vma, loff_t vaddr)
674 * If probe is being deleted, unregister thread could be done with
675 * the vma-rmap-walk through. Adding a probe now can be fatal since
676 * nobody will be able to cleanup. Also we could be from fork or
677 * mremap path, where the probe might have already been inserted.
678 * Hence behave as if probe already existed.
680 if (!uprobe->consumers)
683 addr = (unsigned long)vaddr;
685 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
686 ret = copy_insn(uprobe, vma->vm_file);
690 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
693 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, addr);
697 /* write_opcode() assumes we don't cross page boundary */
698 BUG_ON((uprobe->offset & ~PAGE_MASK) +
699 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
701 uprobe->flags |= UPROBE_COPY_INSN;
705 * Ideally, should be updating the probe count after the breakpoint
706 * has been successfully inserted. However a thread could hit the
707 * breakpoint we just inserted even before the probe count is
708 * incremented. If this is the first breakpoint placed, breakpoint
709 * notifier might ignore uprobes and pass the trap to the thread.
710 * Hence increment before and decrement on failure.
712 atomic_inc(&mm->uprobes_state.count);
713 ret = set_swbp(&uprobe->arch, mm, addr);
715 atomic_dec(&mm->uprobes_state.count);
721 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, loff_t vaddr)
723 if (!set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true))
724 atomic_dec(&mm->uprobes_state.count);
728 * There could be threads that have already hit the breakpoint. They
729 * will recheck the current insn and restart if find_uprobe() fails.
730 * See find_active_uprobe().
732 static void delete_uprobe(struct uprobe *uprobe)
736 spin_lock_irqsave(&uprobes_treelock, flags);
737 rb_erase(&uprobe->rb_node, &uprobes_tree);
738 spin_unlock_irqrestore(&uprobes_treelock, flags);
741 atomic_dec(&uprobe_events);
745 struct map_info *next;
746 struct mm_struct *mm;
750 static inline struct map_info *free_map_info(struct map_info *info)
752 struct map_info *next = info->next;
757 static struct map_info *
758 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
760 unsigned long pgoff = offset >> PAGE_SHIFT;
761 struct prio_tree_iter iter;
762 struct vm_area_struct *vma;
763 struct map_info *curr = NULL;
764 struct map_info *prev = NULL;
765 struct map_info *info;
769 mutex_lock(&mapping->i_mmap_mutex);
770 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
771 if (!valid_vma(vma, is_register))
774 if (!prev && !more) {
776 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
777 * reclaim. This is optimistic, no harm done if it fails.
779 prev = kmalloc(sizeof(struct map_info),
780 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
789 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
797 info->mm = vma->vm_mm;
798 info->vaddr = vma_address(vma, offset);
800 mutex_unlock(&mapping->i_mmap_mutex);
812 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
814 curr = ERR_PTR(-ENOMEM);
824 prev = free_map_info(prev);
828 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
830 struct map_info *info;
833 info = build_map_info(uprobe->inode->i_mapping,
834 uprobe->offset, is_register);
836 return PTR_ERR(info);
839 struct mm_struct *mm = info->mm;
840 struct vm_area_struct *vma;
846 down_write(&mm->mmap_sem);
847 vma = find_vma(mm, (unsigned long)info->vaddr);
848 if (!vma || !valid_vma(vma, is_register))
851 vaddr = vma_address(vma, uprobe->offset);
852 if (vma->vm_file->f_mapping->host != uprobe->inode ||
853 vaddr != info->vaddr)
857 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
859 * We can race against uprobe_mmap(), see the
860 * comment near uprobe_hash().
865 remove_breakpoint(uprobe, mm, info->vaddr);
868 up_write(&mm->mmap_sem);
871 info = free_map_info(info);
877 static int __uprobe_register(struct uprobe *uprobe)
879 return register_for_each_vma(uprobe, true);
882 static void __uprobe_unregister(struct uprobe *uprobe)
884 if (!register_for_each_vma(uprobe, false))
885 delete_uprobe(uprobe);
887 /* TODO : cant unregister? schedule a worker thread */
891 * uprobe_register - register a probe
892 * @inode: the file in which the probe has to be placed.
893 * @offset: offset from the start of the file.
894 * @uc: information on howto handle the probe..
896 * Apart from the access refcount, uprobe_register() takes a creation
897 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
898 * inserted into the rbtree (i.e first consumer for a @inode:@offset
899 * tuple). Creation refcount stops uprobe_unregister from freeing the
900 * @uprobe even before the register operation is complete. Creation
901 * refcount is released when the last @uc for the @uprobe
904 * Return errno if it cannot successully install probes
905 * else return 0 (success)
907 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
909 struct uprobe *uprobe;
912 if (!inode || !uc || uc->next)
915 if (offset > i_size_read(inode))
919 mutex_lock(uprobes_hash(inode));
920 uprobe = alloc_uprobe(inode, offset);
922 if (uprobe && !consumer_add(uprobe, uc)) {
923 ret = __uprobe_register(uprobe);
925 uprobe->consumers = NULL;
926 __uprobe_unregister(uprobe);
928 uprobe->flags |= UPROBE_RUN_HANDLER;
932 mutex_unlock(uprobes_hash(inode));
939 * uprobe_unregister - unregister a already registered probe.
940 * @inode: the file in which the probe has to be removed.
941 * @offset: offset from the start of the file.
942 * @uc: identify which probe if multiple probes are colocated.
944 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
946 struct uprobe *uprobe;
951 uprobe = find_uprobe(inode, offset);
955 mutex_lock(uprobes_hash(inode));
957 if (consumer_del(uprobe, uc)) {
958 if (!uprobe->consumers) {
959 __uprobe_unregister(uprobe);
960 uprobe->flags &= ~UPROBE_RUN_HANDLER;
964 mutex_unlock(uprobes_hash(inode));
970 * Of all the nodes that correspond to the given inode, return the node
971 * with the least offset.
973 static struct rb_node *find_least_offset_node(struct inode *inode)
975 struct uprobe u = { .inode = inode, .offset = 0};
976 struct rb_node *n = uprobes_tree.rb_node;
977 struct rb_node *close_node = NULL;
978 struct uprobe *uprobe;
982 uprobe = rb_entry(n, struct uprobe, rb_node);
983 match = match_uprobe(&u, uprobe);
985 if (uprobe->inode == inode)
1001 * For a given inode, build a list of probes that need to be inserted.
1003 static void build_probe_list(struct inode *inode, struct list_head *head)
1005 struct uprobe *uprobe;
1006 unsigned long flags;
1009 spin_lock_irqsave(&uprobes_treelock, flags);
1011 n = find_least_offset_node(inode);
1013 for (; n; n = rb_next(n)) {
1014 uprobe = rb_entry(n, struct uprobe, rb_node);
1015 if (uprobe->inode != inode)
1018 list_add(&uprobe->pending_list, head);
1019 atomic_inc(&uprobe->ref);
1022 spin_unlock_irqrestore(&uprobes_treelock, flags);
1026 * Called from mmap_region.
1027 * called with mm->mmap_sem acquired.
1029 * Return -ve no if we fail to insert probes and we cannot
1031 * Return 0 otherwise. i.e:
1033 * - successful insertion of probes
1034 * - (or) no possible probes to be inserted.
1035 * - (or) insertion of probes failed but we can bail-out.
1037 int uprobe_mmap(struct vm_area_struct *vma)
1039 struct list_head tmp_list;
1040 struct uprobe *uprobe, *u;
1041 struct inode *inode;
1044 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1047 inode = vma->vm_file->f_mapping->host;
1051 INIT_LIST_HEAD(&tmp_list);
1052 mutex_lock(uprobes_mmap_hash(inode));
1053 build_probe_list(inode, &tmp_list);
1058 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1061 list_del(&uprobe->pending_list);
1063 vaddr = vma_address(vma, uprobe->offset);
1065 if (vaddr < vma->vm_start || vaddr >= vma->vm_end) {
1070 ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1072 * We can race against uprobe_register(), see the
1073 * comment near uprobe_hash().
1075 if (ret == -EEXIST) {
1078 if (!is_swbp_at_addr(vma->vm_mm, vaddr))
1082 * Unable to insert a breakpoint, but
1083 * breakpoint lies underneath. Increment the
1086 atomic_inc(&vma->vm_mm->uprobes_state.count);
1095 mutex_unlock(uprobes_mmap_hash(inode));
1098 atomic_sub(count, &vma->vm_mm->uprobes_state.count);
1104 * Called in context of a munmap of a vma.
1106 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1108 struct list_head tmp_list;
1109 struct uprobe *uprobe, *u;
1110 struct inode *inode;
1112 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1115 if (!atomic_read(&vma->vm_mm->uprobes_state.count))
1118 inode = vma->vm_file->f_mapping->host;
1122 INIT_LIST_HEAD(&tmp_list);
1123 mutex_lock(uprobes_mmap_hash(inode));
1124 build_probe_list(inode, &tmp_list);
1126 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1129 list_del(&uprobe->pending_list);
1130 vaddr = vma_address(vma, uprobe->offset);
1132 if (vaddr >= start && vaddr < end) {
1134 * An unregister could have removed the probe before
1135 * unmap. So check before we decrement the count.
1137 if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1)
1138 atomic_dec(&vma->vm_mm->uprobes_state.count);
1142 mutex_unlock(uprobes_mmap_hash(inode));
1145 /* Slot allocation for XOL */
1146 static int xol_add_vma(struct xol_area *area)
1148 struct mm_struct *mm;
1151 area->page = alloc_page(GFP_HIGHUSER);
1158 down_write(&mm->mmap_sem);
1159 if (mm->uprobes_state.xol_area)
1164 /* Try to map as high as possible, this is only a hint. */
1165 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1166 if (area->vaddr & ~PAGE_MASK) {
1171 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1172 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1176 smp_wmb(); /* pairs with get_xol_area() */
1177 mm->uprobes_state.xol_area = area;
1181 up_write(&mm->mmap_sem);
1183 __free_page(area->page);
1188 static struct xol_area *get_xol_area(struct mm_struct *mm)
1190 struct xol_area *area;
1192 area = mm->uprobes_state.xol_area;
1193 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1199 * xol_alloc_area - Allocate process's xol_area.
1200 * This area will be used for storing instructions for execution out of
1203 * Returns the allocated area or NULL.
1205 static struct xol_area *xol_alloc_area(void)
1207 struct xol_area *area;
1209 area = kzalloc(sizeof(*area), GFP_KERNEL);
1210 if (unlikely(!area))
1213 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1218 init_waitqueue_head(&area->wq);
1219 if (!xol_add_vma(area))
1223 kfree(area->bitmap);
1226 return get_xol_area(current->mm);
1230 * uprobe_clear_state - Free the area allocated for slots.
1232 void uprobe_clear_state(struct mm_struct *mm)
1234 struct xol_area *area = mm->uprobes_state.xol_area;
1239 put_page(area->page);
1240 kfree(area->bitmap);
1245 * uprobe_reset_state - Free the area allocated for slots.
1247 void uprobe_reset_state(struct mm_struct *mm)
1249 mm->uprobes_state.xol_area = NULL;
1250 atomic_set(&mm->uprobes_state.count, 0);
1254 * - search for a free slot.
1256 static unsigned long xol_take_insn_slot(struct xol_area *area)
1258 unsigned long slot_addr;
1262 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1263 if (slot_nr < UINSNS_PER_PAGE) {
1264 if (!test_and_set_bit(slot_nr, area->bitmap))
1267 slot_nr = UINSNS_PER_PAGE;
1270 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1271 } while (slot_nr >= UINSNS_PER_PAGE);
1273 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1274 atomic_inc(&area->slot_count);
1280 * xol_get_insn_slot - If was not allocated a slot, then
1282 * Returns the allocated slot address or 0.
1284 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1286 struct xol_area *area;
1287 unsigned long offset;
1290 area = get_xol_area(current->mm);
1292 area = xol_alloc_area();
1296 current->utask->xol_vaddr = xol_take_insn_slot(area);
1299 * Initialize the slot if xol_vaddr points to valid
1302 if (unlikely(!current->utask->xol_vaddr))
1305 current->utask->vaddr = slot_addr;
1306 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1307 vaddr = kmap_atomic(area->page);
1308 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1309 kunmap_atomic(vaddr);
1311 return current->utask->xol_vaddr;
1315 * xol_free_insn_slot - If slot was earlier allocated by
1316 * @xol_get_insn_slot(), make the slot available for
1317 * subsequent requests.
1319 static void xol_free_insn_slot(struct task_struct *tsk)
1321 struct xol_area *area;
1322 unsigned long vma_end;
1323 unsigned long slot_addr;
1325 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1328 slot_addr = tsk->utask->xol_vaddr;
1330 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1333 area = tsk->mm->uprobes_state.xol_area;
1334 vma_end = area->vaddr + PAGE_SIZE;
1335 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1336 unsigned long offset;
1339 offset = slot_addr - area->vaddr;
1340 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1341 if (slot_nr >= UINSNS_PER_PAGE)
1344 clear_bit(slot_nr, area->bitmap);
1345 atomic_dec(&area->slot_count);
1346 if (waitqueue_active(&area->wq))
1349 tsk->utask->xol_vaddr = 0;
1354 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1355 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1357 * Return the address of the breakpoint instruction.
1359 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1361 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1365 * Called with no locks held.
1366 * Called in context of a exiting or a exec-ing thread.
1368 void uprobe_free_utask(struct task_struct *t)
1370 struct uprobe_task *utask = t->utask;
1375 if (utask->active_uprobe)
1376 put_uprobe(utask->active_uprobe);
1378 xol_free_insn_slot(t);
1384 * Called in context of a new clone/fork from copy_process.
1386 void uprobe_copy_process(struct task_struct *t)
1392 * Allocate a uprobe_task object for the task.
1393 * Called when the thread hits a breakpoint for the first time.
1396 * - pointer to new uprobe_task on success
1399 static struct uprobe_task *add_utask(void)
1401 struct uprobe_task *utask;
1403 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1404 if (unlikely(!utask))
1407 utask->active_uprobe = NULL;
1408 current->utask = utask;
1412 /* Prepare to single-step probed instruction out of line. */
1414 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1416 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1423 * If we are singlestepping, then ensure this thread is not connected to
1424 * non-fatal signals until completion of singlestep. When xol insn itself
1425 * triggers the signal, restart the original insn even if the task is
1426 * already SIGKILL'ed (since coredump should report the correct ip). This
1427 * is even more important if the task has a handler for SIGSEGV/etc, The
1428 * _same_ instruction should be repeated again after return from the signal
1429 * handler, and SSTEP can never finish in this case.
1431 bool uprobe_deny_signal(void)
1433 struct task_struct *t = current;
1434 struct uprobe_task *utask = t->utask;
1436 if (likely(!utask || !utask->active_uprobe))
1439 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1441 if (signal_pending(t)) {
1442 spin_lock_irq(&t->sighand->siglock);
1443 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1444 spin_unlock_irq(&t->sighand->siglock);
1446 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1447 utask->state = UTASK_SSTEP_TRAPPED;
1448 set_tsk_thread_flag(t, TIF_UPROBE);
1449 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1457 * Avoid singlestepping the original instruction if the original instruction
1458 * is a NOP or can be emulated.
1460 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1462 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1465 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1469 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1471 struct mm_struct *mm = current->mm;
1472 struct uprobe *uprobe = NULL;
1473 struct vm_area_struct *vma;
1475 down_read(&mm->mmap_sem);
1476 vma = find_vma(mm, bp_vaddr);
1477 if (vma && vma->vm_start <= bp_vaddr) {
1478 if (valid_vma(vma, false)) {
1479 struct inode *inode;
1482 inode = vma->vm_file->f_mapping->host;
1483 offset = bp_vaddr - vma->vm_start;
1484 offset += (vma->vm_pgoff << PAGE_SHIFT);
1485 uprobe = find_uprobe(inode, offset);
1489 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1493 up_read(&mm->mmap_sem);
1499 * Run handler and ask thread to singlestep.
1500 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1502 static void handle_swbp(struct pt_regs *regs)
1504 struct uprobe_task *utask;
1505 struct uprobe *uprobe;
1506 unsigned long bp_vaddr;
1507 int uninitialized_var(is_swbp);
1509 bp_vaddr = uprobe_get_swbp_addr(regs);
1510 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1514 /* No matching uprobe; signal SIGTRAP. */
1515 send_sig(SIGTRAP, current, 0);
1518 * Either we raced with uprobe_unregister() or we can't
1519 * access this memory. The latter is only possible if
1520 * another thread plays with our ->mm. In both cases
1521 * we can simply restart. If this vma was unmapped we
1522 * can pretend this insn was not executed yet and get
1523 * the (correct) SIGSEGV after restart.
1525 instruction_pointer_set(regs, bp_vaddr);
1530 utask = current->utask;
1532 utask = add_utask();
1533 /* Cannot allocate; re-execute the instruction. */
1537 utask->active_uprobe = uprobe;
1538 handler_chain(uprobe, regs);
1539 if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs))
1542 utask->state = UTASK_SSTEP;
1543 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1544 user_enable_single_step(current);
1550 utask->active_uprobe = NULL;
1551 utask->state = UTASK_RUNNING;
1554 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1557 * cannot singlestep; cannot skip instruction;
1558 * re-execute the instruction.
1560 instruction_pointer_set(regs, bp_vaddr);
1567 * Perform required fix-ups and disable singlestep.
1568 * Allow pending signals to take effect.
1570 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1572 struct uprobe *uprobe;
1574 uprobe = utask->active_uprobe;
1575 if (utask->state == UTASK_SSTEP_ACK)
1576 arch_uprobe_post_xol(&uprobe->arch, regs);
1577 else if (utask->state == UTASK_SSTEP_TRAPPED)
1578 arch_uprobe_abort_xol(&uprobe->arch, regs);
1583 utask->active_uprobe = NULL;
1584 utask->state = UTASK_RUNNING;
1585 user_disable_single_step(current);
1586 xol_free_insn_slot(current);
1588 spin_lock_irq(¤t->sighand->siglock);
1589 recalc_sigpending(); /* see uprobe_deny_signal() */
1590 spin_unlock_irq(¤t->sighand->siglock);
1594 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on
1595 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1596 * allows the thread to return from interrupt.
1598 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1599 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1602 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1603 * uprobe_notify_resume().
1605 void uprobe_notify_resume(struct pt_regs *regs)
1607 struct uprobe_task *utask;
1609 utask = current->utask;
1610 if (!utask || utask->state == UTASK_BP_HIT)
1613 handle_singlestep(utask, regs);
1617 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1618 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1620 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1622 struct uprobe_task *utask;
1624 if (!current->mm || !atomic_read(¤t->mm->uprobes_state.count))
1625 /* task is currently not uprobed */
1628 utask = current->utask;
1630 utask->state = UTASK_BP_HIT;
1632 set_thread_flag(TIF_UPROBE);
1638 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1639 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1641 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1643 struct uprobe_task *utask = current->utask;
1645 if (!current->mm || !utask || !utask->active_uprobe)
1646 /* task is currently not uprobed */
1649 utask->state = UTASK_SSTEP_ACK;
1650 set_thread_flag(TIF_UPROBE);
1654 static struct notifier_block uprobe_exception_nb = {
1655 .notifier_call = arch_uprobe_exception_notify,
1656 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1659 static int __init init_uprobes(void)
1663 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1664 mutex_init(&uprobes_mutex[i]);
1665 mutex_init(&uprobes_mmap_mutex[i]);
1668 return register_die_notifier(&uprobe_exception_nb);
1670 module_init(init_uprobes);
1672 static void __exit exit_uprobes(void)
1675 module_exit(exit_uprobes);