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);
517 /* add to uprobes_tree, sorted on inode:offset */
518 cur_uprobe = insert_uprobe(uprobe);
520 /* a uprobe exists for this inode:offset combination */
526 atomic_inc(&uprobe_events);
532 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
534 struct uprobe_consumer *uc;
536 if (!(uprobe->flags & UPROBE_RUN_HANDLER))
539 down_read(&uprobe->consumer_rwsem);
540 for (uc = uprobe->consumers; uc; uc = uc->next) {
541 if (!uc->filter || uc->filter(uc, current))
542 uc->handler(uc, regs);
544 up_read(&uprobe->consumer_rwsem);
547 /* Returns the previous consumer */
548 static struct uprobe_consumer *
549 consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
551 down_write(&uprobe->consumer_rwsem);
552 uc->next = uprobe->consumers;
553 uprobe->consumers = uc;
554 up_write(&uprobe->consumer_rwsem);
560 * For uprobe @uprobe, delete the consumer @uc.
561 * Return true if the @uc is deleted successfully
564 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
566 struct uprobe_consumer **con;
569 down_write(&uprobe->consumer_rwsem);
570 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
577 up_write(&uprobe->consumer_rwsem);
583 __copy_insn(struct address_space *mapping, struct file *filp, char *insn,
584 unsigned long nbytes, unsigned long offset)
594 if (!mapping->a_ops->readpage)
597 idx = (unsigned long)(offset >> PAGE_CACHE_SHIFT);
598 off1 = offset &= ~PAGE_MASK;
601 * Ensure that the page that has the original instruction is
602 * populated and in page-cache.
604 page = read_mapping_page(mapping, idx, filp);
606 return PTR_ERR(page);
608 vaddr = kmap_atomic(page);
609 memcpy(insn, vaddr + off1, nbytes);
610 kunmap_atomic(vaddr);
611 page_cache_release(page);
616 static int copy_insn(struct uprobe *uprobe, struct file *filp)
618 struct address_space *mapping;
619 unsigned long nbytes;
622 nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
623 mapping = uprobe->inode->i_mapping;
625 /* Instruction at end of binary; copy only available bytes */
626 if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
627 bytes = uprobe->inode->i_size - uprobe->offset;
629 bytes = MAX_UINSN_BYTES;
631 /* Instruction at the page-boundary; copy bytes in second page */
632 if (nbytes < bytes) {
633 int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
634 bytes - nbytes, uprobe->offset + nbytes);
639 return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
643 * How mm->uprobes_state.count gets updated
644 * uprobe_mmap() increments the count if
645 * - it successfully adds a breakpoint.
646 * - it cannot add a breakpoint, but sees that there is a underlying
647 * breakpoint (via a is_swbp_at_addr()).
649 * uprobe_munmap() decrements the count if
650 * - it sees a underlying breakpoint, (via is_swbp_at_addr)
651 * (Subsequent uprobe_unregister wouldnt find the breakpoint
652 * unless a uprobe_mmap kicks in, since the old vma would be
653 * dropped just after uprobe_munmap.)
655 * uprobe_register increments the count if:
656 * - it successfully adds a breakpoint.
658 * uprobe_unregister decrements the count if:
659 * - it sees a underlying breakpoint and removes successfully.
660 * (via is_swbp_at_addr)
661 * (Subsequent uprobe_munmap wouldnt find the breakpoint
662 * since there is no underlying breakpoint after the
663 * breakpoint removal.)
666 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
667 struct vm_area_struct *vma, loff_t vaddr)
673 * If probe is being deleted, unregister thread could be done with
674 * the vma-rmap-walk through. Adding a probe now can be fatal since
675 * nobody will be able to cleanup. Also we could be from fork or
676 * mremap path, where the probe might have already been inserted.
677 * Hence behave as if probe already existed.
679 if (!uprobe->consumers)
682 addr = (unsigned long)vaddr;
684 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
685 ret = copy_insn(uprobe, vma->vm_file);
689 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
692 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, addr);
696 /* write_opcode() assumes we don't cross page boundary */
697 BUG_ON((uprobe->offset & ~PAGE_MASK) +
698 UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
700 uprobe->flags |= UPROBE_COPY_INSN;
704 * Ideally, should be updating the probe count after the breakpoint
705 * has been successfully inserted. However a thread could hit the
706 * breakpoint we just inserted even before the probe count is
707 * incremented. If this is the first breakpoint placed, breakpoint
708 * notifier might ignore uprobes and pass the trap to the thread.
709 * Hence increment before and decrement on failure.
711 atomic_inc(&mm->uprobes_state.count);
712 ret = set_swbp(&uprobe->arch, mm, addr);
714 atomic_dec(&mm->uprobes_state.count);
720 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, loff_t vaddr)
722 if (!set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true))
723 atomic_dec(&mm->uprobes_state.count);
727 * There could be threads that have already hit the breakpoint. They
728 * will recheck the current insn and restart if find_uprobe() fails.
729 * See find_active_uprobe().
731 static void delete_uprobe(struct uprobe *uprobe)
735 spin_lock_irqsave(&uprobes_treelock, flags);
736 rb_erase(&uprobe->rb_node, &uprobes_tree);
737 spin_unlock_irqrestore(&uprobes_treelock, flags);
740 atomic_dec(&uprobe_events);
744 struct map_info *next;
745 struct mm_struct *mm;
749 static inline struct map_info *free_map_info(struct map_info *info)
751 struct map_info *next = info->next;
756 static struct map_info *
757 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
759 unsigned long pgoff = offset >> PAGE_SHIFT;
760 struct prio_tree_iter iter;
761 struct vm_area_struct *vma;
762 struct map_info *curr = NULL;
763 struct map_info *prev = NULL;
764 struct map_info *info;
768 mutex_lock(&mapping->i_mmap_mutex);
769 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
770 if (!valid_vma(vma, is_register))
773 if (!prev && !more) {
775 * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
776 * reclaim. This is optimistic, no harm done if it fails.
778 prev = kmalloc(sizeof(struct map_info),
779 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
788 if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
796 info->mm = vma->vm_mm;
797 info->vaddr = vma_address(vma, offset);
799 mutex_unlock(&mapping->i_mmap_mutex);
811 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
813 curr = ERR_PTR(-ENOMEM);
823 prev = free_map_info(prev);
827 static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
829 struct map_info *info;
832 info = build_map_info(uprobe->inode->i_mapping,
833 uprobe->offset, is_register);
835 return PTR_ERR(info);
838 struct mm_struct *mm = info->mm;
839 struct vm_area_struct *vma;
845 down_write(&mm->mmap_sem);
846 vma = find_vma(mm, (unsigned long)info->vaddr);
847 if (!vma || !valid_vma(vma, is_register))
850 vaddr = vma_address(vma, uprobe->offset);
851 if (vma->vm_file->f_mapping->host != uprobe->inode ||
852 vaddr != info->vaddr)
856 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
858 * We can race against uprobe_mmap(), see the
859 * comment near uprobe_hash().
864 remove_breakpoint(uprobe, mm, info->vaddr);
867 up_write(&mm->mmap_sem);
870 info = free_map_info(info);
876 static int __uprobe_register(struct uprobe *uprobe)
878 return register_for_each_vma(uprobe, true);
881 static void __uprobe_unregister(struct uprobe *uprobe)
883 if (!register_for_each_vma(uprobe, false))
884 delete_uprobe(uprobe);
886 /* TODO : cant unregister? schedule a worker thread */
890 * uprobe_register - register a probe
891 * @inode: the file in which the probe has to be placed.
892 * @offset: offset from the start of the file.
893 * @uc: information on howto handle the probe..
895 * Apart from the access refcount, uprobe_register() takes a creation
896 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
897 * inserted into the rbtree (i.e first consumer for a @inode:@offset
898 * tuple). Creation refcount stops uprobe_unregister from freeing the
899 * @uprobe even before the register operation is complete. Creation
900 * refcount is released when the last @uc for the @uprobe
903 * Return errno if it cannot successully install probes
904 * else return 0 (success)
906 int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
908 struct uprobe *uprobe;
911 if (!inode || !uc || uc->next)
914 if (offset > i_size_read(inode))
918 mutex_lock(uprobes_hash(inode));
919 uprobe = alloc_uprobe(inode, offset);
921 if (uprobe && !consumer_add(uprobe, uc)) {
922 ret = __uprobe_register(uprobe);
924 uprobe->consumers = NULL;
925 __uprobe_unregister(uprobe);
927 uprobe->flags |= UPROBE_RUN_HANDLER;
931 mutex_unlock(uprobes_hash(inode));
938 * uprobe_unregister - unregister a already registered probe.
939 * @inode: the file in which the probe has to be removed.
940 * @offset: offset from the start of the file.
941 * @uc: identify which probe if multiple probes are colocated.
943 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
945 struct uprobe *uprobe;
950 uprobe = find_uprobe(inode, offset);
954 mutex_lock(uprobes_hash(inode));
956 if (consumer_del(uprobe, uc)) {
957 if (!uprobe->consumers) {
958 __uprobe_unregister(uprobe);
959 uprobe->flags &= ~UPROBE_RUN_HANDLER;
963 mutex_unlock(uprobes_hash(inode));
969 * Of all the nodes that correspond to the given inode, return the node
970 * with the least offset.
972 static struct rb_node *find_least_offset_node(struct inode *inode)
974 struct uprobe u = { .inode = inode, .offset = 0};
975 struct rb_node *n = uprobes_tree.rb_node;
976 struct rb_node *close_node = NULL;
977 struct uprobe *uprobe;
981 uprobe = rb_entry(n, struct uprobe, rb_node);
982 match = match_uprobe(&u, uprobe);
984 if (uprobe->inode == inode)
1000 * For a given inode, build a list of probes that need to be inserted.
1002 static void build_probe_list(struct inode *inode, struct list_head *head)
1004 struct uprobe *uprobe;
1005 unsigned long flags;
1008 spin_lock_irqsave(&uprobes_treelock, flags);
1010 n = find_least_offset_node(inode);
1012 for (; n; n = rb_next(n)) {
1013 uprobe = rb_entry(n, struct uprobe, rb_node);
1014 if (uprobe->inode != inode)
1017 list_add(&uprobe->pending_list, head);
1018 atomic_inc(&uprobe->ref);
1021 spin_unlock_irqrestore(&uprobes_treelock, flags);
1025 * Called from mmap_region.
1026 * called with mm->mmap_sem acquired.
1028 * Return -ve no if we fail to insert probes and we cannot
1030 * Return 0 otherwise. i.e:
1032 * - successful insertion of probes
1033 * - (or) no possible probes to be inserted.
1034 * - (or) insertion of probes failed but we can bail-out.
1036 int uprobe_mmap(struct vm_area_struct *vma)
1038 struct list_head tmp_list;
1039 struct uprobe *uprobe;
1040 struct inode *inode;
1043 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1046 inode = vma->vm_file->f_mapping->host;
1050 INIT_LIST_HEAD(&tmp_list);
1051 mutex_lock(uprobes_mmap_hash(inode));
1052 build_probe_list(inode, &tmp_list);
1057 list_for_each_entry(uprobe, &tmp_list, pending_list) {
1061 vaddr = vma_address(vma, uprobe->offset);
1063 if (vaddr < vma->vm_start || vaddr >= vma->vm_end) {
1068 ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1070 * We can race against uprobe_register(), see the
1071 * comment near uprobe_hash().
1073 if (ret == -EEXIST) {
1076 if (!is_swbp_at_addr(vma->vm_mm, vaddr))
1080 * Unable to insert a breakpoint, but
1081 * breakpoint lies underneath. Increment the
1084 atomic_inc(&vma->vm_mm->uprobes_state.count);
1093 mutex_unlock(uprobes_mmap_hash(inode));
1096 atomic_sub(count, &vma->vm_mm->uprobes_state.count);
1102 * Called in context of a munmap of a vma.
1104 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1106 struct list_head tmp_list;
1107 struct uprobe *uprobe;
1108 struct inode *inode;
1110 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1113 if (!atomic_read(&vma->vm_mm->uprobes_state.count))
1116 inode = vma->vm_file->f_mapping->host;
1120 INIT_LIST_HEAD(&tmp_list);
1121 mutex_lock(uprobes_mmap_hash(inode));
1122 build_probe_list(inode, &tmp_list);
1124 list_for_each_entry(uprobe, &tmp_list, pending_list) {
1127 vaddr = vma_address(vma, uprobe->offset);
1128 if (vaddr >= start && vaddr < end) {
1130 * An unregister could have removed the probe before
1131 * unmap. So check before we decrement the count.
1133 if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1)
1134 atomic_dec(&vma->vm_mm->uprobes_state.count);
1138 mutex_unlock(uprobes_mmap_hash(inode));
1141 /* Slot allocation for XOL */
1142 static int xol_add_vma(struct xol_area *area)
1144 struct mm_struct *mm;
1147 area->page = alloc_page(GFP_HIGHUSER);
1154 down_write(&mm->mmap_sem);
1155 if (mm->uprobes_state.xol_area)
1160 /* Try to map as high as possible, this is only a hint. */
1161 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1162 if (area->vaddr & ~PAGE_MASK) {
1167 ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1168 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1172 smp_wmb(); /* pairs with get_xol_area() */
1173 mm->uprobes_state.xol_area = area;
1177 up_write(&mm->mmap_sem);
1179 __free_page(area->page);
1184 static struct xol_area *get_xol_area(struct mm_struct *mm)
1186 struct xol_area *area;
1188 area = mm->uprobes_state.xol_area;
1189 smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */
1195 * xol_alloc_area - Allocate process's xol_area.
1196 * This area will be used for storing instructions for execution out of
1199 * Returns the allocated area or NULL.
1201 static struct xol_area *xol_alloc_area(void)
1203 struct xol_area *area;
1205 area = kzalloc(sizeof(*area), GFP_KERNEL);
1206 if (unlikely(!area))
1209 area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1214 init_waitqueue_head(&area->wq);
1215 if (!xol_add_vma(area))
1219 kfree(area->bitmap);
1222 return get_xol_area(current->mm);
1226 * uprobe_clear_state - Free the area allocated for slots.
1228 void uprobe_clear_state(struct mm_struct *mm)
1230 struct xol_area *area = mm->uprobes_state.xol_area;
1235 put_page(area->page);
1236 kfree(area->bitmap);
1241 * uprobe_reset_state - Free the area allocated for slots.
1243 void uprobe_reset_state(struct mm_struct *mm)
1245 mm->uprobes_state.xol_area = NULL;
1246 atomic_set(&mm->uprobes_state.count, 0);
1250 * - search for a free slot.
1252 static unsigned long xol_take_insn_slot(struct xol_area *area)
1254 unsigned long slot_addr;
1258 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1259 if (slot_nr < UINSNS_PER_PAGE) {
1260 if (!test_and_set_bit(slot_nr, area->bitmap))
1263 slot_nr = UINSNS_PER_PAGE;
1266 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1267 } while (slot_nr >= UINSNS_PER_PAGE);
1269 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1270 atomic_inc(&area->slot_count);
1276 * xol_get_insn_slot - If was not allocated a slot, then
1278 * Returns the allocated slot address or 0.
1280 static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1282 struct xol_area *area;
1283 unsigned long offset;
1286 area = get_xol_area(current->mm);
1288 area = xol_alloc_area();
1292 current->utask->xol_vaddr = xol_take_insn_slot(area);
1295 * Initialize the slot if xol_vaddr points to valid
1298 if (unlikely(!current->utask->xol_vaddr))
1301 current->utask->vaddr = slot_addr;
1302 offset = current->utask->xol_vaddr & ~PAGE_MASK;
1303 vaddr = kmap_atomic(area->page);
1304 memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1305 kunmap_atomic(vaddr);
1307 return current->utask->xol_vaddr;
1311 * xol_free_insn_slot - If slot was earlier allocated by
1312 * @xol_get_insn_slot(), make the slot available for
1313 * subsequent requests.
1315 static void xol_free_insn_slot(struct task_struct *tsk)
1317 struct xol_area *area;
1318 unsigned long vma_end;
1319 unsigned long slot_addr;
1321 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1324 slot_addr = tsk->utask->xol_vaddr;
1326 if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1329 area = tsk->mm->uprobes_state.xol_area;
1330 vma_end = area->vaddr + PAGE_SIZE;
1331 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1332 unsigned long offset;
1335 offset = slot_addr - area->vaddr;
1336 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1337 if (slot_nr >= UINSNS_PER_PAGE)
1340 clear_bit(slot_nr, area->bitmap);
1341 atomic_dec(&area->slot_count);
1342 if (waitqueue_active(&area->wq))
1345 tsk->utask->xol_vaddr = 0;
1350 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1351 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1353 * Return the address of the breakpoint instruction.
1355 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1357 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1361 * Called with no locks held.
1362 * Called in context of a exiting or a exec-ing thread.
1364 void uprobe_free_utask(struct task_struct *t)
1366 struct uprobe_task *utask = t->utask;
1371 if (utask->active_uprobe)
1372 put_uprobe(utask->active_uprobe);
1374 xol_free_insn_slot(t);
1380 * Called in context of a new clone/fork from copy_process.
1382 void uprobe_copy_process(struct task_struct *t)
1388 * Allocate a uprobe_task object for the task.
1389 * Called when the thread hits a breakpoint for the first time.
1392 * - pointer to new uprobe_task on success
1395 static struct uprobe_task *add_utask(void)
1397 struct uprobe_task *utask;
1399 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1400 if (unlikely(!utask))
1403 utask->active_uprobe = NULL;
1404 current->utask = utask;
1408 /* Prepare to single-step probed instruction out of line. */
1410 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1412 if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1419 * If we are singlestepping, then ensure this thread is not connected to
1420 * non-fatal signals until completion of singlestep. When xol insn itself
1421 * triggers the signal, restart the original insn even if the task is
1422 * already SIGKILL'ed (since coredump should report the correct ip). This
1423 * is even more important if the task has a handler for SIGSEGV/etc, The
1424 * _same_ instruction should be repeated again after return from the signal
1425 * handler, and SSTEP can never finish in this case.
1427 bool uprobe_deny_signal(void)
1429 struct task_struct *t = current;
1430 struct uprobe_task *utask = t->utask;
1432 if (likely(!utask || !utask->active_uprobe))
1435 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1437 if (signal_pending(t)) {
1438 spin_lock_irq(&t->sighand->siglock);
1439 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1440 spin_unlock_irq(&t->sighand->siglock);
1442 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1443 utask->state = UTASK_SSTEP_TRAPPED;
1444 set_tsk_thread_flag(t, TIF_UPROBE);
1445 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1453 * Avoid singlestepping the original instruction if the original instruction
1454 * is a NOP or can be emulated.
1456 static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1458 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1461 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1465 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1467 struct mm_struct *mm = current->mm;
1468 struct uprobe *uprobe = NULL;
1469 struct vm_area_struct *vma;
1471 down_read(&mm->mmap_sem);
1472 vma = find_vma(mm, bp_vaddr);
1473 if (vma && vma->vm_start <= bp_vaddr) {
1474 if (valid_vma(vma, false)) {
1475 struct inode *inode;
1478 inode = vma->vm_file->f_mapping->host;
1479 offset = bp_vaddr - vma->vm_start;
1480 offset += (vma->vm_pgoff << PAGE_SHIFT);
1481 uprobe = find_uprobe(inode, offset);
1485 *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1489 up_read(&mm->mmap_sem);
1495 * Run handler and ask thread to singlestep.
1496 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1498 static void handle_swbp(struct pt_regs *regs)
1500 struct uprobe_task *utask;
1501 struct uprobe *uprobe;
1502 unsigned long bp_vaddr;
1503 int uninitialized_var(is_swbp);
1505 bp_vaddr = uprobe_get_swbp_addr(regs);
1506 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1510 /* No matching uprobe; signal SIGTRAP. */
1511 send_sig(SIGTRAP, current, 0);
1514 * Either we raced with uprobe_unregister() or we can't
1515 * access this memory. The latter is only possible if
1516 * another thread plays with our ->mm. In both cases
1517 * we can simply restart. If this vma was unmapped we
1518 * can pretend this insn was not executed yet and get
1519 * the (correct) SIGSEGV after restart.
1521 instruction_pointer_set(regs, bp_vaddr);
1526 utask = current->utask;
1528 utask = add_utask();
1529 /* Cannot allocate; re-execute the instruction. */
1533 utask->active_uprobe = uprobe;
1534 handler_chain(uprobe, regs);
1535 if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs))
1538 utask->state = UTASK_SSTEP;
1539 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1540 user_enable_single_step(current);
1546 utask->active_uprobe = NULL;
1547 utask->state = UTASK_RUNNING;
1550 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1553 * cannot singlestep; cannot skip instruction;
1554 * re-execute the instruction.
1556 instruction_pointer_set(regs, bp_vaddr);
1563 * Perform required fix-ups and disable singlestep.
1564 * Allow pending signals to take effect.
1566 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1568 struct uprobe *uprobe;
1570 uprobe = utask->active_uprobe;
1571 if (utask->state == UTASK_SSTEP_ACK)
1572 arch_uprobe_post_xol(&uprobe->arch, regs);
1573 else if (utask->state == UTASK_SSTEP_TRAPPED)
1574 arch_uprobe_abort_xol(&uprobe->arch, regs);
1579 utask->active_uprobe = NULL;
1580 utask->state = UTASK_RUNNING;
1581 user_disable_single_step(current);
1582 xol_free_insn_slot(current);
1584 spin_lock_irq(¤t->sighand->siglock);
1585 recalc_sigpending(); /* see uprobe_deny_signal() */
1586 spin_unlock_irq(¤t->sighand->siglock);
1590 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on
1591 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1592 * allows the thread to return from interrupt.
1594 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1595 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1598 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1599 * uprobe_notify_resume().
1601 void uprobe_notify_resume(struct pt_regs *regs)
1603 struct uprobe_task *utask;
1605 utask = current->utask;
1606 if (!utask || utask->state == UTASK_BP_HIT)
1609 handle_singlestep(utask, regs);
1613 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1614 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1616 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1618 struct uprobe_task *utask;
1620 if (!current->mm || !atomic_read(¤t->mm->uprobes_state.count))
1621 /* task is currently not uprobed */
1624 utask = current->utask;
1626 utask->state = UTASK_BP_HIT;
1628 set_thread_flag(TIF_UPROBE);
1634 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1635 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1637 int uprobe_post_sstep_notifier(struct pt_regs *regs)
1639 struct uprobe_task *utask = current->utask;
1641 if (!current->mm || !utask || !utask->active_uprobe)
1642 /* task is currently not uprobed */
1645 utask->state = UTASK_SSTEP_ACK;
1646 set_thread_flag(TIF_UPROBE);
1650 static struct notifier_block uprobe_exception_nb = {
1651 .notifier_call = arch_uprobe_exception_notify,
1652 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1655 static int __init init_uprobes(void)
1659 for (i = 0; i < UPROBES_HASH_SZ; i++) {
1660 mutex_init(&uprobes_mutex[i]);
1661 mutex_init(&uprobes_mmap_mutex[i]);
1664 return register_die_notifier(&uprobe_exception_nb);
1666 module_init(init_uprobes);
1668 static void __exit exit_uprobes(void)
1671 module_exit(exit_uprobes);