1a5f4374aSIngo Molnar /* 2a5f4374aSIngo Molnar * User-space Probes (UProbes) 3a5f4374aSIngo Molnar * 4a5f4374aSIngo Molnar * This program is free software; you can redistribute it and/or modify 5a5f4374aSIngo Molnar * it under the terms of the GNU General Public License as published by 6a5f4374aSIngo Molnar * the Free Software Foundation; either version 2 of the License, or 7a5f4374aSIngo Molnar * (at your option) any later version. 8a5f4374aSIngo Molnar * 9a5f4374aSIngo Molnar * This program is distributed in the hope that it will be useful, 10a5f4374aSIngo Molnar * but WITHOUT ANY WARRANTY; without even the implied warranty of 11a5f4374aSIngo Molnar * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12a5f4374aSIngo Molnar * GNU General Public License for more details. 13a5f4374aSIngo Molnar * 14a5f4374aSIngo Molnar * You should have received a copy of the GNU General Public License 15a5f4374aSIngo Molnar * along with this program; if not, write to the Free Software 16a5f4374aSIngo Molnar * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 17a5f4374aSIngo Molnar * 1835aa621bSIngo Molnar * Copyright (C) IBM Corporation, 2008-2012 19a5f4374aSIngo Molnar * Authors: 20a5f4374aSIngo Molnar * Srikar Dronamraju 21a5f4374aSIngo Molnar * Jim Keniston 2235aa621bSIngo Molnar * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> 23a5f4374aSIngo Molnar */ 24a5f4374aSIngo Molnar 25a5f4374aSIngo Molnar #include <linux/kernel.h> 26a5f4374aSIngo Molnar #include <linux/highmem.h> 27a5f4374aSIngo Molnar #include <linux/pagemap.h> /* read_mapping_page */ 28a5f4374aSIngo Molnar #include <linux/slab.h> 29a5f4374aSIngo Molnar #include <linux/sched.h> 30a5f4374aSIngo Molnar #include <linux/rmap.h> /* anon_vma_prepare */ 31a5f4374aSIngo Molnar #include <linux/mmu_notifier.h> /* set_pte_at_notify */ 32a5f4374aSIngo Molnar #include <linux/swap.h> /* try_to_free_swap */ 330326f5a9SSrikar Dronamraju #include <linux/ptrace.h> /* user_enable_single_step */ 340326f5a9SSrikar Dronamraju #include <linux/kdebug.h> /* notifier mechanism */ 35194f8dcbSOleg Nesterov #include "../../mm/internal.h" /* munlock_vma_page */ 3632cdba1eSOleg Nesterov #include <linux/percpu-rwsem.h> 37a5f4374aSIngo Molnar 38a5f4374aSIngo Molnar #include <linux/uprobes.h> 39a5f4374aSIngo Molnar 40d4b3b638SSrikar Dronamraju #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) 41d4b3b638SSrikar Dronamraju #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE 42d4b3b638SSrikar Dronamraju 43a5f4374aSIngo Molnar static struct rb_root uprobes_tree = RB_ROOT; 44a5f4374aSIngo Molnar 45a5f4374aSIngo Molnar static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ 46a5f4374aSIngo Molnar 47a5f4374aSIngo Molnar #define UPROBES_HASH_SZ 13 48a5f4374aSIngo Molnar 49c5784de2SPeter Zijlstra /* 50c5784de2SPeter Zijlstra * We need separate register/unregister and mmap/munmap lock hashes because 51c5784de2SPeter Zijlstra * of mmap_sem nesting. 52c5784de2SPeter Zijlstra * 53c5784de2SPeter Zijlstra * uprobe_register() needs to install probes on (potentially) all processes 54c5784de2SPeter Zijlstra * and thus needs to acquire multiple mmap_sems (consequtively, not 55c5784de2SPeter Zijlstra * concurrently), whereas uprobe_mmap() is called while holding mmap_sem 56c5784de2SPeter Zijlstra * for the particular process doing the mmap. 57c5784de2SPeter Zijlstra * 58c5784de2SPeter Zijlstra * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem 59c5784de2SPeter Zijlstra * because of lock order against i_mmap_mutex. This means there's a hole in 60c5784de2SPeter Zijlstra * the register vma iteration where a mmap() can happen. 61c5784de2SPeter Zijlstra * 62c5784de2SPeter Zijlstra * Thus uprobe_register() can race with uprobe_mmap() and we can try and 63c5784de2SPeter Zijlstra * install a probe where one is already installed. 64c5784de2SPeter Zijlstra */ 65c5784de2SPeter Zijlstra 66a5f4374aSIngo Molnar /* serialize (un)register */ 67a5f4374aSIngo Molnar static struct mutex uprobes_mutex[UPROBES_HASH_SZ]; 68a5f4374aSIngo Molnar 69a5f4374aSIngo Molnar #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) 70a5f4374aSIngo Molnar 71a5f4374aSIngo Molnar /* serialize uprobe->pending_list */ 72a5f4374aSIngo Molnar static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; 73a5f4374aSIngo Molnar #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) 74a5f4374aSIngo Molnar 7532cdba1eSOleg Nesterov static struct percpu_rw_semaphore dup_mmap_sem; 7632cdba1eSOleg Nesterov 77a5f4374aSIngo Molnar /* 78a5f4374aSIngo Molnar * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe 79a5f4374aSIngo Molnar * events active at this time. Probably a fine grained per inode count is 80a5f4374aSIngo Molnar * better? 81a5f4374aSIngo Molnar */ 82a5f4374aSIngo Molnar static atomic_t uprobe_events = ATOMIC_INIT(0); 83a5f4374aSIngo Molnar 84cb9a19feSOleg Nesterov /* Have a copy of original instruction */ 8571434f2fSOleg Nesterov #define UPROBE_COPY_INSN 0 86cb9a19feSOleg Nesterov /* Can skip singlestep */ 87bb929284SOleg Nesterov #define UPROBE_SKIP_SSTEP 1 88cb9a19feSOleg Nesterov 893ff54efdSSrikar Dronamraju struct uprobe { 903ff54efdSSrikar Dronamraju struct rb_node rb_node; /* node in the rb tree */ 913ff54efdSSrikar Dronamraju atomic_t ref; 92e591c8d7SOleg Nesterov struct rw_semaphore register_rwsem; 933ff54efdSSrikar Dronamraju struct rw_semaphore consumer_rwsem; 944710f05fSOleg Nesterov struct mutex copy_mutex; /* TODO: kill me and UPROBE_COPY_INSN */ 953ff54efdSSrikar Dronamraju struct list_head pending_list; 963ff54efdSSrikar Dronamraju struct uprobe_consumer *consumers; 973ff54efdSSrikar Dronamraju struct inode *inode; /* Also hold a ref to inode */ 983ff54efdSSrikar Dronamraju loff_t offset; 9971434f2fSOleg Nesterov unsigned long flags; 1003ff54efdSSrikar Dronamraju struct arch_uprobe arch; 1013ff54efdSSrikar Dronamraju }; 1023ff54efdSSrikar Dronamraju 103a5f4374aSIngo Molnar /* 104a5f4374aSIngo Molnar * valid_vma: Verify if the specified vma is an executable vma 105a5f4374aSIngo Molnar * Relax restrictions while unregistering: vm_flags might have 106a5f4374aSIngo Molnar * changed after breakpoint was inserted. 107a5f4374aSIngo Molnar * - is_register: indicates if we are in register context. 108a5f4374aSIngo Molnar * - Return 1 if the specified virtual address is in an 109a5f4374aSIngo Molnar * executable vma. 110a5f4374aSIngo Molnar */ 111a5f4374aSIngo Molnar static bool valid_vma(struct vm_area_struct *vma, bool is_register) 112a5f4374aSIngo Molnar { 113e40cfce6SOleg Nesterov vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_SHARED; 114a5f4374aSIngo Molnar 115e40cfce6SOleg Nesterov if (is_register) 116e40cfce6SOleg Nesterov flags |= VM_WRITE; 117a5f4374aSIngo Molnar 118e40cfce6SOleg Nesterov return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; 119a5f4374aSIngo Molnar } 120a5f4374aSIngo Molnar 12157683f72SOleg Nesterov static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) 122a5f4374aSIngo Molnar { 12357683f72SOleg Nesterov return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 124a5f4374aSIngo Molnar } 125a5f4374aSIngo Molnar 126cb113b47SOleg Nesterov static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) 127cb113b47SOleg Nesterov { 128cb113b47SOleg Nesterov return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); 129cb113b47SOleg Nesterov } 130cb113b47SOleg Nesterov 131a5f4374aSIngo Molnar /** 132a5f4374aSIngo Molnar * __replace_page - replace page in vma by new page. 133a5f4374aSIngo Molnar * based on replace_page in mm/ksm.c 134a5f4374aSIngo Molnar * 135a5f4374aSIngo Molnar * @vma: vma that holds the pte pointing to page 136c517ee74SOleg Nesterov * @addr: address the old @page is mapped at 137a5f4374aSIngo Molnar * @page: the cowed page we are replacing by kpage 138a5f4374aSIngo Molnar * @kpage: the modified page we replace page by 139a5f4374aSIngo Molnar * 140a5f4374aSIngo Molnar * Returns 0 on success, -EFAULT on failure. 141a5f4374aSIngo Molnar */ 142c517ee74SOleg Nesterov static int __replace_page(struct vm_area_struct *vma, unsigned long addr, 143c517ee74SOleg Nesterov struct page *page, struct page *kpage) 144a5f4374aSIngo Molnar { 145a5f4374aSIngo Molnar struct mm_struct *mm = vma->vm_mm; 1465323ce71SOleg Nesterov spinlock_t *ptl; 1475323ce71SOleg Nesterov pte_t *ptep; 1489f92448cSOleg Nesterov int err; 1496bdb913fSHaggai Eran /* For mmu_notifiers */ 1506bdb913fSHaggai Eran const unsigned long mmun_start = addr; 1516bdb913fSHaggai Eran const unsigned long mmun_end = addr + PAGE_SIZE; 152a5f4374aSIngo Molnar 153194f8dcbSOleg Nesterov /* For try_to_free_swap() and munlock_vma_page() below */ 1549f92448cSOleg Nesterov lock_page(page); 1559f92448cSOleg Nesterov 1566bdb913fSHaggai Eran mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); 1579f92448cSOleg Nesterov err = -EAGAIN; 1585323ce71SOleg Nesterov ptep = page_check_address(page, mm, addr, &ptl, 0); 159a5f4374aSIngo Molnar if (!ptep) 1609f92448cSOleg Nesterov goto unlock; 161a5f4374aSIngo Molnar 162a5f4374aSIngo Molnar get_page(kpage); 163a5f4374aSIngo Molnar page_add_new_anon_rmap(kpage, vma, addr); 164a5f4374aSIngo Molnar 1657396fa81SSrikar Dronamraju if (!PageAnon(page)) { 1667396fa81SSrikar Dronamraju dec_mm_counter(mm, MM_FILEPAGES); 1677396fa81SSrikar Dronamraju inc_mm_counter(mm, MM_ANONPAGES); 1687396fa81SSrikar Dronamraju } 1697396fa81SSrikar Dronamraju 170a5f4374aSIngo Molnar flush_cache_page(vma, addr, pte_pfn(*ptep)); 171a5f4374aSIngo Molnar ptep_clear_flush(vma, addr, ptep); 172a5f4374aSIngo Molnar set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); 173a5f4374aSIngo Molnar 174a5f4374aSIngo Molnar page_remove_rmap(page); 175a5f4374aSIngo Molnar if (!page_mapped(page)) 176a5f4374aSIngo Molnar try_to_free_swap(page); 177a5f4374aSIngo Molnar pte_unmap_unlock(ptep, ptl); 178a5f4374aSIngo Molnar 179194f8dcbSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 180194f8dcbSOleg Nesterov munlock_vma_page(page); 181194f8dcbSOleg Nesterov put_page(page); 182194f8dcbSOleg Nesterov 1839f92448cSOleg Nesterov err = 0; 1849f92448cSOleg Nesterov unlock: 1856bdb913fSHaggai Eran mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); 1869f92448cSOleg Nesterov unlock_page(page); 1879f92448cSOleg Nesterov return err; 188a5f4374aSIngo Molnar } 189a5f4374aSIngo Molnar 190a5f4374aSIngo Molnar /** 1915cb4ac3aSSrikar Dronamraju * is_swbp_insn - check if instruction is breakpoint instruction. 192a5f4374aSIngo Molnar * @insn: instruction to be checked. 1935cb4ac3aSSrikar Dronamraju * Default implementation of is_swbp_insn 194a5f4374aSIngo Molnar * Returns true if @insn is a breakpoint instruction. 195a5f4374aSIngo Molnar */ 1965cb4ac3aSSrikar Dronamraju bool __weak is_swbp_insn(uprobe_opcode_t *insn) 197a5f4374aSIngo Molnar { 1985cb4ac3aSSrikar Dronamraju return *insn == UPROBE_SWBP_INSN; 199a5f4374aSIngo Molnar } 200a5f4374aSIngo Molnar 201cceb55aaSOleg Nesterov static void copy_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *opcode) 202cceb55aaSOleg Nesterov { 203cceb55aaSOleg Nesterov void *kaddr = kmap_atomic(page); 204cceb55aaSOleg Nesterov memcpy(opcode, kaddr + (vaddr & ~PAGE_MASK), UPROBE_SWBP_INSN_SIZE); 205cceb55aaSOleg Nesterov kunmap_atomic(kaddr); 206cceb55aaSOleg Nesterov } 207cceb55aaSOleg Nesterov 208ed6f6a50SOleg Nesterov static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) 209ed6f6a50SOleg Nesterov { 210ed6f6a50SOleg Nesterov uprobe_opcode_t old_opcode; 211ed6f6a50SOleg Nesterov bool is_swbp; 212ed6f6a50SOleg Nesterov 213ed6f6a50SOleg Nesterov copy_opcode(page, vaddr, &old_opcode); 214ed6f6a50SOleg Nesterov is_swbp = is_swbp_insn(&old_opcode); 215ed6f6a50SOleg Nesterov 216ed6f6a50SOleg Nesterov if (is_swbp_insn(new_opcode)) { 217ed6f6a50SOleg Nesterov if (is_swbp) /* register: already installed? */ 218ed6f6a50SOleg Nesterov return 0; 219ed6f6a50SOleg Nesterov } else { 220ed6f6a50SOleg Nesterov if (!is_swbp) /* unregister: was it changed by us? */ 221076a365bSOleg Nesterov return 0; 222ed6f6a50SOleg Nesterov } 223ed6f6a50SOleg Nesterov 224ed6f6a50SOleg Nesterov return 1; 225ed6f6a50SOleg Nesterov } 226ed6f6a50SOleg Nesterov 227a5f4374aSIngo Molnar /* 228a5f4374aSIngo Molnar * NOTE: 229a5f4374aSIngo Molnar * Expect the breakpoint instruction to be the smallest size instruction for 230a5f4374aSIngo Molnar * the architecture. If an arch has variable length instruction and the 231a5f4374aSIngo Molnar * breakpoint instruction is not of the smallest length instruction 232cceb55aaSOleg Nesterov * supported by that architecture then we need to modify is_swbp_at_addr and 233a5f4374aSIngo Molnar * write_opcode accordingly. This would never be a problem for archs that 234a5f4374aSIngo Molnar * have fixed length instructions. 235a5f4374aSIngo Molnar */ 236a5f4374aSIngo Molnar 237a5f4374aSIngo Molnar /* 238a5f4374aSIngo Molnar * write_opcode - write the opcode at a given virtual address. 239a5f4374aSIngo Molnar * @mm: the probed process address space. 240a5f4374aSIngo Molnar * @vaddr: the virtual address to store the opcode. 241a5f4374aSIngo Molnar * @opcode: opcode to be written at @vaddr. 242a5f4374aSIngo Molnar * 243a5f4374aSIngo Molnar * Called with mm->mmap_sem held (for read and with a reference to 244a5f4374aSIngo Molnar * mm). 245a5f4374aSIngo Molnar * 246a5f4374aSIngo Molnar * For mm @mm, write the opcode at @vaddr. 247a5f4374aSIngo Molnar * Return 0 (success) or a negative errno. 248a5f4374aSIngo Molnar */ 249cceb55aaSOleg Nesterov static int write_opcode(struct mm_struct *mm, unsigned long vaddr, 250cceb55aaSOleg Nesterov uprobe_opcode_t opcode) 251a5f4374aSIngo Molnar { 252a5f4374aSIngo Molnar struct page *old_page, *new_page; 253a5f4374aSIngo Molnar void *vaddr_old, *vaddr_new; 254a5f4374aSIngo Molnar struct vm_area_struct *vma; 255a5f4374aSIngo Molnar int ret; 256f403072cSOleg Nesterov 2575323ce71SOleg Nesterov retry: 258a5f4374aSIngo Molnar /* Read the page with vaddr into memory */ 25975ed82eaSOleg Nesterov ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &old_page, &vma); 260a5f4374aSIngo Molnar if (ret <= 0) 261a5f4374aSIngo Molnar return ret; 262a5f4374aSIngo Molnar 263ed6f6a50SOleg Nesterov ret = verify_opcode(old_page, vaddr, &opcode); 264ed6f6a50SOleg Nesterov if (ret <= 0) 265ed6f6a50SOleg Nesterov goto put_old; 266ed6f6a50SOleg Nesterov 267a5f4374aSIngo Molnar ret = -ENOMEM; 268a5f4374aSIngo Molnar new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); 269a5f4374aSIngo Molnar if (!new_page) 2709f92448cSOleg Nesterov goto put_old; 271a5f4374aSIngo Molnar 272a5f4374aSIngo Molnar __SetPageUptodate(new_page); 273a5f4374aSIngo Molnar 274a5f4374aSIngo Molnar /* copy the page now that we've got it stable */ 275a5f4374aSIngo Molnar vaddr_old = kmap_atomic(old_page); 276a5f4374aSIngo Molnar vaddr_new = kmap_atomic(new_page); 277a5f4374aSIngo Molnar 278a5f4374aSIngo Molnar memcpy(vaddr_new, vaddr_old, PAGE_SIZE); 279d9c4a30eSOleg Nesterov memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE); 280a5f4374aSIngo Molnar 281a5f4374aSIngo Molnar kunmap_atomic(vaddr_new); 282a5f4374aSIngo Molnar kunmap_atomic(vaddr_old); 283a5f4374aSIngo Molnar 284a5f4374aSIngo Molnar ret = anon_vma_prepare(vma); 285a5f4374aSIngo Molnar if (ret) 2869f92448cSOleg Nesterov goto put_new; 287a5f4374aSIngo Molnar 288c517ee74SOleg Nesterov ret = __replace_page(vma, vaddr, old_page, new_page); 289a5f4374aSIngo Molnar 2909f92448cSOleg Nesterov put_new: 291a5f4374aSIngo Molnar page_cache_release(new_page); 2929f92448cSOleg Nesterov put_old: 293a5f4374aSIngo Molnar put_page(old_page); 294a5f4374aSIngo Molnar 2955323ce71SOleg Nesterov if (unlikely(ret == -EAGAIN)) 2965323ce71SOleg Nesterov goto retry; 297a5f4374aSIngo Molnar return ret; 298a5f4374aSIngo Molnar } 299a5f4374aSIngo Molnar 300a5f4374aSIngo Molnar /** 3015cb4ac3aSSrikar Dronamraju * set_swbp - store breakpoint at a given address. 302e3343e6aSSrikar Dronamraju * @auprobe: arch specific probepoint information. 303a5f4374aSIngo Molnar * @mm: the probed process address space. 304a5f4374aSIngo Molnar * @vaddr: the virtual address to insert the opcode. 305a5f4374aSIngo Molnar * 306a5f4374aSIngo Molnar * For mm @mm, store the breakpoint instruction at @vaddr. 307a5f4374aSIngo Molnar * Return 0 (success) or a negative errno. 308a5f4374aSIngo Molnar */ 3095cb4ac3aSSrikar Dronamraju int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) 310a5f4374aSIngo Molnar { 311cceb55aaSOleg Nesterov return write_opcode(mm, vaddr, UPROBE_SWBP_INSN); 312a5f4374aSIngo Molnar } 313a5f4374aSIngo Molnar 314a5f4374aSIngo Molnar /** 315a5f4374aSIngo Molnar * set_orig_insn - Restore the original instruction. 316a5f4374aSIngo Molnar * @mm: the probed process address space. 317e3343e6aSSrikar Dronamraju * @auprobe: arch specific probepoint information. 318a5f4374aSIngo Molnar * @vaddr: the virtual address to insert the opcode. 319a5f4374aSIngo Molnar * 320a5f4374aSIngo Molnar * For mm @mm, restore the original opcode (opcode) at @vaddr. 321a5f4374aSIngo Molnar * Return 0 (success) or a negative errno. 322a5f4374aSIngo Molnar */ 323a5f4374aSIngo Molnar int __weak 324ded86e7cSOleg Nesterov set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) 325a5f4374aSIngo Molnar { 326cceb55aaSOleg Nesterov return write_opcode(mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); 327a5f4374aSIngo Molnar } 328a5f4374aSIngo Molnar 329a5f4374aSIngo Molnar static int match_uprobe(struct uprobe *l, struct uprobe *r) 330a5f4374aSIngo Molnar { 331a5f4374aSIngo Molnar if (l->inode < r->inode) 332a5f4374aSIngo Molnar return -1; 333a5f4374aSIngo Molnar 334a5f4374aSIngo Molnar if (l->inode > r->inode) 335a5f4374aSIngo Molnar return 1; 336a5f4374aSIngo Molnar 337a5f4374aSIngo Molnar if (l->offset < r->offset) 338a5f4374aSIngo Molnar return -1; 339a5f4374aSIngo Molnar 340a5f4374aSIngo Molnar if (l->offset > r->offset) 341a5f4374aSIngo Molnar return 1; 342a5f4374aSIngo Molnar 343a5f4374aSIngo Molnar return 0; 344a5f4374aSIngo Molnar } 345a5f4374aSIngo Molnar 346a5f4374aSIngo Molnar static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) 347a5f4374aSIngo Molnar { 348a5f4374aSIngo Molnar struct uprobe u = { .inode = inode, .offset = offset }; 349a5f4374aSIngo Molnar struct rb_node *n = uprobes_tree.rb_node; 350a5f4374aSIngo Molnar struct uprobe *uprobe; 351a5f4374aSIngo Molnar int match; 352a5f4374aSIngo Molnar 353a5f4374aSIngo Molnar while (n) { 354a5f4374aSIngo Molnar uprobe = rb_entry(n, struct uprobe, rb_node); 355a5f4374aSIngo Molnar match = match_uprobe(&u, uprobe); 356a5f4374aSIngo Molnar if (!match) { 357a5f4374aSIngo Molnar atomic_inc(&uprobe->ref); 358a5f4374aSIngo Molnar return uprobe; 359a5f4374aSIngo Molnar } 360a5f4374aSIngo Molnar 361a5f4374aSIngo Molnar if (match < 0) 362a5f4374aSIngo Molnar n = n->rb_left; 363a5f4374aSIngo Molnar else 364a5f4374aSIngo Molnar n = n->rb_right; 365a5f4374aSIngo Molnar } 366a5f4374aSIngo Molnar return NULL; 367a5f4374aSIngo Molnar } 368a5f4374aSIngo Molnar 369a5f4374aSIngo Molnar /* 370a5f4374aSIngo Molnar * Find a uprobe corresponding to a given inode:offset 371a5f4374aSIngo Molnar * Acquires uprobes_treelock 372a5f4374aSIngo Molnar */ 373a5f4374aSIngo Molnar static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) 374a5f4374aSIngo Molnar { 375a5f4374aSIngo Molnar struct uprobe *uprobe; 376a5f4374aSIngo Molnar 3776f47caa0SOleg Nesterov spin_lock(&uprobes_treelock); 378a5f4374aSIngo Molnar uprobe = __find_uprobe(inode, offset); 3796f47caa0SOleg Nesterov spin_unlock(&uprobes_treelock); 380a5f4374aSIngo Molnar 381a5f4374aSIngo Molnar return uprobe; 382a5f4374aSIngo Molnar } 383a5f4374aSIngo Molnar 384a5f4374aSIngo Molnar static struct uprobe *__insert_uprobe(struct uprobe *uprobe) 385a5f4374aSIngo Molnar { 386a5f4374aSIngo Molnar struct rb_node **p = &uprobes_tree.rb_node; 387a5f4374aSIngo Molnar struct rb_node *parent = NULL; 388a5f4374aSIngo Molnar struct uprobe *u; 389a5f4374aSIngo Molnar int match; 390a5f4374aSIngo Molnar 391a5f4374aSIngo Molnar while (*p) { 392a5f4374aSIngo Molnar parent = *p; 393a5f4374aSIngo Molnar u = rb_entry(parent, struct uprobe, rb_node); 394a5f4374aSIngo Molnar match = match_uprobe(uprobe, u); 395a5f4374aSIngo Molnar if (!match) { 396a5f4374aSIngo Molnar atomic_inc(&u->ref); 397a5f4374aSIngo Molnar return u; 398a5f4374aSIngo Molnar } 399a5f4374aSIngo Molnar 400a5f4374aSIngo Molnar if (match < 0) 401a5f4374aSIngo Molnar p = &parent->rb_left; 402a5f4374aSIngo Molnar else 403a5f4374aSIngo Molnar p = &parent->rb_right; 404a5f4374aSIngo Molnar 405a5f4374aSIngo Molnar } 406a5f4374aSIngo Molnar 407a5f4374aSIngo Molnar u = NULL; 408a5f4374aSIngo Molnar rb_link_node(&uprobe->rb_node, parent, p); 409a5f4374aSIngo Molnar rb_insert_color(&uprobe->rb_node, &uprobes_tree); 410a5f4374aSIngo Molnar /* get access + creation ref */ 411a5f4374aSIngo Molnar atomic_set(&uprobe->ref, 2); 412a5f4374aSIngo Molnar 413a5f4374aSIngo Molnar return u; 414a5f4374aSIngo Molnar } 415a5f4374aSIngo Molnar 416a5f4374aSIngo Molnar /* 417a5f4374aSIngo Molnar * Acquire uprobes_treelock. 418a5f4374aSIngo Molnar * Matching uprobe already exists in rbtree; 419a5f4374aSIngo Molnar * increment (access refcount) and return the matching uprobe. 420a5f4374aSIngo Molnar * 421a5f4374aSIngo Molnar * No matching uprobe; insert the uprobe in rb_tree; 422a5f4374aSIngo Molnar * get a double refcount (access + creation) and return NULL. 423a5f4374aSIngo Molnar */ 424a5f4374aSIngo Molnar static struct uprobe *insert_uprobe(struct uprobe *uprobe) 425a5f4374aSIngo Molnar { 426a5f4374aSIngo Molnar struct uprobe *u; 427a5f4374aSIngo Molnar 4286f47caa0SOleg Nesterov spin_lock(&uprobes_treelock); 429a5f4374aSIngo Molnar u = __insert_uprobe(uprobe); 4306f47caa0SOleg Nesterov spin_unlock(&uprobes_treelock); 431a5f4374aSIngo Molnar 432a5f4374aSIngo Molnar return u; 433a5f4374aSIngo Molnar } 434a5f4374aSIngo Molnar 435a5f4374aSIngo Molnar static void put_uprobe(struct uprobe *uprobe) 436a5f4374aSIngo Molnar { 437a5f4374aSIngo Molnar if (atomic_dec_and_test(&uprobe->ref)) 438a5f4374aSIngo Molnar kfree(uprobe); 439a5f4374aSIngo Molnar } 440a5f4374aSIngo Molnar 441a5f4374aSIngo Molnar static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset) 442a5f4374aSIngo Molnar { 443a5f4374aSIngo Molnar struct uprobe *uprobe, *cur_uprobe; 444a5f4374aSIngo Molnar 445a5f4374aSIngo Molnar uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); 446a5f4374aSIngo Molnar if (!uprobe) 447a5f4374aSIngo Molnar return NULL; 448a5f4374aSIngo Molnar 449a5f4374aSIngo Molnar uprobe->inode = igrab(inode); 450a5f4374aSIngo Molnar uprobe->offset = offset; 451e591c8d7SOleg Nesterov init_rwsem(&uprobe->register_rwsem); 452a5f4374aSIngo Molnar init_rwsem(&uprobe->consumer_rwsem); 4534710f05fSOleg Nesterov mutex_init(&uprobe->copy_mutex); 454bbc33d05SOleg Nesterov /* For now assume that the instruction need not be single-stepped */ 455bbc33d05SOleg Nesterov __set_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); 456a5f4374aSIngo Molnar 457a5f4374aSIngo Molnar /* add to uprobes_tree, sorted on inode:offset */ 458a5f4374aSIngo Molnar cur_uprobe = insert_uprobe(uprobe); 459a5f4374aSIngo Molnar 460a5f4374aSIngo Molnar /* a uprobe exists for this inode:offset combination */ 461a5f4374aSIngo Molnar if (cur_uprobe) { 462a5f4374aSIngo Molnar kfree(uprobe); 463a5f4374aSIngo Molnar uprobe = cur_uprobe; 464a5f4374aSIngo Molnar iput(inode); 465a5f4374aSIngo Molnar } else { 466a5f4374aSIngo Molnar atomic_inc(&uprobe_events); 467a5f4374aSIngo Molnar } 468a5f4374aSIngo Molnar 469a5f4374aSIngo Molnar return uprobe; 470a5f4374aSIngo Molnar } 471a5f4374aSIngo Molnar 4720326f5a9SSrikar Dronamraju static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) 4730326f5a9SSrikar Dronamraju { 4740326f5a9SSrikar Dronamraju struct uprobe_consumer *uc; 4750326f5a9SSrikar Dronamraju 476e591c8d7SOleg Nesterov down_read(&uprobe->register_rwsem); 477fe20d71fSOleg Nesterov for (uc = uprobe->consumers; uc; uc = uc->next) 4780326f5a9SSrikar Dronamraju uc->handler(uc, regs); 479e591c8d7SOleg Nesterov up_read(&uprobe->register_rwsem); 4800326f5a9SSrikar Dronamraju } 4810326f5a9SSrikar Dronamraju 4829a98e03cSOleg Nesterov static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) 483a5f4374aSIngo Molnar { 484a5f4374aSIngo Molnar down_write(&uprobe->consumer_rwsem); 485e3343e6aSSrikar Dronamraju uc->next = uprobe->consumers; 486e3343e6aSSrikar Dronamraju uprobe->consumers = uc; 487a5f4374aSIngo Molnar up_write(&uprobe->consumer_rwsem); 488a5f4374aSIngo Molnar } 489a5f4374aSIngo Molnar 490a5f4374aSIngo Molnar /* 491e3343e6aSSrikar Dronamraju * For uprobe @uprobe, delete the consumer @uc. 492e3343e6aSSrikar Dronamraju * Return true if the @uc is deleted successfully 493a5f4374aSIngo Molnar * or return false. 494a5f4374aSIngo Molnar */ 495e3343e6aSSrikar Dronamraju static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) 496a5f4374aSIngo Molnar { 497a5f4374aSIngo Molnar struct uprobe_consumer **con; 498a5f4374aSIngo Molnar bool ret = false; 499a5f4374aSIngo Molnar 500a5f4374aSIngo Molnar down_write(&uprobe->consumer_rwsem); 501a5f4374aSIngo Molnar for (con = &uprobe->consumers; *con; con = &(*con)->next) { 502e3343e6aSSrikar Dronamraju if (*con == uc) { 503e3343e6aSSrikar Dronamraju *con = uc->next; 504a5f4374aSIngo Molnar ret = true; 505a5f4374aSIngo Molnar break; 506a5f4374aSIngo Molnar } 507a5f4374aSIngo Molnar } 508a5f4374aSIngo Molnar up_write(&uprobe->consumer_rwsem); 509a5f4374aSIngo Molnar 510a5f4374aSIngo Molnar return ret; 511a5f4374aSIngo Molnar } 512a5f4374aSIngo Molnar 513e3343e6aSSrikar Dronamraju static int 514d436615eSOleg Nesterov __copy_insn(struct address_space *mapping, struct file *filp, char *insn, 515593609a5SOleg Nesterov unsigned long nbytes, loff_t offset) 516a5f4374aSIngo Molnar { 517a5f4374aSIngo Molnar struct page *page; 518a5f4374aSIngo Molnar void *vaddr; 519593609a5SOleg Nesterov unsigned long off; 520593609a5SOleg Nesterov pgoff_t idx; 521a5f4374aSIngo Molnar 522a5f4374aSIngo Molnar if (!filp) 523a5f4374aSIngo Molnar return -EINVAL; 524a5f4374aSIngo Molnar 525cc359d18SOleg Nesterov if (!mapping->a_ops->readpage) 526cc359d18SOleg Nesterov return -EIO; 527cc359d18SOleg Nesterov 528593609a5SOleg Nesterov idx = offset >> PAGE_CACHE_SHIFT; 529593609a5SOleg Nesterov off = offset & ~PAGE_MASK; 530a5f4374aSIngo Molnar 531a5f4374aSIngo Molnar /* 532a5f4374aSIngo Molnar * Ensure that the page that has the original instruction is 533a5f4374aSIngo Molnar * populated and in page-cache. 534a5f4374aSIngo Molnar */ 535a5f4374aSIngo Molnar page = read_mapping_page(mapping, idx, filp); 536a5f4374aSIngo Molnar if (IS_ERR(page)) 537a5f4374aSIngo Molnar return PTR_ERR(page); 538a5f4374aSIngo Molnar 539a5f4374aSIngo Molnar vaddr = kmap_atomic(page); 540593609a5SOleg Nesterov memcpy(insn, vaddr + off, nbytes); 541a5f4374aSIngo Molnar kunmap_atomic(vaddr); 542a5f4374aSIngo Molnar page_cache_release(page); 543a5f4374aSIngo Molnar 544a5f4374aSIngo Molnar return 0; 545a5f4374aSIngo Molnar } 546a5f4374aSIngo Molnar 547d436615eSOleg Nesterov static int copy_insn(struct uprobe *uprobe, struct file *filp) 548a5f4374aSIngo Molnar { 549a5f4374aSIngo Molnar struct address_space *mapping; 550a5f4374aSIngo Molnar unsigned long nbytes; 551a5f4374aSIngo Molnar int bytes; 552a5f4374aSIngo Molnar 553d436615eSOleg Nesterov nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK); 554a5f4374aSIngo Molnar mapping = uprobe->inode->i_mapping; 555a5f4374aSIngo Molnar 556a5f4374aSIngo Molnar /* Instruction at end of binary; copy only available bytes */ 557a5f4374aSIngo Molnar if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size) 558a5f4374aSIngo Molnar bytes = uprobe->inode->i_size - uprobe->offset; 559a5f4374aSIngo Molnar else 560a5f4374aSIngo Molnar bytes = MAX_UINSN_BYTES; 561a5f4374aSIngo Molnar 562a5f4374aSIngo Molnar /* Instruction at the page-boundary; copy bytes in second page */ 563a5f4374aSIngo Molnar if (nbytes < bytes) { 564fc36f595SOleg Nesterov int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes, 565fc36f595SOleg Nesterov bytes - nbytes, uprobe->offset + nbytes); 566fc36f595SOleg Nesterov if (err) 567fc36f595SOleg Nesterov return err; 568a5f4374aSIngo Molnar bytes = nbytes; 569a5f4374aSIngo Molnar } 570d436615eSOleg Nesterov return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset); 571a5f4374aSIngo Molnar } 572a5f4374aSIngo Molnar 573cb9a19feSOleg Nesterov static int prepare_uprobe(struct uprobe *uprobe, struct file *file, 574cb9a19feSOleg Nesterov struct mm_struct *mm, unsigned long vaddr) 575cb9a19feSOleg Nesterov { 576cb9a19feSOleg Nesterov int ret = 0; 577cb9a19feSOleg Nesterov 57871434f2fSOleg Nesterov if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) 579cb9a19feSOleg Nesterov return ret; 580cb9a19feSOleg Nesterov 5814710f05fSOleg Nesterov mutex_lock(&uprobe->copy_mutex); 58271434f2fSOleg Nesterov if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) 5834710f05fSOleg Nesterov goto out; 5844710f05fSOleg Nesterov 585cb9a19feSOleg Nesterov ret = copy_insn(uprobe, file); 586cb9a19feSOleg Nesterov if (ret) 587cb9a19feSOleg Nesterov goto out; 588cb9a19feSOleg Nesterov 589cb9a19feSOleg Nesterov ret = -ENOTSUPP; 590cb9a19feSOleg Nesterov if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) 591cb9a19feSOleg Nesterov goto out; 592cb9a19feSOleg Nesterov 593cb9a19feSOleg Nesterov ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); 594cb9a19feSOleg Nesterov if (ret) 595cb9a19feSOleg Nesterov goto out; 596cb9a19feSOleg Nesterov 597cb9a19feSOleg Nesterov /* write_opcode() assumes we don't cross page boundary */ 598cb9a19feSOleg Nesterov BUG_ON((uprobe->offset & ~PAGE_MASK) + 599cb9a19feSOleg Nesterov UPROBE_SWBP_INSN_SIZE > PAGE_SIZE); 600cb9a19feSOleg Nesterov 601cb9a19feSOleg Nesterov smp_wmb(); /* pairs with rmb() in find_active_uprobe() */ 60271434f2fSOleg Nesterov set_bit(UPROBE_COPY_INSN, &uprobe->flags); 603cb9a19feSOleg Nesterov 604cb9a19feSOleg Nesterov out: 6054710f05fSOleg Nesterov mutex_unlock(&uprobe->copy_mutex); 6064710f05fSOleg Nesterov 607cb9a19feSOleg Nesterov return ret; 608cb9a19feSOleg Nesterov } 609cb9a19feSOleg Nesterov 61063633cbfSOleg Nesterov static bool filter_chain(struct uprobe *uprobe) 61163633cbfSOleg Nesterov { 6121ff6fee5SOleg Nesterov struct uprobe_consumer *uc; 6131ff6fee5SOleg Nesterov bool ret = false; 6141ff6fee5SOleg Nesterov 6151ff6fee5SOleg Nesterov down_read(&uprobe->consumer_rwsem); 6161ff6fee5SOleg Nesterov for (uc = uprobe->consumers; uc; uc = uc->next) { 6171ff6fee5SOleg Nesterov /* TODO: ret = uc->filter(...) */ 6181ff6fee5SOleg Nesterov ret = true; 6191ff6fee5SOleg Nesterov if (ret) 6201ff6fee5SOleg Nesterov break; 6211ff6fee5SOleg Nesterov } 6221ff6fee5SOleg Nesterov up_read(&uprobe->consumer_rwsem); 6231ff6fee5SOleg Nesterov 6241ff6fee5SOleg Nesterov return ret; 62563633cbfSOleg Nesterov } 62663633cbfSOleg Nesterov 627e3343e6aSSrikar Dronamraju static int 628e3343e6aSSrikar Dronamraju install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, 629816c03fbSOleg Nesterov struct vm_area_struct *vma, unsigned long vaddr) 630a5f4374aSIngo Molnar { 631f8ac4ec9SOleg Nesterov bool first_uprobe; 632a5f4374aSIngo Molnar int ret; 633a5f4374aSIngo Molnar 634a5f4374aSIngo Molnar /* 635a5f4374aSIngo Molnar * If probe is being deleted, unregister thread could be done with 636a5f4374aSIngo Molnar * the vma-rmap-walk through. Adding a probe now can be fatal since 63763633cbfSOleg Nesterov * nobody will be able to cleanup. But in this case filter_chain() 63863633cbfSOleg Nesterov * must return false, all consumers have gone away. 639a5f4374aSIngo Molnar */ 64063633cbfSOleg Nesterov if (!filter_chain(uprobe)) 64178f74116SOleg Nesterov return 0; 642a5f4374aSIngo Molnar 643cb9a19feSOleg Nesterov ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); 644a5f4374aSIngo Molnar if (ret) 645a5f4374aSIngo Molnar return ret; 646a5f4374aSIngo Molnar 647f8ac4ec9SOleg Nesterov /* 648f8ac4ec9SOleg Nesterov * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), 649f8ac4ec9SOleg Nesterov * the task can hit this breakpoint right after __replace_page(). 650f8ac4ec9SOleg Nesterov */ 651f8ac4ec9SOleg Nesterov first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); 652f8ac4ec9SOleg Nesterov if (first_uprobe) 653f8ac4ec9SOleg Nesterov set_bit(MMF_HAS_UPROBES, &mm->flags); 654f8ac4ec9SOleg Nesterov 655816c03fbSOleg Nesterov ret = set_swbp(&uprobe->arch, mm, vaddr); 6569f68f672SOleg Nesterov if (!ret) 6579f68f672SOleg Nesterov clear_bit(MMF_RECALC_UPROBES, &mm->flags); 6589f68f672SOleg Nesterov else if (first_uprobe) 659f8ac4ec9SOleg Nesterov clear_bit(MMF_HAS_UPROBES, &mm->flags); 660a5f4374aSIngo Molnar 661a5f4374aSIngo Molnar return ret; 662a5f4374aSIngo Molnar } 663a5f4374aSIngo Molnar 664076a365bSOleg Nesterov static int 665816c03fbSOleg Nesterov remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) 666a5f4374aSIngo Molnar { 6679f68f672SOleg Nesterov if (!test_bit(MMF_HAS_UPROBES, &mm->flags)) 668076a365bSOleg Nesterov return 0; 6699f68f672SOleg Nesterov 67063633cbfSOleg Nesterov if (filter_chain(uprobe)) 67163633cbfSOleg Nesterov return 0; 67263633cbfSOleg Nesterov 6739f68f672SOleg Nesterov set_bit(MMF_RECALC_UPROBES, &mm->flags); 674076a365bSOleg Nesterov return set_orig_insn(&uprobe->arch, mm, vaddr); 675a5f4374aSIngo Molnar } 676a5f4374aSIngo Molnar 6770326f5a9SSrikar Dronamraju /* 678778b032dSOleg Nesterov * There could be threads that have already hit the breakpoint. They 679778b032dSOleg Nesterov * will recheck the current insn and restart if find_uprobe() fails. 680778b032dSOleg Nesterov * See find_active_uprobe(). 6810326f5a9SSrikar Dronamraju */ 682a5f4374aSIngo Molnar static void delete_uprobe(struct uprobe *uprobe) 683a5f4374aSIngo Molnar { 6846f47caa0SOleg Nesterov spin_lock(&uprobes_treelock); 685a5f4374aSIngo Molnar rb_erase(&uprobe->rb_node, &uprobes_tree); 6866f47caa0SOleg Nesterov spin_unlock(&uprobes_treelock); 687a5f4374aSIngo Molnar iput(uprobe->inode); 688a5f4374aSIngo Molnar put_uprobe(uprobe); 689a5f4374aSIngo Molnar atomic_dec(&uprobe_events); 690a5f4374aSIngo Molnar } 691a5f4374aSIngo Molnar 69226872090SOleg Nesterov struct map_info { 69326872090SOleg Nesterov struct map_info *next; 69426872090SOleg Nesterov struct mm_struct *mm; 695816c03fbSOleg Nesterov unsigned long vaddr; 69626872090SOleg Nesterov }; 69726872090SOleg Nesterov 69826872090SOleg Nesterov static inline struct map_info *free_map_info(struct map_info *info) 699a5f4374aSIngo Molnar { 70026872090SOleg Nesterov struct map_info *next = info->next; 70126872090SOleg Nesterov kfree(info); 70226872090SOleg Nesterov return next; 70326872090SOleg Nesterov } 70426872090SOleg Nesterov 70526872090SOleg Nesterov static struct map_info * 70626872090SOleg Nesterov build_map_info(struct address_space *mapping, loff_t offset, bool is_register) 70726872090SOleg Nesterov { 70826872090SOleg Nesterov unsigned long pgoff = offset >> PAGE_SHIFT; 709a5f4374aSIngo Molnar struct vm_area_struct *vma; 71026872090SOleg Nesterov struct map_info *curr = NULL; 71126872090SOleg Nesterov struct map_info *prev = NULL; 71226872090SOleg Nesterov struct map_info *info; 71326872090SOleg Nesterov int more = 0; 714a5f4374aSIngo Molnar 71526872090SOleg Nesterov again: 71626872090SOleg Nesterov mutex_lock(&mapping->i_mmap_mutex); 7176b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { 718a5f4374aSIngo Molnar if (!valid_vma(vma, is_register)) 719a5f4374aSIngo Molnar continue; 720a5f4374aSIngo Molnar 7217a5bfb66SOleg Nesterov if (!prev && !more) { 7227a5bfb66SOleg Nesterov /* 7237a5bfb66SOleg Nesterov * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through 7247a5bfb66SOleg Nesterov * reclaim. This is optimistic, no harm done if it fails. 7257a5bfb66SOleg Nesterov */ 7267a5bfb66SOleg Nesterov prev = kmalloc(sizeof(struct map_info), 7277a5bfb66SOleg Nesterov GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); 7287a5bfb66SOleg Nesterov if (prev) 7297a5bfb66SOleg Nesterov prev->next = NULL; 7307a5bfb66SOleg Nesterov } 73126872090SOleg Nesterov if (!prev) { 73226872090SOleg Nesterov more++; 73326872090SOleg Nesterov continue; 734a5f4374aSIngo Molnar } 735a5f4374aSIngo Molnar 73626872090SOleg Nesterov if (!atomic_inc_not_zero(&vma->vm_mm->mm_users)) 73726872090SOleg Nesterov continue; 738a5f4374aSIngo Molnar 73926872090SOleg Nesterov info = prev; 74026872090SOleg Nesterov prev = prev->next; 74126872090SOleg Nesterov info->next = curr; 74226872090SOleg Nesterov curr = info; 74326872090SOleg Nesterov 74426872090SOleg Nesterov info->mm = vma->vm_mm; 74557683f72SOleg Nesterov info->vaddr = offset_to_vaddr(vma, offset); 746a5f4374aSIngo Molnar } 747a5f4374aSIngo Molnar mutex_unlock(&mapping->i_mmap_mutex); 748a5f4374aSIngo Molnar 74926872090SOleg Nesterov if (!more) 75026872090SOleg Nesterov goto out; 751a5f4374aSIngo Molnar 75226872090SOleg Nesterov prev = curr; 75326872090SOleg Nesterov while (curr) { 75426872090SOleg Nesterov mmput(curr->mm); 75526872090SOleg Nesterov curr = curr->next; 75626872090SOleg Nesterov } 75726872090SOleg Nesterov 75826872090SOleg Nesterov do { 75926872090SOleg Nesterov info = kmalloc(sizeof(struct map_info), GFP_KERNEL); 76026872090SOleg Nesterov if (!info) { 76126872090SOleg Nesterov curr = ERR_PTR(-ENOMEM); 76226872090SOleg Nesterov goto out; 76326872090SOleg Nesterov } 76426872090SOleg Nesterov info->next = prev; 76526872090SOleg Nesterov prev = info; 76626872090SOleg Nesterov } while (--more); 76726872090SOleg Nesterov 76826872090SOleg Nesterov goto again; 76926872090SOleg Nesterov out: 77026872090SOleg Nesterov while (prev) 77126872090SOleg Nesterov prev = free_map_info(prev); 77226872090SOleg Nesterov return curr; 773a5f4374aSIngo Molnar } 774a5f4374aSIngo Molnar 775a5f4374aSIngo Molnar static int register_for_each_vma(struct uprobe *uprobe, bool is_register) 776a5f4374aSIngo Molnar { 77726872090SOleg Nesterov struct map_info *info; 77826872090SOleg Nesterov int err = 0; 77926872090SOleg Nesterov 78032cdba1eSOleg Nesterov percpu_down_write(&dup_mmap_sem); 78126872090SOleg Nesterov info = build_map_info(uprobe->inode->i_mapping, 78226872090SOleg Nesterov uprobe->offset, is_register); 78332cdba1eSOleg Nesterov if (IS_ERR(info)) { 78432cdba1eSOleg Nesterov err = PTR_ERR(info); 78532cdba1eSOleg Nesterov goto out; 78632cdba1eSOleg Nesterov } 78726872090SOleg Nesterov 78826872090SOleg Nesterov while (info) { 78926872090SOleg Nesterov struct mm_struct *mm = info->mm; 790a5f4374aSIngo Molnar struct vm_area_struct *vma; 791a5f4374aSIngo Molnar 792076a365bSOleg Nesterov if (err && is_register) 79326872090SOleg Nesterov goto free; 794a5f4374aSIngo Molnar 79577fc4af1SOleg Nesterov down_write(&mm->mmap_sem); 796f4d6dfe5SOleg Nesterov vma = find_vma(mm, info->vaddr); 797f4d6dfe5SOleg Nesterov if (!vma || !valid_vma(vma, is_register) || 798f4d6dfe5SOleg Nesterov vma->vm_file->f_mapping->host != uprobe->inode) 79926872090SOleg Nesterov goto unlock; 80026872090SOleg Nesterov 801f4d6dfe5SOleg Nesterov if (vma->vm_start > info->vaddr || 802f4d6dfe5SOleg Nesterov vaddr_to_offset(vma, info->vaddr) != uprobe->offset) 80326872090SOleg Nesterov goto unlock; 804a5f4374aSIngo Molnar 80578f74116SOleg Nesterov if (is_register) 80626872090SOleg Nesterov err = install_breakpoint(uprobe, mm, vma, info->vaddr); 80778f74116SOleg Nesterov else 808076a365bSOleg Nesterov err |= remove_breakpoint(uprobe, mm, info->vaddr); 80978f74116SOleg Nesterov 81026872090SOleg Nesterov unlock: 81126872090SOleg Nesterov up_write(&mm->mmap_sem); 81226872090SOleg Nesterov free: 81326872090SOleg Nesterov mmput(mm); 81426872090SOleg Nesterov info = free_map_info(info); 815a5f4374aSIngo Molnar } 81632cdba1eSOleg Nesterov out: 81732cdba1eSOleg Nesterov percpu_up_write(&dup_mmap_sem); 81826872090SOleg Nesterov return err; 819a5f4374aSIngo Molnar } 820a5f4374aSIngo Molnar 8219a98e03cSOleg Nesterov static int __uprobe_register(struct uprobe *uprobe, struct uprobe_consumer *uc) 822a5f4374aSIngo Molnar { 8239a98e03cSOleg Nesterov consumer_add(uprobe, uc); 824bb929284SOleg Nesterov return register_for_each_vma(uprobe, true); 825a5f4374aSIngo Molnar } 826a5f4374aSIngo Molnar 82704aab9b2SOleg Nesterov static void __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) 828a5f4374aSIngo Molnar { 82904aab9b2SOleg Nesterov int err; 830a5f4374aSIngo Molnar 83104aab9b2SOleg Nesterov if (!consumer_del(uprobe, uc)) /* WARN? */ 83204aab9b2SOleg Nesterov return; 83304aab9b2SOleg Nesterov 83404aab9b2SOleg Nesterov err = register_for_each_vma(uprobe, false); 835a5f4374aSIngo Molnar /* TODO : cant unregister? schedule a worker thread */ 836bb929284SOleg Nesterov if (!uprobe->consumers && !err) 83704aab9b2SOleg Nesterov delete_uprobe(uprobe); 83804aab9b2SOleg Nesterov } 839a5f4374aSIngo Molnar 840a5f4374aSIngo Molnar /* 841a5f4374aSIngo Molnar * uprobe_register - register a probe 842a5f4374aSIngo Molnar * @inode: the file in which the probe has to be placed. 843a5f4374aSIngo Molnar * @offset: offset from the start of the file. 844e3343e6aSSrikar Dronamraju * @uc: information on howto handle the probe.. 845a5f4374aSIngo Molnar * 846a5f4374aSIngo Molnar * Apart from the access refcount, uprobe_register() takes a creation 847a5f4374aSIngo Molnar * refcount (thro alloc_uprobe) if and only if this @uprobe is getting 848a5f4374aSIngo Molnar * inserted into the rbtree (i.e first consumer for a @inode:@offset 849a5f4374aSIngo Molnar * tuple). Creation refcount stops uprobe_unregister from freeing the 850a5f4374aSIngo Molnar * @uprobe even before the register operation is complete. Creation 851e3343e6aSSrikar Dronamraju * refcount is released when the last @uc for the @uprobe 852a5f4374aSIngo Molnar * unregisters. 853a5f4374aSIngo Molnar * 854a5f4374aSIngo Molnar * Return errno if it cannot successully install probes 855a5f4374aSIngo Molnar * else return 0 (success) 856a5f4374aSIngo Molnar */ 857e3343e6aSSrikar Dronamraju int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) 858a5f4374aSIngo Molnar { 859a5f4374aSIngo Molnar struct uprobe *uprobe; 860a5f4374aSIngo Molnar int ret; 861a5f4374aSIngo Molnar 862f0744af7SOleg Nesterov /* Racy, just to catch the obvious mistakes */ 863a5f4374aSIngo Molnar if (offset > i_size_read(inode)) 864a5f4374aSIngo Molnar return -EINVAL; 865a5f4374aSIngo Molnar 8669a98e03cSOleg Nesterov ret = -ENOMEM; 867a5f4374aSIngo Molnar mutex_lock(uprobes_hash(inode)); 868a5f4374aSIngo Molnar uprobe = alloc_uprobe(inode, offset); 8699a98e03cSOleg Nesterov if (uprobe) { 870e591c8d7SOleg Nesterov down_write(&uprobe->register_rwsem); 8719a98e03cSOleg Nesterov ret = __uprobe_register(uprobe, uc); 8729a98e03cSOleg Nesterov if (ret) 87304aab9b2SOleg Nesterov __uprobe_unregister(uprobe, uc); 874e591c8d7SOleg Nesterov up_write(&uprobe->register_rwsem); 875a5f4374aSIngo Molnar } 876a5f4374aSIngo Molnar mutex_unlock(uprobes_hash(inode)); 8776d1d8dfaSSebastian Andrzej Siewior if (uprobe) 878a5f4374aSIngo Molnar put_uprobe(uprobe); 879a5f4374aSIngo Molnar 880a5f4374aSIngo Molnar return ret; 881a5f4374aSIngo Molnar } 882a5f4374aSIngo Molnar 883a5f4374aSIngo Molnar /* 884a5f4374aSIngo Molnar * uprobe_unregister - unregister a already registered probe. 885a5f4374aSIngo Molnar * @inode: the file in which the probe has to be removed. 886a5f4374aSIngo Molnar * @offset: offset from the start of the file. 887e3343e6aSSrikar Dronamraju * @uc: identify which probe if multiple probes are colocated. 888a5f4374aSIngo Molnar */ 889e3343e6aSSrikar Dronamraju void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) 890a5f4374aSIngo Molnar { 891a5f4374aSIngo Molnar struct uprobe *uprobe; 892a5f4374aSIngo Molnar 893a5f4374aSIngo Molnar uprobe = find_uprobe(inode, offset); 894a5f4374aSIngo Molnar if (!uprobe) 895a5f4374aSIngo Molnar return; 896a5f4374aSIngo Molnar 897a5f4374aSIngo Molnar mutex_lock(uprobes_hash(inode)); 898e591c8d7SOleg Nesterov down_write(&uprobe->register_rwsem); 89904aab9b2SOleg Nesterov __uprobe_unregister(uprobe, uc); 900e591c8d7SOleg Nesterov up_write(&uprobe->register_rwsem); 901a5f4374aSIngo Molnar mutex_unlock(uprobes_hash(inode)); 902a5f4374aSIngo Molnar put_uprobe(uprobe); 903a5f4374aSIngo Molnar } 904a5f4374aSIngo Molnar 905891c3970SOleg Nesterov static struct rb_node * 906891c3970SOleg Nesterov find_node_in_range(struct inode *inode, loff_t min, loff_t max) 907a5f4374aSIngo Molnar { 908a5f4374aSIngo Molnar struct rb_node *n = uprobes_tree.rb_node; 909a5f4374aSIngo Molnar 910a5f4374aSIngo Molnar while (n) { 911891c3970SOleg Nesterov struct uprobe *u = rb_entry(n, struct uprobe, rb_node); 912a5f4374aSIngo Molnar 913891c3970SOleg Nesterov if (inode < u->inode) { 914a5f4374aSIngo Molnar n = n->rb_left; 915891c3970SOleg Nesterov } else if (inode > u->inode) { 916a5f4374aSIngo Molnar n = n->rb_right; 917891c3970SOleg Nesterov } else { 918891c3970SOleg Nesterov if (max < u->offset) 919891c3970SOleg Nesterov n = n->rb_left; 920891c3970SOleg Nesterov else if (min > u->offset) 921891c3970SOleg Nesterov n = n->rb_right; 922891c3970SOleg Nesterov else 923891c3970SOleg Nesterov break; 924891c3970SOleg Nesterov } 925a5f4374aSIngo Molnar } 926a5f4374aSIngo Molnar 927891c3970SOleg Nesterov return n; 928a5f4374aSIngo Molnar } 929a5f4374aSIngo Molnar 930a5f4374aSIngo Molnar /* 931891c3970SOleg Nesterov * For a given range in vma, build a list of probes that need to be inserted. 932a5f4374aSIngo Molnar */ 933891c3970SOleg Nesterov static void build_probe_list(struct inode *inode, 934891c3970SOleg Nesterov struct vm_area_struct *vma, 935891c3970SOleg Nesterov unsigned long start, unsigned long end, 936891c3970SOleg Nesterov struct list_head *head) 937a5f4374aSIngo Molnar { 938891c3970SOleg Nesterov loff_t min, max; 939891c3970SOleg Nesterov struct rb_node *n, *t; 940891c3970SOleg Nesterov struct uprobe *u; 941891c3970SOleg Nesterov 942891c3970SOleg Nesterov INIT_LIST_HEAD(head); 943cb113b47SOleg Nesterov min = vaddr_to_offset(vma, start); 944891c3970SOleg Nesterov max = min + (end - start) - 1; 945a5f4374aSIngo Molnar 9466f47caa0SOleg Nesterov spin_lock(&uprobes_treelock); 947891c3970SOleg Nesterov n = find_node_in_range(inode, min, max); 948891c3970SOleg Nesterov if (n) { 949891c3970SOleg Nesterov for (t = n; t; t = rb_prev(t)) { 950891c3970SOleg Nesterov u = rb_entry(t, struct uprobe, rb_node); 951891c3970SOleg Nesterov if (u->inode != inode || u->offset < min) 952a5f4374aSIngo Molnar break; 953891c3970SOleg Nesterov list_add(&u->pending_list, head); 954891c3970SOleg Nesterov atomic_inc(&u->ref); 955a5f4374aSIngo Molnar } 956891c3970SOleg Nesterov for (t = n; (t = rb_next(t)); ) { 957891c3970SOleg Nesterov u = rb_entry(t, struct uprobe, rb_node); 958891c3970SOleg Nesterov if (u->inode != inode || u->offset > max) 959891c3970SOleg Nesterov break; 960891c3970SOleg Nesterov list_add(&u->pending_list, head); 961891c3970SOleg Nesterov atomic_inc(&u->ref); 962891c3970SOleg Nesterov } 963891c3970SOleg Nesterov } 9646f47caa0SOleg Nesterov spin_unlock(&uprobes_treelock); 965a5f4374aSIngo Molnar } 966a5f4374aSIngo Molnar 967a5f4374aSIngo Molnar /* 9685e5be71aSOleg Nesterov * Called from mmap_region/vma_adjust with mm->mmap_sem acquired. 969a5f4374aSIngo Molnar * 9705e5be71aSOleg Nesterov * Currently we ignore all errors and always return 0, the callers 9715e5be71aSOleg Nesterov * can't handle the failure anyway. 972a5f4374aSIngo Molnar */ 973a5f4374aSIngo Molnar int uprobe_mmap(struct vm_area_struct *vma) 974a5f4374aSIngo Molnar { 975a5f4374aSIngo Molnar struct list_head tmp_list; 976665605a2SOleg Nesterov struct uprobe *uprobe, *u; 977a5f4374aSIngo Molnar struct inode *inode; 978a5f4374aSIngo Molnar 979a5f4374aSIngo Molnar if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) 980a5f4374aSIngo Molnar return 0; 981a5f4374aSIngo Molnar 982a5f4374aSIngo Molnar inode = vma->vm_file->f_mapping->host; 983a5f4374aSIngo Molnar if (!inode) 984a5f4374aSIngo Molnar return 0; 985a5f4374aSIngo Molnar 986a5f4374aSIngo Molnar mutex_lock(uprobes_mmap_hash(inode)); 987891c3970SOleg Nesterov build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); 988a5f4374aSIngo Molnar 989665605a2SOleg Nesterov list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { 9905e5be71aSOleg Nesterov if (!fatal_signal_pending(current)) { 99157683f72SOleg Nesterov unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); 9925e5be71aSOleg Nesterov install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); 993a5f4374aSIngo Molnar } 994a5f4374aSIngo Molnar put_uprobe(uprobe); 995a5f4374aSIngo Molnar } 996a5f4374aSIngo Molnar mutex_unlock(uprobes_mmap_hash(inode)); 997a5f4374aSIngo Molnar 9985e5be71aSOleg Nesterov return 0; 999a5f4374aSIngo Molnar } 1000a5f4374aSIngo Molnar 10019f68f672SOleg Nesterov static bool 10029f68f672SOleg Nesterov vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) 10039f68f672SOleg Nesterov { 10049f68f672SOleg Nesterov loff_t min, max; 10059f68f672SOleg Nesterov struct inode *inode; 10069f68f672SOleg Nesterov struct rb_node *n; 10079f68f672SOleg Nesterov 10089f68f672SOleg Nesterov inode = vma->vm_file->f_mapping->host; 10099f68f672SOleg Nesterov 10109f68f672SOleg Nesterov min = vaddr_to_offset(vma, start); 10119f68f672SOleg Nesterov max = min + (end - start) - 1; 10129f68f672SOleg Nesterov 10139f68f672SOleg Nesterov spin_lock(&uprobes_treelock); 10149f68f672SOleg Nesterov n = find_node_in_range(inode, min, max); 10159f68f672SOleg Nesterov spin_unlock(&uprobes_treelock); 10169f68f672SOleg Nesterov 10179f68f672SOleg Nesterov return !!n; 10189f68f672SOleg Nesterov } 10199f68f672SOleg Nesterov 1020682968e0SSrikar Dronamraju /* 1021682968e0SSrikar Dronamraju * Called in context of a munmap of a vma. 1022682968e0SSrikar Dronamraju */ 1023cbc91f71SSrikar Dronamraju void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) 1024682968e0SSrikar Dronamraju { 1025682968e0SSrikar Dronamraju if (!atomic_read(&uprobe_events) || !valid_vma(vma, false)) 1026682968e0SSrikar Dronamraju return; 1027682968e0SSrikar Dronamraju 10282fd611a9SOleg Nesterov if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ 10292fd611a9SOleg Nesterov return; 10302fd611a9SOleg Nesterov 10319f68f672SOleg Nesterov if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || 10329f68f672SOleg Nesterov test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) 1033f8ac4ec9SOleg Nesterov return; 1034f8ac4ec9SOleg Nesterov 10359f68f672SOleg Nesterov if (vma_has_uprobes(vma, start, end)) 10369f68f672SOleg Nesterov set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); 1037682968e0SSrikar Dronamraju } 1038682968e0SSrikar Dronamraju 1039d4b3b638SSrikar Dronamraju /* Slot allocation for XOL */ 1040d4b3b638SSrikar Dronamraju static int xol_add_vma(struct xol_area *area) 1041d4b3b638SSrikar Dronamraju { 1042d4b3b638SSrikar Dronamraju struct mm_struct *mm; 1043d4b3b638SSrikar Dronamraju int ret; 1044d4b3b638SSrikar Dronamraju 1045d4b3b638SSrikar Dronamraju area->page = alloc_page(GFP_HIGHUSER); 1046d4b3b638SSrikar Dronamraju if (!area->page) 1047d4b3b638SSrikar Dronamraju return -ENOMEM; 1048d4b3b638SSrikar Dronamraju 1049d4b3b638SSrikar Dronamraju ret = -EALREADY; 1050d4b3b638SSrikar Dronamraju mm = current->mm; 1051d4b3b638SSrikar Dronamraju 1052d4b3b638SSrikar Dronamraju down_write(&mm->mmap_sem); 1053d4b3b638SSrikar Dronamraju if (mm->uprobes_state.xol_area) 1054d4b3b638SSrikar Dronamraju goto fail; 1055d4b3b638SSrikar Dronamraju 1056d4b3b638SSrikar Dronamraju ret = -ENOMEM; 1057d4b3b638SSrikar Dronamraju 1058d4b3b638SSrikar Dronamraju /* Try to map as high as possible, this is only a hint. */ 1059d4b3b638SSrikar Dronamraju area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0); 1060d4b3b638SSrikar Dronamraju if (area->vaddr & ~PAGE_MASK) { 1061d4b3b638SSrikar Dronamraju ret = area->vaddr; 1062d4b3b638SSrikar Dronamraju goto fail; 1063d4b3b638SSrikar Dronamraju } 1064d4b3b638SSrikar Dronamraju 1065d4b3b638SSrikar Dronamraju ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE, 1066d4b3b638SSrikar Dronamraju VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page); 1067d4b3b638SSrikar Dronamraju if (ret) 1068d4b3b638SSrikar Dronamraju goto fail; 1069d4b3b638SSrikar Dronamraju 1070d4b3b638SSrikar Dronamraju smp_wmb(); /* pairs with get_xol_area() */ 1071d4b3b638SSrikar Dronamraju mm->uprobes_state.xol_area = area; 1072d4b3b638SSrikar Dronamraju ret = 0; 1073d4b3b638SSrikar Dronamraju 1074d4b3b638SSrikar Dronamraju fail: 1075d4b3b638SSrikar Dronamraju up_write(&mm->mmap_sem); 1076d4b3b638SSrikar Dronamraju if (ret) 1077d4b3b638SSrikar Dronamraju __free_page(area->page); 1078d4b3b638SSrikar Dronamraju 1079d4b3b638SSrikar Dronamraju return ret; 1080d4b3b638SSrikar Dronamraju } 1081d4b3b638SSrikar Dronamraju 1082d4b3b638SSrikar Dronamraju static struct xol_area *get_xol_area(struct mm_struct *mm) 1083d4b3b638SSrikar Dronamraju { 1084d4b3b638SSrikar Dronamraju struct xol_area *area; 1085d4b3b638SSrikar Dronamraju 1086d4b3b638SSrikar Dronamraju area = mm->uprobes_state.xol_area; 1087d4b3b638SSrikar Dronamraju smp_read_barrier_depends(); /* pairs with wmb in xol_add_vma() */ 1088d4b3b638SSrikar Dronamraju 1089d4b3b638SSrikar Dronamraju return area; 1090d4b3b638SSrikar Dronamraju } 1091d4b3b638SSrikar Dronamraju 1092d4b3b638SSrikar Dronamraju /* 1093d4b3b638SSrikar Dronamraju * xol_alloc_area - Allocate process's xol_area. 1094d4b3b638SSrikar Dronamraju * This area will be used for storing instructions for execution out of 1095d4b3b638SSrikar Dronamraju * line. 1096d4b3b638SSrikar Dronamraju * 1097d4b3b638SSrikar Dronamraju * Returns the allocated area or NULL. 1098d4b3b638SSrikar Dronamraju */ 1099d4b3b638SSrikar Dronamraju static struct xol_area *xol_alloc_area(void) 1100d4b3b638SSrikar Dronamraju { 1101d4b3b638SSrikar Dronamraju struct xol_area *area; 1102d4b3b638SSrikar Dronamraju 1103d4b3b638SSrikar Dronamraju area = kzalloc(sizeof(*area), GFP_KERNEL); 1104d4b3b638SSrikar Dronamraju if (unlikely(!area)) 1105d4b3b638SSrikar Dronamraju return NULL; 1106d4b3b638SSrikar Dronamraju 1107d4b3b638SSrikar Dronamraju area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL); 1108d4b3b638SSrikar Dronamraju 1109d4b3b638SSrikar Dronamraju if (!area->bitmap) 1110d4b3b638SSrikar Dronamraju goto fail; 1111d4b3b638SSrikar Dronamraju 1112d4b3b638SSrikar Dronamraju init_waitqueue_head(&area->wq); 1113d4b3b638SSrikar Dronamraju if (!xol_add_vma(area)) 1114d4b3b638SSrikar Dronamraju return area; 1115d4b3b638SSrikar Dronamraju 1116d4b3b638SSrikar Dronamraju fail: 1117d4b3b638SSrikar Dronamraju kfree(area->bitmap); 1118d4b3b638SSrikar Dronamraju kfree(area); 1119d4b3b638SSrikar Dronamraju 1120d4b3b638SSrikar Dronamraju return get_xol_area(current->mm); 1121d4b3b638SSrikar Dronamraju } 1122d4b3b638SSrikar Dronamraju 1123d4b3b638SSrikar Dronamraju /* 1124d4b3b638SSrikar Dronamraju * uprobe_clear_state - Free the area allocated for slots. 1125d4b3b638SSrikar Dronamraju */ 1126d4b3b638SSrikar Dronamraju void uprobe_clear_state(struct mm_struct *mm) 1127d4b3b638SSrikar Dronamraju { 1128d4b3b638SSrikar Dronamraju struct xol_area *area = mm->uprobes_state.xol_area; 1129d4b3b638SSrikar Dronamraju 1130d4b3b638SSrikar Dronamraju if (!area) 1131d4b3b638SSrikar Dronamraju return; 1132d4b3b638SSrikar Dronamraju 1133d4b3b638SSrikar Dronamraju put_page(area->page); 1134d4b3b638SSrikar Dronamraju kfree(area->bitmap); 1135d4b3b638SSrikar Dronamraju kfree(area); 1136d4b3b638SSrikar Dronamraju } 1137d4b3b638SSrikar Dronamraju 113832cdba1eSOleg Nesterov void uprobe_start_dup_mmap(void) 113932cdba1eSOleg Nesterov { 114032cdba1eSOleg Nesterov percpu_down_read(&dup_mmap_sem); 114132cdba1eSOleg Nesterov } 114232cdba1eSOleg Nesterov 114332cdba1eSOleg Nesterov void uprobe_end_dup_mmap(void) 114432cdba1eSOleg Nesterov { 114532cdba1eSOleg Nesterov percpu_up_read(&dup_mmap_sem); 114632cdba1eSOleg Nesterov } 114732cdba1eSOleg Nesterov 1148f8ac4ec9SOleg Nesterov void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) 1149f8ac4ec9SOleg Nesterov { 115061559a81SOleg Nesterov newmm->uprobes_state.xol_area = NULL; 115161559a81SOleg Nesterov 11529f68f672SOleg Nesterov if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { 1153f8ac4ec9SOleg Nesterov set_bit(MMF_HAS_UPROBES, &newmm->flags); 11549f68f672SOleg Nesterov /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ 11559f68f672SOleg Nesterov set_bit(MMF_RECALC_UPROBES, &newmm->flags); 11569f68f672SOleg Nesterov } 1157f8ac4ec9SOleg Nesterov } 1158f8ac4ec9SOleg Nesterov 1159d4b3b638SSrikar Dronamraju /* 1160d4b3b638SSrikar Dronamraju * - search for a free slot. 1161d4b3b638SSrikar Dronamraju */ 1162d4b3b638SSrikar Dronamraju static unsigned long xol_take_insn_slot(struct xol_area *area) 1163d4b3b638SSrikar Dronamraju { 1164d4b3b638SSrikar Dronamraju unsigned long slot_addr; 1165d4b3b638SSrikar Dronamraju int slot_nr; 1166d4b3b638SSrikar Dronamraju 1167d4b3b638SSrikar Dronamraju do { 1168d4b3b638SSrikar Dronamraju slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); 1169d4b3b638SSrikar Dronamraju if (slot_nr < UINSNS_PER_PAGE) { 1170d4b3b638SSrikar Dronamraju if (!test_and_set_bit(slot_nr, area->bitmap)) 1171d4b3b638SSrikar Dronamraju break; 1172d4b3b638SSrikar Dronamraju 1173d4b3b638SSrikar Dronamraju slot_nr = UINSNS_PER_PAGE; 1174d4b3b638SSrikar Dronamraju continue; 1175d4b3b638SSrikar Dronamraju } 1176d4b3b638SSrikar Dronamraju wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); 1177d4b3b638SSrikar Dronamraju } while (slot_nr >= UINSNS_PER_PAGE); 1178d4b3b638SSrikar Dronamraju 1179d4b3b638SSrikar Dronamraju slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); 1180d4b3b638SSrikar Dronamraju atomic_inc(&area->slot_count); 1181d4b3b638SSrikar Dronamraju 1182d4b3b638SSrikar Dronamraju return slot_addr; 1183d4b3b638SSrikar Dronamraju } 1184d4b3b638SSrikar Dronamraju 1185d4b3b638SSrikar Dronamraju /* 1186d4b3b638SSrikar Dronamraju * xol_get_insn_slot - If was not allocated a slot, then 1187d4b3b638SSrikar Dronamraju * allocate a slot. 1188d4b3b638SSrikar Dronamraju * Returns the allocated slot address or 0. 1189d4b3b638SSrikar Dronamraju */ 1190d4b3b638SSrikar Dronamraju static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr) 1191d4b3b638SSrikar Dronamraju { 1192d4b3b638SSrikar Dronamraju struct xol_area *area; 1193d4b3b638SSrikar Dronamraju unsigned long offset; 1194d4b3b638SSrikar Dronamraju void *vaddr; 1195d4b3b638SSrikar Dronamraju 1196d4b3b638SSrikar Dronamraju area = get_xol_area(current->mm); 1197d4b3b638SSrikar Dronamraju if (!area) { 1198d4b3b638SSrikar Dronamraju area = xol_alloc_area(); 1199d4b3b638SSrikar Dronamraju if (!area) 1200d4b3b638SSrikar Dronamraju return 0; 1201d4b3b638SSrikar Dronamraju } 1202d4b3b638SSrikar Dronamraju current->utask->xol_vaddr = xol_take_insn_slot(area); 1203d4b3b638SSrikar Dronamraju 1204d4b3b638SSrikar Dronamraju /* 1205d4b3b638SSrikar Dronamraju * Initialize the slot if xol_vaddr points to valid 1206d4b3b638SSrikar Dronamraju * instruction slot. 1207d4b3b638SSrikar Dronamraju */ 1208d4b3b638SSrikar Dronamraju if (unlikely(!current->utask->xol_vaddr)) 1209d4b3b638SSrikar Dronamraju return 0; 1210d4b3b638SSrikar Dronamraju 1211d4b3b638SSrikar Dronamraju current->utask->vaddr = slot_addr; 1212d4b3b638SSrikar Dronamraju offset = current->utask->xol_vaddr & ~PAGE_MASK; 1213d4b3b638SSrikar Dronamraju vaddr = kmap_atomic(area->page); 1214d4b3b638SSrikar Dronamraju memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES); 1215d4b3b638SSrikar Dronamraju kunmap_atomic(vaddr); 121665b6ecc0SRabin Vincent /* 121765b6ecc0SRabin Vincent * We probably need flush_icache_user_range() but it needs vma. 121865b6ecc0SRabin Vincent * This should work on supported architectures too. 121965b6ecc0SRabin Vincent */ 122065b6ecc0SRabin Vincent flush_dcache_page(area->page); 1221d4b3b638SSrikar Dronamraju 1222d4b3b638SSrikar Dronamraju return current->utask->xol_vaddr; 1223d4b3b638SSrikar Dronamraju } 1224d4b3b638SSrikar Dronamraju 1225d4b3b638SSrikar Dronamraju /* 1226d4b3b638SSrikar Dronamraju * xol_free_insn_slot - If slot was earlier allocated by 1227d4b3b638SSrikar Dronamraju * @xol_get_insn_slot(), make the slot available for 1228d4b3b638SSrikar Dronamraju * subsequent requests. 1229d4b3b638SSrikar Dronamraju */ 1230d4b3b638SSrikar Dronamraju static void xol_free_insn_slot(struct task_struct *tsk) 1231d4b3b638SSrikar Dronamraju { 1232d4b3b638SSrikar Dronamraju struct xol_area *area; 1233d4b3b638SSrikar Dronamraju unsigned long vma_end; 1234d4b3b638SSrikar Dronamraju unsigned long slot_addr; 1235d4b3b638SSrikar Dronamraju 1236d4b3b638SSrikar Dronamraju if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) 1237d4b3b638SSrikar Dronamraju return; 1238d4b3b638SSrikar Dronamraju 1239d4b3b638SSrikar Dronamraju slot_addr = tsk->utask->xol_vaddr; 1240d4b3b638SSrikar Dronamraju 1241d4b3b638SSrikar Dronamraju if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr))) 1242d4b3b638SSrikar Dronamraju return; 1243d4b3b638SSrikar Dronamraju 1244d4b3b638SSrikar Dronamraju area = tsk->mm->uprobes_state.xol_area; 1245d4b3b638SSrikar Dronamraju vma_end = area->vaddr + PAGE_SIZE; 1246d4b3b638SSrikar Dronamraju if (area->vaddr <= slot_addr && slot_addr < vma_end) { 1247d4b3b638SSrikar Dronamraju unsigned long offset; 1248d4b3b638SSrikar Dronamraju int slot_nr; 1249d4b3b638SSrikar Dronamraju 1250d4b3b638SSrikar Dronamraju offset = slot_addr - area->vaddr; 1251d4b3b638SSrikar Dronamraju slot_nr = offset / UPROBE_XOL_SLOT_BYTES; 1252d4b3b638SSrikar Dronamraju if (slot_nr >= UINSNS_PER_PAGE) 1253d4b3b638SSrikar Dronamraju return; 1254d4b3b638SSrikar Dronamraju 1255d4b3b638SSrikar Dronamraju clear_bit(slot_nr, area->bitmap); 1256d4b3b638SSrikar Dronamraju atomic_dec(&area->slot_count); 1257d4b3b638SSrikar Dronamraju if (waitqueue_active(&area->wq)) 1258d4b3b638SSrikar Dronamraju wake_up(&area->wq); 1259d4b3b638SSrikar Dronamraju 1260d4b3b638SSrikar Dronamraju tsk->utask->xol_vaddr = 0; 1261d4b3b638SSrikar Dronamraju } 1262d4b3b638SSrikar Dronamraju } 1263d4b3b638SSrikar Dronamraju 12640326f5a9SSrikar Dronamraju /** 12650326f5a9SSrikar Dronamraju * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs 12660326f5a9SSrikar Dronamraju * @regs: Reflects the saved state of the task after it has hit a breakpoint 12670326f5a9SSrikar Dronamraju * instruction. 12680326f5a9SSrikar Dronamraju * Return the address of the breakpoint instruction. 12690326f5a9SSrikar Dronamraju */ 12700326f5a9SSrikar Dronamraju unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) 12710326f5a9SSrikar Dronamraju { 12720326f5a9SSrikar Dronamraju return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; 12730326f5a9SSrikar Dronamraju } 12740326f5a9SSrikar Dronamraju 12750326f5a9SSrikar Dronamraju /* 12760326f5a9SSrikar Dronamraju * Called with no locks held. 12770326f5a9SSrikar Dronamraju * Called in context of a exiting or a exec-ing thread. 12780326f5a9SSrikar Dronamraju */ 12790326f5a9SSrikar Dronamraju void uprobe_free_utask(struct task_struct *t) 12800326f5a9SSrikar Dronamraju { 12810326f5a9SSrikar Dronamraju struct uprobe_task *utask = t->utask; 12820326f5a9SSrikar Dronamraju 12830326f5a9SSrikar Dronamraju if (!utask) 12840326f5a9SSrikar Dronamraju return; 12850326f5a9SSrikar Dronamraju 12860326f5a9SSrikar Dronamraju if (utask->active_uprobe) 12870326f5a9SSrikar Dronamraju put_uprobe(utask->active_uprobe); 12880326f5a9SSrikar Dronamraju 1289d4b3b638SSrikar Dronamraju xol_free_insn_slot(t); 12900326f5a9SSrikar Dronamraju kfree(utask); 12910326f5a9SSrikar Dronamraju t->utask = NULL; 12920326f5a9SSrikar Dronamraju } 12930326f5a9SSrikar Dronamraju 12940326f5a9SSrikar Dronamraju /* 12950326f5a9SSrikar Dronamraju * Called in context of a new clone/fork from copy_process. 12960326f5a9SSrikar Dronamraju */ 12970326f5a9SSrikar Dronamraju void uprobe_copy_process(struct task_struct *t) 12980326f5a9SSrikar Dronamraju { 12990326f5a9SSrikar Dronamraju t->utask = NULL; 13000326f5a9SSrikar Dronamraju } 13010326f5a9SSrikar Dronamraju 13020326f5a9SSrikar Dronamraju /* 13030326f5a9SSrikar Dronamraju * Allocate a uprobe_task object for the task. 13040326f5a9SSrikar Dronamraju * Called when the thread hits a breakpoint for the first time. 13050326f5a9SSrikar Dronamraju * 13060326f5a9SSrikar Dronamraju * Returns: 13070326f5a9SSrikar Dronamraju * - pointer to new uprobe_task on success 13080326f5a9SSrikar Dronamraju * - NULL otherwise 13090326f5a9SSrikar Dronamraju */ 13100326f5a9SSrikar Dronamraju static struct uprobe_task *add_utask(void) 13110326f5a9SSrikar Dronamraju { 13120326f5a9SSrikar Dronamraju struct uprobe_task *utask; 13130326f5a9SSrikar Dronamraju 13140326f5a9SSrikar Dronamraju utask = kzalloc(sizeof *utask, GFP_KERNEL); 13150326f5a9SSrikar Dronamraju if (unlikely(!utask)) 13160326f5a9SSrikar Dronamraju return NULL; 13170326f5a9SSrikar Dronamraju 13180326f5a9SSrikar Dronamraju current->utask = utask; 13190326f5a9SSrikar Dronamraju return utask; 13200326f5a9SSrikar Dronamraju } 13210326f5a9SSrikar Dronamraju 13220326f5a9SSrikar Dronamraju /* Prepare to single-step probed instruction out of line. */ 13230326f5a9SSrikar Dronamraju static int 13240326f5a9SSrikar Dronamraju pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr) 13250326f5a9SSrikar Dronamraju { 1326d4b3b638SSrikar Dronamraju if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs)) 1327d4b3b638SSrikar Dronamraju return 0; 1328d4b3b638SSrikar Dronamraju 13290326f5a9SSrikar Dronamraju return -EFAULT; 13300326f5a9SSrikar Dronamraju } 13310326f5a9SSrikar Dronamraju 13320326f5a9SSrikar Dronamraju /* 13330326f5a9SSrikar Dronamraju * If we are singlestepping, then ensure this thread is not connected to 13340326f5a9SSrikar Dronamraju * non-fatal signals until completion of singlestep. When xol insn itself 13350326f5a9SSrikar Dronamraju * triggers the signal, restart the original insn even if the task is 13360326f5a9SSrikar Dronamraju * already SIGKILL'ed (since coredump should report the correct ip). This 13370326f5a9SSrikar Dronamraju * is even more important if the task has a handler for SIGSEGV/etc, The 13380326f5a9SSrikar Dronamraju * _same_ instruction should be repeated again after return from the signal 13390326f5a9SSrikar Dronamraju * handler, and SSTEP can never finish in this case. 13400326f5a9SSrikar Dronamraju */ 13410326f5a9SSrikar Dronamraju bool uprobe_deny_signal(void) 13420326f5a9SSrikar Dronamraju { 13430326f5a9SSrikar Dronamraju struct task_struct *t = current; 13440326f5a9SSrikar Dronamraju struct uprobe_task *utask = t->utask; 13450326f5a9SSrikar Dronamraju 13460326f5a9SSrikar Dronamraju if (likely(!utask || !utask->active_uprobe)) 13470326f5a9SSrikar Dronamraju return false; 13480326f5a9SSrikar Dronamraju 13490326f5a9SSrikar Dronamraju WARN_ON_ONCE(utask->state != UTASK_SSTEP); 13500326f5a9SSrikar Dronamraju 13510326f5a9SSrikar Dronamraju if (signal_pending(t)) { 13520326f5a9SSrikar Dronamraju spin_lock_irq(&t->sighand->siglock); 13530326f5a9SSrikar Dronamraju clear_tsk_thread_flag(t, TIF_SIGPENDING); 13540326f5a9SSrikar Dronamraju spin_unlock_irq(&t->sighand->siglock); 13550326f5a9SSrikar Dronamraju 13560326f5a9SSrikar Dronamraju if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { 13570326f5a9SSrikar Dronamraju utask->state = UTASK_SSTEP_TRAPPED; 13580326f5a9SSrikar Dronamraju set_tsk_thread_flag(t, TIF_UPROBE); 13590326f5a9SSrikar Dronamraju set_tsk_thread_flag(t, TIF_NOTIFY_RESUME); 13600326f5a9SSrikar Dronamraju } 13610326f5a9SSrikar Dronamraju } 13620326f5a9SSrikar Dronamraju 13630326f5a9SSrikar Dronamraju return true; 13640326f5a9SSrikar Dronamraju } 13650326f5a9SSrikar Dronamraju 13660326f5a9SSrikar Dronamraju /* 13670326f5a9SSrikar Dronamraju * Avoid singlestepping the original instruction if the original instruction 13680326f5a9SSrikar Dronamraju * is a NOP or can be emulated. 13690326f5a9SSrikar Dronamraju */ 13700326f5a9SSrikar Dronamraju static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs) 13710326f5a9SSrikar Dronamraju { 137271434f2fSOleg Nesterov if (test_bit(UPROBE_SKIP_SSTEP, &uprobe->flags)) { 13730326f5a9SSrikar Dronamraju if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) 13740326f5a9SSrikar Dronamraju return true; 137571434f2fSOleg Nesterov clear_bit(UPROBE_SKIP_SSTEP, &uprobe->flags); 13760578a970SOleg Nesterov } 13770326f5a9SSrikar Dronamraju return false; 13780326f5a9SSrikar Dronamraju } 13790326f5a9SSrikar Dronamraju 1380499a4f3eSOleg Nesterov static void mmf_recalc_uprobes(struct mm_struct *mm) 1381499a4f3eSOleg Nesterov { 1382499a4f3eSOleg Nesterov struct vm_area_struct *vma; 1383499a4f3eSOleg Nesterov 1384499a4f3eSOleg Nesterov for (vma = mm->mmap; vma; vma = vma->vm_next) { 1385499a4f3eSOleg Nesterov if (!valid_vma(vma, false)) 1386499a4f3eSOleg Nesterov continue; 1387499a4f3eSOleg Nesterov /* 1388499a4f3eSOleg Nesterov * This is not strictly accurate, we can race with 1389499a4f3eSOleg Nesterov * uprobe_unregister() and see the already removed 1390499a4f3eSOleg Nesterov * uprobe if delete_uprobe() was not yet called. 139163633cbfSOleg Nesterov * Or this uprobe can be filtered out. 1392499a4f3eSOleg Nesterov */ 1393499a4f3eSOleg Nesterov if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) 1394499a4f3eSOleg Nesterov return; 1395499a4f3eSOleg Nesterov } 1396499a4f3eSOleg Nesterov 1397499a4f3eSOleg Nesterov clear_bit(MMF_HAS_UPROBES, &mm->flags); 1398499a4f3eSOleg Nesterov } 1399499a4f3eSOleg Nesterov 1400ec75fba9SOleg Nesterov static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr) 1401ec75fba9SOleg Nesterov { 1402ec75fba9SOleg Nesterov struct page *page; 1403ec75fba9SOleg Nesterov uprobe_opcode_t opcode; 1404ec75fba9SOleg Nesterov int result; 1405ec75fba9SOleg Nesterov 1406ec75fba9SOleg Nesterov pagefault_disable(); 1407ec75fba9SOleg Nesterov result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr, 1408ec75fba9SOleg Nesterov sizeof(opcode)); 1409ec75fba9SOleg Nesterov pagefault_enable(); 1410ec75fba9SOleg Nesterov 1411ec75fba9SOleg Nesterov if (likely(result == 0)) 1412ec75fba9SOleg Nesterov goto out; 1413ec75fba9SOleg Nesterov 1414ec75fba9SOleg Nesterov result = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL); 1415ec75fba9SOleg Nesterov if (result < 0) 1416ec75fba9SOleg Nesterov return result; 1417ec75fba9SOleg Nesterov 1418ec75fba9SOleg Nesterov copy_opcode(page, vaddr, &opcode); 1419ec75fba9SOleg Nesterov put_page(page); 1420ec75fba9SOleg Nesterov out: 1421ec75fba9SOleg Nesterov return is_swbp_insn(&opcode); 1422ec75fba9SOleg Nesterov } 1423ec75fba9SOleg Nesterov 1424d790d346SOleg Nesterov static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) 14250326f5a9SSrikar Dronamraju { 14263a9ea052SOleg Nesterov struct mm_struct *mm = current->mm; 14273a9ea052SOleg Nesterov struct uprobe *uprobe = NULL; 14280326f5a9SSrikar Dronamraju struct vm_area_struct *vma; 14290326f5a9SSrikar Dronamraju 14300326f5a9SSrikar Dronamraju down_read(&mm->mmap_sem); 14310326f5a9SSrikar Dronamraju vma = find_vma(mm, bp_vaddr); 14323a9ea052SOleg Nesterov if (vma && vma->vm_start <= bp_vaddr) { 14333a9ea052SOleg Nesterov if (valid_vma(vma, false)) { 1434cb113b47SOleg Nesterov struct inode *inode = vma->vm_file->f_mapping->host; 1435cb113b47SOleg Nesterov loff_t offset = vaddr_to_offset(vma, bp_vaddr); 14360326f5a9SSrikar Dronamraju 14370326f5a9SSrikar Dronamraju uprobe = find_uprobe(inode, offset); 14380326f5a9SSrikar Dronamraju } 1439d790d346SOleg Nesterov 1440d790d346SOleg Nesterov if (!uprobe) 1441d790d346SOleg Nesterov *is_swbp = is_swbp_at_addr(mm, bp_vaddr); 1442d790d346SOleg Nesterov } else { 1443d790d346SOleg Nesterov *is_swbp = -EFAULT; 14443a9ea052SOleg Nesterov } 1445499a4f3eSOleg Nesterov 1446499a4f3eSOleg Nesterov if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) 1447499a4f3eSOleg Nesterov mmf_recalc_uprobes(mm); 14480326f5a9SSrikar Dronamraju up_read(&mm->mmap_sem); 14490326f5a9SSrikar Dronamraju 14503a9ea052SOleg Nesterov return uprobe; 14513a9ea052SOleg Nesterov } 14523a9ea052SOleg Nesterov 14533a9ea052SOleg Nesterov /* 14543a9ea052SOleg Nesterov * Run handler and ask thread to singlestep. 14553a9ea052SOleg Nesterov * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. 14563a9ea052SOleg Nesterov */ 14573a9ea052SOleg Nesterov static void handle_swbp(struct pt_regs *regs) 14583a9ea052SOleg Nesterov { 14593a9ea052SOleg Nesterov struct uprobe_task *utask; 14603a9ea052SOleg Nesterov struct uprobe *uprobe; 14613a9ea052SOleg Nesterov unsigned long bp_vaddr; 146256bb4cf6SOleg Nesterov int uninitialized_var(is_swbp); 14633a9ea052SOleg Nesterov 14643a9ea052SOleg Nesterov bp_vaddr = uprobe_get_swbp_addr(regs); 1465d790d346SOleg Nesterov uprobe = find_active_uprobe(bp_vaddr, &is_swbp); 14663a9ea052SOleg Nesterov 14670326f5a9SSrikar Dronamraju if (!uprobe) { 146856bb4cf6SOleg Nesterov if (is_swbp > 0) { 14690326f5a9SSrikar Dronamraju /* No matching uprobe; signal SIGTRAP. */ 14700326f5a9SSrikar Dronamraju send_sig(SIGTRAP, current, 0); 147156bb4cf6SOleg Nesterov } else { 147256bb4cf6SOleg Nesterov /* 147356bb4cf6SOleg Nesterov * Either we raced with uprobe_unregister() or we can't 147456bb4cf6SOleg Nesterov * access this memory. The latter is only possible if 147556bb4cf6SOleg Nesterov * another thread plays with our ->mm. In both cases 147656bb4cf6SOleg Nesterov * we can simply restart. If this vma was unmapped we 147756bb4cf6SOleg Nesterov * can pretend this insn was not executed yet and get 147856bb4cf6SOleg Nesterov * the (correct) SIGSEGV after restart. 147956bb4cf6SOleg Nesterov */ 148056bb4cf6SOleg Nesterov instruction_pointer_set(regs, bp_vaddr); 148156bb4cf6SOleg Nesterov } 14820326f5a9SSrikar Dronamraju return; 14830326f5a9SSrikar Dronamraju } 1484142b18ddSOleg Nesterov /* 1485142b18ddSOleg Nesterov * TODO: move copy_insn/etc into _register and remove this hack. 1486142b18ddSOleg Nesterov * After we hit the bp, _unregister + _register can install the 1487142b18ddSOleg Nesterov * new and not-yet-analyzed uprobe at the same address, restart. 1488142b18ddSOleg Nesterov */ 1489142b18ddSOleg Nesterov smp_rmb(); /* pairs with wmb() in install_breakpoint() */ 149071434f2fSOleg Nesterov if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) 1491142b18ddSOleg Nesterov goto restart; 14920326f5a9SSrikar Dronamraju 14930326f5a9SSrikar Dronamraju utask = current->utask; 14940326f5a9SSrikar Dronamraju if (!utask) { 14950326f5a9SSrikar Dronamraju utask = add_utask(); 14960326f5a9SSrikar Dronamraju /* Cannot allocate; re-execute the instruction. */ 14970326f5a9SSrikar Dronamraju if (!utask) 14980578a970SOleg Nesterov goto restart; 14990326f5a9SSrikar Dronamraju } 15000326f5a9SSrikar Dronamraju 15010326f5a9SSrikar Dronamraju handler_chain(uprobe, regs); 15020578a970SOleg Nesterov if (can_skip_sstep(uprobe, regs)) 15030578a970SOleg Nesterov goto out; 15040326f5a9SSrikar Dronamraju 15050326f5a9SSrikar Dronamraju if (!pre_ssout(uprobe, regs, bp_vaddr)) { 1506746a9e6bSOleg Nesterov utask->active_uprobe = uprobe; 1507746a9e6bSOleg Nesterov utask->state = UTASK_SSTEP; 15080326f5a9SSrikar Dronamraju return; 15090326f5a9SSrikar Dronamraju } 15100326f5a9SSrikar Dronamraju 15110578a970SOleg Nesterov restart: 15120326f5a9SSrikar Dronamraju /* 15130326f5a9SSrikar Dronamraju * cannot singlestep; cannot skip instruction; 15140326f5a9SSrikar Dronamraju * re-execute the instruction. 15150326f5a9SSrikar Dronamraju */ 15160326f5a9SSrikar Dronamraju instruction_pointer_set(regs, bp_vaddr); 15170578a970SOleg Nesterov out: 15180326f5a9SSrikar Dronamraju put_uprobe(uprobe); 15190326f5a9SSrikar Dronamraju } 15200326f5a9SSrikar Dronamraju 15210326f5a9SSrikar Dronamraju /* 15220326f5a9SSrikar Dronamraju * Perform required fix-ups and disable singlestep. 15230326f5a9SSrikar Dronamraju * Allow pending signals to take effect. 15240326f5a9SSrikar Dronamraju */ 15250326f5a9SSrikar Dronamraju static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) 15260326f5a9SSrikar Dronamraju { 15270326f5a9SSrikar Dronamraju struct uprobe *uprobe; 15280326f5a9SSrikar Dronamraju 15290326f5a9SSrikar Dronamraju uprobe = utask->active_uprobe; 15300326f5a9SSrikar Dronamraju if (utask->state == UTASK_SSTEP_ACK) 15310326f5a9SSrikar Dronamraju arch_uprobe_post_xol(&uprobe->arch, regs); 15320326f5a9SSrikar Dronamraju else if (utask->state == UTASK_SSTEP_TRAPPED) 15330326f5a9SSrikar Dronamraju arch_uprobe_abort_xol(&uprobe->arch, regs); 15340326f5a9SSrikar Dronamraju else 15350326f5a9SSrikar Dronamraju WARN_ON_ONCE(1); 15360326f5a9SSrikar Dronamraju 15370326f5a9SSrikar Dronamraju put_uprobe(uprobe); 15380326f5a9SSrikar Dronamraju utask->active_uprobe = NULL; 15390326f5a9SSrikar Dronamraju utask->state = UTASK_RUNNING; 1540d4b3b638SSrikar Dronamraju xol_free_insn_slot(current); 15410326f5a9SSrikar Dronamraju 15420326f5a9SSrikar Dronamraju spin_lock_irq(¤t->sighand->siglock); 15430326f5a9SSrikar Dronamraju recalc_sigpending(); /* see uprobe_deny_signal() */ 15440326f5a9SSrikar Dronamraju spin_unlock_irq(¤t->sighand->siglock); 15450326f5a9SSrikar Dronamraju } 15460326f5a9SSrikar Dronamraju 15470326f5a9SSrikar Dronamraju /* 15481b08e907SOleg Nesterov * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and 15491b08e907SOleg Nesterov * allows the thread to return from interrupt. After that handle_swbp() 15501b08e907SOleg Nesterov * sets utask->active_uprobe. 15510326f5a9SSrikar Dronamraju * 15521b08e907SOleg Nesterov * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag 15531b08e907SOleg Nesterov * and allows the thread to return from interrupt. 15540326f5a9SSrikar Dronamraju * 15550326f5a9SSrikar Dronamraju * While returning to userspace, thread notices the TIF_UPROBE flag and calls 15560326f5a9SSrikar Dronamraju * uprobe_notify_resume(). 15570326f5a9SSrikar Dronamraju */ 15580326f5a9SSrikar Dronamraju void uprobe_notify_resume(struct pt_regs *regs) 15590326f5a9SSrikar Dronamraju { 15600326f5a9SSrikar Dronamraju struct uprobe_task *utask; 15610326f5a9SSrikar Dronamraju 1562db023ea5SOleg Nesterov clear_thread_flag(TIF_UPROBE); 1563db023ea5SOleg Nesterov 15640326f5a9SSrikar Dronamraju utask = current->utask; 15651b08e907SOleg Nesterov if (utask && utask->active_uprobe) 15660326f5a9SSrikar Dronamraju handle_singlestep(utask, regs); 15671b08e907SOleg Nesterov else 15681b08e907SOleg Nesterov handle_swbp(regs); 15690326f5a9SSrikar Dronamraju } 15700326f5a9SSrikar Dronamraju 15710326f5a9SSrikar Dronamraju /* 15720326f5a9SSrikar Dronamraju * uprobe_pre_sstep_notifier gets called from interrupt context as part of 15730326f5a9SSrikar Dronamraju * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. 15740326f5a9SSrikar Dronamraju */ 15750326f5a9SSrikar Dronamraju int uprobe_pre_sstep_notifier(struct pt_regs *regs) 15760326f5a9SSrikar Dronamraju { 1577f8ac4ec9SOleg Nesterov if (!current->mm || !test_bit(MMF_HAS_UPROBES, ¤t->mm->flags)) 15780326f5a9SSrikar Dronamraju return 0; 15790326f5a9SSrikar Dronamraju 15800326f5a9SSrikar Dronamraju set_thread_flag(TIF_UPROBE); 15810326f5a9SSrikar Dronamraju return 1; 15820326f5a9SSrikar Dronamraju } 15830326f5a9SSrikar Dronamraju 15840326f5a9SSrikar Dronamraju /* 15850326f5a9SSrikar Dronamraju * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier 15860326f5a9SSrikar Dronamraju * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. 15870326f5a9SSrikar Dronamraju */ 15880326f5a9SSrikar Dronamraju int uprobe_post_sstep_notifier(struct pt_regs *regs) 15890326f5a9SSrikar Dronamraju { 15900326f5a9SSrikar Dronamraju struct uprobe_task *utask = current->utask; 15910326f5a9SSrikar Dronamraju 15920326f5a9SSrikar Dronamraju if (!current->mm || !utask || !utask->active_uprobe) 15930326f5a9SSrikar Dronamraju /* task is currently not uprobed */ 15940326f5a9SSrikar Dronamraju return 0; 15950326f5a9SSrikar Dronamraju 15960326f5a9SSrikar Dronamraju utask->state = UTASK_SSTEP_ACK; 15970326f5a9SSrikar Dronamraju set_thread_flag(TIF_UPROBE); 15980326f5a9SSrikar Dronamraju return 1; 15990326f5a9SSrikar Dronamraju } 16000326f5a9SSrikar Dronamraju 16010326f5a9SSrikar Dronamraju static struct notifier_block uprobe_exception_nb = { 16020326f5a9SSrikar Dronamraju .notifier_call = arch_uprobe_exception_notify, 16030326f5a9SSrikar Dronamraju .priority = INT_MAX-1, /* notified after kprobes, kgdb */ 16040326f5a9SSrikar Dronamraju }; 16050326f5a9SSrikar Dronamraju 1606a5f4374aSIngo Molnar static int __init init_uprobes(void) 1607a5f4374aSIngo Molnar { 1608a5f4374aSIngo Molnar int i; 1609a5f4374aSIngo Molnar 1610a5f4374aSIngo Molnar for (i = 0; i < UPROBES_HASH_SZ; i++) { 1611a5f4374aSIngo Molnar mutex_init(&uprobes_mutex[i]); 1612a5f4374aSIngo Molnar mutex_init(&uprobes_mmap_mutex[i]); 1613a5f4374aSIngo Molnar } 16140326f5a9SSrikar Dronamraju 161532cdba1eSOleg Nesterov if (percpu_init_rwsem(&dup_mmap_sem)) 161632cdba1eSOleg Nesterov return -ENOMEM; 161732cdba1eSOleg Nesterov 16180326f5a9SSrikar Dronamraju return register_die_notifier(&uprobe_exception_nb); 1619a5f4374aSIngo Molnar } 16200326f5a9SSrikar Dronamraju module_init(init_uprobes); 1621a5f4374aSIngo Molnar 1622a5f4374aSIngo Molnar static void __exit exit_uprobes(void) 1623a5f4374aSIngo Molnar { 1624a5f4374aSIngo Molnar } 1625a5f4374aSIngo Molnar module_exit(exit_uprobes); 1626