bsd/dev/dtrace/dtrace_ptss.c

   1 /*
   2  * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28
  29 #include <sys/types.h>
  30 #include <sys/proc.h>
  31 #include <sys/proc_internal.h>
  32 #include <sys/systm.h>
  33 #include <sys/user.h>
  34 #include <sys/dtrace_ptss.h>
  35
  36 #include <mach/vm_param.h>
  37 #include <mach/mach_vm.h>
  38
  39 #include <kern/task.h>
  40
  41 #include <vm/vm_map.h>
  42
  43 /*
  44  * This function requires the sprlock to be held
  45  *
  46  * In general, it will not block. If it needs to allocate a new
  47  * page of memory, the underlying kernel _MALLOC may block.
  48  */
  49 struct dtrace_ptss_page_entry*
  50 dtrace_ptss_claim_entry_locked(struct proc* p) {
  51         lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
  52
  53         struct dtrace_ptss_page_entry* entry = NULL;
  54
  55         while (TRUE) {
  56                 struct dtrace_ptss_page_entry* temp = p->p_dtrace_ptss_free_list;
  57
  58                 if (temp == NULL) {
  59                         // Nothing on the free list. Allocate a new page, its okay if multiple threads race here.
  60                         struct dtrace_ptss_page* page = dtrace_ptss_allocate_page(p);
  61
  62                         // Make sure we actually got a page
  63                         if (page == NULL)
  64                                 return NULL;
  65
  66                         // Add the page to the page list
  67                         page->next = p->p_dtrace_ptss_pages;
  68                         p->p_dtrace_ptss_pages = page;
  69
  70                         // CAS the entries onto the free list.
  71                         do {
  72                                 page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next = p->p_dtrace_ptss_free_list;
  73                         } while (!OSCompareAndSwapPtr((void *)page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next,
  74                                                    (void *)&page->entries[0],
  75                                                    (void * volatile *)&p->p_dtrace_ptss_free_list));
  76
  77                         // Now that we've added to the free list, try again.
  78                         continue;
  79                 }
  80
  81                 // Claim temp
  82                 if (!OSCompareAndSwapPtr((void *)temp, (void *)temp->next, (void * volatile *)&p->p_dtrace_ptss_free_list))
  83                         continue;
  84
  85                 // At this point, we own temp.
  86                 entry = temp;
  87
  88                 break;
  89         }
  90
  91         return entry;
  92 }
  93
  94 /*
  95  * This function does not require any locks to be held on entry.
  96  */
  97 struct dtrace_ptss_page_entry*
  98 dtrace_ptss_claim_entry(struct proc* p) {
  99         // Verify no locks held on entry
 100         lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_NOTOWNED);
 101         lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
 102
 103         struct dtrace_ptss_page_entry* entry = NULL;
 104
 105         while (TRUE) {
 106                 struct dtrace_ptss_page_entry* temp = p->p_dtrace_ptss_free_list;
 107
 108                 if (temp == NULL) {
 109                         lck_mtx_lock(&p->p_dtrace_sprlock);
 110                         temp = dtrace_ptss_claim_entry_locked(p);
 111                         lck_mtx_unlock(&p->p_dtrace_sprlock);
 112                         return temp;
 113                 }
 114
 115                 // Claim temp
 116                 if (!OSCompareAndSwapPtr((void *)temp, (void *)temp->next, (void * volatile *)&p->p_dtrace_ptss_free_list))
 117                         continue;
 118
 119                 // At this point, we own temp.
 120                 entry = temp;
 121
 122                 break;
 123         }
 124
 125         return entry;
 126 }
 127
 128 /*
 129  * This function does not require any locks to be held on entry.
 130  */
 131 void
 132 dtrace_ptss_release_entry(struct proc* p, struct dtrace_ptss_page_entry* e) {
 133         if (p && e) {
 134                 do {
 135                         e->next = p->p_dtrace_ptss_free_list;
 136                 } while (!OSCompareAndSwapPtr((void *)e->next, (void *)e, (void * volatile *)&p->p_dtrace_ptss_free_list));
 137         }
 138 }
 139
 140 /*
 141  * This function allocates a new page in the target process's address space.
 142  *
 143  * It returns a dtrace_ptss_page that has its entries chained, with the last
 144  * entries next field set to NULL. It does not add the page or the entries to
 145  * the process's page/entry lists.
 146  *
 147  * This function does not require that any locks be held when it is invoked.
 148  */
 149 struct dtrace_ptss_page*
 150 dtrace_ptss_allocate_page(struct proc* p)
 151 {
 152         // Allocate the kernel side data
 153         struct dtrace_ptss_page* ptss_page = _MALLOC(sizeof(struct dtrace_ptss_page), M_TEMP, M_ZERO | M_WAITOK);
 154         if (ptss_page == NULL)
 155                 return NULL;
 156
 157         // Now allocate a page in user space and set its protections to allow execute.
 158         task_t task = p->task;
 159         vm_map_t map = get_task_map_reference(task);
 160
 161         mach_vm_address_t addr = 0LL;
 162         mach_vm_size_t size = PAGE_SIZE; // We need some way to assert that this matches vm_map_round_page() !!!
 163
 164 #if CONFIG_EMBEDDED
 165         /* The embedded OS has extra permissions for writable and executable pages. We can't pass in the flags
 166          * we need for the correct permissions from mach_vm_allocate, so need to call mach_vm_map directly. */
 167         vm_map_offset_t map_addr = 0;
 168         kern_return_t kr = mach_vm_map(map, &map_addr, size, 0, VM_FLAGS_ANYWHERE, IPC_PORT_NULL, 0, FALSE, VM_PROT_READ|VM_PROT_EXECUTE, VM_PROT_READ|VM_PROT_EXECUTE, VM_INHERIT_DEFAULT);
 169         if (kr != KERN_SUCCESS) {
 170                 goto err;
 171         }
 172         addr = map_addr;
 173 #else
 174         kern_return_t kr = mach_vm_allocate(map, &addr, size, VM_FLAGS_ANYWHERE);
 175         if (kr != KERN_SUCCESS) {
 176                 goto err;
 177         }
 178
 179         kr = mach_vm_protect(map, addr, size, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
 180         if (kr != KERN_SUCCESS) {
 181                 mach_vm_deallocate(map, addr, size);
 182                 goto err;
 183         }
 184 #endif
 185
 186         // Chain the page entries.
 187         int i;
 188         for (i=0; i<DTRACE_PTSS_ENTRIES_PER_PAGE; i++) {
 189                 ptss_page->entries[i].addr = addr + (i * DTRACE_PTSS_SCRATCH_SPACE_PER_THREAD);
 190                 ptss_page->entries[i].next = &ptss_page->entries[i+1];
 191         }
 192
 193         // The last entry should point to NULL
 194         ptss_page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next = NULL;
 195
 196         vm_map_deallocate(map);
 197
 198         return ptss_page;
 199
 200 err:
 201         _FREE(ptss_page, M_TEMP);
 202
 203         vm_map_deallocate(map);
 204
 205         return NULL;
 206 }
 207
 208 /*
 209  * This function frees an existing page in the target process's address space.
 210  *
 211  * It does not alter any of the process's page/entry lists.
 212  *
 213  * TODO: Inline in dtrace_ptrace_exec_exit?
 214  */
 215 void
 216 dtrace_ptss_free_page(struct proc* p, struct dtrace_ptss_page* ptss_page)
 217 {
 218         // Grab the task and get a reference to its vm_map
 219         task_t task = p->task;
 220         vm_map_t map = get_task_map_reference(task);
 221
 222         mach_vm_address_t addr = ptss_page->entries[0].addr;
 223         mach_vm_size_t size = PAGE_SIZE; // We need some way to assert that this matches vm_map_round_page() !!!
 224
 225         // Silent failures, no point in checking return code.
 226         mach_vm_deallocate(map, addr, size);
 227
 228         vm_map_deallocate(map);
 229 }
 230
 231 /*
 232  * This function assumes that the target process has been
 233  * suspended, and the proc_lock & sprlock is held
 234  */
 235 void
 236 dtrace_ptss_enable(struct proc* p) {
 237         lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
 238         lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_OWNED);
 239
 240         struct uthread* uth;
 241         /*
 242          * XXX There has been a concern raised about holding the proc_lock
 243          * while calling dtrace_ptss_claim_entry(), due to the fact
 244          * that dtrace_ptss_claim_entry() can potentially malloc.
 245          */
 246         TAILQ_FOREACH(uth, &p->p_uthlist, uu_list) {
 247                 uth->t_dtrace_scratch = dtrace_ptss_claim_entry_locked(p);
 248         }
 249 }
 250
 251 /*
 252  * This function is not thread safe.
 253  *
 254  * It assumes the sprlock is held, and the proc_lock is not.
 255  */
 256 void
 257 dtrace_ptss_exec_exit(struct proc* p) {
 258         /*
 259          * Should hold sprlock to touch the pages list. Must not
 260          * hold the proc lock to avoid deadlock.
 261          */
 262         lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
 263         lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
 264
 265         p->p_dtrace_ptss_free_list = NULL;
 266
 267         struct dtrace_ptss_page* temp = p->p_dtrace_ptss_pages;
 268         p->p_dtrace_ptss_pages = NULL;
 269
 270         while (temp != NULL) {
 271                 struct dtrace_ptss_page* next = temp->next;
 272
 273                 // Do we need to specifically mach_vm_deallocate the user pages?
 274                 // This can be called when the process is exiting, I believe the proc's
 275                 // vm_map_t may already be toast.
 276
 277                 // Must be certain to free the kernel memory!
 278                 _FREE(temp, M_TEMP);
 279                 temp = next;
 280         }
 281 }
 282
 283 /*
 284  * This function is not thread safe. It is not used for vfork.
 285  *
 286  * The child proc ptss fields are initialized to NULL at fork time.
 287  * Pages allocated in the parent are copied as part of the vm_map copy, though.
 288  * We need to deallocate those pages.
 289  *
 290  * Parent and child sprlock should be held, and proc_lock must NOT be held.
 291  */
 292 void
 293 dtrace_ptss_fork(struct proc* parent, struct proc* child) {
 294         // The child should not have any pages/entries allocated at this point.
 295         // ASSERT(child->p_dtrace_ptss_pages == NULL);
 296         // ASSERT(child->p_dtrace_ptss_free_list == NULL);
 297
 298         /*
 299          * The parent's sprlock should be held, to protect its pages list
 300          * from changing while the child references it. The child's sprlock
 301          * must also be held, because we are modifying its pages list.
 302          * Finally, to prevent a deadlock with the fasttrap cleanup code,
 303          * neither the parent or child proc_lock should be held.
 304          */
 305         lck_mtx_assert(&parent->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
 306         lck_mtx_assert(&parent->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
 307         lck_mtx_assert(&child->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
 308         lck_mtx_assert(&child->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
 309
 310         // Get page list from *PARENT*
 311         struct dtrace_ptss_page* temp = parent->p_dtrace_ptss_pages;
 312
 313         while (temp != NULL) {
 314                 // Freeing the page in the *CHILD*
 315                 dtrace_ptss_free_page(child, temp);
 316
 317                 // Do not free the kernel memory, it belong to the parent.
 318                 temp = temp->next;
 319         }
 320 }