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_map.h>
  37 #include <mach/vm_param.h>
  38 #include <mach/mach_vm.h>
  39
  40 #include <kern/task.h>
  41
  42 #include <vm/vm_map.h>
  43
  44 /*
  45  * This function requires the sprlock to be held
  46  *
  47  * In general, it will not block. If it needs to allocate a new
  48  * page of memory, the underlying kernel _MALLOC may block.
  49  */
  50 struct dtrace_ptss_page_entry*
  51 dtrace_ptss_claim_entry_locked(struct proc* p) {
  52         LCK_MTX_ASSERT(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
  53
  54         struct dtrace_ptss_page_entry* entry = NULL;
  55
  56         while (TRUE) {
  57                 struct dtrace_ptss_page_entry* temp = p->p_dtrace_ptss_free_list;
  58
  59                 if (temp == NULL) {
  60                         // Nothing on the free list. Allocate a new page, its okay if multiple threads race here.
  61                         struct dtrace_ptss_page* page = dtrace_ptss_allocate_page(p);
  62
  63                         // Make sure we actually got a page
  64                         if (page == NULL)
  65                                 return NULL;
  66
  67                         // Add the page to the page list
  68                         page->next = p->p_dtrace_ptss_pages;
  69                         p->p_dtrace_ptss_pages = page;
  70
  71                         // CAS the entries onto the free list.
  72                         do {
  73                                 page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next = p->p_dtrace_ptss_free_list;
  74                         } while (!OSCompareAndSwapPtr((void *)page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next,
  75                                                    (void *)&page->entries[0],
  76                                                    (void * volatile *)&p->p_dtrace_ptss_free_list));
  77
  78                         // Now that we've added to the free list, try again.
  79                         continue;
  80                 }
  81
  82                 // Claim temp
  83                 if (!OSCompareAndSwapPtr((void *)temp, (void *)temp->next, (void * volatile *)&p->p_dtrace_ptss_free_list))
  84                         continue;
  85
  86                 // At this point, we own temp.
  87                 entry = temp;
  88
  89                 break;
  90         }
  91
  92         return entry;
  93 }
  94
  95 /*
  96  * This function does not require any locks to be held on entry.
  97  */
  98 struct dtrace_ptss_page_entry*
  99 dtrace_ptss_claim_entry(struct proc* p) {
 100         // Verify no locks held on entry
 101         LCK_MTX_ASSERT(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_NOTOWNED);
 102         LCK_MTX_ASSERT(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
 103
 104         struct dtrace_ptss_page_entry* entry = NULL;
 105
 106         while (TRUE) {
 107                 struct dtrace_ptss_page_entry* temp = p->p_dtrace_ptss_free_list;
 108
 109                 if (temp == NULL) {
 110                         lck_mtx_lock(&p->p_dtrace_sprlock);
 111                         temp = dtrace_ptss_claim_entry_locked(p);
 112                         lck_mtx_unlock(&p->p_dtrace_sprlock);
 113                         return temp;
 114                 }
 115
 116                 // Claim temp
 117                 if (!OSCompareAndSwapPtr((void *)temp, (void *)temp->next, (void * volatile *)&p->p_dtrace_ptss_free_list))
 118                         continue;
 119
 120                 // At this point, we own temp.
 121                 entry = temp;
 122
 123                 break;
 124         }
 125
 126         return entry;
 127 }
 128
 129 /*
 130  * This function does not require any locks to be held on entry.
 131  *
 132  * (PR-11138709) A NULL p->p_dtrace_ptss_pages means the entry can
 133  * no longer be referenced safely. When found in this state, the chore
 134  * of releasing an entry to the free list is ignored.
 135  */
 136 void
 137 dtrace_ptss_release_entry(struct proc* p, struct dtrace_ptss_page_entry* e) {
 138         if (p && p->p_dtrace_ptss_pages && e) {
 139                 do {
 140                         e->next = p->p_dtrace_ptss_free_list;
 141                 } while (!OSCompareAndSwapPtr((void *)e->next, (void *)e, (void * volatile *)&p->p_dtrace_ptss_free_list));
 142         }
 143 }
 144
 145 /*
 146  * This function allocates a new page in the target process's address space.
 147  *
 148  * It returns a dtrace_ptss_page that has its entries chained, with the last
 149  * entries next field set to NULL. It does not add the page or the entries to
 150  * the process's page/entry lists.
 151  *
 152  * This function does not require that any locks be held when it is invoked.
 153  */
 154 struct dtrace_ptss_page*
 155 dtrace_ptss_allocate_page(struct proc* p)
 156 {
 157         // Allocate the kernel side data
 158         struct dtrace_ptss_page* ptss_page = _MALLOC(sizeof(struct dtrace_ptss_page), M_TEMP, M_ZERO | M_WAITOK);
 159         if (ptss_page == NULL)
 160                 return NULL;
 161
 162         // Now allocate a page in user space and set its protections to allow execute.
 163         task_t task = p->task;
 164         vm_map_t map = get_task_map_reference(task);
 165         if (map == NULL)
 166           goto err;
 167
 168         mach_vm_size_t size = PAGE_MAX_SIZE;
 169         mach_vm_offset_t addr = 0;
 170         mach_vm_offset_t write_addr = 0;
 171         /*
 172          * The embedded OS has extra permissions for writable and executable pages.
 173          * To ensure correct permissions, we must set the page protections separately.
 174          */
 175         vm_prot_t cur_protection = VM_PROT_READ|VM_PROT_EXECUTE;
 176         vm_prot_t max_protection = VM_PROT_READ|VM_PROT_EXECUTE|VM_PROT_WRITE;
 177
 178         kern_return_t kr = mach_vm_map_kernel(map, &addr, size, 0, VM_FLAGS_ANYWHERE, VM_MAP_KERNEL_FLAGS_NONE, VM_KERN_MEMORY_NONE, IPC_PORT_NULL, 0, FALSE, cur_protection, max_protection, VM_INHERIT_DEFAULT);
 179         if (kr != KERN_SUCCESS) {
 180                 goto err;
 181         }
 182         /*
 183          * If on embedded, remap the scratch space as writable at another
 184          * virtual address
 185          */
 186          kr = mach_vm_remap_kernel(map, &write_addr, size, 0, VM_FLAGS_ANYWHERE, VM_KERN_MEMORY_NONE, map, addr, FALSE, &cur_protection, &max_protection, VM_INHERIT_DEFAULT);
 187         if (kr != KERN_SUCCESS || !(max_protection & VM_PROT_WRITE))
 188                 goto err;
 189
 190         kr = mach_vm_protect (map, (mach_vm_offset_t)write_addr, (mach_vm_size_t)size, 0, VM_PROT_READ | VM_PROT_WRITE);
 191         if (kr != KERN_SUCCESS)
 192                 goto err;
 193
 194         // Chain the page entries.
 195         int i;
 196         for (i=0; i<DTRACE_PTSS_ENTRIES_PER_PAGE; i++) {
 197                 ptss_page->entries[i].addr = addr + (i * DTRACE_PTSS_SCRATCH_SPACE_PER_THREAD);
 198                 ptss_page->entries[i].write_addr = write_addr + (i * DTRACE_PTSS_SCRATCH_SPACE_PER_THREAD);
 199                 ptss_page->entries[i].next = &ptss_page->entries[i+1];
 200         }
 201
 202         // The last entry should point to NULL
 203         ptss_page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next = NULL;
 204
 205         vm_map_deallocate(map);
 206
 207         return ptss_page;
 208
 209 err:
 210         _FREE(ptss_page, M_TEMP);
 211
 212         if (map)
 213           vm_map_deallocate(map);
 214
 215         return NULL;
 216 }
 217
 218 /*
 219  * This function frees an existing page in the target process's address space.
 220  *
 221  * It does not alter any of the process's page/entry lists.
 222  *
 223  * TODO: Inline in dtrace_ptrace_exec_exit?
 224  */
 225 void
 226 dtrace_ptss_free_page(struct proc* p, struct dtrace_ptss_page* ptss_page)
 227 {
 228         // Grab the task and get a reference to its vm_map
 229         task_t task = p->task;
 230         vm_map_t map = get_task_map_reference(task);
 231
 232         mach_vm_address_t addr = ptss_page->entries[0].addr;
 233         mach_vm_size_t size = PAGE_SIZE; // We need some way to assert that this matches vm_map_round_page() !!!
 234
 235         // Silent failures, no point in checking return code.
 236         mach_vm_deallocate(map, addr, size);
 237
 238         mach_vm_address_t write_addr = ptss_page->entries[0].write_addr;
 239         mach_vm_deallocate(map, write_addr, size);
 240
 241         vm_map_deallocate(map);
 242 }
 243
 244 /*
 245  * This function assumes that the target process has been
 246  * suspended, and the proc_lock & sprlock is held
 247  */
 248 void
 249 dtrace_ptss_enable(struct proc* p) {
 250         LCK_MTX_ASSERT(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
 251         LCK_MTX_ASSERT(&p->p_mlock, LCK_MTX_ASSERT_OWNED);
 252
 253         struct uthread* uth;
 254         /*
 255          * XXX There has been a concern raised about holding the proc_lock
 256          * while calling dtrace_ptss_claim_entry(), due to the fact
 257          * that dtrace_ptss_claim_entry() can potentially malloc.
 258          */
 259         TAILQ_FOREACH(uth, &p->p_uthlist, uu_list) {
 260                 uth->t_dtrace_scratch = dtrace_ptss_claim_entry_locked(p);
 261         }
 262 }
 263
 264 /*
 265  * This function is not thread safe.
 266  *
 267  * It assumes the sprlock is held, and the proc_lock is not.
 268  */
 269 void
 270 dtrace_ptss_exec_exit(struct proc* p) {
 271         /*
 272          * Should hold sprlock to touch the pages list. Must not
 273          * hold the proc lock to avoid deadlock.
 274          */
 275         LCK_MTX_ASSERT(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
 276         LCK_MTX_ASSERT(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
 277
 278         p->p_dtrace_ptss_free_list = NULL;
 279
 280         struct dtrace_ptss_page* temp = p->p_dtrace_ptss_pages;
 281         p->p_dtrace_ptss_pages = NULL;
 282
 283         while (temp != NULL) {
 284                 struct dtrace_ptss_page* next = temp->next;
 285
 286                 // Do we need to specifically mach_vm_deallocate the user pages?
 287                 // This can be called when the process is exiting, I believe the proc's
 288                 // vm_map_t may already be toast.
 289
 290                 // Must be certain to free the kernel memory!
 291                 _FREE(temp, M_TEMP);
 292                 temp = next;
 293         }
 294 }
 295
 296 /*
 297  * This function is not thread safe. It is not used for vfork.
 298  *
 299  * The child proc ptss fields are initialized to NULL at fork time.
 300  * Pages allocated in the parent are copied as part of the vm_map copy, though.
 301  * We need to deallocate those pages.
 302  *
 303  * Parent and child sprlock should be held, and proc_lock must NOT be held.
 304  */
 305 void
 306 dtrace_ptss_fork(struct proc* parent, struct proc* child) {
 307         // The child should not have any pages/entries allocated at this point.
 308         // ASSERT(child->p_dtrace_ptss_pages == NULL);
 309         // ASSERT(child->p_dtrace_ptss_free_list == NULL);
 310
 311         /*
 312          * The parent's sprlock should be held, to protect its pages list
 313          * from changing while the child references it. The child's sprlock
 314          * must also be held, because we are modifying its pages list.
 315          * Finally, to prevent a deadlock with the fasttrap cleanup code,
 316          * neither the parent or child proc_lock should be held.
 317          */
 318         LCK_MTX_ASSERT(&parent->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
 319         LCK_MTX_ASSERT(&parent->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
 320         LCK_MTX_ASSERT(&child->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
 321         LCK_MTX_ASSERT(&child->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
 322
 323         // Get page list from *PARENT*
 324         struct dtrace_ptss_page* temp = parent->p_dtrace_ptss_pages;
 325
 326         while (temp != NULL) {
 327                 // Freeing the page in the *CHILD*
 328                 dtrace_ptss_free_page(child, temp);
 329
 330                 // Do not free the kernel memory, it belong to the parent.
 331                 temp = temp->next;
 332         }
 333 }