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 */ 49struct dtrace_ptss_page_entry* 50dtrace_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 */ 97struct dtrace_ptss_page_entry* 98dtrace_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 * (PR-11138709) A NULL p->p_dtrace_ptss_pages means the entry can 132 * no longer be referenced safely. When found in this state, the chore 133 * of releasing an entry to the free list is ignored. 134 */ 135void 136dtrace_ptss_release_entry(struct proc* p, struct dtrace_ptss_page_entry* e) { 137 if (p && p->p_dtrace_ptss_pages && e) { 138 do { 139 e->next = p->p_dtrace_ptss_free_list; 140 } while (!OSCompareAndSwapPtr((void *)e->next, (void *)e, (void * volatile *)&p->p_dtrace_ptss_free_list)); 141 } 142} 143 144/* 145 * This function allocates a new page in the target process's address space. 146 * 147 * It returns a dtrace_ptss_page that has its entries chained, with the last 148 * entries next field set to NULL. It does not add the page or the entries to 149 * the process's page/entry lists. 150 * 151 * This function does not require that any locks be held when it is invoked. 152 */ 153struct dtrace_ptss_page* 154dtrace_ptss_allocate_page(struct proc* p) 155{ 156 // Allocate the kernel side data 157 struct dtrace_ptss_page* ptss_page = _MALLOC(sizeof(struct dtrace_ptss_page), M_TEMP, M_ZERO | M_WAITOK); 158 if (ptss_page == NULL) 159 return NULL; 160 161 // Now allocate a page in user space and set its protections to allow execute. 162 task_t task = p->task; 163 vm_map_t map = get_task_map_reference(task); 164 165 mach_vm_address_t addr = 0LL; 166 mach_vm_size_t size = PAGE_SIZE; // We need some way to assert that this matches vm_map_round_page() !!! 167 168#if CONFIG_EMBEDDED 169 /* The embedded OS has extra permissions for writable and executable pages. We can't pass in the flags 170 * we need for the correct permissions from mach_vm_allocate, so need to call mach_vm_map directly. */ 171 mach_vm_offset_t map_addr = 0; 172 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); 173 if (kr != KERN_SUCCESS) { 174 goto err; 175 } 176 addr = map_addr; 177#else 178 kern_return_t kr = mach_vm_allocate(map, &addr, size, VM_FLAGS_ANYWHERE); 179 if (kr != KERN_SUCCESS) { 180 goto err; 181 } 182 183 kr = mach_vm_protect(map, addr, size, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 184 if (kr != KERN_SUCCESS) { 185 mach_vm_deallocate(map, addr, size); 186 goto err; 187 } 188#endif 189 190 // Chain the page entries. 191 int i; 192 for (i=0; i<DTRACE_PTSS_ENTRIES_PER_PAGE; i++) { 193 ptss_page->entries[i].addr = addr + (i * DTRACE_PTSS_SCRATCH_SPACE_PER_THREAD); 194 ptss_page->entries[i].next = &ptss_page->entries[i+1]; 195 } 196 197 // The last entry should point to NULL 198 ptss_page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next = NULL; 199 200 vm_map_deallocate(map); 201 202 return ptss_page; 203 204err: 205 _FREE(ptss_page, M_TEMP); 206 207 vm_map_deallocate(map); 208 209 return NULL; 210} 211 212/* 213 * This function frees an existing page in the target process's address space. 214 * 215 * It does not alter any of the process's page/entry lists. 216 * 217 * TODO: Inline in dtrace_ptrace_exec_exit? 218 */ 219void 220dtrace_ptss_free_page(struct proc* p, struct dtrace_ptss_page* ptss_page) 221{ 222 // Grab the task and get a reference to its vm_map 223 task_t task = p->task; 224 vm_map_t map = get_task_map_reference(task); 225 226 mach_vm_address_t addr = ptss_page->entries[0].addr; 227 mach_vm_size_t size = PAGE_SIZE; // We need some way to assert that this matches vm_map_round_page() !!! 228 229 // Silent failures, no point in checking return code. 230 mach_vm_deallocate(map, addr, size); 231 232 vm_map_deallocate(map); 233} 234 235/* 236 * This function assumes that the target process has been 237 * suspended, and the proc_lock & sprlock is held 238 */ 239void 240dtrace_ptss_enable(struct proc* p) { 241 lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED); 242 lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_OWNED); 243 244 struct uthread* uth; 245 /* 246 * XXX There has been a concern raised about holding the proc_lock 247 * while calling dtrace_ptss_claim_entry(), due to the fact 248 * that dtrace_ptss_claim_entry() can potentially malloc. 249 */ 250 TAILQ_FOREACH(uth, &p->p_uthlist, uu_list) { 251 uth->t_dtrace_scratch = dtrace_ptss_claim_entry_locked(p); 252 } 253} 254 255/* 256 * This function is not thread safe. 257 * 258 * It assumes the sprlock is held, and the proc_lock is not. 259 */ 260void 261dtrace_ptss_exec_exit(struct proc* p) { 262 /* 263 * Should hold sprlock to touch the pages list. Must not 264 * hold the proc lock to avoid deadlock. 265 */ 266 lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED); 267 lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED); 268 269 p->p_dtrace_ptss_free_list = NULL; 270 271 struct dtrace_ptss_page* temp = p->p_dtrace_ptss_pages; 272 p->p_dtrace_ptss_pages = NULL; 273 274 while (temp != NULL) { 275 struct dtrace_ptss_page* next = temp->next; 276 277 // Do we need to specifically mach_vm_deallocate the user pages? 278 // This can be called when the process is exiting, I believe the proc's 279 // vm_map_t may already be toast. 280 281 // Must be certain to free the kernel memory! 282 _FREE(temp, M_TEMP); 283 temp = next; 284 } 285} 286 287/* 288 * This function is not thread safe. It is not used for vfork. 289 * 290 * The child proc ptss fields are initialized to NULL at fork time. 291 * Pages allocated in the parent are copied as part of the vm_map copy, though. 292 * We need to deallocate those pages. 293 * 294 * Parent and child sprlock should be held, and proc_lock must NOT be held. 295 */ 296void 297dtrace_ptss_fork(struct proc* parent, struct proc* child) { 298 // The child should not have any pages/entries allocated at this point. 299 // ASSERT(child->p_dtrace_ptss_pages == NULL); 300 // ASSERT(child->p_dtrace_ptss_free_list == NULL); 301 302 /* 303 * The parent's sprlock should be held, to protect its pages list 304 * from changing while the child references it. The child's sprlock 305 * must also be held, because we are modifying its pages list. 306 * Finally, to prevent a deadlock with the fasttrap cleanup code, 307 * neither the parent or child proc_lock should be held. 308 */ 309 lck_mtx_assert(&parent->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED); 310 lck_mtx_assert(&parent->p_mlock, LCK_MTX_ASSERT_NOTOWNED); 311 lck_mtx_assert(&child->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED); 312 lck_mtx_assert(&child->p_mlock, LCK_MTX_ASSERT_NOTOWNED); 313 314 // Get page list from *PARENT* 315 struct dtrace_ptss_page* temp = parent->p_dtrace_ptss_pages; 316 317 while (temp != NULL) { 318 // Freeing the page in the *CHILD* 319 dtrace_ptss_free_page(child, temp); 320 321 // Do not free the kernel memory, it belong to the parent. 322 temp = temp->next; 323 } 324} 325