1 /* $OpenBSD: exec_subr.c,v 1.66 2023/03/19 20:32:13 kettenis Exp $ */ 2 /* $NetBSD: exec_subr.c,v 1.9 1994/12/04 03:10:42 mycroft Exp $ */ 3 4 /* 5 * Copyright (c) 1993, 1994 Christopher G. Demetriou 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by Christopher G. Demetriou. 19 * 4. The name of the author may not be used to endorse or promote products 20 * derived from this software without specific prior written permission 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/proc.h> 37 #include <sys/malloc.h> 38 #include <sys/vnode.h> 39 #include <sys/exec.h> 40 #include <sys/mman.h> 41 #include <sys/resourcevar.h> 42 43 #include <uvm/uvm_extern.h> 44 45 #ifdef DEBUG 46 /* 47 * new_vmcmd(): 48 * create a new vmcmd structure and fill in its fields based 49 * on function call arguments. make sure objects ref'd by 50 * the vmcmd are 'held'. 51 * 52 * If not debugging, this is a macro, so it's expanded inline. 53 */ 54 55 void 56 new_vmcmd(struct exec_vmcmd_set *evsp, 57 int (*proc)(struct proc *, struct exec_vmcmd *), u_long len, u_long addr, 58 struct vnode *vp, u_long offset, u_int prot, int flags) 59 { 60 struct exec_vmcmd *vcp; 61 62 if (evsp->evs_used >= evsp->evs_cnt) 63 vmcmdset_extend(evsp); 64 vcp = &evsp->evs_cmds[evsp->evs_used++]; 65 vcp->ev_proc = proc; 66 vcp->ev_len = len; 67 vcp->ev_addr = addr; 68 if ((vcp->ev_vp = vp) != NULL) 69 vref(vp); 70 vcp->ev_offset = offset; 71 vcp->ev_prot = prot; 72 vcp->ev_flags = flags; 73 } 74 #endif /* DEBUG */ 75 76 void 77 vmcmdset_extend(struct exec_vmcmd_set *evsp) 78 { 79 struct exec_vmcmd *nvcp; 80 u_int ocnt; 81 82 #ifdef DIAGNOSTIC 83 if (evsp->evs_used < evsp->evs_cnt) 84 panic("vmcmdset_extend: not necessary"); 85 #endif 86 87 ocnt = evsp->evs_cnt; 88 KASSERT(ocnt > 0); 89 /* figure out number of entries in new set */ 90 evsp->evs_cnt += ocnt; 91 92 /* reallocate the command set */ 93 nvcp = mallocarray(evsp->evs_cnt, sizeof(*nvcp), M_EXEC, 94 M_WAITOK); 95 memcpy(nvcp, evsp->evs_cmds, ocnt * sizeof(*nvcp)); 96 if (evsp->evs_cmds != evsp->evs_start) 97 free(evsp->evs_cmds, M_EXEC, ocnt * sizeof(*nvcp)); 98 evsp->evs_cmds = nvcp; 99 } 100 101 void 102 kill_vmcmds(struct exec_vmcmd_set *evsp) 103 { 104 struct exec_vmcmd *vcp; 105 int i; 106 107 for (i = 0; i < evsp->evs_used; i++) { 108 vcp = &evsp->evs_cmds[i]; 109 if (vcp->ev_vp != NULLVP) 110 vrele(vcp->ev_vp); 111 } 112 113 /* 114 * Free old vmcmds and reset the array. 115 */ 116 evsp->evs_used = 0; 117 if (evsp->evs_cmds != evsp->evs_start) 118 free(evsp->evs_cmds, M_EXEC, 119 evsp->evs_cnt * sizeof(struct exec_vmcmd)); 120 evsp->evs_cmds = evsp->evs_start; 121 evsp->evs_cnt = EXEC_DEFAULT_VMCMD_SETSIZE; 122 } 123 124 int 125 exec_process_vmcmds(struct proc *p, struct exec_package *epp) 126 { 127 struct exec_vmcmd *base_vc = NULL; 128 int error = 0; 129 int i; 130 131 for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) { 132 struct exec_vmcmd *vcp; 133 134 vcp = &epp->ep_vmcmds.evs_cmds[i]; 135 136 if (vcp->ev_flags & VMCMD_RELATIVE) { 137 #ifdef DIAGNOSTIC 138 if (base_vc == NULL) 139 panic("exec_process_vmcmds: RELATIVE no base"); 140 #endif 141 vcp->ev_addr += base_vc->ev_addr; 142 } 143 error = (*vcp->ev_proc)(p, vcp); 144 if (vcp->ev_flags & VMCMD_BASE) { 145 base_vc = vcp; 146 } 147 } 148 149 kill_vmcmds(&epp->ep_vmcmds); 150 151 return (error); 152 } 153 154 /* 155 * vmcmd_map_pagedvn(): 156 * handle vmcmd which specifies that a vnode should be mmap'd. 157 * appropriate for handling demand-paged text and data segments. 158 */ 159 160 int 161 vmcmd_map_pagedvn(struct proc *p, struct exec_vmcmd *cmd) 162 { 163 /* 164 * note that if you're going to map part of a process as being 165 * paged from a vnode, that vnode had damn well better be marked as 166 * VTEXT. that's handled in the routine which sets up the vmcmd to 167 * call this routine. 168 */ 169 struct uvm_object *uobj; 170 unsigned int flags = UVM_FLAG_COPYONW | UVM_FLAG_FIXED; 171 int error; 172 173 /* 174 * map the vnode in using uvm_map. 175 */ 176 177 if (cmd->ev_len == 0) 178 return (0); 179 if (cmd->ev_offset & PAGE_MASK) 180 return (EINVAL); 181 if (cmd->ev_addr & PAGE_MASK) 182 return (EINVAL); 183 if (cmd->ev_len & PAGE_MASK) 184 return (EINVAL); 185 186 /* 187 * first, attach to the object 188 */ 189 190 uobj = uvn_attach(cmd->ev_vp, PROT_READ | PROT_EXEC); 191 if (uobj == NULL) 192 return (ENOMEM); 193 194 /* 195 * do the map 196 */ 197 if ((cmd->ev_flags & VMCMD_SYSCALL) && (cmd->ev_prot & PROT_EXEC)) 198 flags |= UVM_FLAG_SYSCALL; 199 200 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, cmd->ev_len, 201 uobj, cmd->ev_offset, 0, 202 UVM_MAPFLAG(cmd->ev_prot, PROT_MASK, MAP_INHERIT_COPY, 203 MADV_NORMAL, flags)); 204 205 /* 206 * check for error 207 */ 208 209 if (error) { 210 /* 211 * error: detach from object 212 */ 213 uobj->pgops->pgo_detach(uobj); 214 } else { 215 if (cmd->ev_flags & VMCMD_IMMUTABLE) 216 uvm_map_immutable(&p->p_vmspace->vm_map, cmd->ev_addr, 217 round_page(cmd->ev_addr + cmd->ev_len), 1); 218 #ifdef PMAP_CHECK_COPYIN 219 if (PMAP_CHECK_COPYIN && 220 ((flags & UVM_FLAG_SYSCALL) || 221 ((cmd->ev_flags & VMCMD_IMMUTABLE) && (cmd->ev_prot & PROT_EXEC)))) 222 uvm_map_check_copyin_add(&p->p_vmspace->vm_map, 223 cmd->ev_addr, round_page(cmd->ev_addr + cmd->ev_len)); 224 #endif 225 } 226 227 return (error); 228 } 229 230 /* 231 * vmcmd_map_readvn(): 232 * handle vmcmd which specifies that a vnode should be read from. 233 * appropriate for non-demand-paged text/data segments, i.e. impure 234 * objects (a la OMAGIC and NMAGIC). 235 */ 236 237 int 238 vmcmd_map_readvn(struct proc *p, struct exec_vmcmd *cmd) 239 { 240 int error; 241 vm_prot_t prot; 242 243 if (cmd->ev_len == 0) 244 return (0); 245 246 prot = cmd->ev_prot; 247 248 KASSERT((cmd->ev_addr & PAGE_MASK) == 0); 249 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, 250 round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0, 251 UVM_MAPFLAG(prot | PROT_WRITE, PROT_MASK, MAP_INHERIT_COPY, 252 MADV_NORMAL, UVM_FLAG_FIXED|UVM_FLAG_OVERLAY|UVM_FLAG_COPYONW)); 253 254 if (error) 255 return (error); 256 257 error = vn_rdwr(UIO_READ, cmd->ev_vp, (caddr_t)cmd->ev_addr, 258 cmd->ev_len, cmd->ev_offset, UIO_USERSPACE, IO_UNIT, 259 p->p_ucred, NULL, p); 260 if (error) 261 return (error); 262 263 if ((prot & PROT_WRITE) == 0) { 264 /* 265 * we had to map in the area at PROT_WRITE so that vn_rdwr() 266 * could write to it. however, the caller seems to want 267 * it mapped read-only, so now we are going to have to call 268 * uvm_map_protect() to fix up the protection. ICK. 269 */ 270 error = (uvm_map_protect(&p->p_vmspace->vm_map, 271 cmd->ev_addr, round_page(cmd->ev_len), 272 prot, 0, FALSE, TRUE)); 273 } 274 if (error == 0) { 275 if (cmd->ev_flags & VMCMD_IMMUTABLE) 276 uvm_map_immutable(&p->p_vmspace->vm_map, cmd->ev_addr, 277 round_page(cmd->ev_addr + cmd->ev_len), 1); 278 } 279 return (error); 280 } 281 282 /* 283 * vmcmd_map_zero(): 284 * handle vmcmd which specifies a zero-filled address space region. 285 */ 286 287 int 288 vmcmd_map_zero(struct proc *p, struct exec_vmcmd *cmd) 289 { 290 int error; 291 292 if (cmd->ev_len == 0) 293 return (0); 294 295 KASSERT((cmd->ev_addr & PAGE_MASK) == 0); 296 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, 297 round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0, 298 UVM_MAPFLAG(cmd->ev_prot, PROT_MASK, MAP_INHERIT_COPY, 299 MADV_NORMAL, UVM_FLAG_FIXED|UVM_FLAG_COPYONW | 300 (cmd->ev_flags & VMCMD_STACK ? UVM_FLAG_STACK : 0))); 301 if (cmd->ev_flags & VMCMD_IMMUTABLE) 302 uvm_map_immutable(&p->p_vmspace->vm_map, cmd->ev_addr, 303 round_page(cmd->ev_addr + cmd->ev_len), 1); 304 return error; 305 } 306 307 /* 308 * vmcmd_mutable(): 309 * handle vmcmd which changes an address space region.back to mutable 310 */ 311 312 int 313 vmcmd_mutable(struct proc *p, struct exec_vmcmd *cmd) 314 { 315 if (cmd->ev_len == 0) 316 return (0); 317 318 /* ev_addr, ev_len may be misaligned, so maximize the region */ 319 uvm_map_immutable(&p->p_vmspace->vm_map, trunc_page(cmd->ev_addr), 320 round_page(cmd->ev_addr + cmd->ev_len), 0); 321 return 0; 322 } 323 324 /* 325 * vmcmd_randomize(): 326 * handle vmcmd which specifies a randomized address space region. 327 */ 328 #define RANDOMIZE_CTX_THRESHOLD 512 329 int 330 vmcmd_randomize(struct proc *p, struct exec_vmcmd *cmd) 331 { 332 int error; 333 struct arc4random_ctx *ctx; 334 char *buf; 335 size_t sublen, off = 0; 336 size_t len = cmd->ev_len; 337 338 if (len == 0) 339 return (0); 340 if (len > ELF_RANDOMIZE_LIMIT) 341 return (EINVAL); 342 343 buf = malloc(PAGE_SIZE, M_TEMP, M_WAITOK); 344 if (len < RANDOMIZE_CTX_THRESHOLD) { 345 arc4random_buf(buf, len); 346 error = copyout(buf, (void *)cmd->ev_addr, len); 347 explicit_bzero(buf, len); 348 } else { 349 ctx = arc4random_ctx_new(); 350 do { 351 sublen = MIN(len, PAGE_SIZE); 352 arc4random_ctx_buf(ctx, buf, sublen); 353 error = copyout(buf, (void *)cmd->ev_addr + off, sublen); 354 if (error) 355 break; 356 off += sublen; 357 len -= sublen; 358 sched_pause(yield); 359 } while (len); 360 arc4random_ctx_free(ctx); 361 explicit_bzero(buf, PAGE_SIZE); 362 } 363 free(buf, M_TEMP, PAGE_SIZE); 364 return (error); 365 } 366 367 #ifndef MAXSSIZ_GUARD 368 #define MAXSSIZ_GUARD (1024 * 1024) 369 #endif 370 371 /* 372 * exec_setup_stack(): Set up the stack segment for an executable. 373 * 374 * Note that the ep_ssize parameter must be set to be the current stack 375 * limit; this is adjusted in the body of execve() to yield the 376 * appropriate stack segment usage once the argument length is 377 * calculated. 378 * 379 * This function returns an int for uniformity with other (future) formats' 380 * stack setup functions. They might have errors to return. 381 */ 382 383 int 384 exec_setup_stack(struct proc *p, struct exec_package *epp) 385 { 386 vsize_t dist = 0; 387 388 #ifdef MACHINE_STACK_GROWS_UP 389 epp->ep_maxsaddr = USRSTACK; 390 epp->ep_minsaddr = USRSTACK + MAXSSIZ; 391 #else 392 epp->ep_maxsaddr = USRSTACK - MAXSSIZ - MAXSSIZ_GUARD; 393 epp->ep_minsaddr = USRSTACK; 394 #endif 395 epp->ep_ssize = round_page(lim_cur(RLIMIT_STACK)); 396 397 #ifdef VM_MIN_STACK_ADDRESS 398 dist = USRSTACK - MAXSSIZ - MAXSSIZ_GUARD - VM_MIN_STACK_ADDRESS; 399 if (dist >> PAGE_SHIFT > 0xffffffff) 400 dist = (vsize_t)arc4random() << PAGE_SHIFT; 401 else 402 dist = (vsize_t)arc4random_uniform(dist >> PAGE_SHIFT) << PAGE_SHIFT; 403 #else 404 if (stackgap_random != 0) { 405 dist = arc4random() & (stackgap_random - 1); 406 dist = trunc_page(dist); 407 } 408 #endif 409 410 #ifdef MACHINE_STACK_GROWS_UP 411 epp->ep_maxsaddr += dist; 412 epp->ep_minsaddr += dist; 413 #else 414 epp->ep_maxsaddr -= dist; 415 epp->ep_minsaddr -= dist; 416 #endif 417 418 /* 419 * set up commands for stack. note that this takes *two*, one to 420 * map the part of the stack which we can access, and one to map 421 * the part which we can't. 422 * 423 * arguably, it could be made into one, but that would require the 424 * addition of another mapping proc, which is unnecessary 425 * 426 * note that in memory, things assumed to be: 0 ....... ep_maxsaddr 427 * <stack> ep_minsaddr 428 */ 429 #ifdef MACHINE_STACK_GROWS_UP 430 NEW_VMCMD2(&epp->ep_vmcmds, vmcmd_map_zero, 431 ((epp->ep_minsaddr - epp->ep_ssize) - epp->ep_maxsaddr), 432 epp->ep_maxsaddr + epp->ep_ssize, 433 NULLVP, 0, PROT_NONE, VMCMD_IMMUTABLE); 434 NEW_VMCMD2(&epp->ep_vmcmds, vmcmd_map_zero, epp->ep_ssize, 435 epp->ep_maxsaddr, 436 NULLVP, 0, PROT_READ | PROT_WRITE, VMCMD_STACK | VMCMD_IMMUTABLE); 437 #else 438 NEW_VMCMD2(&epp->ep_vmcmds, vmcmd_map_zero, 439 ((epp->ep_minsaddr - epp->ep_ssize) - epp->ep_maxsaddr), 440 epp->ep_maxsaddr, 441 NULLVP, 0, PROT_NONE, VMCMD_IMMUTABLE); 442 NEW_VMCMD2(&epp->ep_vmcmds, vmcmd_map_zero, epp->ep_ssize, 443 (epp->ep_minsaddr - epp->ep_ssize), 444 NULLVP, 0, PROT_READ | PROT_WRITE, VMCMD_STACK | VMCMD_IMMUTABLE); 445 #endif 446 447 return (0); 448 } 449