1 /* $NetBSD: exec_subr.c,v 1.62 2009/03/29 01:02:50 mrg Exp $ */ 2 3 /* 4 * Copyright (c) 1993, 1994, 1996 Christopher G. Demetriou 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by Christopher G. Demetriou. 18 * 4. The name of the author may not be used to endorse or promote products 19 * derived from this software without specific prior written permission 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 22 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 23 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 24 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 26 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 30 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 #include <sys/cdefs.h> 34 __KERNEL_RCSID(0, "$NetBSD: exec_subr.c,v 1.62 2009/03/29 01:02:50 mrg Exp $"); 35 36 #include "opt_pax.h" 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/proc.h> 41 #include <sys/kmem.h> 42 #include <sys/vnode.h> 43 #include <sys/filedesc.h> 44 #include <sys/exec.h> 45 #include <sys/mman.h> 46 #include <sys/resourcevar.h> 47 #include <sys/device.h> 48 49 #ifdef PAX_MPROTECT 50 #include <sys/pax.h> 51 #endif /* PAX_MPROTECT */ 52 53 #include <uvm/uvm.h> 54 55 #define VMCMD_EVCNT_DECL(name) \ 56 static struct evcnt vmcmd_ev_##name = \ 57 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, "vmcmd", #name); \ 58 EVCNT_ATTACH_STATIC(vmcmd_ev_##name) 59 60 #define VMCMD_EVCNT_INCR(name) \ 61 vmcmd_ev_##name.ev_count++ 62 63 VMCMD_EVCNT_DECL(calls); 64 VMCMD_EVCNT_DECL(extends); 65 VMCMD_EVCNT_DECL(kills); 66 67 /* 68 * new_vmcmd(): 69 * create a new vmcmd structure and fill in its fields based 70 * on function call arguments. make sure objects ref'd by 71 * the vmcmd are 'held'. 72 */ 73 74 void 75 new_vmcmd(struct exec_vmcmd_set *evsp, 76 int (*proc)(struct lwp * l, struct exec_vmcmd *), 77 u_long len, u_long addr, struct vnode *vp, u_long offset, 78 u_int prot, int flags) 79 { 80 struct exec_vmcmd *vcp; 81 82 VMCMD_EVCNT_INCR(calls); 83 84 if (evsp->evs_used >= evsp->evs_cnt) 85 vmcmdset_extend(evsp); 86 vcp = &evsp->evs_cmds[evsp->evs_used++]; 87 vcp->ev_proc = proc; 88 vcp->ev_len = len; 89 vcp->ev_addr = addr; 90 if ((vcp->ev_vp = vp) != NULL) 91 vref(vp); 92 vcp->ev_offset = offset; 93 vcp->ev_prot = prot; 94 vcp->ev_flags = flags; 95 } 96 97 void 98 vmcmdset_extend(struct exec_vmcmd_set *evsp) 99 { 100 struct exec_vmcmd *nvcp; 101 u_int ocnt; 102 103 #ifdef DIAGNOSTIC 104 if (evsp->evs_used < evsp->evs_cnt) 105 panic("vmcmdset_extend: not necessary"); 106 #endif 107 108 /* figure out number of entries in new set */ 109 if ((ocnt = evsp->evs_cnt) != 0) { 110 evsp->evs_cnt += ocnt; 111 VMCMD_EVCNT_INCR(extends); 112 } else 113 evsp->evs_cnt = EXEC_DEFAULT_VMCMD_SETSIZE; 114 115 /* allocate it */ 116 nvcp = kmem_alloc(evsp->evs_cnt * sizeof(struct exec_vmcmd), KM_SLEEP); 117 118 /* free the old struct, if there was one, and record the new one */ 119 if (ocnt) { 120 memcpy(nvcp, evsp->evs_cmds, 121 (ocnt * sizeof(struct exec_vmcmd))); 122 kmem_free(evsp->evs_cmds, ocnt * sizeof(struct exec_vmcmd)); 123 } 124 evsp->evs_cmds = nvcp; 125 } 126 127 void 128 kill_vmcmds(struct exec_vmcmd_set *evsp) 129 { 130 struct exec_vmcmd *vcp; 131 u_int i; 132 133 VMCMD_EVCNT_INCR(kills); 134 135 if (evsp->evs_cnt == 0) 136 return; 137 138 for (i = 0; i < evsp->evs_used; i++) { 139 vcp = &evsp->evs_cmds[i]; 140 if (vcp->ev_vp != NULL) 141 vrele(vcp->ev_vp); 142 } 143 kmem_free(evsp->evs_cmds, evsp->evs_cnt * sizeof(struct exec_vmcmd)); 144 evsp->evs_used = evsp->evs_cnt = 0; 145 } 146 147 /* 148 * vmcmd_map_pagedvn(): 149 * handle vmcmd which specifies that a vnode should be mmap'd. 150 * appropriate for handling demand-paged text and data segments. 151 */ 152 153 int 154 vmcmd_map_pagedvn(struct lwp *l, struct exec_vmcmd *cmd) 155 { 156 struct uvm_object *uobj; 157 struct vnode *vp = cmd->ev_vp; 158 struct proc *p = l->l_proc; 159 int error; 160 vm_prot_t prot, maxprot; 161 162 KASSERT(vp->v_iflag & VI_TEXT); 163 164 /* 165 * map the vnode in using uvm_map. 166 */ 167 168 if (cmd->ev_len == 0) 169 return(0); 170 if (cmd->ev_offset & PAGE_MASK) 171 return(EINVAL); 172 if (cmd->ev_addr & PAGE_MASK) 173 return(EINVAL); 174 if (cmd->ev_len & PAGE_MASK) 175 return(EINVAL); 176 177 prot = cmd->ev_prot; 178 maxprot = UVM_PROT_ALL; 179 #ifdef PAX_MPROTECT 180 pax_mprotect(l, &prot, &maxprot); 181 #endif /* PAX_MPROTECT */ 182 183 /* 184 * check the file system's opinion about mmapping the file 185 */ 186 187 error = VOP_MMAP(vp, prot, l->l_cred); 188 if (error) 189 return error; 190 191 if ((vp->v_vflag & VV_MAPPED) == 0) { 192 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 193 vp->v_vflag |= VV_MAPPED; 194 VOP_UNLOCK(vp, 0); 195 } 196 197 /* 198 * do the map, reference the object for this map entry 199 */ 200 uobj = &vp->v_uobj; 201 vref(vp); 202 203 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, cmd->ev_len, 204 uobj, cmd->ev_offset, 0, 205 UVM_MAPFLAG(prot, maxprot, UVM_INH_COPY, 206 UVM_ADV_NORMAL, UVM_FLAG_COPYONW|UVM_FLAG_FIXED)); 207 if (error) { 208 uobj->pgops->pgo_detach(uobj); 209 } 210 return error; 211 } 212 213 /* 214 * vmcmd_map_readvn(): 215 * handle vmcmd which specifies that a vnode should be read from. 216 * appropriate for non-demand-paged text/data segments, i.e. impure 217 * objects (a la OMAGIC and NMAGIC). 218 */ 219 int 220 vmcmd_map_readvn(struct lwp *l, struct exec_vmcmd *cmd) 221 { 222 struct proc *p = l->l_proc; 223 int error; 224 long diff; 225 226 if (cmd->ev_len == 0) 227 return 0; 228 229 diff = cmd->ev_addr - trunc_page(cmd->ev_addr); 230 cmd->ev_addr -= diff; /* required by uvm_map */ 231 cmd->ev_offset -= diff; 232 cmd->ev_len += diff; 233 234 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, 235 round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0, 236 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_COPY, 237 UVM_ADV_NORMAL, 238 UVM_FLAG_FIXED|UVM_FLAG_OVERLAY|UVM_FLAG_COPYONW)); 239 240 if (error) 241 return error; 242 243 return vmcmd_readvn(l, cmd); 244 } 245 246 int 247 vmcmd_readvn(struct lwp *l, struct exec_vmcmd *cmd) 248 { 249 struct proc *p = l->l_proc; 250 int error; 251 vm_prot_t prot, maxprot; 252 253 error = vn_rdwr(UIO_READ, cmd->ev_vp, (void *)cmd->ev_addr, 254 cmd->ev_len, cmd->ev_offset, UIO_USERSPACE, IO_UNIT, 255 l->l_cred, NULL, l); 256 if (error) 257 return error; 258 259 prot = cmd->ev_prot; 260 maxprot = VM_PROT_ALL; 261 #ifdef PAX_MPROTECT 262 pax_mprotect(l, &prot, &maxprot); 263 #endif /* PAX_MPROTECT */ 264 265 #ifdef PMAP_NEED_PROCWR 266 /* 267 * we had to write the process, make sure the pages are synched 268 * with the instruction cache. 269 */ 270 if (prot & VM_PROT_EXECUTE) 271 pmap_procwr(p, cmd->ev_addr, cmd->ev_len); 272 #endif 273 274 /* 275 * we had to map in the area at PROT_ALL so that vn_rdwr() 276 * could write to it. however, the caller seems to want 277 * it mapped read-only, so now we are going to have to call 278 * uvm_map_protect() to fix up the protection. ICK. 279 */ 280 if (maxprot != VM_PROT_ALL) { 281 error = uvm_map_protect(&p->p_vmspace->vm_map, 282 trunc_page(cmd->ev_addr), 283 round_page(cmd->ev_addr + cmd->ev_len), 284 maxprot, true); 285 if (error) 286 return (error); 287 } 288 289 if (prot != maxprot) { 290 error = uvm_map_protect(&p->p_vmspace->vm_map, 291 trunc_page(cmd->ev_addr), 292 round_page(cmd->ev_addr + cmd->ev_len), 293 prot, false); 294 if (error) 295 return (error); 296 } 297 298 return 0; 299 } 300 301 /* 302 * vmcmd_map_zero(): 303 * handle vmcmd which specifies a zero-filled address space region. The 304 * address range must be first allocated, then protected appropriately. 305 */ 306 307 int 308 vmcmd_map_zero(struct lwp *l, struct exec_vmcmd *cmd) 309 { 310 struct proc *p = l->l_proc; 311 int error; 312 long diff; 313 vm_prot_t prot, maxprot; 314 315 diff = cmd->ev_addr - trunc_page(cmd->ev_addr); 316 cmd->ev_addr -= diff; /* required by uvm_map */ 317 cmd->ev_len += diff; 318 319 prot = cmd->ev_prot; 320 maxprot = UVM_PROT_ALL; 321 #ifdef PAX_MPROTECT 322 pax_mprotect(l, &prot, &maxprot); 323 #endif /* PAX_MPROTECT */ 324 325 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, 326 round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0, 327 UVM_MAPFLAG(prot, maxprot, UVM_INH_COPY, 328 UVM_ADV_NORMAL, 329 UVM_FLAG_FIXED|UVM_FLAG_COPYONW)); 330 if (cmd->ev_flags & VMCMD_STACK) 331 curproc->p_vmspace->vm_issize += atop(round_page(cmd->ev_len)); 332 return error; 333 } 334 335 /* 336 * exec_read_from(): 337 * 338 * Read from vnode into buffer at offset. 339 */ 340 int 341 exec_read_from(struct lwp *l, struct vnode *vp, u_long off, void *bf, 342 size_t size) 343 { 344 int error; 345 size_t resid; 346 347 if ((error = vn_rdwr(UIO_READ, vp, bf, size, off, UIO_SYSSPACE, 348 0, l->l_cred, &resid, NULL)) != 0) 349 return error; 350 /* 351 * See if we got all of it 352 */ 353 if (resid != 0) 354 return ENOEXEC; 355 return 0; 356 } 357 358 /* 359 * exec_setup_stack(): Set up the stack segment for an elf 360 * executable. 361 * 362 * Note that the ep_ssize parameter must be set to be the current stack 363 * limit; this is adjusted in the body of execve() to yield the 364 * appropriate stack segment usage once the argument length is 365 * calculated. 366 * 367 * This function returns an int for uniformity with other (future) formats' 368 * stack setup functions. They might have errors to return. 369 */ 370 371 int 372 exec_setup_stack(struct lwp *l, struct exec_package *epp) 373 { 374 u_long max_stack_size; 375 u_long access_linear_min, access_size; 376 u_long noaccess_linear_min, noaccess_size; 377 378 #ifndef USRSTACK32 379 #define USRSTACK32 (0x00000000ffffffffL&~PGOFSET) 380 #endif 381 382 if (epp->ep_flags & EXEC_32) { 383 epp->ep_minsaddr = USRSTACK32; 384 max_stack_size = MAXSSIZ; 385 } else { 386 epp->ep_minsaddr = USRSTACK; 387 max_stack_size = MAXSSIZ; 388 } 389 390 #ifdef PAX_ASLR 391 pax_aslr_stack(l, epp, &max_stack_size); 392 #endif /* PAX_ASLR */ 393 394 l->l_proc->p_stackbase = epp->ep_minsaddr; 395 396 epp->ep_maxsaddr = (u_long)STACK_GROW(epp->ep_minsaddr, 397 max_stack_size); 398 epp->ep_ssize = l->l_proc->p_rlimit[RLIMIT_STACK].rlim_cur; 399 400 /* 401 * set up commands for stack. note that this takes *two*, one to 402 * map the part of the stack which we can access, and one to map 403 * the part which we can't. 404 * 405 * arguably, it could be made into one, but that would require the 406 * addition of another mapping proc, which is unnecessary 407 */ 408 access_size = epp->ep_ssize; 409 access_linear_min = (u_long)STACK_ALLOC(epp->ep_minsaddr, access_size); 410 noaccess_size = max_stack_size - access_size; 411 noaccess_linear_min = (u_long)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr, 412 access_size), noaccess_size); 413 if (noaccess_size > 0) { 414 NEW_VMCMD2(&epp->ep_vmcmds, vmcmd_map_zero, noaccess_size, 415 noaccess_linear_min, NULL, 0, VM_PROT_NONE, VMCMD_STACK); 416 } 417 KASSERT(access_size > 0); 418 NEW_VMCMD2(&epp->ep_vmcmds, vmcmd_map_zero, access_size, 419 access_linear_min, NULL, 0, VM_PROT_READ | VM_PROT_WRITE, 420 VMCMD_STACK); 421 422 return 0; 423 } 424