1 /* $NetBSD: exec_subr.c,v 1.42 2004/09/17 14:11:25 skrll 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 1.42 2004/09/17 14:11:25 skrll Exp $"); 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/proc.h> 39 #include <sys/malloc.h> 40 #include <sys/vnode.h> 41 #include <sys/filedesc.h> 42 #include <sys/exec.h> 43 #include <sys/mman.h> 44 #include <sys/resourcevar.h> 45 46 #include <uvm/uvm.h> 47 48 /* 49 * XXX cgd 960926: this module should collect simple statistics 50 * (calls, extends, kills). 51 */ 52 53 /* 54 * new_vmcmd(): 55 * create a new vmcmd structure and fill in its fields based 56 * on function call arguments. make sure objects ref'd by 57 * the vmcmd are 'held'. 58 */ 59 60 void 61 new_vmcmd(struct exec_vmcmd_set *evsp, 62 int (*proc)(struct proc * p, struct exec_vmcmd *), 63 u_long len, u_long addr, struct vnode *vp, u_long offset, 64 u_int prot, int flags) 65 { 66 struct exec_vmcmd *vcp; 67 68 if (evsp->evs_used >= evsp->evs_cnt) 69 vmcmdset_extend(evsp); 70 vcp = &evsp->evs_cmds[evsp->evs_used++]; 71 vcp->ev_proc = proc; 72 vcp->ev_len = len; 73 vcp->ev_addr = addr; 74 if ((vcp->ev_vp = vp) != NULL) 75 vref(vp); 76 vcp->ev_offset = offset; 77 vcp->ev_prot = prot; 78 vcp->ev_flags = flags; 79 } 80 81 void 82 vmcmdset_extend(struct exec_vmcmd_set *evsp) 83 { 84 struct exec_vmcmd *nvcp; 85 u_int ocnt; 86 87 #ifdef DIAGNOSTIC 88 if (evsp->evs_used < evsp->evs_cnt) 89 panic("vmcmdset_extend: not necessary"); 90 #endif 91 92 /* figure out number of entries in new set */ 93 ocnt = evsp->evs_cnt; 94 evsp->evs_cnt += ocnt ? ocnt : EXEC_DEFAULT_VMCMD_SETSIZE; 95 96 /* allocate it */ 97 nvcp = malloc(evsp->evs_cnt * sizeof(struct exec_vmcmd), 98 M_EXEC, M_WAITOK); 99 100 /* free the old struct, if there was one, and record the new one */ 101 if (ocnt) { 102 memcpy(nvcp, evsp->evs_cmds, 103 (ocnt * sizeof(struct exec_vmcmd))); 104 free(evsp->evs_cmds, M_EXEC); 105 } 106 evsp->evs_cmds = nvcp; 107 } 108 109 void 110 kill_vmcmds(struct exec_vmcmd_set *evsp) 111 { 112 struct exec_vmcmd *vcp; 113 u_int i; 114 115 if (evsp->evs_cnt == 0) 116 return; 117 118 for (i = 0; i < evsp->evs_used; i++) { 119 vcp = &evsp->evs_cmds[i]; 120 if (vcp->ev_vp != NULL) 121 vrele(vcp->ev_vp); 122 } 123 evsp->evs_used = evsp->evs_cnt = 0; 124 free(evsp->evs_cmds, M_EXEC); 125 } 126 127 /* 128 * vmcmd_map_pagedvn(): 129 * handle vmcmd which specifies that a vnode should be mmap'd. 130 * appropriate for handling demand-paged text and data segments. 131 */ 132 133 int 134 vmcmd_map_pagedvn(struct proc *p, struct exec_vmcmd *cmd) 135 { 136 struct uvm_object *uobj; 137 int error; 138 139 KASSERT(cmd->ev_vp->v_flag & VTEXT); 140 141 /* 142 * map the vnode in using uvm_map. 143 */ 144 145 if (cmd->ev_len == 0) 146 return(0); 147 if (cmd->ev_offset & PAGE_MASK) 148 return(EINVAL); 149 if (cmd->ev_addr & PAGE_MASK) 150 return(EINVAL); 151 if (cmd->ev_len & PAGE_MASK) 152 return(EINVAL); 153 154 /* 155 * first, attach to the object 156 */ 157 158 uobj = uvn_attach(cmd->ev_vp, VM_PROT_READ|VM_PROT_EXECUTE); 159 if (uobj == NULL) 160 return(ENOMEM); 161 VREF(cmd->ev_vp); 162 163 /* 164 * do the map 165 */ 166 167 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, cmd->ev_len, 168 uobj, cmd->ev_offset, 0, 169 UVM_MAPFLAG(cmd->ev_prot, VM_PROT_ALL, UVM_INH_COPY, 170 UVM_ADV_NORMAL, UVM_FLAG_COPYONW|UVM_FLAG_FIXED)); 171 if (error) { 172 uobj->pgops->pgo_detach(uobj); 173 } 174 return error; 175 } 176 177 /* 178 * vmcmd_map_readvn(): 179 * handle vmcmd which specifies that a vnode should be read from. 180 * appropriate for non-demand-paged text/data segments, i.e. impure 181 * objects (a la OMAGIC and NMAGIC). 182 */ 183 int 184 vmcmd_map_readvn(struct proc *p, struct exec_vmcmd *cmd) 185 { 186 int error; 187 long diff; 188 189 if (cmd->ev_len == 0) 190 return 0; 191 192 diff = cmd->ev_addr - trunc_page(cmd->ev_addr); 193 cmd->ev_addr -= diff; /* required by uvm_map */ 194 cmd->ev_offset -= diff; 195 cmd->ev_len += diff; 196 197 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, 198 round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0, 199 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_COPY, 200 UVM_ADV_NORMAL, 201 UVM_FLAG_FIXED|UVM_FLAG_OVERLAY|UVM_FLAG_COPYONW)); 202 203 if (error) 204 return error; 205 206 return vmcmd_readvn(p, cmd); 207 } 208 209 int 210 vmcmd_readvn(struct proc *p, struct exec_vmcmd *cmd) 211 { 212 int error; 213 214 error = vn_rdwr(UIO_READ, cmd->ev_vp, (caddr_t)cmd->ev_addr, 215 cmd->ev_len, cmd->ev_offset, UIO_USERSPACE, IO_UNIT, 216 p->p_ucred, NULL, p); 217 if (error) 218 return error; 219 220 #ifdef PMAP_NEED_PROCWR 221 /* 222 * we had to write the process, make sure the pages are synched 223 * with the instruction cache. 224 */ 225 if (cmd->ev_prot & VM_PROT_EXECUTE) 226 pmap_procwr(p, cmd->ev_addr, cmd->ev_len); 227 #endif 228 229 if (cmd->ev_prot != (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE)) { 230 231 /* 232 * we had to map in the area at PROT_ALL so that vn_rdwr() 233 * could write to it. however, the caller seems to want 234 * it mapped read-only, so now we are going to have to call 235 * uvm_map_protect() to fix up the protection. ICK. 236 */ 237 238 return uvm_map_protect(&p->p_vmspace->vm_map, 239 trunc_page(cmd->ev_addr), 240 round_page(cmd->ev_addr + cmd->ev_len), 241 cmd->ev_prot, FALSE); 242 } 243 return 0; 244 } 245 246 /* 247 * vmcmd_map_zero(): 248 * handle vmcmd which specifies a zero-filled address space region. The 249 * address range must be first allocated, then protected appropriately. 250 */ 251 252 int 253 vmcmd_map_zero(struct proc *p, struct exec_vmcmd *cmd) 254 { 255 int error; 256 long diff; 257 258 diff = cmd->ev_addr - trunc_page(cmd->ev_addr); 259 cmd->ev_addr -= diff; /* required by uvm_map */ 260 cmd->ev_len += diff; 261 262 error = uvm_map(&p->p_vmspace->vm_map, &cmd->ev_addr, 263 round_page(cmd->ev_len), NULL, UVM_UNKNOWN_OFFSET, 0, 264 UVM_MAPFLAG(cmd->ev_prot, UVM_PROT_ALL, UVM_INH_COPY, 265 UVM_ADV_NORMAL, 266 UVM_FLAG_FIXED|UVM_FLAG_COPYONW)); 267 return error; 268 } 269 270 /* 271 * exec_read_from(): 272 * 273 * Read from vnode into buffer at offset. 274 */ 275 int 276 exec_read_from(struct proc *p, struct vnode *vp, u_long off, void *buf, 277 size_t size) 278 { 279 int error; 280 size_t resid; 281 282 if ((error = vn_rdwr(UIO_READ, vp, buf, size, off, UIO_SYSSPACE, 283 0, p->p_ucred, &resid, NULL)) != 0) 284 return error; 285 /* 286 * See if we got all of it 287 */ 288 if (resid != 0) 289 return ENOEXEC; 290 return 0; 291 } 292 293 /* 294 * exec_setup_stack(): Set up the stack segment for an elf 295 * executable. 296 * 297 * Note that the ep_ssize parameter must be set to be the current stack 298 * limit; this is adjusted in the body of execve() to yield the 299 * appropriate stack segment usage once the argument length is 300 * calculated. 301 * 302 * This function returns an int for uniformity with other (future) formats' 303 * stack setup functions. They might have errors to return. 304 */ 305 306 int 307 exec_setup_stack(struct proc *p, struct exec_package *epp) 308 { 309 u_long max_stack_size; 310 u_long access_linear_min, access_size; 311 u_long noaccess_linear_min, noaccess_size; 312 313 #ifndef USRSTACK32 314 #define USRSTACK32 (0x00000000ffffffffL&~PGOFSET) 315 #endif 316 317 if (epp->ep_flags & EXEC_32) { 318 epp->ep_minsaddr = USRSTACK32; 319 max_stack_size = MAXSSIZ; 320 } else { 321 epp->ep_minsaddr = USRSTACK; 322 max_stack_size = MAXSSIZ; 323 } 324 epp->ep_maxsaddr = (u_long)STACK_GROW(epp->ep_minsaddr, 325 max_stack_size); 326 epp->ep_ssize = p->p_rlimit[RLIMIT_STACK].rlim_cur; 327 328 /* 329 * set up commands for stack. note that this takes *two*, one to 330 * map the part of the stack which we can access, and one to map 331 * the part which we can't. 332 * 333 * arguably, it could be made into one, but that would require the 334 * addition of another mapping proc, which is unnecessary 335 */ 336 access_size = epp->ep_ssize; 337 access_linear_min = (u_long)STACK_ALLOC(epp->ep_minsaddr, access_size); 338 noaccess_size = max_stack_size - access_size; 339 noaccess_linear_min = (u_long)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr, 340 access_size), noaccess_size); 341 if (noaccess_size > 0) { 342 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, noaccess_size, 343 noaccess_linear_min, NULL, 0, VM_PROT_NONE); 344 } 345 KASSERT(access_size > 0); 346 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, access_size, 347 access_linear_min, NULL, 0, VM_PROT_READ | VM_PROT_WRITE); 348 349 return 0; 350 } 351