xref: /freebsd/sys/compat/linux/linux_mmap.c (revision 907b59d76938e654f0d040a888e8dfca3de1e222)
1 /*-
2  * Copyright (c) 2004 Tim J. Robbins
3  * Copyright (c) 2002 Doug Rabson
4  * Copyright (c) 2000 Marcel Moolenaar
5  * Copyright (c) 1994-1995 Søren Schmidt
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  *    in this position and unchanged.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. The name of the author may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  *
31  * $FreeBSD$
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/capsicum.h>
38 #include <sys/file.h>
39 #include <sys/imgact.h>
40 #include <sys/ktr.h>
41 #include <sys/mman.h>
42 #include <sys/proc.h>
43 #include <sys/resourcevar.h>
44 #include <sys/sysent.h>
45 #include <sys/sysproto.h>
46 
47 #include <vm/pmap.h>
48 #include <vm/vm_map.h>
49 
50 #include <compat/linux/linux_emul.h>
51 #include <compat/linux/linux_mmap.h>
52 #include <compat/linux/linux_persona.h>
53 #include <compat/linux/linux_util.h>
54 
55 
56 #define STACK_SIZE  (2 * 1024 * 1024)
57 #define GUARD_SIZE  (4 * PAGE_SIZE)
58 
59 #if defined(__amd64__)
60 static void linux_fixup_prot(struct thread *td, int *prot);
61 #endif
62 
63 
64 int
65 linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
66     int flags, int fd, off_t pos)
67 {
68 	struct proc *p = td->td_proc;
69 	struct vmspace *vms = td->td_proc->p_vmspace;
70 	struct mmap_args /* {
71 		caddr_t addr;
72 		size_t len;
73 		int prot;
74 		int flags;
75 		int fd;
76 		off_t pos;
77 	} */ bsd_args;
78 	int error;
79 	struct file *fp;
80 
81 	cap_rights_t rights;
82 	LINUX_CTR6(mmap2, "0x%lx, %ld, %ld, 0x%08lx, %ld, 0x%lx",
83 	    addr, len, prot, flags, fd, pos);
84 
85 	error = 0;
86 	bsd_args.flags = 0;
87 	fp = NULL;
88 
89 	/*
90 	 * Linux mmap(2):
91 	 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE
92 	 */
93 	if (!((flags & LINUX_MAP_SHARED) ^ (flags & LINUX_MAP_PRIVATE)))
94 		return (EINVAL);
95 
96 	if (flags & LINUX_MAP_SHARED)
97 		bsd_args.flags |= MAP_SHARED;
98 	if (flags & LINUX_MAP_PRIVATE)
99 		bsd_args.flags |= MAP_PRIVATE;
100 	if (flags & LINUX_MAP_FIXED)
101 		bsd_args.flags |= MAP_FIXED;
102 	if (flags & LINUX_MAP_ANON) {
103 		/* Enforce pos to be on page boundary, then ignore. */
104 		if ((pos & PAGE_MASK) != 0)
105 			return (EINVAL);
106 		pos = 0;
107 		bsd_args.flags |= MAP_ANON;
108 	} else
109 		bsd_args.flags |= MAP_NOSYNC;
110 	if (flags & LINUX_MAP_GROWSDOWN)
111 		bsd_args.flags |= MAP_STACK;
112 
113 	/*
114 	 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC
115 	 * on Linux/i386 if the binary requires executable stack.
116 	 * We do this only for IA32 emulation as on native i386 this is does not
117 	 * make sense without PAE.
118 	 *
119 	 * XXX. Linux checks that the file system is not mounted with noexec.
120 	 */
121 	bsd_args.prot = prot;
122 #if defined(__amd64__)
123 	linux_fixup_prot(td, &bsd_args.prot);
124 #endif
125 
126 	/* Linux does not check file descriptor when MAP_ANONYMOUS is set. */
127 	bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : fd;
128 	if (bsd_args.fd != -1) {
129 		/*
130 		 * Linux follows Solaris mmap(2) description:
131 		 * The file descriptor fildes is opened with
132 		 * read permission, regardless of the
133 		 * protection options specified.
134 		 */
135 
136 		error = fget(td, bsd_args.fd,
137 		    cap_rights_init(&rights, CAP_MMAP), &fp);
138 		if (error != 0)
139 			return (error);
140 		if (fp->f_type != DTYPE_VNODE) {
141 			fdrop(fp, td);
142 			return (EINVAL);
143 		}
144 
145 		/* Linux mmap() just fails for O_WRONLY files */
146 		if (!(fp->f_flag & FREAD)) {
147 			fdrop(fp, td);
148 			return (EACCES);
149 		}
150 
151 		fdrop(fp, td);
152 	}
153 
154 	if (flags & LINUX_MAP_GROWSDOWN) {
155 		/*
156 		 * The Linux MAP_GROWSDOWN option does not limit auto
157 		 * growth of the region.  Linux mmap with this option
158 		 * takes as addr the initial BOS, and as len, the initial
159 		 * region size.  It can then grow down from addr without
160 		 * limit.  However, Linux threads has an implicit internal
161 		 * limit to stack size of STACK_SIZE.  Its just not
162 		 * enforced explicitly in Linux.  But, here we impose
163 		 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack
164 		 * region, since we can do this with our mmap.
165 		 *
166 		 * Our mmap with MAP_STACK takes addr as the maximum
167 		 * downsize limit on BOS, and as len the max size of
168 		 * the region.  It then maps the top SGROWSIZ bytes,
169 		 * and auto grows the region down, up to the limit
170 		 * in addr.
171 		 *
172 		 * If we don't use the MAP_STACK option, the effect
173 		 * of this code is to allocate a stack region of a
174 		 * fixed size of (STACK_SIZE - GUARD_SIZE).
175 		 */
176 
177 		if ((caddr_t)addr + len > vms->vm_maxsaddr) {
178 			/*
179 			 * Some Linux apps will attempt to mmap
180 			 * thread stacks near the top of their
181 			 * address space.  If their TOS is greater
182 			 * than vm_maxsaddr, vm_map_growstack()
183 			 * will confuse the thread stack with the
184 			 * process stack and deliver a SEGV if they
185 			 * attempt to grow the thread stack past their
186 			 * current stacksize rlimit.  To avoid this,
187 			 * adjust vm_maxsaddr upwards to reflect
188 			 * the current stacksize rlimit rather
189 			 * than the maximum possible stacksize.
190 			 * It would be better to adjust the
191 			 * mmap'ed region, but some apps do not check
192 			 * mmap's return value.
193 			 */
194 			PROC_LOCK(p);
195 			vms->vm_maxsaddr = (char *)p->p_sysent->sv_usrstack -
196 			    lim_cur_proc(p, RLIMIT_STACK);
197 			PROC_UNLOCK(p);
198 		}
199 
200 		/*
201 		 * This gives us our maximum stack size and a new BOS.
202 		 * If we're using VM_STACK, then mmap will just map
203 		 * the top SGROWSIZ bytes, and let the stack grow down
204 		 * to the limit at BOS.  If we're not using VM_STACK
205 		 * we map the full stack, since we don't have a way
206 		 * to autogrow it.
207 		 */
208 		if (len > STACK_SIZE - GUARD_SIZE) {
209 			bsd_args.addr = (caddr_t)addr;
210 			bsd_args.len = len;
211 		} else {
212 			bsd_args.addr = (caddr_t)addr -
213 			    (STACK_SIZE - GUARD_SIZE - len);
214 			bsd_args.len = STACK_SIZE - GUARD_SIZE;
215 		}
216 	} else {
217 		bsd_args.addr = (caddr_t)addr;
218 		bsd_args.len  = len;
219 	}
220 	bsd_args.pos = pos;
221 
222 	error = sys_mmap(td, &bsd_args);
223 
224 	LINUX_CTR2(mmap2, "return: %d (%p)", error, td->td_retval[0]);
225 
226 	return (error);
227 }
228 
229 int
230 linux_mprotect_common(struct thread *td, uintptr_t addr, size_t len, int prot)
231 {
232 	struct mprotect_args bsd_args;
233 
234 	bsd_args.addr = (void *)addr;
235 	bsd_args.len = len;
236 	bsd_args.prot = prot;
237 
238 #if defined(__amd64__)
239 	linux_fixup_prot(td, &bsd_args.prot);
240 #endif
241 	return (sys_mprotect(td, &bsd_args));
242 }
243 
244 #if defined(__amd64__)
245 static void
246 linux_fixup_prot(struct thread *td, int *prot)
247 {
248 	struct linux_pemuldata *pem;
249 
250 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32) && *prot & PROT_READ) {
251 		pem = pem_find(td->td_proc);
252 		if (pem->persona & LINUX_READ_IMPLIES_EXEC)
253 			*prot |= PROT_EXEC;
254 	}
255 
256 }
257 #endif
258