xref: /dragonfly/sys/vfs/procfs/procfs_mem.c (revision 0de61e28)
1 /*
2  * Copyright (c) 1993 Jan-Simon Pendry
3  * Copyright (c) 1993 Sean Eric Fagan
4  * Copyright (c) 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Jan-Simon Pendry and Sean Eric Fagan.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)procfs_mem.c	8.5 (Berkeley) 6/15/94
35  *
36  * $FreeBSD: src/sys/miscfs/procfs/procfs_mem.c,v 1.46.2.3 2002/01/22 17:22:59 nectar Exp $
37  */
38 
39 /*
40  * This is a lightly hacked and merged version
41  * of sef's pread/pwrite functions
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/uio.h>
47 #include <sys/proc.h>
48 #include <sys/priv.h>
49 #include <sys/vnode.h>
50 #include <vfs/procfs/procfs.h>
51 #include <vm/vm.h>
52 #include <vm/vm_param.h>
53 #include <sys/lock.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_extern.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_kern.h>
58 #include <vm/vm_object.h>
59 #include <vm/vm_page.h>
60 #include <sys/ptrace.h>
61 
62 #include <machine/vmm.h>
63 
64 static int	procfs_rwmem (struct proc *curp,
65 				  struct proc *p, struct uio *uio);
66 
67 /*
68  * p->p_token is held on entry.
69  */
70 static int
71 procfs_rwmem(struct proc *curp, struct proc *p, struct uio *uio)
72 {
73 	int error;
74 	int writing;
75 	struct vmspace *vm;
76 	vm_map_t map;
77 	vm_offset_t pageno = 0;		/* page number */
78 	vm_prot_t reqprot;
79 	vm_offset_t kva;
80 
81 	/*
82 	 * if the vmspace is in the midst of being allocated or deallocated,
83 	 * or the process is exiting, don't try to grab anything.  The
84 	 * page table usage in that process may be messed up.
85 	 */
86 	vm = p->p_vmspace;
87 	if (p->p_stat == SIDL || p->p_stat == SZOMB)
88 		return EFAULT;
89 	if ((p->p_flags & (P_WEXIT | P_INEXEC)) || vmspace_getrefs(vm) < 0)
90 		return EFAULT;
91 
92 	/*
93 	 * The map we want...
94 	 */
95 	vmspace_hold(vm);
96 	map = &vm->vm_map;
97 
98 	writing = (uio->uio_rw == UIO_WRITE);
99 	reqprot = VM_PROT_READ;
100 	if (writing)
101 		reqprot |= VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE;
102 
103 	kva = kmem_alloc_pageable(&kernel_map, PAGE_SIZE, VM_SUBSYS_PROC);
104 
105 	/*
106 	 * Only map in one page at a time.  We don't have to, but it
107 	 * makes things easier.  This way is trivial - right?
108 	 */
109 	do {
110 		vm_offset_t uva;
111 		vm_offset_t page_offset;	/* offset into page */
112 		size_t len;
113 		vm_page_t m;
114 		int busy;
115 
116 		uva = (vm_offset_t) uio->uio_offset;
117 
118 		/*
119 		 * Get the page number of this segment.
120 		 */
121 		pageno = trunc_page(uva);
122 		page_offset = uva - pageno;
123 
124 		/*
125 		 * If the target process is running in VMM mode
126 		 * translate the address into a GPA (Guest Physical
127 		 * Address) via the EPT before doing the lookup.
128 		 */
129 		if (p->p_vmm) {
130 			register_t gpa;
131 			vmm_vm_get_gpa(p, &gpa, (register_t) pageno);
132 			pageno = (vm_offset_t)gpa;
133 		}
134 
135 		/*
136 		 * How many bytes to copy
137 		 */
138 		len = szmin(PAGE_SIZE - page_offset, uio->uio_resid);
139 
140 		/*
141 		 * Fault the page on behalf of the process
142 		 *
143 		 * XXX busied page on write fault can deadlock against our
144 		 *     uiomove.
145 		 */
146 		m = vm_fault_page(map, pageno, reqprot,
147 				  VM_FAULT_NORMAL,
148 				  &error, &busy);
149 		if (error) {
150 			KKASSERT(m == NULL);
151 			error = EFAULT;
152 			break;
153 		}
154 
155 		/*
156 		 * Cleanup pmap then create a temporary KVA mapping and
157 		 * do the I/O.  We can switch between cpus so don't bother
158 		 * synchronizing across all cores.
159 		 */
160 		pmap_kenter_quick(kva, VM_PAGE_TO_PHYS(m));
161 		error = uiomove((caddr_t)(kva + page_offset), len, uio);
162 		pmap_kremove_quick(kva);
163 
164 		/*
165 		 * Release the page and we are done
166 		 */
167 		if (busy)
168 			vm_page_wakeup(m);
169 		else
170 			vm_page_unhold(m);
171 	} while (error == 0 && uio->uio_resid > 0);
172 
173 	vmspace_drop(vm);
174 	kmem_free(&kernel_map, kva, PAGE_SIZE);
175 
176 	return (error);
177 }
178 
179 /*
180  * Copy data in and out of the target process.
181  * We do this by mapping the process's page into
182  * the kernel and then doing a uiomove direct
183  * from the kernel address space.
184  *
185  * lp->lwp_proc->p_token is held on entry.
186  */
187 int
188 procfs_domem(struct proc *curp, struct lwp *lp, struct pfsnode *pfs,
189 	     struct uio *uio)
190 {
191 	struct proc *p = lp->lwp_proc;
192 	int error;
193 
194 	if (uio->uio_resid == 0)
195 		return (0);
196 
197 	if ((p->p_flags & P_INEXEC) != 0) {
198 		/*
199 		 * Can't trace a process that's currently exec'ing.
200 		 */
201 		error = EAGAIN;
202 	} else if (!CHECKIO(curp, p) || p_trespass(curp->p_ucred, p->p_ucred)) {
203 		/*
204 		 * Can't trace processes outside our jail
205 		 */
206 		error = EPERM;
207 	} else {
208 		error = procfs_rwmem(curp, p, uio);
209 	}
210 	return(error);
211 }
212 
213 /*
214  * Given process (p), find the vnode from which
215  * its text segment is being executed.
216  *
217  * It would be nice to grab this information from
218  * the VM system, however, there is no sure-fire
219  * way of doing that.  Instead, fork(), exec() and
220  * wait() all maintain the p_textvp field in the
221  * process proc structure which contains a held
222  * reference to the exec'ed vnode.
223  *
224  * XXX - Currently, this is not not used, as the
225  * /proc/pid/file object exposes an information leak
226  * that shouldn't happen.  Using a mount option would
227  * make it configurable on a per-system (or, at least,
228  * per-mount) basis; however, that's not really best.
229  * The best way to do it, I think, would be as an
230  * ioctl; this would restrict it to the uid running
231  * program, or root, which seems a reasonable compromise.
232  * However, the number of applications for this is
233  * minimal, if it can't be seen in the filesytem space,
234  * and doint it as an ioctl makes it somewhat less
235  * useful due to the, well, inelegance.
236  *
237  */
238 struct vnode *
239 procfs_findtextvp(struct proc *p)
240 {
241 	return (p->p_textvp);
242 }
243