xref: /dflybsd-src/sys/vfs/procfs/procfs_mem.c (revision bf22d4c1f95f57623b2b3030738e116d3a547284)
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. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)procfs_mem.c	8.5 (Berkeley) 6/15/94
39  *
40  * $FreeBSD: src/sys/miscfs/procfs/procfs_mem.c,v 1.46.2.3 2002/01/22 17:22:59 nectar Exp $
41  * $DragonFly: src/sys/vfs/procfs/procfs_mem.c,v 1.9 2004/05/13 17:40:19 dillon Exp $
42  */
43 
44 /*
45  * This is a lightly hacked and merged version
46  * of sef's pread/pwrite functions
47  */
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/proc.h>
52 #include <sys/vnode.h>
53 #include <vfs/procfs/procfs.h>
54 #include <vm/vm.h>
55 #include <vm/vm_param.h>
56 #include <sys/lock.h>
57 #include <vm/pmap.h>
58 #include <vm/vm_extern.h>
59 #include <vm/vm_map.h>
60 #include <vm/vm_kern.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_page.h>
63 #include <sys/user.h>
64 #include <sys/ptrace.h>
65 
66 static int	procfs_rwmem (struct proc *curp,
67 				  struct proc *p, struct uio *uio);
68 
69 static int
70 procfs_rwmem(struct proc *curp, struct proc *p, struct uio *uio)
71 {
72 	int error;
73 	int writing;
74 	struct vmspace *vm;
75 	vm_map_t map;
76 	vm_offset_t pageno = 0;		/* page number */
77 	vm_prot_t reqprot;
78 	vm_offset_t kva;
79 
80 	/*
81 	 * if the vmspace is in the midst of being deallocated or the
82 	 * process is exiting, don't try to grab anything.  The page table
83 	 * usage in that process can be messed up.
84 	 */
85 	vm = p->p_vmspace;
86 	if ((p->p_flag & P_WEXIT) || (vm->vm_refcnt < 1))
87 		return EFAULT;
88 	++vm->vm_refcnt;
89 	/*
90 	 * The map we want...
91 	 */
92 	map = &vm->vm_map;
93 
94 	writing = uio->uio_rw == UIO_WRITE;
95 	reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) : VM_PROT_READ;
96 
97 	kva = kmem_alloc_pageable(kernel_map, PAGE_SIZE);
98 
99 	/*
100 	 * Only map in one page at a time.  We don't have to, but it
101 	 * makes things easier.  This way is trivial - right?
102 	 */
103 	do {
104 		vm_map_t tmap;
105 		vm_offset_t uva;
106 		int page_offset;		/* offset into page */
107 		vm_map_entry_t out_entry;
108 		vm_prot_t out_prot;
109 		boolean_t wired;
110 		vm_pindex_t pindex;
111 		vm_object_t object;
112 		u_int len;
113 		vm_page_t m;
114 		int s;
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 		 * How many bytes to copy
126 		 */
127 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
128 
129 		/*
130 		 * Fault the page on behalf of the process
131 		 */
132 		error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
133 		if (error) {
134 			error = EFAULT;
135 			break;
136 		}
137 
138 		/*
139 		 * Now we need to get the page.  out_entry, out_prot, wired,
140 		 * and single_use aren't used.  One would think the vm code
141 		 * would be a *bit* nicer...  We use tmap because
142 		 * vm_map_lookup() can change the map argument.
143 		 */
144 		tmap = map;
145 		error = vm_map_lookup(&tmap, pageno, reqprot,
146 			      &out_entry, &object, &pindex, &out_prot,
147 			      &wired);
148 
149 		if (error) {
150 			error = EFAULT;
151 			break;
152 		}
153 
154 		/*
155 		 * spl protection is required to avoid interrupt freeing
156 		 * races, reference the object to avoid it being ripped
157 		 * out from under us if we block.
158 		 */
159 		s = splvm();
160 		vm_object_reference(object);
161 again:
162 		m = vm_page_lookup(object, pindex);
163 
164 		/*
165 		 * Allow fallback to backing objects if we are reading
166 		 */
167 		while (m == NULL && !writing && object->backing_object) {
168 			pindex += OFF_TO_IDX(object->backing_object_offset);
169 			object = object->backing_object;
170 			m = vm_page_lookup(object, pindex);
171 		}
172 
173 		/*
174 		 * Wait for any I/O's to complete, then hold the page
175 		 * so we can release the spl.
176 		 */
177 		if (m) {
178 			if (vm_page_sleep_busy(m, FALSE, "rwmem"))
179 				goto again;
180 			vm_page_hold(m);
181 		}
182 		splx(s);
183 
184 		/*
185 		 * We no longer need the object.  If we do not have a page
186 		 * then cleanup.
187 		 */
188 		vm_object_deallocate(object);
189 		if (m == NULL) {
190 			vm_map_lookup_done(tmap, out_entry, 0);
191 			error = EFAULT;
192 			break;
193 		}
194 
195 		/*
196 		 * Cleanup tmap then create a temporary KVA mapping and
197 		 * do the I/O.
198 		 */
199 		vm_map_lookup_done(tmap, out_entry, 0);
200 		pmap_kenter(kva, VM_PAGE_TO_PHYS(m));
201 		error = uiomove((caddr_t)(kva + page_offset), len, uio);
202 		pmap_kremove(kva);
203 
204 		/*
205 		 * release the page and we are done
206 		 */
207 		s = splbio();
208 		vm_page_unhold(m);
209 		splx(s);
210 	} while (error == 0 && uio->uio_resid > 0);
211 
212 	kmem_free(kernel_map, kva, PAGE_SIZE);
213 	vmspace_free(vm);
214 	return (error);
215 }
216 
217 /*
218  * Copy data in and out of the target process.
219  * We do this by mapping the process's page into
220  * the kernel and then doing a uiomove direct
221  * from the kernel address space.
222  */
223 int
224 procfs_domem(struct proc *curp, struct proc *p, struct pfsnode *pfs,
225 	     struct uio *uio)
226 {
227 	if (uio->uio_resid == 0)
228 		return (0);
229 
230 	/* Can't trace a process that's currently exec'ing. */
231 	if ((p->p_flag & P_INEXEC) != 0)
232 		return EAGAIN;
233  	if (!CHECKIO(curp, p) || p_trespass(curp->p_ucred, p->p_ucred))
234  		return EPERM;
235 
236 	return (procfs_rwmem(curp, p, uio));
237 }
238 
239 /*
240  * Given process (p), find the vnode from which
241  * its text segment is being executed.
242  *
243  * It would be nice to grab this information from
244  * the VM system, however, there is no sure-fire
245  * way of doing that.  Instead, fork(), exec() and
246  * wait() all maintain the p_textvp field in the
247  * process proc structure which contains a held
248  * reference to the exec'ed vnode.
249  *
250  * XXX - Currently, this is not not used, as the
251  * /proc/pid/file object exposes an information leak
252  * that shouldn't happen.  Using a mount option would
253  * make it configurable on a per-system (or, at least,
254  * per-mount) basis; however, that's not really best.
255  * The best way to do it, I think, would be as an
256  * ioctl; this would restrict it to the uid running
257  * program, or root, which seems a reasonable compromise.
258  * However, the number of applications for this is
259  * minimal, if it can't be seen in the filesytem space,
260  * and doint it as an ioctl makes it somewhat less
261  * useful due to the, well, inelegance.
262  *
263  */
264 struct vnode *
265 procfs_findtextvp(struct proc *p)
266 {
267 	return (p->p_textvp);
268 }
269