kernel - Major SMP performance patch / VM system, bus-fault/seg-fault fixes
[dragonfly.git] / sys / vfs / procfs / procfs_map.c
1 /*
2  * Copyright (c) 1993 Jan-Simon Pendry
3  * Copyright (c) 1993
4  *      The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Jan-Simon Pendry.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
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  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *      @(#)procfs_status.c     8.3 (Berkeley) 2/17/94
34  *
35  * $FreeBSD: src/sys/miscfs/procfs/procfs_map.c,v 1.24.2.1 2001/08/04 13:12:24 rwatson Exp $
36  * $DragonFly: src/sys/vfs/procfs/procfs_map.c,v 1.7 2007/02/19 01:14:24 corecode Exp $
37  */
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/proc.h>
42 #include <sys/vnode.h>
43 #include <vfs/procfs/procfs.h>
44
45 #include <vm/vm.h>
46 #include <sys/lock.h>
47 #include <vm/pmap.h>
48 #include <vm/vm_map.h>
49 #include <vm/vm_page.h>
50 #include <vm/vm_object.h>
51
52
53 #define MEBUFFERSIZE 256
54
55 /*
56  * The map entries can *almost* be read with programs like cat.  However,
57  * large maps need special programs to read.  It is not easy to implement
58  * a program that can sense the required size of the buffer, and then
59  * subsequently do a read with the appropriate size.  This operation cannot
60  * be atomic.  The best that we can do is to allow the program to do a read
61  * with an arbitrarily large buffer, and return as much as we can.  We can
62  * return an error code if the buffer is too small (EFBIG), then the program
63  * can try a bigger buffer.
64  */
65 int
66 procfs_domap(struct proc *curp, struct lwp *lp, struct pfsnode *pfs,
67              struct uio *uio)
68 {
69         struct proc *p = lp->lwp_proc;
70         int len;
71         struct vnode *vp;
72         char *fullpath, *freepath;
73         int error;
74         vm_map_t map = &p->p_vmspace->vm_map;
75         pmap_t pmap = vmspace_pmap(p->p_vmspace);
76         vm_map_entry_t entry;
77         char mebuffer[MEBUFFERSIZE];
78
79         if (uio->uio_rw != UIO_READ)
80                 return (EOPNOTSUPP);
81
82         if (uio->uio_offset != 0)
83                 return (0);
84         
85         error = 0;
86         vm_map_lock_read(map);
87         for (entry = map->header.next;
88                 ((uio->uio_resid > 0) && (entry != &map->header));
89                 entry = entry->next) {
90                 vm_object_t obj, tobj, lobj;
91                 int ref_count, shadow_count, flags;
92                 vm_offset_t addr;
93                 vm_offset_t ostart;
94                 int resident, privateresident;
95                 char *type;
96
97                 if (entry->maptype != VM_MAPTYPE_NORMAL &&
98                     entry->maptype != VM_MAPTYPE_VPAGETABLE) {
99                         continue;
100                 }
101
102                 obj = entry->object.vm_object;
103                 if (obj)
104                         vm_object_hold(obj);
105
106                 if (obj && (obj->shadow_count == 1))
107                         privateresident = obj->resident_page_count;
108                 else
109                         privateresident = 0;
110
111                 /*
112                  * Use map->hint as a poor man's ripout detector.
113                  */
114                 map->hint = entry;
115                 ostart = entry->start;
116
117                 /*
118                  * Count resident pages (XXX can be horrible on 64-bit)
119                  */
120                 resident = 0;
121                 addr = entry->start;
122                 while (addr < entry->end) {
123                         if (pmap_extract(pmap, addr))
124                                 resident++;
125                         addr += PAGE_SIZE;
126                 }
127                 if (obj) {
128                         lobj = obj;
129                         while ((tobj = lobj->backing_object) != NULL) {
130                                 KKASSERT(tobj != obj);
131                                 vm_object_hold(tobj);
132                                 if (tobj == lobj->backing_object) {
133                                         if (lobj != obj) {
134                                                 vm_object_lock_swap();
135                                                 vm_object_drop(lobj);
136                                         }
137                                         lobj = tobj;
138                                 } else {
139                                         vm_object_drop(tobj);
140                                 }
141                         }
142                 } else {
143                         lobj = NULL;
144                 }
145
146                 freepath = NULL;
147                 fullpath = "-";
148                 if (lobj) {
149                         switch(lobj->type) {
150                         default:
151                         case OBJT_DEFAULT:
152                                 type = "default";
153                                 vp = NULL;
154                                 break;
155                         case OBJT_VNODE:
156                                 type = "vnode";
157                                 vp = lobj->handle;
158                                 vref(vp);
159                                 break;
160                         case OBJT_SWAP:
161                                 type = "swap";
162                                 vp = NULL;
163                                 break;
164                         case OBJT_DEVICE:
165                                 type = "device";
166                                 vp = NULL;
167                                 break;
168                         }
169                         
170                         flags = obj->flags;
171                         ref_count = obj->ref_count;
172                         shadow_count = obj->shadow_count;
173                         if (vp != NULL) {
174                                 vn_fullpath(p, vp, &fullpath, &freepath, 1);
175                                 vrele(vp);
176                         }
177                         if (lobj != obj)
178                                 vm_object_drop(lobj);
179                 } else {
180                         type = "none";
181                         flags = 0;
182                         ref_count = 0;
183                         shadow_count = 0;
184                 }
185
186                 /*
187                  * format:
188                  *  start, end, res, priv res, cow, access, type, (fullpath).
189                  */
190                 ksnprintf(mebuffer, sizeof(mebuffer),
191                     "0x%lx 0x%lx %d %d %p %s%s%s %d %d 0x%x %s %s %s %s\n",
192                         (u_long)entry->start, (u_long)entry->end,
193                         resident, privateresident, obj,
194                         (entry->protection & VM_PROT_READ)?"r":"-",
195                         (entry->protection & VM_PROT_WRITE)?"w":"-",
196                         (entry->protection & VM_PROT_EXECUTE)?"x":"-",
197                         ref_count, shadow_count, flags,
198                         (entry->eflags & MAP_ENTRY_COW)?"COW":"NCOW",
199                         (entry->eflags & MAP_ENTRY_NEEDS_COPY)?"NC":"NNC",
200                         type, fullpath);
201
202                 if (obj)
203                         vm_object_drop(obj);
204
205                 if (freepath != NULL) {
206                         kfree(freepath, M_TEMP);
207                         freepath = NULL;
208                 }
209
210                 len = strlen(mebuffer);
211                 if (len > uio->uio_resid) {
212                         error = EFBIG;
213                         break;
214                 }
215
216                 /*
217                  * We cannot safely hold the map locked while accessing
218                  * userspace as a VM fault might recurse the locked map.
219                  */
220                 vm_map_unlock_read(map);
221                 error = uiomove(mebuffer, len, uio);
222                 vm_map_lock_read(map);
223                 if (error)
224                         break;
225
226                 /*
227                  * We use map->hint as a poor man's ripout detector.  If
228                  * it does not match the entry we set it to prior to
229                  * unlocking the map the entry MIGHT now be stale.  In
230                  * this case we do an expensive lookup to find our place
231                  * in the iteration again.
232                  */
233                 if (map->hint != entry) {
234                         vm_map_entry_t reentry;
235
236                         vm_map_lookup_entry(map, ostart, &reentry);
237                         entry = reentry;
238                 }
239         }
240         vm_map_unlock_read(map);
241
242         return error;
243 }
244
245 int
246 procfs_validmap(struct lwp *lp)
247 {
248         return ((lp->lwp_proc->p_flag & P_SYSTEM) == 0);
249 }