proc: move sysrq-trigger out of fs/proc/
[safe/jmp/linux-2.6] / fs / proc / kcore.c
1 /*
2  *      fs/proc/kcore.c kernel ELF core dumper
3  *
4  *      Modelled on fs/exec.c:aout_core_dump()
5  *      Jeremy Fitzhardinge <jeremy@sw.oz.au>
6  *      ELF version written by David Howells <David.Howells@nexor.co.uk>
7  *      Modified and incorporated into 2.3.x by Tigran Aivazian <tigran@veritas.com>
8  *      Support to dump vmalloc'd areas (ELF only), Tigran Aivazian <tigran@veritas.com>
9  *      Safe accesses to vmalloc/direct-mapped discontiguous areas, Kanoj Sarcar <kanoj@sgi.com>
10  */
11
12 #include <linux/mm.h>
13 #include <linux/proc_fs.h>
14 #include <linux/user.h>
15 #include <linux/capability.h>
16 #include <linux/elf.h>
17 #include <linux/elfcore.h>
18 #include <linux/vmalloc.h>
19 #include <linux/highmem.h>
20 #include <linux/init.h>
21 #include <asm/uaccess.h>
22 #include <asm/io.h>
23
24 #define CORE_STR "CORE"
25
26 #ifndef ELF_CORE_EFLAGS
27 #define ELF_CORE_EFLAGS 0
28 #endif
29
30 static int open_kcore(struct inode * inode, struct file * filp)
31 {
32         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
33 }
34
35 static ssize_t read_kcore(struct file *, char __user *, size_t, loff_t *);
36
37 const struct file_operations proc_kcore_operations = {
38         .read           = read_kcore,
39         .open           = open_kcore,
40 };
41
42 #ifndef kc_vaddr_to_offset
43 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
44 #endif
45 #ifndef kc_offset_to_vaddr
46 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
47 #endif
48
49 /* An ELF note in memory */
50 struct memelfnote
51 {
52         const char *name;
53         int type;
54         unsigned int datasz;
55         void *data;
56 };
57
58 static struct kcore_list *kclist;
59 static DEFINE_RWLOCK(kclist_lock);
60
61 void
62 kclist_add(struct kcore_list *new, void *addr, size_t size)
63 {
64         new->addr = (unsigned long)addr;
65         new->size = size;
66
67         write_lock(&kclist_lock);
68         new->next = kclist;
69         kclist = new;
70         write_unlock(&kclist_lock);
71 }
72
73 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
74 {
75         size_t try, size;
76         struct kcore_list *m;
77
78         *nphdr = 1; /* PT_NOTE */
79         size = 0;
80
81         for (m=kclist; m; m=m->next) {
82                 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
83                 if (try > size)
84                         size = try;
85                 *nphdr = *nphdr + 1;
86         }
87         *elf_buflen =   sizeof(struct elfhdr) + 
88                         (*nphdr + 2)*sizeof(struct elf_phdr) + 
89                         3 * ((sizeof(struct elf_note)) +
90                              roundup(sizeof(CORE_STR), 4)) +
91                         roundup(sizeof(struct elf_prstatus), 4) +
92                         roundup(sizeof(struct elf_prpsinfo), 4) +
93                         roundup(sizeof(struct task_struct), 4);
94         *elf_buflen = PAGE_ALIGN(*elf_buflen);
95         return size + *elf_buflen;
96 }
97
98
99 /*****************************************************************************/
100 /*
101  * determine size of ELF note
102  */
103 static int notesize(struct memelfnote *en)
104 {
105         int sz;
106
107         sz = sizeof(struct elf_note);
108         sz += roundup((strlen(en->name) + 1), 4);
109         sz += roundup(en->datasz, 4);
110
111         return sz;
112 } /* end notesize() */
113
114 /*****************************************************************************/
115 /*
116  * store a note in the header buffer
117  */
118 static char *storenote(struct memelfnote *men, char *bufp)
119 {
120         struct elf_note en;
121
122 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
123
124         en.n_namesz = strlen(men->name) + 1;
125         en.n_descsz = men->datasz;
126         en.n_type = men->type;
127
128         DUMP_WRITE(&en, sizeof(en));
129         DUMP_WRITE(men->name, en.n_namesz);
130
131         /* XXX - cast from long long to long to avoid need for libgcc.a */
132         bufp = (char*) roundup((unsigned long)bufp,4);
133         DUMP_WRITE(men->data, men->datasz);
134         bufp = (char*) roundup((unsigned long)bufp,4);
135
136 #undef DUMP_WRITE
137
138         return bufp;
139 } /* end storenote() */
140
141 /*
142  * store an ELF coredump header in the supplied buffer
143  * nphdr is the number of elf_phdr to insert
144  */
145 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
146 {
147         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
148         struct elf_prpsinfo prpsinfo;   /* NT_PRPSINFO */
149         struct elf_phdr *nhdr, *phdr;
150         struct elfhdr *elf;
151         struct memelfnote notes[3];
152         off_t offset = 0;
153         struct kcore_list *m;
154
155         /* setup ELF header */
156         elf = (struct elfhdr *) bufp;
157         bufp += sizeof(struct elfhdr);
158         offset += sizeof(struct elfhdr);
159         memcpy(elf->e_ident, ELFMAG, SELFMAG);
160         elf->e_ident[EI_CLASS]  = ELF_CLASS;
161         elf->e_ident[EI_DATA]   = ELF_DATA;
162         elf->e_ident[EI_VERSION]= EV_CURRENT;
163         elf->e_ident[EI_OSABI] = ELF_OSABI;
164         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
165         elf->e_type     = ET_CORE;
166         elf->e_machine  = ELF_ARCH;
167         elf->e_version  = EV_CURRENT;
168         elf->e_entry    = 0;
169         elf->e_phoff    = sizeof(struct elfhdr);
170         elf->e_shoff    = 0;
171         elf->e_flags    = ELF_CORE_EFLAGS;
172         elf->e_ehsize   = sizeof(struct elfhdr);
173         elf->e_phentsize= sizeof(struct elf_phdr);
174         elf->e_phnum    = nphdr;
175         elf->e_shentsize= 0;
176         elf->e_shnum    = 0;
177         elf->e_shstrndx = 0;
178
179         /* setup ELF PT_NOTE program header */
180         nhdr = (struct elf_phdr *) bufp;
181         bufp += sizeof(struct elf_phdr);
182         offset += sizeof(struct elf_phdr);
183         nhdr->p_type    = PT_NOTE;
184         nhdr->p_offset  = 0;
185         nhdr->p_vaddr   = 0;
186         nhdr->p_paddr   = 0;
187         nhdr->p_filesz  = 0;
188         nhdr->p_memsz   = 0;
189         nhdr->p_flags   = 0;
190         nhdr->p_align   = 0;
191
192         /* setup ELF PT_LOAD program header for every area */
193         for (m=kclist; m; m=m->next) {
194                 phdr = (struct elf_phdr *) bufp;
195                 bufp += sizeof(struct elf_phdr);
196                 offset += sizeof(struct elf_phdr);
197
198                 phdr->p_type    = PT_LOAD;
199                 phdr->p_flags   = PF_R|PF_W|PF_X;
200                 phdr->p_offset  = kc_vaddr_to_offset(m->addr) + dataoff;
201                 phdr->p_vaddr   = (size_t)m->addr;
202                 phdr->p_paddr   = 0;
203                 phdr->p_filesz  = phdr->p_memsz = m->size;
204                 phdr->p_align   = PAGE_SIZE;
205         }
206
207         /*
208          * Set up the notes in similar form to SVR4 core dumps made
209          * with info from their /proc.
210          */
211         nhdr->p_offset  = offset;
212
213         /* set up the process status */
214         notes[0].name = CORE_STR;
215         notes[0].type = NT_PRSTATUS;
216         notes[0].datasz = sizeof(struct elf_prstatus);
217         notes[0].data = &prstatus;
218
219         memset(&prstatus, 0, sizeof(struct elf_prstatus));
220
221         nhdr->p_filesz  = notesize(&notes[0]);
222         bufp = storenote(&notes[0], bufp);
223
224         /* set up the process info */
225         notes[1].name   = CORE_STR;
226         notes[1].type   = NT_PRPSINFO;
227         notes[1].datasz = sizeof(struct elf_prpsinfo);
228         notes[1].data   = &prpsinfo;
229
230         memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
231         prpsinfo.pr_state       = 0;
232         prpsinfo.pr_sname       = 'R';
233         prpsinfo.pr_zomb        = 0;
234
235         strcpy(prpsinfo.pr_fname, "vmlinux");
236         strncpy(prpsinfo.pr_psargs, saved_command_line, ELF_PRARGSZ);
237
238         nhdr->p_filesz  += notesize(&notes[1]);
239         bufp = storenote(&notes[1], bufp);
240
241         /* set up the task structure */
242         notes[2].name   = CORE_STR;
243         notes[2].type   = NT_TASKSTRUCT;
244         notes[2].datasz = sizeof(struct task_struct);
245         notes[2].data   = current;
246
247         nhdr->p_filesz  += notesize(&notes[2]);
248         bufp = storenote(&notes[2], bufp);
249
250 } /* end elf_kcore_store_hdr() */
251
252 /*****************************************************************************/
253 /*
254  * read from the ELF header and then kernel memory
255  */
256 static ssize_t
257 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
258 {
259         ssize_t acc = 0;
260         size_t size, tsz;
261         size_t elf_buflen;
262         int nphdr;
263         unsigned long start;
264
265         read_lock(&kclist_lock);
266         proc_root_kcore->size = size = get_kcore_size(&nphdr, &elf_buflen);
267         if (buflen == 0 || *fpos >= size) {
268                 read_unlock(&kclist_lock);
269                 return 0;
270         }
271
272         /* trim buflen to not go beyond EOF */
273         if (buflen > size - *fpos)
274                 buflen = size - *fpos;
275
276         /* construct an ELF core header if we'll need some of it */
277         if (*fpos < elf_buflen) {
278                 char * elf_buf;
279
280                 tsz = elf_buflen - *fpos;
281                 if (buflen < tsz)
282                         tsz = buflen;
283                 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
284                 if (!elf_buf) {
285                         read_unlock(&kclist_lock);
286                         return -ENOMEM;
287                 }
288                 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
289                 read_unlock(&kclist_lock);
290                 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
291                         kfree(elf_buf);
292                         return -EFAULT;
293                 }
294                 kfree(elf_buf);
295                 buflen -= tsz;
296                 *fpos += tsz;
297                 buffer += tsz;
298                 acc += tsz;
299
300                 /* leave now if filled buffer already */
301                 if (buflen == 0)
302                         return acc;
303         } else
304                 read_unlock(&kclist_lock);
305
306         /*
307          * Check to see if our file offset matches with any of
308          * the addresses in the elf_phdr on our list.
309          */
310         start = kc_offset_to_vaddr(*fpos - elf_buflen);
311         if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
312                 tsz = buflen;
313                 
314         while (buflen) {
315                 struct kcore_list *m;
316
317                 read_lock(&kclist_lock);
318                 for (m=kclist; m; m=m->next) {
319                         if (start >= m->addr && start < (m->addr+m->size))
320                                 break;
321                 }
322                 read_unlock(&kclist_lock);
323
324                 if (m == NULL) {
325                         if (clear_user(buffer, tsz))
326                                 return -EFAULT;
327                 } else if (is_vmalloc_addr((void *)start)) {
328                         char * elf_buf;
329                         struct vm_struct *m;
330                         unsigned long curstart = start;
331                         unsigned long cursize = tsz;
332
333                         elf_buf = kzalloc(tsz, GFP_KERNEL);
334                         if (!elf_buf)
335                                 return -ENOMEM;
336
337                         read_lock(&vmlist_lock);
338                         for (m=vmlist; m && cursize; m=m->next) {
339                                 unsigned long vmstart;
340                                 unsigned long vmsize;
341                                 unsigned long msize = m->size - PAGE_SIZE;
342
343                                 if (((unsigned long)m->addr + msize) < 
344                                                                 curstart)
345                                         continue;
346                                 if ((unsigned long)m->addr > (curstart + 
347                                                                 cursize))
348                                         break;
349                                 vmstart = (curstart < (unsigned long)m->addr ? 
350                                         (unsigned long)m->addr : curstart);
351                                 if (((unsigned long)m->addr + msize) > 
352                                                         (curstart + cursize))
353                                         vmsize = curstart + cursize - vmstart;
354                                 else
355                                         vmsize = (unsigned long)m->addr + 
356                                                         msize - vmstart;
357                                 curstart = vmstart + vmsize;
358                                 cursize -= vmsize;
359                                 /* don't dump ioremap'd stuff! (TA) */
360                                 if (m->flags & VM_IOREMAP)
361                                         continue;
362                                 memcpy(elf_buf + (vmstart - start),
363                                         (char *)vmstart, vmsize);
364                         }
365                         read_unlock(&vmlist_lock);
366                         if (copy_to_user(buffer, elf_buf, tsz)) {
367                                 kfree(elf_buf);
368                                 return -EFAULT;
369                         }
370                         kfree(elf_buf);
371                 } else {
372                         if (kern_addr_valid(start)) {
373                                 unsigned long n;
374
375                                 n = copy_to_user(buffer, (char *)start, tsz);
376                                 /*
377                                  * We cannot distingush between fault on source
378                                  * and fault on destination. When this happens
379                                  * we clear too and hope it will trigger the
380                                  * EFAULT again.
381                                  */
382                                 if (n) { 
383                                         if (clear_user(buffer + tsz - n,
384                                                                 n))
385                                                 return -EFAULT;
386                                 }
387                         } else {
388                                 if (clear_user(buffer, tsz))
389                                         return -EFAULT;
390                         }
391                 }
392                 buflen -= tsz;
393                 *fpos += tsz;
394                 buffer += tsz;
395                 acc += tsz;
396                 start += tsz;
397                 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
398         }
399
400         return acc;
401 }