blob: 17f6e8fa13978801e939a2a9cf749fcae8d7bf4f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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/config.h>
13#include <linux/mm.h>
14#include <linux/proc_fs.h>
15#include <linux/user.h>
16#include <linux/a.out.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/elf.h>
19#include <linux/elfcore.h>
20#include <linux/vmalloc.h>
21#include <linux/highmem.h>
22#include <linux/init.h>
23#include <asm/uaccess.h>
24#include <asm/io.h>
25
26
27static int open_kcore(struct inode * inode, struct file * filp)
28{
29 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
30}
31
32static ssize_t read_kcore(struct file *, char __user *, size_t, loff_t *);
33
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080034const struct file_operations proc_kcore_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 .read = read_kcore,
36 .open = open_kcore,
37};
38
39#ifndef kc_vaddr_to_offset
40#define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
41#endif
42#ifndef kc_offset_to_vaddr
43#define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
44#endif
45
46#define roundup(x, y) ((((x)+((y)-1))/(y))*(y))
47
48/* An ELF note in memory */
49struct memelfnote
50{
51 const char *name;
52 int type;
53 unsigned int datasz;
54 void *data;
55};
56
57static struct kcore_list *kclist;
58static DEFINE_RWLOCK(kclist_lock);
59
60void
61kclist_add(struct kcore_list *new, void *addr, size_t size)
62{
63 new->addr = (unsigned long)addr;
64 new->size = size;
65
66 write_lock(&kclist_lock);
67 new->next = kclist;
68 kclist = new;
69 write_unlock(&kclist_lock);
70}
71
72static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
73{
74 size_t try, size;
75 struct kcore_list *m;
76
77 *nphdr = 1; /* PT_NOTE */
78 size = 0;
79
80 for (m=kclist; m; m=m->next) {
81 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
82 if (try > size)
83 size = try;
84 *nphdr = *nphdr + 1;
85 }
86 *elf_buflen = sizeof(struct elfhdr) +
87 (*nphdr + 2)*sizeof(struct elf_phdr) +
88 3 * (sizeof(struct elf_note) + 4) +
89 sizeof(struct elf_prstatus) +
90 sizeof(struct elf_prpsinfo) +
91 sizeof(struct task_struct);
92 *elf_buflen = PAGE_ALIGN(*elf_buflen);
93 return size + *elf_buflen;
94}
95
96
97/*****************************************************************************/
98/*
99 * determine size of ELF note
100 */
101static int notesize(struct memelfnote *en)
102{
103 int sz;
104
105 sz = sizeof(struct elf_note);
106 sz += roundup(strlen(en->name), 4);
107 sz += roundup(en->datasz, 4);
108
109 return sz;
110} /* end notesize() */
111
112/*****************************************************************************/
113/*
114 * store a note in the header buffer
115 */
116static char *storenote(struct memelfnote *men, char *bufp)
117{
118 struct elf_note en;
119
120#define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
121
122 en.n_namesz = strlen(men->name);
123 en.n_descsz = men->datasz;
124 en.n_type = men->type;
125
126 DUMP_WRITE(&en, sizeof(en));
127 DUMP_WRITE(men->name, en.n_namesz);
128
129 /* XXX - cast from long long to long to avoid need for libgcc.a */
130 bufp = (char*) roundup((unsigned long)bufp,4);
131 DUMP_WRITE(men->data, men->datasz);
132 bufp = (char*) roundup((unsigned long)bufp,4);
133
134#undef DUMP_WRITE
135
136 return bufp;
137} /* end storenote() */
138
139/*
140 * store an ELF coredump header in the supplied buffer
141 * nphdr is the number of elf_phdr to insert
142 */
143static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
144{
145 struct elf_prstatus prstatus; /* NT_PRSTATUS */
146 struct elf_prpsinfo prpsinfo; /* NT_PRPSINFO */
147 struct elf_phdr *nhdr, *phdr;
148 struct elfhdr *elf;
149 struct memelfnote notes[3];
150 off_t offset = 0;
151 struct kcore_list *m;
152
153 /* setup ELF header */
154 elf = (struct elfhdr *) bufp;
155 bufp += sizeof(struct elfhdr);
156 offset += sizeof(struct elfhdr);
157 memcpy(elf->e_ident, ELFMAG, SELFMAG);
158 elf->e_ident[EI_CLASS] = ELF_CLASS;
159 elf->e_ident[EI_DATA] = ELF_DATA;
160 elf->e_ident[EI_VERSION]= EV_CURRENT;
161 elf->e_ident[EI_OSABI] = ELF_OSABI;
162 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
163 elf->e_type = ET_CORE;
164 elf->e_machine = ELF_ARCH;
165 elf->e_version = EV_CURRENT;
166 elf->e_entry = 0;
167 elf->e_phoff = sizeof(struct elfhdr);
168 elf->e_shoff = 0;
169#if defined(CONFIG_H8300)
170 elf->e_flags = ELF_FLAGS;
171#else
172 elf->e_flags = 0;
173#endif
174 elf->e_ehsize = sizeof(struct elfhdr);
175 elf->e_phentsize= sizeof(struct elf_phdr);
176 elf->e_phnum = nphdr;
177 elf->e_shentsize= 0;
178 elf->e_shnum = 0;
179 elf->e_shstrndx = 0;
180
181 /* setup ELF PT_NOTE program header */
182 nhdr = (struct elf_phdr *) bufp;
183 bufp += sizeof(struct elf_phdr);
184 offset += sizeof(struct elf_phdr);
185 nhdr->p_type = PT_NOTE;
186 nhdr->p_offset = 0;
187 nhdr->p_vaddr = 0;
188 nhdr->p_paddr = 0;
189 nhdr->p_filesz = 0;
190 nhdr->p_memsz = 0;
191 nhdr->p_flags = 0;
192 nhdr->p_align = 0;
193
194 /* setup ELF PT_LOAD program header for every area */
195 for (m=kclist; m; m=m->next) {
196 phdr = (struct elf_phdr *) bufp;
197 bufp += sizeof(struct elf_phdr);
198 offset += sizeof(struct elf_phdr);
199
200 phdr->p_type = PT_LOAD;
201 phdr->p_flags = PF_R|PF_W|PF_X;
202 phdr->p_offset = kc_vaddr_to_offset(m->addr) + dataoff;
203 phdr->p_vaddr = (size_t)m->addr;
204 phdr->p_paddr = 0;
205 phdr->p_filesz = phdr->p_memsz = m->size;
206 phdr->p_align = PAGE_SIZE;
207 }
208
209 /*
210 * Set up the notes in similar form to SVR4 core dumps made
211 * with info from their /proc.
212 */
213 nhdr->p_offset = offset;
214
215 /* set up the process status */
216 notes[0].name = "CORE";
217 notes[0].type = NT_PRSTATUS;
218 notes[0].datasz = sizeof(struct elf_prstatus);
219 notes[0].data = &prstatus;
220
221 memset(&prstatus, 0, sizeof(struct elf_prstatus));
222
223 nhdr->p_filesz = notesize(&notes[0]);
224 bufp = storenote(&notes[0], bufp);
225
226 /* set up the process info */
227 notes[1].name = "CORE";
228 notes[1].type = NT_PRPSINFO;
229 notes[1].datasz = sizeof(struct elf_prpsinfo);
230 notes[1].data = &prpsinfo;
231
232 memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
233 prpsinfo.pr_state = 0;
234 prpsinfo.pr_sname = 'R';
235 prpsinfo.pr_zomb = 0;
236
237 strcpy(prpsinfo.pr_fname, "vmlinux");
238 strncpy(prpsinfo.pr_psargs, saved_command_line, ELF_PRARGSZ);
239
240 nhdr->p_filesz += notesize(&notes[1]);
241 bufp = storenote(&notes[1], bufp);
242
243 /* set up the task structure */
244 notes[2].name = "CORE";
245 notes[2].type = NT_TASKSTRUCT;
246 notes[2].datasz = sizeof(struct task_struct);
247 notes[2].data = current;
248
249 nhdr->p_filesz += notesize(&notes[2]);
250 bufp = storenote(&notes[2], bufp);
251
252} /* end elf_kcore_store_hdr() */
253
254/*****************************************************************************/
255/*
256 * read from the ELF header and then kernel memory
257 */
258static ssize_t
259read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
260{
261 ssize_t acc = 0;
262 size_t size, tsz;
263 size_t elf_buflen;
264 int nphdr;
265 unsigned long start;
266
267 read_lock(&kclist_lock);
268 proc_root_kcore->size = size = get_kcore_size(&nphdr, &elf_buflen);
269 if (buflen == 0 || *fpos >= size) {
270 read_unlock(&kclist_lock);
271 return 0;
272 }
273
274 /* trim buflen to not go beyond EOF */
275 if (buflen > size - *fpos)
276 buflen = size - *fpos;
277
278 /* construct an ELF core header if we'll need some of it */
279 if (*fpos < elf_buflen) {
280 char * elf_buf;
281
282 tsz = elf_buflen - *fpos;
283 if (buflen < tsz)
284 tsz = buflen;
285 elf_buf = kmalloc(elf_buflen, GFP_ATOMIC);
286 if (!elf_buf) {
287 read_unlock(&kclist_lock);
288 return -ENOMEM;
289 }
290 memset(elf_buf, 0, elf_buflen);
291 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
292 read_unlock(&kclist_lock);
293 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
294 kfree(elf_buf);
295 return -EFAULT;
296 }
297 kfree(elf_buf);
298 buflen -= tsz;
299 *fpos += tsz;
300 buffer += tsz;
301 acc += tsz;
302
303 /* leave now if filled buffer already */
304 if (buflen == 0)
305 return acc;
306 } else
307 read_unlock(&kclist_lock);
308
309 /*
310 * Check to see if our file offset matches with any of
311 * the addresses in the elf_phdr on our list.
312 */
313 start = kc_offset_to_vaddr(*fpos - elf_buflen);
314 if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
315 tsz = buflen;
316
317 while (buflen) {
318 struct kcore_list *m;
319
320 read_lock(&kclist_lock);
321 for (m=kclist; m; m=m->next) {
322 if (start >= m->addr && start < (m->addr+m->size))
323 break;
324 }
325 read_unlock(&kclist_lock);
326
327 if (m == NULL) {
328 if (clear_user(buffer, tsz))
329 return -EFAULT;
330 } else if ((start >= VMALLOC_START) && (start < VMALLOC_END)) {
331 char * elf_buf;
332 struct vm_struct *m;
333 unsigned long curstart = start;
334 unsigned long cursize = tsz;
335
336 elf_buf = kmalloc(tsz, GFP_KERNEL);
337 if (!elf_buf)
338 return -ENOMEM;
339 memset(elf_buf, 0, tsz);
340
341 read_lock(&vmlist_lock);
342 for (m=vmlist; m && cursize; m=m->next) {
343 unsigned long vmstart;
344 unsigned long vmsize;
345 unsigned long msize = m->size - PAGE_SIZE;
346
347 if (((unsigned long)m->addr + msize) <
348 curstart)
349 continue;
350 if ((unsigned long)m->addr > (curstart +
351 cursize))
352 break;
353 vmstart = (curstart < (unsigned long)m->addr ?
354 (unsigned long)m->addr : curstart);
355 if (((unsigned long)m->addr + msize) >
356 (curstart + cursize))
357 vmsize = curstart + cursize - vmstart;
358 else
359 vmsize = (unsigned long)m->addr +
360 msize - vmstart;
361 curstart = vmstart + vmsize;
362 cursize -= vmsize;
363 /* don't dump ioremap'd stuff! (TA) */
364 if (m->flags & VM_IOREMAP)
365 continue;
366 memcpy(elf_buf + (vmstart - start),
367 (char *)vmstart, vmsize);
368 }
369 read_unlock(&vmlist_lock);
370 if (copy_to_user(buffer, elf_buf, tsz)) {
371 kfree(elf_buf);
372 return -EFAULT;
373 }
374 kfree(elf_buf);
375 } else {
376 if (kern_addr_valid(start)) {
377 unsigned long n;
378
379 n = copy_to_user(buffer, (char *)start, tsz);
380 /*
381 * We cannot distingush between fault on source
382 * and fault on destination. When this happens
383 * we clear too and hope it will trigger the
384 * EFAULT again.
385 */
386 if (n) {
387 if (clear_user(buffer + tsz - n,
388 tsz - n))
389 return -EFAULT;
390 }
391 } else {
392 if (clear_user(buffer, tsz))
393 return -EFAULT;
394 }
395 }
396 buflen -= tsz;
397 *fpos += tsz;
398 buffer += tsz;
399 acc += tsz;
400 start += tsz;
401 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
402 }
403
404 return acc;
405}