fdde1cc78392cf92040c357e86271c15f4992459
[linux-2.6.git] / 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 #include <linux/list.h>
24 #include <asm/sections.h>
25
26 #define CORE_STR "CORE"
27
28 #ifndef ELF_CORE_EFLAGS
29 #define ELF_CORE_EFLAGS 0
30 #endif
31
32 static struct proc_dir_entry *proc_root_kcore;
33
34 static int open_kcore(struct inode * inode, struct file * filp)
35 {
36         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
37 }
38
39 static ssize_t read_kcore(struct file *, char __user *, size_t, loff_t *);
40
41 static const struct file_operations proc_kcore_operations = {
42         .read           = read_kcore,
43         .open           = open_kcore,
44 };
45
46 #ifndef kc_vaddr_to_offset
47 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
48 #endif
49 #ifndef kc_offset_to_vaddr
50 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
51 #endif
52
53 /* An ELF note in memory */
54 struct memelfnote
55 {
56         const char *name;
57         int type;
58         unsigned int datasz;
59         void *data;
60 };
61
62 static LIST_HEAD(kclist_head);
63 static DEFINE_RWLOCK(kclist_lock);
64
65 void
66 kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
67 {
68         new->addr = (unsigned long)addr;
69         new->size = size;
70         new->type = type;
71
72         write_lock(&kclist_lock);
73         list_add_tail(&new->list, &kclist_head);
74         write_unlock(&kclist_lock);
75 }
76
77 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
78 {
79         size_t try, size;
80         struct kcore_list *m;
81
82         *nphdr = 1; /* PT_NOTE */
83         size = 0;
84
85         list_for_each_entry(m, &kclist_head, list) {
86                 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
87                 if (try > size)
88                         size = try;
89                 *nphdr = *nphdr + 1;
90         }
91         *elf_buflen =   sizeof(struct elfhdr) + 
92                         (*nphdr + 2)*sizeof(struct elf_phdr) + 
93                         3 * ((sizeof(struct elf_note)) +
94                              roundup(sizeof(CORE_STR), 4)) +
95                         roundup(sizeof(struct elf_prstatus), 4) +
96                         roundup(sizeof(struct elf_prpsinfo), 4) +
97                         roundup(sizeof(struct task_struct), 4);
98         *elf_buflen = PAGE_ALIGN(*elf_buflen);
99         return size + *elf_buflen;
100 }
101
102
103 /*****************************************************************************/
104 /*
105  * determine size of ELF note
106  */
107 static int notesize(struct memelfnote *en)
108 {
109         int sz;
110
111         sz = sizeof(struct elf_note);
112         sz += roundup((strlen(en->name) + 1), 4);
113         sz += roundup(en->datasz, 4);
114
115         return sz;
116 } /* end notesize() */
117
118 /*****************************************************************************/
119 /*
120  * store a note in the header buffer
121  */
122 static char *storenote(struct memelfnote *men, char *bufp)
123 {
124         struct elf_note en;
125
126 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
127
128         en.n_namesz = strlen(men->name) + 1;
129         en.n_descsz = men->datasz;
130         en.n_type = men->type;
131
132         DUMP_WRITE(&en, sizeof(en));
133         DUMP_WRITE(men->name, en.n_namesz);
134
135         /* XXX - cast from long long to long to avoid need for libgcc.a */
136         bufp = (char*) roundup((unsigned long)bufp,4);
137         DUMP_WRITE(men->data, men->datasz);
138         bufp = (char*) roundup((unsigned long)bufp,4);
139
140 #undef DUMP_WRITE
141
142         return bufp;
143 } /* end storenote() */
144
145 /*
146  * store an ELF coredump header in the supplied buffer
147  * nphdr is the number of elf_phdr to insert
148  */
149 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
150 {
151         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
152         struct elf_prpsinfo prpsinfo;   /* NT_PRPSINFO */
153         struct elf_phdr *nhdr, *phdr;
154         struct elfhdr *elf;
155         struct memelfnote notes[3];
156         off_t offset = 0;
157         struct kcore_list *m;
158
159         /* setup ELF header */
160         elf = (struct elfhdr *) bufp;
161         bufp += sizeof(struct elfhdr);
162         offset += sizeof(struct elfhdr);
163         memcpy(elf->e_ident, ELFMAG, SELFMAG);
164         elf->e_ident[EI_CLASS]  = ELF_CLASS;
165         elf->e_ident[EI_DATA]   = ELF_DATA;
166         elf->e_ident[EI_VERSION]= EV_CURRENT;
167         elf->e_ident[EI_OSABI] = ELF_OSABI;
168         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
169         elf->e_type     = ET_CORE;
170         elf->e_machine  = ELF_ARCH;
171         elf->e_version  = EV_CURRENT;
172         elf->e_entry    = 0;
173         elf->e_phoff    = sizeof(struct elfhdr);
174         elf->e_shoff    = 0;
175         elf->e_flags    = ELF_CORE_EFLAGS;
176         elf->e_ehsize   = sizeof(struct elfhdr);
177         elf->e_phentsize= sizeof(struct elf_phdr);
178         elf->e_phnum    = nphdr;
179         elf->e_shentsize= 0;
180         elf->e_shnum    = 0;
181         elf->e_shstrndx = 0;
182
183         /* setup ELF PT_NOTE program header */
184         nhdr = (struct elf_phdr *) bufp;
185         bufp += sizeof(struct elf_phdr);
186         offset += sizeof(struct elf_phdr);
187         nhdr->p_type    = PT_NOTE;
188         nhdr->p_offset  = 0;
189         nhdr->p_vaddr   = 0;
190         nhdr->p_paddr   = 0;
191         nhdr->p_filesz  = 0;
192         nhdr->p_memsz   = 0;
193         nhdr->p_flags   = 0;
194         nhdr->p_align   = 0;
195
196         /* setup ELF PT_LOAD program header for every area */
197         list_for_each_entry(m, &kclist_head, list) {
198                 phdr = (struct elf_phdr *) bufp;
199                 bufp += sizeof(struct elf_phdr);
200                 offset += sizeof(struct elf_phdr);
201
202                 phdr->p_type    = PT_LOAD;
203                 phdr->p_flags   = PF_R|PF_W|PF_X;
204                 phdr->p_offset  = kc_vaddr_to_offset(m->addr) + dataoff;
205                 phdr->p_vaddr   = (size_t)m->addr;
206                 phdr->p_paddr   = 0;
207                 phdr->p_filesz  = phdr->p_memsz = m->size;
208                 phdr->p_align   = PAGE_SIZE;
209         }
210
211         /*
212          * Set up the notes in similar form to SVR4 core dumps made
213          * with info from their /proc.
214          */
215         nhdr->p_offset  = offset;
216
217         /* set up the process status */
218         notes[0].name = CORE_STR;
219         notes[0].type = NT_PRSTATUS;
220         notes[0].datasz = sizeof(struct elf_prstatus);
221         notes[0].data = &prstatus;
222
223         memset(&prstatus, 0, sizeof(struct elf_prstatus));
224
225         nhdr->p_filesz  = notesize(&notes[0]);
226         bufp = storenote(&notes[0], bufp);
227
228         /* set up the process info */
229         notes[1].name   = CORE_STR;
230         notes[1].type   = NT_PRPSINFO;
231         notes[1].datasz = sizeof(struct elf_prpsinfo);
232         notes[1].data   = &prpsinfo;
233
234         memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
235         prpsinfo.pr_state       = 0;
236         prpsinfo.pr_sname       = 'R';
237         prpsinfo.pr_zomb        = 0;
238
239         strcpy(prpsinfo.pr_fname, "vmlinux");
240         strncpy(prpsinfo.pr_psargs, saved_command_line, ELF_PRARGSZ);
241
242         nhdr->p_filesz  += notesize(&notes[1]);
243         bufp = storenote(&notes[1], bufp);
244
245         /* set up the task structure */
246         notes[2].name   = CORE_STR;
247         notes[2].type   = NT_TASKSTRUCT;
248         notes[2].datasz = sizeof(struct task_struct);
249         notes[2].data   = current;
250
251         nhdr->p_filesz  += notesize(&notes[2]);
252         bufp = storenote(&notes[2], bufp);
253
254 } /* end elf_kcore_store_hdr() */
255
256 /*****************************************************************************/
257 /*
258  * read from the ELF header and then kernel memory
259  */
260 static ssize_t
261 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
262 {
263         ssize_t acc = 0;
264         size_t size, tsz;
265         size_t elf_buflen;
266         int nphdr;
267         unsigned long start;
268
269         read_lock(&kclist_lock);
270         proc_root_kcore->size = size = get_kcore_size(&nphdr, &elf_buflen);
271         if (buflen == 0 || *fpos >= size) {
272                 read_unlock(&kclist_lock);
273                 return 0;
274         }
275
276         /* trim buflen to not go beyond EOF */
277         if (buflen > size - *fpos)
278                 buflen = size - *fpos;
279
280         /* construct an ELF core header if we'll need some of it */
281         if (*fpos < elf_buflen) {
282                 char * elf_buf;
283
284                 tsz = elf_buflen - *fpos;
285                 if (buflen < tsz)
286                         tsz = buflen;
287                 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
288                 if (!elf_buf) {
289                         read_unlock(&kclist_lock);
290                         return -ENOMEM;
291                 }
292                 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
293                 read_unlock(&kclist_lock);
294                 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
295                         kfree(elf_buf);
296                         return -EFAULT;
297                 }
298                 kfree(elf_buf);
299                 buflen -= tsz;
300                 *fpos += tsz;
301                 buffer += tsz;
302                 acc += tsz;
303
304                 /* leave now if filled buffer already */
305                 if (buflen == 0)
306                         return acc;
307         } else
308                 read_unlock(&kclist_lock);
309
310         /*
311          * Check to see if our file offset matches with any of
312          * the addresses in the elf_phdr on our list.
313          */
314         start = kc_offset_to_vaddr(*fpos - elf_buflen);
315         if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
316                 tsz = buflen;
317                 
318         while (buflen) {
319                 struct kcore_list *m;
320
321                 read_lock(&kclist_lock);
322                 list_for_each_entry(m, &kclist_head, list) {
323                         if (start >= m->addr && start < (m->addr+m->size))
324                                 break;
325                 }
326                 read_unlock(&kclist_lock);
327
328                 if (m == NULL) {
329                         if (clear_user(buffer, tsz))
330                                 return -EFAULT;
331                 } else if (is_vmalloc_addr((void *)start)) {
332                         char * elf_buf;
333
334                         elf_buf = kzalloc(tsz, GFP_KERNEL);
335                         if (!elf_buf)
336                                 return -ENOMEM;
337                         vread(elf_buf, (char *)start, tsz);
338                         /* we have to zero-fill user buffer even if no read */
339                         if (copy_to_user(buffer, elf_buf, tsz)) {
340                                 kfree(elf_buf);
341                                 return -EFAULT;
342                         }
343                         kfree(elf_buf);
344                 } else {
345                         if (kern_addr_valid(start)) {
346                                 unsigned long n;
347
348                                 n = copy_to_user(buffer, (char *)start, tsz);
349                                 /*
350                                  * We cannot distingush between fault on source
351                                  * and fault on destination. When this happens
352                                  * we clear too and hope it will trigger the
353                                  * EFAULT again.
354                                  */
355                                 if (n) { 
356                                         if (clear_user(buffer + tsz - n,
357                                                                 n))
358                                                 return -EFAULT;
359                                 }
360                         } else {
361                                 if (clear_user(buffer, tsz))
362                                         return -EFAULT;
363                         }
364                 }
365                 buflen -= tsz;
366                 *fpos += tsz;
367                 buffer += tsz;
368                 acc += tsz;
369                 start += tsz;
370                 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
371         }
372
373         return acc;
374 }
375
376 static struct kcore_list kcore_vmalloc;
377
378 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
379 static struct kcore_list kcore_text;
380 /*
381  * If defined, special segment is used for mapping kernel text instead of
382  * direct-map area. We need to create special TEXT section.
383  */
384 static void __init proc_kcore_text_init(void)
385 {
386         kclist_add(&kcore_text, _stext, _end - _stext, KCORE_TEXT);
387 }
388 #else
389 static void __init proc_kcore_text_init(void)
390 {
391 }
392 #endif
393
394 static int __init proc_kcore_init(void)
395 {
396         proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, &proc_kcore_operations);
397         proc_kcore_text_init();
398         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
399                 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
400         return 0;
401 }
402 module_init(proc_kcore_init);