kcore: add kclist types
[linux-3.10.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
25 #define CORE_STR "CORE"
26
27 #ifndef ELF_CORE_EFLAGS
28 #define ELF_CORE_EFLAGS 0
29 #endif
30
31 static struct proc_dir_entry *proc_root_kcore;
32
33 static int open_kcore(struct inode * inode, struct file * filp)
34 {
35         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
36 }
37
38 static ssize_t read_kcore(struct file *, char __user *, size_t, loff_t *);
39
40 static const struct file_operations proc_kcore_operations = {
41         .read           = read_kcore,
42         .open           = open_kcore,
43 };
44
45 #ifndef kc_vaddr_to_offset
46 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
47 #endif
48 #ifndef kc_offset_to_vaddr
49 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
50 #endif
51
52 /* An ELF note in memory */
53 struct memelfnote
54 {
55         const char *name;
56         int type;
57         unsigned int datasz;
58         void *data;
59 };
60
61 static LIST_HEAD(kclist_head);
62 static DEFINE_RWLOCK(kclist_lock);
63
64 void
65 kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
66 {
67         new->addr = (unsigned long)addr;
68         new->size = size;
69         new->type = type;
70
71         write_lock(&kclist_lock);
72         list_add_tail(&new->list, &kclist_head);
73         write_unlock(&kclist_lock);
74 }
75
76 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
77 {
78         size_t try, size;
79         struct kcore_list *m;
80
81         *nphdr = 1; /* PT_NOTE */
82         size = 0;
83
84         list_for_each_entry(m, &kclist_head, list) {
85                 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
86                 if (try > size)
87                         size = try;
88                 *nphdr = *nphdr + 1;
89         }
90         *elf_buflen =   sizeof(struct elfhdr) + 
91                         (*nphdr + 2)*sizeof(struct elf_phdr) + 
92                         3 * ((sizeof(struct elf_note)) +
93                              roundup(sizeof(CORE_STR), 4)) +
94                         roundup(sizeof(struct elf_prstatus), 4) +
95                         roundup(sizeof(struct elf_prpsinfo), 4) +
96                         roundup(sizeof(struct task_struct), 4);
97         *elf_buflen = PAGE_ALIGN(*elf_buflen);
98         return size + *elf_buflen;
99 }
100
101
102 /*****************************************************************************/
103 /*
104  * determine size of ELF note
105  */
106 static int notesize(struct memelfnote *en)
107 {
108         int sz;
109
110         sz = sizeof(struct elf_note);
111         sz += roundup((strlen(en->name) + 1), 4);
112         sz += roundup(en->datasz, 4);
113
114         return sz;
115 } /* end notesize() */
116
117 /*****************************************************************************/
118 /*
119  * store a note in the header buffer
120  */
121 static char *storenote(struct memelfnote *men, char *bufp)
122 {
123         struct elf_note en;
124
125 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
126
127         en.n_namesz = strlen(men->name) + 1;
128         en.n_descsz = men->datasz;
129         en.n_type = men->type;
130
131         DUMP_WRITE(&en, sizeof(en));
132         DUMP_WRITE(men->name, en.n_namesz);
133
134         /* XXX - cast from long long to long to avoid need for libgcc.a */
135         bufp = (char*) roundup((unsigned long)bufp,4);
136         DUMP_WRITE(men->data, men->datasz);
137         bufp = (char*) roundup((unsigned long)bufp,4);
138
139 #undef DUMP_WRITE
140
141         return bufp;
142 } /* end storenote() */
143
144 /*
145  * store an ELF coredump header in the supplied buffer
146  * nphdr is the number of elf_phdr to insert
147  */
148 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
149 {
150         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
151         struct elf_prpsinfo prpsinfo;   /* NT_PRPSINFO */
152         struct elf_phdr *nhdr, *phdr;
153         struct elfhdr *elf;
154         struct memelfnote notes[3];
155         off_t offset = 0;
156         struct kcore_list *m;
157
158         /* setup ELF header */
159         elf = (struct elfhdr *) bufp;
160         bufp += sizeof(struct elfhdr);
161         offset += sizeof(struct elfhdr);
162         memcpy(elf->e_ident, ELFMAG, SELFMAG);
163         elf->e_ident[EI_CLASS]  = ELF_CLASS;
164         elf->e_ident[EI_DATA]   = ELF_DATA;
165         elf->e_ident[EI_VERSION]= EV_CURRENT;
166         elf->e_ident[EI_OSABI] = ELF_OSABI;
167         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
168         elf->e_type     = ET_CORE;
169         elf->e_machine  = ELF_ARCH;
170         elf->e_version  = EV_CURRENT;
171         elf->e_entry    = 0;
172         elf->e_phoff    = sizeof(struct elfhdr);
173         elf->e_shoff    = 0;
174         elf->e_flags    = ELF_CORE_EFLAGS;
175         elf->e_ehsize   = sizeof(struct elfhdr);
176         elf->e_phentsize= sizeof(struct elf_phdr);
177         elf->e_phnum    = nphdr;
178         elf->e_shentsize= 0;
179         elf->e_shnum    = 0;
180         elf->e_shstrndx = 0;
181
182         /* setup ELF PT_NOTE program header */
183         nhdr = (struct elf_phdr *) bufp;
184         bufp += sizeof(struct elf_phdr);
185         offset += sizeof(struct elf_phdr);
186         nhdr->p_type    = PT_NOTE;
187         nhdr->p_offset  = 0;
188         nhdr->p_vaddr   = 0;
189         nhdr->p_paddr   = 0;
190         nhdr->p_filesz  = 0;
191         nhdr->p_memsz   = 0;
192         nhdr->p_flags   = 0;
193         nhdr->p_align   = 0;
194
195         /* setup ELF PT_LOAD program header for every area */
196         list_for_each_entry(m, &kclist_head, list) {
197                 phdr = (struct elf_phdr *) bufp;
198                 bufp += sizeof(struct elf_phdr);
199                 offset += sizeof(struct elf_phdr);
200
201                 phdr->p_type    = PT_LOAD;
202                 phdr->p_flags   = PF_R|PF_W|PF_X;
203                 phdr->p_offset  = kc_vaddr_to_offset(m->addr) + dataoff;
204                 phdr->p_vaddr   = (size_t)m->addr;
205                 phdr->p_paddr   = 0;
206                 phdr->p_filesz  = phdr->p_memsz = m->size;
207                 phdr->p_align   = PAGE_SIZE;
208         }
209
210         /*
211          * Set up the notes in similar form to SVR4 core dumps made
212          * with info from their /proc.
213          */
214         nhdr->p_offset  = offset;
215
216         /* set up the process status */
217         notes[0].name = CORE_STR;
218         notes[0].type = NT_PRSTATUS;
219         notes[0].datasz = sizeof(struct elf_prstatus);
220         notes[0].data = &prstatus;
221
222         memset(&prstatus, 0, sizeof(struct elf_prstatus));
223
224         nhdr->p_filesz  = notesize(&notes[0]);
225         bufp = storenote(&notes[0], bufp);
226
227         /* set up the process info */
228         notes[1].name   = CORE_STR;
229         notes[1].type   = NT_PRPSINFO;
230         notes[1].datasz = sizeof(struct elf_prpsinfo);
231         notes[1].data   = &prpsinfo;
232
233         memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
234         prpsinfo.pr_state       = 0;
235         prpsinfo.pr_sname       = 'R';
236         prpsinfo.pr_zomb        = 0;
237
238         strcpy(prpsinfo.pr_fname, "vmlinux");
239         strncpy(prpsinfo.pr_psargs, saved_command_line, ELF_PRARGSZ);
240
241         nhdr->p_filesz  += notesize(&notes[1]);
242         bufp = storenote(&notes[1], bufp);
243
244         /* set up the task structure */
245         notes[2].name   = CORE_STR;
246         notes[2].type   = NT_TASKSTRUCT;
247         notes[2].datasz = sizeof(struct task_struct);
248         notes[2].data   = current;
249
250         nhdr->p_filesz  += notesize(&notes[2]);
251         bufp = storenote(&notes[2], bufp);
252
253 } /* end elf_kcore_store_hdr() */
254
255 /*****************************************************************************/
256 /*
257  * read from the ELF header and then kernel memory
258  */
259 static ssize_t
260 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
261 {
262         ssize_t acc = 0;
263         size_t size, tsz;
264         size_t elf_buflen;
265         int nphdr;
266         unsigned long start;
267
268         read_lock(&kclist_lock);
269         proc_root_kcore->size = size = get_kcore_size(&nphdr, &elf_buflen);
270         if (buflen == 0 || *fpos >= size) {
271                 read_unlock(&kclist_lock);
272                 return 0;
273         }
274
275         /* trim buflen to not go beyond EOF */
276         if (buflen > size - *fpos)
277                 buflen = size - *fpos;
278
279         /* construct an ELF core header if we'll need some of it */
280         if (*fpos < elf_buflen) {
281                 char * elf_buf;
282
283                 tsz = elf_buflen - *fpos;
284                 if (buflen < tsz)
285                         tsz = buflen;
286                 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
287                 if (!elf_buf) {
288                         read_unlock(&kclist_lock);
289                         return -ENOMEM;
290                 }
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                 list_for_each_entry(m, &kclist_head, list) {
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 (is_vmalloc_addr((void *)start)) {
331                         char * elf_buf;
332
333                         elf_buf = kzalloc(tsz, GFP_KERNEL);
334                         if (!elf_buf)
335                                 return -ENOMEM;
336                         vread(elf_buf, (char *)start, tsz);
337                         /* we have to zero-fill user buffer even if no read */
338                         if (copy_to_user(buffer, elf_buf, tsz)) {
339                                 kfree(elf_buf);
340                                 return -EFAULT;
341                         }
342                         kfree(elf_buf);
343                 } else {
344                         if (kern_addr_valid(start)) {
345                                 unsigned long n;
346
347                                 n = copy_to_user(buffer, (char *)start, tsz);
348                                 /*
349                                  * We cannot distingush between fault on source
350                                  * and fault on destination. When this happens
351                                  * we clear too and hope it will trigger the
352                                  * EFAULT again.
353                                  */
354                                 if (n) { 
355                                         if (clear_user(buffer + tsz - n,
356                                                                 n))
357                                                 return -EFAULT;
358                                 }
359                         } else {
360                                 if (clear_user(buffer, tsz))
361                                         return -EFAULT;
362                         }
363                 }
364                 buflen -= tsz;
365                 *fpos += tsz;
366                 buffer += tsz;
367                 acc += tsz;
368                 start += tsz;
369                 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
370         }
371
372         return acc;
373 }
374
375 static int __init proc_kcore_init(void)
376 {
377         proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, &proc_kcore_operations);
378         return 0;
379 }
380 module_init(proc_kcore_init);