Merge branch 'bkl/procfs' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
[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/bootmem.h>
21 #include <linux/init.h>
22 #include <linux/slab.h>
23 #include <asm/uaccess.h>
24 #include <asm/io.h>
25 #include <linux/list.h>
26 #include <linux/ioport.h>
27 #include <linux/memory.h>
28 #include <asm/sections.h>
29
30 #define CORE_STR "CORE"
31
32 #ifndef ELF_CORE_EFLAGS
33 #define ELF_CORE_EFLAGS 0
34 #endif
35
36 static struct proc_dir_entry *proc_root_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 /* An ELF note in memory */
47 struct memelfnote
48 {
49         const char *name;
50         int type;
51         unsigned int datasz;
52         void *data;
53 };
54
55 static LIST_HEAD(kclist_head);
56 static DEFINE_RWLOCK(kclist_lock);
57 static int kcore_need_update = 1;
58
59 void
60 kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
61 {
62         new->addr = (unsigned long)addr;
63         new->size = size;
64         new->type = type;
65
66         write_lock(&kclist_lock);
67         list_add_tail(&new->list, &kclist_head);
68         write_unlock(&kclist_lock);
69 }
70
71 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
72 {
73         size_t try, size;
74         struct kcore_list *m;
75
76         *nphdr = 1; /* PT_NOTE */
77         size = 0;
78
79         list_for_each_entry(m, &kclist_head, list) {
80                 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
81                 if (try > size)
82                         size = try;
83                 *nphdr = *nphdr + 1;
84         }
85         *elf_buflen =   sizeof(struct elfhdr) + 
86                         (*nphdr + 2)*sizeof(struct elf_phdr) + 
87                         3 * ((sizeof(struct elf_note)) +
88                              roundup(sizeof(CORE_STR), 4)) +
89                         roundup(sizeof(struct elf_prstatus), 4) +
90                         roundup(sizeof(struct elf_prpsinfo), 4) +
91                         roundup(sizeof(struct task_struct), 4);
92         *elf_buflen = PAGE_ALIGN(*elf_buflen);
93         return size + *elf_buflen;
94 }
95
96 static void free_kclist_ents(struct list_head *head)
97 {
98         struct kcore_list *tmp, *pos;
99
100         list_for_each_entry_safe(pos, tmp, head, list) {
101                 list_del(&pos->list);
102                 kfree(pos);
103         }
104 }
105 /*
106  * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
107  */
108 static void __kcore_update_ram(struct list_head *list)
109 {
110         int nphdr;
111         size_t size;
112         struct kcore_list *tmp, *pos;
113         LIST_HEAD(garbage);
114
115         write_lock(&kclist_lock);
116         if (kcore_need_update) {
117                 list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
118                         if (pos->type == KCORE_RAM
119                                 || pos->type == KCORE_VMEMMAP)
120                                 list_move(&pos->list, &garbage);
121                 }
122                 list_splice_tail(list, &kclist_head);
123         } else
124                 list_splice(list, &garbage);
125         kcore_need_update = 0;
126         proc_root_kcore->size = get_kcore_size(&nphdr, &size);
127         write_unlock(&kclist_lock);
128
129         free_kclist_ents(&garbage);
130 }
131
132
133 #ifdef CONFIG_HIGHMEM
134 /*
135  * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
136  * because memory hole is not as big as !HIGHMEM case.
137  * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
138  */
139 static int kcore_update_ram(void)
140 {
141         LIST_HEAD(head);
142         struct kcore_list *ent;
143         int ret = 0;
144
145         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
146         if (!ent)
147                 return -ENOMEM;
148         ent->addr = (unsigned long)__va(0);
149         ent->size = max_low_pfn << PAGE_SHIFT;
150         ent->type = KCORE_RAM;
151         list_add(&ent->list, &head);
152         __kcore_update_ram(&head);
153         return ret;
154 }
155
156 #else /* !CONFIG_HIGHMEM */
157
158 #ifdef CONFIG_SPARSEMEM_VMEMMAP
159 /* calculate vmemmap's address from given system ram pfn and register it */
160 int get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
161 {
162         unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
163         unsigned long nr_pages = ent->size >> PAGE_SHIFT;
164         unsigned long start, end;
165         struct kcore_list *vmm, *tmp;
166
167
168         start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
169         end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
170         end = ALIGN(end, PAGE_SIZE);
171         /* overlap check (because we have to align page */
172         list_for_each_entry(tmp, head, list) {
173                 if (tmp->type != KCORE_VMEMMAP)
174                         continue;
175                 if (start < tmp->addr + tmp->size)
176                         if (end > tmp->addr)
177                                 end = tmp->addr;
178         }
179         if (start < end) {
180                 vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
181                 if (!vmm)
182                         return 0;
183                 vmm->addr = start;
184                 vmm->size = end - start;
185                 vmm->type = KCORE_VMEMMAP;
186                 list_add_tail(&vmm->list, head);
187         }
188         return 1;
189
190 }
191 #else
192 int get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
193 {
194         return 1;
195 }
196
197 #endif
198
199 static int
200 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
201 {
202         struct list_head *head = (struct list_head *)arg;
203         struct kcore_list *ent;
204
205         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
206         if (!ent)
207                 return -ENOMEM;
208         ent->addr = (unsigned long)__va((pfn << PAGE_SHIFT));
209         ent->size = nr_pages << PAGE_SHIFT;
210
211         /* Sanity check: Can happen in 32bit arch...maybe */
212         if (ent->addr < (unsigned long) __va(0))
213                 goto free_out;
214
215         /* cut not-mapped area. ....from ppc-32 code. */
216         if (ULONG_MAX - ent->addr < ent->size)
217                 ent->size = ULONG_MAX - ent->addr;
218
219         /* cut when vmalloc() area is higher than direct-map area */
220         if (VMALLOC_START > (unsigned long)__va(0)) {
221                 if (ent->addr > VMALLOC_START)
222                         goto free_out;
223                 if (VMALLOC_START - ent->addr < ent->size)
224                         ent->size = VMALLOC_START - ent->addr;
225         }
226
227         ent->type = KCORE_RAM;
228         list_add_tail(&ent->list, head);
229
230         if (!get_sparsemem_vmemmap_info(ent, head)) {
231                 list_del(&ent->list);
232                 goto free_out;
233         }
234
235         return 0;
236 free_out:
237         kfree(ent);
238         return 1;
239 }
240
241 static int kcore_update_ram(void)
242 {
243         int nid, ret;
244         unsigned long end_pfn;
245         LIST_HEAD(head);
246
247         /* Not inialized....update now */
248         /* find out "max pfn" */
249         end_pfn = 0;
250         for_each_node_state(nid, N_HIGH_MEMORY) {
251                 unsigned long node_end;
252                 node_end  = NODE_DATA(nid)->node_start_pfn +
253                         NODE_DATA(nid)->node_spanned_pages;
254                 if (end_pfn < node_end)
255                         end_pfn = node_end;
256         }
257         /* scan 0 to max_pfn */
258         ret = walk_system_ram_range(0, end_pfn, &head, kclist_add_private);
259         if (ret) {
260                 free_kclist_ents(&head);
261                 return -ENOMEM;
262         }
263         __kcore_update_ram(&head);
264         return ret;
265 }
266 #endif /* CONFIG_HIGHMEM */
267
268 /*****************************************************************************/
269 /*
270  * determine size of ELF note
271  */
272 static int notesize(struct memelfnote *en)
273 {
274         int sz;
275
276         sz = sizeof(struct elf_note);
277         sz += roundup((strlen(en->name) + 1), 4);
278         sz += roundup(en->datasz, 4);
279
280         return sz;
281 } /* end notesize() */
282
283 /*****************************************************************************/
284 /*
285  * store a note in the header buffer
286  */
287 static char *storenote(struct memelfnote *men, char *bufp)
288 {
289         struct elf_note en;
290
291 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
292
293         en.n_namesz = strlen(men->name) + 1;
294         en.n_descsz = men->datasz;
295         en.n_type = men->type;
296
297         DUMP_WRITE(&en, sizeof(en));
298         DUMP_WRITE(men->name, en.n_namesz);
299
300         /* XXX - cast from long long to long to avoid need for libgcc.a */
301         bufp = (char*) roundup((unsigned long)bufp,4);
302         DUMP_WRITE(men->data, men->datasz);
303         bufp = (char*) roundup((unsigned long)bufp,4);
304
305 #undef DUMP_WRITE
306
307         return bufp;
308 } /* end storenote() */
309
310 /*
311  * store an ELF coredump header in the supplied buffer
312  * nphdr is the number of elf_phdr to insert
313  */
314 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
315 {
316         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
317         struct elf_prpsinfo prpsinfo;   /* NT_PRPSINFO */
318         struct elf_phdr *nhdr, *phdr;
319         struct elfhdr *elf;
320         struct memelfnote notes[3];
321         off_t offset = 0;
322         struct kcore_list *m;
323
324         /* setup ELF header */
325         elf = (struct elfhdr *) bufp;
326         bufp += sizeof(struct elfhdr);
327         offset += sizeof(struct elfhdr);
328         memcpy(elf->e_ident, ELFMAG, SELFMAG);
329         elf->e_ident[EI_CLASS]  = ELF_CLASS;
330         elf->e_ident[EI_DATA]   = ELF_DATA;
331         elf->e_ident[EI_VERSION]= EV_CURRENT;
332         elf->e_ident[EI_OSABI] = ELF_OSABI;
333         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
334         elf->e_type     = ET_CORE;
335         elf->e_machine  = ELF_ARCH;
336         elf->e_version  = EV_CURRENT;
337         elf->e_entry    = 0;
338         elf->e_phoff    = sizeof(struct elfhdr);
339         elf->e_shoff    = 0;
340         elf->e_flags    = ELF_CORE_EFLAGS;
341         elf->e_ehsize   = sizeof(struct elfhdr);
342         elf->e_phentsize= sizeof(struct elf_phdr);
343         elf->e_phnum    = nphdr;
344         elf->e_shentsize= 0;
345         elf->e_shnum    = 0;
346         elf->e_shstrndx = 0;
347
348         /* setup ELF PT_NOTE program header */
349         nhdr = (struct elf_phdr *) bufp;
350         bufp += sizeof(struct elf_phdr);
351         offset += sizeof(struct elf_phdr);
352         nhdr->p_type    = PT_NOTE;
353         nhdr->p_offset  = 0;
354         nhdr->p_vaddr   = 0;
355         nhdr->p_paddr   = 0;
356         nhdr->p_filesz  = 0;
357         nhdr->p_memsz   = 0;
358         nhdr->p_flags   = 0;
359         nhdr->p_align   = 0;
360
361         /* setup ELF PT_LOAD program header for every area */
362         list_for_each_entry(m, &kclist_head, list) {
363                 phdr = (struct elf_phdr *) bufp;
364                 bufp += sizeof(struct elf_phdr);
365                 offset += sizeof(struct elf_phdr);
366
367                 phdr->p_type    = PT_LOAD;
368                 phdr->p_flags   = PF_R|PF_W|PF_X;
369                 phdr->p_offset  = kc_vaddr_to_offset(m->addr) + dataoff;
370                 phdr->p_vaddr   = (size_t)m->addr;
371                 phdr->p_paddr   = 0;
372                 phdr->p_filesz  = phdr->p_memsz = m->size;
373                 phdr->p_align   = PAGE_SIZE;
374         }
375
376         /*
377          * Set up the notes in similar form to SVR4 core dumps made
378          * with info from their /proc.
379          */
380         nhdr->p_offset  = offset;
381
382         /* set up the process status */
383         notes[0].name = CORE_STR;
384         notes[0].type = NT_PRSTATUS;
385         notes[0].datasz = sizeof(struct elf_prstatus);
386         notes[0].data = &prstatus;
387
388         memset(&prstatus, 0, sizeof(struct elf_prstatus));
389
390         nhdr->p_filesz  = notesize(&notes[0]);
391         bufp = storenote(&notes[0], bufp);
392
393         /* set up the process info */
394         notes[1].name   = CORE_STR;
395         notes[1].type   = NT_PRPSINFO;
396         notes[1].datasz = sizeof(struct elf_prpsinfo);
397         notes[1].data   = &prpsinfo;
398
399         memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
400         prpsinfo.pr_state       = 0;
401         prpsinfo.pr_sname       = 'R';
402         prpsinfo.pr_zomb        = 0;
403
404         strcpy(prpsinfo.pr_fname, "vmlinux");
405         strncpy(prpsinfo.pr_psargs, saved_command_line, ELF_PRARGSZ);
406
407         nhdr->p_filesz  += notesize(&notes[1]);
408         bufp = storenote(&notes[1], bufp);
409
410         /* set up the task structure */
411         notes[2].name   = CORE_STR;
412         notes[2].type   = NT_TASKSTRUCT;
413         notes[2].datasz = sizeof(struct task_struct);
414         notes[2].data   = current;
415
416         nhdr->p_filesz  += notesize(&notes[2]);
417         bufp = storenote(&notes[2], bufp);
418
419 } /* end elf_kcore_store_hdr() */
420
421 /*****************************************************************************/
422 /*
423  * read from the ELF header and then kernel memory
424  */
425 static ssize_t
426 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
427 {
428         ssize_t acc = 0;
429         size_t size, tsz;
430         size_t elf_buflen;
431         int nphdr;
432         unsigned long start;
433
434         read_lock(&kclist_lock);
435         size = get_kcore_size(&nphdr, &elf_buflen);
436
437         if (buflen == 0 || *fpos >= size) {
438                 read_unlock(&kclist_lock);
439                 return 0;
440         }
441
442         /* trim buflen to not go beyond EOF */
443         if (buflen > size - *fpos)
444                 buflen = size - *fpos;
445
446         /* construct an ELF core header if we'll need some of it */
447         if (*fpos < elf_buflen) {
448                 char * elf_buf;
449
450                 tsz = elf_buflen - *fpos;
451                 if (buflen < tsz)
452                         tsz = buflen;
453                 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
454                 if (!elf_buf) {
455                         read_unlock(&kclist_lock);
456                         return -ENOMEM;
457                 }
458                 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
459                 read_unlock(&kclist_lock);
460                 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
461                         kfree(elf_buf);
462                         return -EFAULT;
463                 }
464                 kfree(elf_buf);
465                 buflen -= tsz;
466                 *fpos += tsz;
467                 buffer += tsz;
468                 acc += tsz;
469
470                 /* leave now if filled buffer already */
471                 if (buflen == 0)
472                         return acc;
473         } else
474                 read_unlock(&kclist_lock);
475
476         /*
477          * Check to see if our file offset matches with any of
478          * the addresses in the elf_phdr on our list.
479          */
480         start = kc_offset_to_vaddr(*fpos - elf_buflen);
481         if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
482                 tsz = buflen;
483                 
484         while (buflen) {
485                 struct kcore_list *m;
486
487                 read_lock(&kclist_lock);
488                 list_for_each_entry(m, &kclist_head, list) {
489                         if (start >= m->addr && start < (m->addr+m->size))
490                                 break;
491                 }
492                 read_unlock(&kclist_lock);
493
494                 if (&m->list == &kclist_head) {
495                         if (clear_user(buffer, tsz))
496                                 return -EFAULT;
497                 } else if (is_vmalloc_or_module_addr((void *)start)) {
498                         char * elf_buf;
499
500                         elf_buf = kzalloc(tsz, GFP_KERNEL);
501                         if (!elf_buf)
502                                 return -ENOMEM;
503                         vread(elf_buf, (char *)start, tsz);
504                         /* we have to zero-fill user buffer even if no read */
505                         if (copy_to_user(buffer, elf_buf, tsz)) {
506                                 kfree(elf_buf);
507                                 return -EFAULT;
508                         }
509                         kfree(elf_buf);
510                 } else {
511                         if (kern_addr_valid(start)) {
512                                 unsigned long n;
513
514                                 n = copy_to_user(buffer, (char *)start, tsz);
515                                 /*
516                                  * We cannot distingush between fault on source
517                                  * and fault on destination. When this happens
518                                  * we clear too and hope it will trigger the
519                                  * EFAULT again.
520                                  */
521                                 if (n) { 
522                                         if (clear_user(buffer + tsz - n,
523                                                                 n))
524                                                 return -EFAULT;
525                                 }
526                         } else {
527                                 if (clear_user(buffer, tsz))
528                                         return -EFAULT;
529                         }
530                 }
531                 buflen -= tsz;
532                 *fpos += tsz;
533                 buffer += tsz;
534                 acc += tsz;
535                 start += tsz;
536                 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
537         }
538
539         return acc;
540 }
541
542
543 static int open_kcore(struct inode *inode, struct file *filp)
544 {
545         if (!capable(CAP_SYS_RAWIO))
546                 return -EPERM;
547         if (kcore_need_update)
548                 kcore_update_ram();
549         if (i_size_read(inode) != proc_root_kcore->size) {
550                 mutex_lock(&inode->i_mutex);
551                 i_size_write(inode, proc_root_kcore->size);
552                 mutex_unlock(&inode->i_mutex);
553         }
554         return 0;
555 }
556
557
558 static const struct file_operations proc_kcore_operations = {
559         .read           = read_kcore,
560         .open           = open_kcore,
561         .llseek         = generic_file_llseek,
562 };
563
564 #ifdef CONFIG_MEMORY_HOTPLUG
565 /* just remember that we have to update kcore */
566 static int __meminit kcore_callback(struct notifier_block *self,
567                                     unsigned long action, void *arg)
568 {
569         switch (action) {
570         case MEM_ONLINE:
571         case MEM_OFFLINE:
572                 write_lock(&kclist_lock);
573                 kcore_need_update = 1;
574                 write_unlock(&kclist_lock);
575         }
576         return NOTIFY_OK;
577 }
578 #endif
579
580
581 static struct kcore_list kcore_vmalloc;
582
583 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
584 static struct kcore_list kcore_text;
585 /*
586  * If defined, special segment is used for mapping kernel text instead of
587  * direct-map area. We need to create special TEXT section.
588  */
589 static void __init proc_kcore_text_init(void)
590 {
591         kclist_add(&kcore_text, _stext, _end - _stext, KCORE_TEXT);
592 }
593 #else
594 static void __init proc_kcore_text_init(void)
595 {
596 }
597 #endif
598
599 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
600 /*
601  * MODULES_VADDR has no intersection with VMALLOC_ADDR.
602  */
603 struct kcore_list kcore_modules;
604 static void __init add_modules_range(void)
605 {
606         kclist_add(&kcore_modules, (void *)MODULES_VADDR,
607                         MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
608 }
609 #else
610 static void __init add_modules_range(void)
611 {
612 }
613 #endif
614
615 static int __init proc_kcore_init(void)
616 {
617         proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
618                                       &proc_kcore_operations);
619         if (!proc_root_kcore) {
620                 printk(KERN_ERR "couldn't create /proc/kcore\n");
621                 return 0; /* Always returns 0. */
622         }
623         /* Store text area if it's special */
624         proc_kcore_text_init();
625         /* Store vmalloc area */
626         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
627                 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
628         add_modules_range();
629         /* Store direct-map area from physical memory map */
630         kcore_update_ram();
631         hotplug_memory_notifier(kcore_callback, 0);
632
633         return 0;
634 }
635 module_init(proc_kcore_init);