Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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/kcore.h>
15 #include <linux/user.h>
16 #include <linux/capability.h>
17 #include <linux/elf.h>
18 #include <linux/elfcore.h>
19 #include <linux/notifier.h>
20 #include <linux/vmalloc.h>
21 #include <linux/highmem.h>
22 #include <linux/printk.h>
23 #include <linux/bootmem.h>
24 #include <linux/init.h>
25 #include <linux/slab.h>
26 #include <asm/uaccess.h>
27 #include <asm/io.h>
28 #include <linux/list.h>
29 #include <linux/ioport.h>
30 #include <linux/memory.h>
31 #include <asm/sections.h>
32 #include "internal.h"
33
34 #define CORE_STR "CORE"
35
36 #ifndef ELF_CORE_EFLAGS
37 #define ELF_CORE_EFLAGS 0
38 #endif
39
40 static struct proc_dir_entry *proc_root_kcore;
41
42
43 #ifndef kc_vaddr_to_offset
44 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
45 #endif
46 #ifndef kc_offset_to_vaddr
47 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
48 #endif
49
50 /* An ELF note in memory */
51 struct memelfnote
52 {
53         const char *name;
54         int type;
55         unsigned int datasz;
56         void *data;
57 };
58
59 static LIST_HEAD(kclist_head);
60 static DEFINE_RWLOCK(kclist_lock);
61 static int kcore_need_update = 1;
62
63 void
64 kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
65 {
66         new->addr = (unsigned long)addr;
67         new->size = size;
68         new->type = type;
69
70         write_lock(&kclist_lock);
71         list_add_tail(&new->list, &kclist_head);
72         write_unlock(&kclist_lock);
73 }
74
75 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
76 {
77         size_t try, size;
78         struct kcore_list *m;
79
80         *nphdr = 1; /* PT_NOTE */
81         size = 0;
82
83         list_for_each_entry(m, &kclist_head, list) {
84                 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
85                 if (try > size)
86                         size = try;
87                 *nphdr = *nphdr + 1;
88         }
89         *elf_buflen =   sizeof(struct elfhdr) + 
90                         (*nphdr + 2)*sizeof(struct elf_phdr) + 
91                         3 * ((sizeof(struct elf_note)) +
92                              roundup(sizeof(CORE_STR), 4)) +
93                         roundup(sizeof(struct elf_prstatus), 4) +
94                         roundup(sizeof(struct elf_prpsinfo), 4) +
95                         roundup(sizeof(struct task_struct), 4);
96         *elf_buflen = PAGE_ALIGN(*elf_buflen);
97         return size + *elf_buflen;
98 }
99
100 static void free_kclist_ents(struct list_head *head)
101 {
102         struct kcore_list *tmp, *pos;
103
104         list_for_each_entry_safe(pos, tmp, head, list) {
105                 list_del(&pos->list);
106                 kfree(pos);
107         }
108 }
109 /*
110  * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
111  */
112 static void __kcore_update_ram(struct list_head *list)
113 {
114         int nphdr;
115         size_t size;
116         struct kcore_list *tmp, *pos;
117         LIST_HEAD(garbage);
118
119         write_lock(&kclist_lock);
120         if (kcore_need_update) {
121                 list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
122                         if (pos->type == KCORE_RAM
123                                 || pos->type == KCORE_VMEMMAP)
124                                 list_move(&pos->list, &garbage);
125                 }
126                 list_splice_tail(list, &kclist_head);
127         } else
128                 list_splice(list, &garbage);
129         kcore_need_update = 0;
130         proc_root_kcore->size = get_kcore_size(&nphdr, &size);
131         write_unlock(&kclist_lock);
132
133         free_kclist_ents(&garbage);
134 }
135
136
137 #ifdef CONFIG_HIGHMEM
138 /*
139  * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
140  * because memory hole is not as big as !HIGHMEM case.
141  * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
142  */
143 static int kcore_update_ram(void)
144 {
145         LIST_HEAD(head);
146         struct kcore_list *ent;
147         int ret = 0;
148
149         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
150         if (!ent)
151                 return -ENOMEM;
152         ent->addr = (unsigned long)__va(0);
153         ent->size = max_low_pfn << PAGE_SHIFT;
154         ent->type = KCORE_RAM;
155         list_add(&ent->list, &head);
156         __kcore_update_ram(&head);
157         return ret;
158 }
159
160 #else /* !CONFIG_HIGHMEM */
161
162 #ifdef CONFIG_SPARSEMEM_VMEMMAP
163 /* calculate vmemmap's address from given system ram pfn and register it */
164 static int
165 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
166 {
167         unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
168         unsigned long nr_pages = ent->size >> PAGE_SHIFT;
169         unsigned long start, end;
170         struct kcore_list *vmm, *tmp;
171
172
173         start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
174         end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
175         end = PAGE_ALIGN(end);
176         /* overlap check (because we have to align page */
177         list_for_each_entry(tmp, head, list) {
178                 if (tmp->type != KCORE_VMEMMAP)
179                         continue;
180                 if (start < tmp->addr + tmp->size)
181                         if (end > tmp->addr)
182                                 end = tmp->addr;
183         }
184         if (start < end) {
185                 vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
186                 if (!vmm)
187                         return 0;
188                 vmm->addr = start;
189                 vmm->size = end - start;
190                 vmm->type = KCORE_VMEMMAP;
191                 list_add_tail(&vmm->list, head);
192         }
193         return 1;
194
195 }
196 #else
197 static int
198 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
199 {
200         return 1;
201 }
202
203 #endif
204
205 static int
206 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
207 {
208         struct list_head *head = (struct list_head *)arg;
209         struct kcore_list *ent;
210
211         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
212         if (!ent)
213                 return -ENOMEM;
214         ent->addr = (unsigned long)__va((pfn << PAGE_SHIFT));
215         ent->size = nr_pages << PAGE_SHIFT;
216
217         /* Sanity check: Can happen in 32bit arch...maybe */
218         if (ent->addr < (unsigned long) __va(0))
219                 goto free_out;
220
221         /* cut not-mapped area. ....from ppc-32 code. */
222         if (ULONG_MAX - ent->addr < ent->size)
223                 ent->size = ULONG_MAX - ent->addr;
224
225         /* cut when vmalloc() area is higher than direct-map area */
226         if (VMALLOC_START > (unsigned long)__va(0)) {
227                 if (ent->addr > VMALLOC_START)
228                         goto free_out;
229                 if (VMALLOC_START - ent->addr < ent->size)
230                         ent->size = VMALLOC_START - ent->addr;
231         }
232
233         ent->type = KCORE_RAM;
234         list_add_tail(&ent->list, head);
235
236         if (!get_sparsemem_vmemmap_info(ent, head)) {
237                 list_del(&ent->list);
238                 goto free_out;
239         }
240
241         return 0;
242 free_out:
243         kfree(ent);
244         return 1;
245 }
246
247 static int kcore_update_ram(void)
248 {
249         int nid, ret;
250         unsigned long end_pfn;
251         LIST_HEAD(head);
252
253         /* Not inialized....update now */
254         /* find out "max pfn" */
255         end_pfn = 0;
256         for_each_node_state(nid, N_MEMORY) {
257                 unsigned long node_end;
258                 node_end = node_end_pfn(nid);
259                 if (end_pfn < node_end)
260                         end_pfn = node_end;
261         }
262         /* scan 0 to max_pfn */
263         ret = walk_system_ram_range(0, end_pfn, &head, kclist_add_private);
264         if (ret) {
265                 free_kclist_ents(&head);
266                 return -ENOMEM;
267         }
268         __kcore_update_ram(&head);
269         return ret;
270 }
271 #endif /* CONFIG_HIGHMEM */
272
273 /*****************************************************************************/
274 /*
275  * determine size of ELF note
276  */
277 static int notesize(struct memelfnote *en)
278 {
279         int sz;
280
281         sz = sizeof(struct elf_note);
282         sz += roundup((strlen(en->name) + 1), 4);
283         sz += roundup(en->datasz, 4);
284
285         return sz;
286 } /* end notesize() */
287
288 /*****************************************************************************/
289 /*
290  * store a note in the header buffer
291  */
292 static char *storenote(struct memelfnote *men, char *bufp)
293 {
294         struct elf_note en;
295
296 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
297
298         en.n_namesz = strlen(men->name) + 1;
299         en.n_descsz = men->datasz;
300         en.n_type = men->type;
301
302         DUMP_WRITE(&en, sizeof(en));
303         DUMP_WRITE(men->name, en.n_namesz);
304
305         /* XXX - cast from long long to long to avoid need for libgcc.a */
306         bufp = (char*) roundup((unsigned long)bufp,4);
307         DUMP_WRITE(men->data, men->datasz);
308         bufp = (char*) roundup((unsigned long)bufp,4);
309
310 #undef DUMP_WRITE
311
312         return bufp;
313 } /* end storenote() */
314
315 /*
316  * store an ELF coredump header in the supplied buffer
317  * nphdr is the number of elf_phdr to insert
318  */
319 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
320 {
321         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
322         struct elf_prpsinfo prpsinfo;   /* NT_PRPSINFO */
323         struct elf_phdr *nhdr, *phdr;
324         struct elfhdr *elf;
325         struct memelfnote notes[3];
326         off_t offset = 0;
327         struct kcore_list *m;
328
329         /* setup ELF header */
330         elf = (struct elfhdr *) bufp;
331         bufp += sizeof(struct elfhdr);
332         offset += sizeof(struct elfhdr);
333         memcpy(elf->e_ident, ELFMAG, SELFMAG);
334         elf->e_ident[EI_CLASS]  = ELF_CLASS;
335         elf->e_ident[EI_DATA]   = ELF_DATA;
336         elf->e_ident[EI_VERSION]= EV_CURRENT;
337         elf->e_ident[EI_OSABI] = ELF_OSABI;
338         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
339         elf->e_type     = ET_CORE;
340         elf->e_machine  = ELF_ARCH;
341         elf->e_version  = EV_CURRENT;
342         elf->e_entry    = 0;
343         elf->e_phoff    = sizeof(struct elfhdr);
344         elf->e_shoff    = 0;
345         elf->e_flags    = ELF_CORE_EFLAGS;
346         elf->e_ehsize   = sizeof(struct elfhdr);
347         elf->e_phentsize= sizeof(struct elf_phdr);
348         elf->e_phnum    = nphdr;
349         elf->e_shentsize= 0;
350         elf->e_shnum    = 0;
351         elf->e_shstrndx = 0;
352
353         /* setup ELF PT_NOTE program header */
354         nhdr = (struct elf_phdr *) bufp;
355         bufp += sizeof(struct elf_phdr);
356         offset += sizeof(struct elf_phdr);
357         nhdr->p_type    = PT_NOTE;
358         nhdr->p_offset  = 0;
359         nhdr->p_vaddr   = 0;
360         nhdr->p_paddr   = 0;
361         nhdr->p_filesz  = 0;
362         nhdr->p_memsz   = 0;
363         nhdr->p_flags   = 0;
364         nhdr->p_align   = 0;
365
366         /* setup ELF PT_LOAD program header for every area */
367         list_for_each_entry(m, &kclist_head, list) {
368                 phdr = (struct elf_phdr *) bufp;
369                 bufp += sizeof(struct elf_phdr);
370                 offset += sizeof(struct elf_phdr);
371
372                 phdr->p_type    = PT_LOAD;
373                 phdr->p_flags   = PF_R|PF_W|PF_X;
374                 phdr->p_offset  = kc_vaddr_to_offset(m->addr) + dataoff;
375                 phdr->p_vaddr   = (size_t)m->addr;
376                 phdr->p_paddr   = 0;
377                 phdr->p_filesz  = phdr->p_memsz = m->size;
378                 phdr->p_align   = PAGE_SIZE;
379         }
380
381         /*
382          * Set up the notes in similar form to SVR4 core dumps made
383          * with info from their /proc.
384          */
385         nhdr->p_offset  = offset;
386
387         /* set up the process status */
388         notes[0].name = CORE_STR;
389         notes[0].type = NT_PRSTATUS;
390         notes[0].datasz = sizeof(struct elf_prstatus);
391         notes[0].data = &prstatus;
392
393         memset(&prstatus, 0, sizeof(struct elf_prstatus));
394
395         nhdr->p_filesz  = notesize(&notes[0]);
396         bufp = storenote(&notes[0], bufp);
397
398         /* set up the process info */
399         notes[1].name   = CORE_STR;
400         notes[1].type   = NT_PRPSINFO;
401         notes[1].datasz = sizeof(struct elf_prpsinfo);
402         notes[1].data   = &prpsinfo;
403
404         memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
405         prpsinfo.pr_state       = 0;
406         prpsinfo.pr_sname       = 'R';
407         prpsinfo.pr_zomb        = 0;
408
409         strcpy(prpsinfo.pr_fname, "vmlinux");
410         strlcpy(prpsinfo.pr_psargs, saved_command_line, sizeof(prpsinfo.pr_psargs));
411
412         nhdr->p_filesz  += notesize(&notes[1]);
413         bufp = storenote(&notes[1], bufp);
414
415         /* set up the task structure */
416         notes[2].name   = CORE_STR;
417         notes[2].type   = NT_TASKSTRUCT;
418         notes[2].datasz = sizeof(struct task_struct);
419         notes[2].data   = current;
420
421         nhdr->p_filesz  += notesize(&notes[2]);
422         bufp = storenote(&notes[2], bufp);
423
424 } /* end elf_kcore_store_hdr() */
425
426 /*****************************************************************************/
427 /*
428  * read from the ELF header and then kernel memory
429  */
430 static ssize_t
431 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
432 {
433         ssize_t acc = 0;
434         size_t size, tsz;
435         size_t elf_buflen;
436         int nphdr;
437         unsigned long start;
438
439         read_lock(&kclist_lock);
440         size = get_kcore_size(&nphdr, &elf_buflen);
441
442         if (buflen == 0 || *fpos >= size) {
443                 read_unlock(&kclist_lock);
444                 return 0;
445         }
446
447         /* trim buflen to not go beyond EOF */
448         if (buflen > size - *fpos)
449                 buflen = size - *fpos;
450
451         /* construct an ELF core header if we'll need some of it */
452         if (*fpos < elf_buflen) {
453                 char * elf_buf;
454
455                 tsz = elf_buflen - *fpos;
456                 if (buflen < tsz)
457                         tsz = buflen;
458                 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
459                 if (!elf_buf) {
460                         read_unlock(&kclist_lock);
461                         return -ENOMEM;
462                 }
463                 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
464                 read_unlock(&kclist_lock);
465                 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
466                         kfree(elf_buf);
467                         return -EFAULT;
468                 }
469                 kfree(elf_buf);
470                 buflen -= tsz;
471                 *fpos += tsz;
472                 buffer += tsz;
473                 acc += tsz;
474
475                 /* leave now if filled buffer already */
476                 if (buflen == 0)
477                         return acc;
478         } else
479                 read_unlock(&kclist_lock);
480
481         /*
482          * Check to see if our file offset matches with any of
483          * the addresses in the elf_phdr on our list.
484          */
485         start = kc_offset_to_vaddr(*fpos - elf_buflen);
486         if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
487                 tsz = buflen;
488                 
489         while (buflen) {
490                 struct kcore_list *m;
491
492                 read_lock(&kclist_lock);
493                 list_for_each_entry(m, &kclist_head, list) {
494                         if (start >= m->addr && start < (m->addr+m->size))
495                                 break;
496                 }
497                 read_unlock(&kclist_lock);
498
499                 if (&m->list == &kclist_head) {
500                         if (clear_user(buffer, tsz))
501                                 return -EFAULT;
502                 } else if (is_vmalloc_or_module_addr((void *)start)) {
503                         char * elf_buf;
504
505                         elf_buf = kzalloc(tsz, GFP_KERNEL);
506                         if (!elf_buf)
507                                 return -ENOMEM;
508                         vread(elf_buf, (char *)start, tsz);
509                         /* we have to zero-fill user buffer even if no read */
510                         if (copy_to_user(buffer, elf_buf, tsz)) {
511                                 kfree(elf_buf);
512                                 return -EFAULT;
513                         }
514                         kfree(elf_buf);
515                 } else {
516                         if (kern_addr_valid(start)) {
517                                 unsigned long n;
518
519                                 n = copy_to_user(buffer, (char *)start, tsz);
520                                 /*
521                                  * We cannot distinguish between fault on source
522                                  * and fault on destination. When this happens
523                                  * we clear too and hope it will trigger the
524                                  * EFAULT again.
525                                  */
526                                 if (n) { 
527                                         if (clear_user(buffer + tsz - n,
528                                                                 n))
529                                                 return -EFAULT;
530                                 }
531                         } else {
532                                 if (clear_user(buffer, tsz))
533                                         return -EFAULT;
534                         }
535                 }
536                 buflen -= tsz;
537                 *fpos += tsz;
538                 buffer += tsz;
539                 acc += tsz;
540                 start += tsz;
541                 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
542         }
543
544         return acc;
545 }
546
547
548 static int open_kcore(struct inode *inode, struct file *filp)
549 {
550         if (!capable(CAP_SYS_RAWIO))
551                 return -EPERM;
552         if (kcore_need_update)
553                 kcore_update_ram();
554         if (i_size_read(inode) != proc_root_kcore->size) {
555                 mutex_lock(&inode->i_mutex);
556                 i_size_write(inode, proc_root_kcore->size);
557                 mutex_unlock(&inode->i_mutex);
558         }
559         return 0;
560 }
561
562
563 static const struct file_operations proc_kcore_operations = {
564         .read           = read_kcore,
565         .open           = open_kcore,
566         .llseek         = default_llseek,
567 };
568
569 /* just remember that we have to update kcore */
570 static int __meminit kcore_callback(struct notifier_block *self,
571                                     unsigned long action, void *arg)
572 {
573         switch (action) {
574         case MEM_ONLINE:
575         case MEM_OFFLINE:
576                 write_lock(&kclist_lock);
577                 kcore_need_update = 1;
578                 write_unlock(&kclist_lock);
579         }
580         return NOTIFY_OK;
581 }
582
583 static struct notifier_block kcore_callback_nb __meminitdata = {
584         .notifier_call = kcore_callback,
585         .priority = 0,
586 };
587
588 static struct kcore_list kcore_vmalloc;
589
590 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
591 static struct kcore_list kcore_text;
592 /*
593  * If defined, special segment is used for mapping kernel text instead of
594  * direct-map area. We need to create special TEXT section.
595  */
596 static void __init proc_kcore_text_init(void)
597 {
598         kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
599 }
600 #else
601 static void __init proc_kcore_text_init(void)
602 {
603 }
604 #endif
605
606 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
607 /*
608  * MODULES_VADDR has no intersection with VMALLOC_ADDR.
609  */
610 struct kcore_list kcore_modules;
611 static void __init add_modules_range(void)
612 {
613         if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
614                 kclist_add(&kcore_modules, (void *)MODULES_VADDR,
615                         MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
616         }
617 }
618 #else
619 static void __init add_modules_range(void)
620 {
621 }
622 #endif
623
624 static int __init proc_kcore_init(void)
625 {
626         proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
627                                       &proc_kcore_operations);
628         if (!proc_root_kcore) {
629                 pr_err("couldn't create /proc/kcore\n");
630                 return 0; /* Always returns 0. */
631         }
632         /* Store text area if it's special */
633         proc_kcore_text_init();
634         /* Store vmalloc area */
635         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
636                 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
637         add_modules_range();
638         /* Store direct-map area from physical memory map */
639         kcore_update_ram();
640         register_hotmemory_notifier(&kcore_callback_nb);
641
642         return 0;
643 }
644 fs_initcall(proc_kcore_init);