Linux 4.8.3
[linux/fpc-iii.git] / fs / proc / kcore.c
blob5c89a07e3d7f540b48757b907557e5f5df300ff2
1 /*
2 * fs/proc/kcore.c kernel ELF core dumper
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>
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"
34 #define CORE_STR "CORE"
36 #ifndef ELF_CORE_EFLAGS
37 #define ELF_CORE_EFLAGS 0
38 #endif
40 static struct proc_dir_entry *proc_root_kcore;
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
50 /* An ELF note in memory */
51 struct memelfnote
53 const char *name;
54 int type;
55 unsigned int datasz;
56 void *data;
59 static LIST_HEAD(kclist_head);
60 static DEFINE_RWLOCK(kclist_lock);
61 static int kcore_need_update = 1;
63 void
64 kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
66 new->addr = (unsigned long)addr;
67 new->size = size;
68 new->type = type;
70 write_lock(&kclist_lock);
71 list_add_tail(&new->list, &kclist_head);
72 write_unlock(&kclist_lock);
75 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
77 size_t try, size;
78 struct kcore_list *m;
80 *nphdr = 1; /* PT_NOTE */
81 size = 0;
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;
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(arch_task_struct_size, 4);
96 *elf_buflen = PAGE_ALIGN(*elf_buflen);
97 return size + *elf_buflen;
100 static void free_kclist_ents(struct list_head *head)
102 struct kcore_list *tmp, *pos;
104 list_for_each_entry_safe(pos, tmp, head, list) {
105 list_del(&pos->list);
106 kfree(pos);
110 * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
112 static void __kcore_update_ram(struct list_head *list)
114 int nphdr;
115 size_t size;
116 struct kcore_list *tmp, *pos;
117 LIST_HEAD(garbage);
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);
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);
133 free_kclist_ents(&garbage);
137 #ifdef CONFIG_HIGHMEM
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.)
143 static int kcore_update_ram(void)
145 LIST_HEAD(head);
146 struct kcore_list *ent;
147 int ret = 0;
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;
160 #else /* !CONFIG_HIGHMEM */
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)
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;
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;
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);
193 return 1;
196 #else
197 static int
198 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
200 return 1;
203 #endif
205 static int
206 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
208 struct list_head *head = (struct list_head *)arg;
209 struct kcore_list *ent;
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;
217 /* Sanity check: Can happen in 32bit arch...maybe */
218 if (ent->addr < (unsigned long) __va(0))
219 goto free_out;
221 /* cut not-mapped area. ....from ppc-32 code. */
222 if (ULONG_MAX - ent->addr < ent->size)
223 ent->size = ULONG_MAX - ent->addr;
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;
233 ent->type = KCORE_RAM;
234 list_add_tail(&ent->list, head);
236 if (!get_sparsemem_vmemmap_info(ent, head)) {
237 list_del(&ent->list);
238 goto free_out;
241 return 0;
242 free_out:
243 kfree(ent);
244 return 1;
247 static int kcore_update_ram(void)
249 int nid, ret;
250 unsigned long end_pfn;
251 LIST_HEAD(head);
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;
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;
268 __kcore_update_ram(&head);
269 return ret;
271 #endif /* CONFIG_HIGHMEM */
273 /*****************************************************************************/
275 * determine size of ELF note
277 static int notesize(struct memelfnote *en)
279 int sz;
281 sz = sizeof(struct elf_note);
282 sz += roundup((strlen(en->name) + 1), 4);
283 sz += roundup(en->datasz, 4);
285 return sz;
286 } /* end notesize() */
288 /*****************************************************************************/
290 * store a note in the header buffer
292 static char *storenote(struct memelfnote *men, char *bufp)
294 struct elf_note en;
296 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
298 en.n_namesz = strlen(men->name) + 1;
299 en.n_descsz = men->datasz;
300 en.n_type = men->type;
302 DUMP_WRITE(&en, sizeof(en));
303 DUMP_WRITE(men->name, en.n_namesz);
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);
310 #undef DUMP_WRITE
312 return bufp;
313 } /* end storenote() */
316 * store an ELF coredump header in the supplied buffer
317 * nphdr is the number of elf_phdr to insert
319 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
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;
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;
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;
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);
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;
382 * Set up the notes in similar form to SVR4 core dumps made
383 * with info from their /proc.
385 nhdr->p_offset = offset;
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;
393 memset(&prstatus, 0, sizeof(struct elf_prstatus));
395 nhdr->p_filesz = notesize(&notes[0]);
396 bufp = storenote(&notes[0], bufp);
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;
404 memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
405 prpsinfo.pr_state = 0;
406 prpsinfo.pr_sname = 'R';
407 prpsinfo.pr_zomb = 0;
409 strcpy(prpsinfo.pr_fname, "vmlinux");
410 strlcpy(prpsinfo.pr_psargs, saved_command_line, sizeof(prpsinfo.pr_psargs));
412 nhdr->p_filesz += notesize(&notes[1]);
413 bufp = storenote(&notes[1], bufp);
415 /* set up the task structure */
416 notes[2].name = CORE_STR;
417 notes[2].type = NT_TASKSTRUCT;
418 notes[2].datasz = arch_task_struct_size;
419 notes[2].data = current;
421 nhdr->p_filesz += notesize(&notes[2]);
422 bufp = storenote(&notes[2], bufp);
424 } /* end elf_kcore_store_hdr() */
426 /*****************************************************************************/
428 * read from the ELF header and then kernel memory
430 static ssize_t
431 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
433 char *buf = file->private_data;
434 ssize_t acc = 0;
435 size_t size, tsz;
436 size_t elf_buflen;
437 int nphdr;
438 unsigned long start;
440 read_lock(&kclist_lock);
441 size = get_kcore_size(&nphdr, &elf_buflen);
443 if (buflen == 0 || *fpos >= size) {
444 read_unlock(&kclist_lock);
445 return 0;
448 /* trim buflen to not go beyond EOF */
449 if (buflen > size - *fpos)
450 buflen = size - *fpos;
452 /* construct an ELF core header if we'll need some of it */
453 if (*fpos < elf_buflen) {
454 char * elf_buf;
456 tsz = elf_buflen - *fpos;
457 if (buflen < tsz)
458 tsz = buflen;
459 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
460 if (!elf_buf) {
461 read_unlock(&kclist_lock);
462 return -ENOMEM;
464 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
465 read_unlock(&kclist_lock);
466 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
467 kfree(elf_buf);
468 return -EFAULT;
470 kfree(elf_buf);
471 buflen -= tsz;
472 *fpos += tsz;
473 buffer += tsz;
474 acc += tsz;
476 /* leave now if filled buffer already */
477 if (buflen == 0)
478 return acc;
479 } else
480 read_unlock(&kclist_lock);
483 * Check to see if our file offset matches with any of
484 * the addresses in the elf_phdr on our list.
486 start = kc_offset_to_vaddr(*fpos - elf_buflen);
487 if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
488 tsz = buflen;
490 while (buflen) {
491 struct kcore_list *m;
493 read_lock(&kclist_lock);
494 list_for_each_entry(m, &kclist_head, list) {
495 if (start >= m->addr && start < (m->addr+m->size))
496 break;
498 read_unlock(&kclist_lock);
500 if (&m->list == &kclist_head) {
501 if (clear_user(buffer, tsz))
502 return -EFAULT;
503 } else if (is_vmalloc_or_module_addr((void *)start)) {
504 vread(buf, (char *)start, tsz);
505 /* we have to zero-fill user buffer even if no read */
506 if (copy_to_user(buffer, buf, tsz))
507 return -EFAULT;
508 } else {
509 if (kern_addr_valid(start)) {
510 unsigned long n;
513 * Using bounce buffer to bypass the
514 * hardened user copy kernel text checks.
516 memcpy(buf, (char *) start, tsz);
517 n = copy_to_user(buffer, buf, tsz);
519 * We cannot distinguish between fault on source
520 * and fault on destination. When this happens
521 * we clear too and hope it will trigger the
522 * EFAULT again.
524 if (n) {
525 if (clear_user(buffer + tsz - n,
527 return -EFAULT;
529 } else {
530 if (clear_user(buffer, tsz))
531 return -EFAULT;
534 buflen -= tsz;
535 *fpos += tsz;
536 buffer += tsz;
537 acc += tsz;
538 start += tsz;
539 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
542 return acc;
546 static int open_kcore(struct inode *inode, struct file *filp)
548 if (!capable(CAP_SYS_RAWIO))
549 return -EPERM;
551 filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
552 if (!filp->private_data)
553 return -ENOMEM;
555 if (kcore_need_update)
556 kcore_update_ram();
557 if (i_size_read(inode) != proc_root_kcore->size) {
558 inode_lock(inode);
559 i_size_write(inode, proc_root_kcore->size);
560 inode_unlock(inode);
562 return 0;
565 static int release_kcore(struct inode *inode, struct file *file)
567 kfree(file->private_data);
568 return 0;
571 static const struct file_operations proc_kcore_operations = {
572 .read = read_kcore,
573 .open = open_kcore,
574 .release = release_kcore,
575 .llseek = default_llseek,
578 /* just remember that we have to update kcore */
579 static int __meminit kcore_callback(struct notifier_block *self,
580 unsigned long action, void *arg)
582 switch (action) {
583 case MEM_ONLINE:
584 case MEM_OFFLINE:
585 write_lock(&kclist_lock);
586 kcore_need_update = 1;
587 write_unlock(&kclist_lock);
589 return NOTIFY_OK;
592 static struct notifier_block kcore_callback_nb __meminitdata = {
593 .notifier_call = kcore_callback,
594 .priority = 0,
597 static struct kcore_list kcore_vmalloc;
599 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
600 static struct kcore_list kcore_text;
602 * If defined, special segment is used for mapping kernel text instead of
603 * direct-map area. We need to create special TEXT section.
605 static void __init proc_kcore_text_init(void)
607 kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
609 #else
610 static void __init proc_kcore_text_init(void)
613 #endif
615 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
617 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
619 struct kcore_list kcore_modules;
620 static void __init add_modules_range(void)
622 if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
623 kclist_add(&kcore_modules, (void *)MODULES_VADDR,
624 MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
627 #else
628 static void __init add_modules_range(void)
631 #endif
633 static int __init proc_kcore_init(void)
635 proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
636 &proc_kcore_operations);
637 if (!proc_root_kcore) {
638 pr_err("couldn't create /proc/kcore\n");
639 return 0; /* Always returns 0. */
641 /* Store text area if it's special */
642 proc_kcore_text_init();
643 /* Store vmalloc area */
644 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
645 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
646 add_modules_range();
647 /* Store direct-map area from physical memory map */
648 kcore_update_ram();
649 register_hotmemory_notifier(&kcore_callback_nb);
651 return 0;
653 fs_initcall(proc_kcore_init);