Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/proc/kcore.c kernel ELF core dumper
4 *
5 * Modelled on fs/exec.c:aout_core_dump()
6 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
7 * ELF version written by David Howells <David.Howells@nexor.co.uk>
8 * Modified and incorporated into 2.3.x by Tigran Aivazian <tigran@veritas.com>
9 * Support to dump vmalloc'd areas (ELF only), Tigran Aivazian <tigran@veritas.com>
10 * Safe accesses to vmalloc/direct-mapped discontiguous areas, Kanoj Sarcar <kanoj@sgi.com>
11 */
12
13#include <linux/mm.h>
14#include <linux/proc_fs.h>
15#include <linux/kcore.h>
16#include <linux/user.h>
17#include <linux/capability.h>
18#include <linux/elf.h>
19#include <linux/elfcore.h>
20#include <linux/notifier.h>
21#include <linux/vmalloc.h>
22#include <linux/highmem.h>
23#include <linux/printk.h>
24#include <linux/bootmem.h>
25#include <linux/init.h>
26#include <linux/slab.h>
27#include <linux/uaccess.h>
28#include <asm/io.h>
29#include <linux/list.h>
30#include <linux/ioport.h>
31#include <linux/memory.h>
32#include <linux/sched/task.h>
33#include <asm/sections.h>
34#include "internal.h"
35
36#define CORE_STR "CORE"
37
38#ifndef ELF_CORE_EFLAGS
39#define ELF_CORE_EFLAGS 0
40#endif
41
42static struct proc_dir_entry *proc_root_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 */
53struct memelfnote
54{
55 const char *name;
56 int type;
57 unsigned int datasz;
58 void *data;
59};
60
61static LIST_HEAD(kclist_head);
62static DEFINE_RWLOCK(kclist_lock);
63static int kcore_need_update = 1;
64
65void
66kclist_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
77static 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(arch_task_struct_size, 4);
98 *elf_buflen = PAGE_ALIGN(*elf_buflen);
99 return size + *elf_buflen;
100}
101
102static void free_kclist_ents(struct list_head *head)
103{
104 struct kcore_list *tmp, *pos;
105
106 list_for_each_entry_safe(pos, tmp, head, list) {
107 list_del(&pos->list);
108 kfree(pos);
109 }
110}
111/*
112 * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
113 */
114static void __kcore_update_ram(struct list_head *list)
115{
116 int nphdr;
117 size_t size;
118 struct kcore_list *tmp, *pos;
119 LIST_HEAD(garbage);
120
121 write_lock(&kclist_lock);
122 if (kcore_need_update) {
123 list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
124 if (pos->type == KCORE_RAM
125 || pos->type == KCORE_VMEMMAP)
126 list_move(&pos->list, &garbage);
127 }
128 list_splice_tail(list, &kclist_head);
129 } else
130 list_splice(list, &garbage);
131 kcore_need_update = 0;
132 proc_root_kcore->size = get_kcore_size(&nphdr, &size);
133 write_unlock(&kclist_lock);
134
135 free_kclist_ents(&garbage);
136}
137
138
139#ifdef CONFIG_HIGHMEM
140/*
141 * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
142 * because memory hole is not as big as !HIGHMEM case.
143 * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
144 */
145static int kcore_update_ram(void)
146{
147 LIST_HEAD(head);
148 struct kcore_list *ent;
149 int ret = 0;
150
151 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
152 if (!ent)
153 return -ENOMEM;
154 ent->addr = (unsigned long)__va(0);
155 ent->size = max_low_pfn << PAGE_SHIFT;
156 ent->type = KCORE_RAM;
157 list_add(&ent->list, &head);
158 __kcore_update_ram(&head);
159 return ret;
160}
161
162#else /* !CONFIG_HIGHMEM */
163
164#ifdef CONFIG_SPARSEMEM_VMEMMAP
165/* calculate vmemmap's address from given system ram pfn and register it */
166static int
167get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
168{
169 unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
170 unsigned long nr_pages = ent->size >> PAGE_SHIFT;
171 unsigned long start, end;
172 struct kcore_list *vmm, *tmp;
173
174
175 start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
176 end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
177 end = PAGE_ALIGN(end);
178 /* overlap check (because we have to align page */
179 list_for_each_entry(tmp, head, list) {
180 if (tmp->type != KCORE_VMEMMAP)
181 continue;
182 if (start < tmp->addr + tmp->size)
183 if (end > tmp->addr)
184 end = tmp->addr;
185 }
186 if (start < end) {
187 vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
188 if (!vmm)
189 return 0;
190 vmm->addr = start;
191 vmm->size = end - start;
192 vmm->type = KCORE_VMEMMAP;
193 list_add_tail(&vmm->list, head);
194 }
195 return 1;
196
197}
198#else
199static int
200get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
201{
202 return 1;
203}
204
205#endif
206
207static int
208kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
209{
210 struct list_head *head = (struct list_head *)arg;
211 struct kcore_list *ent;
212 struct page *p;
213
214 if (!pfn_valid(pfn))
215 return 1;
216
217 p = pfn_to_page(pfn);
218 if (!memmap_valid_within(pfn, p, page_zone(p)))
219 return 1;
220
221 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
222 if (!ent)
223 return -ENOMEM;
224 ent->addr = (unsigned long)page_to_virt(p);
225 ent->size = nr_pages << PAGE_SHIFT;
226
227 if (!virt_addr_valid(ent->addr))
228 goto free_out;
229
230 /* cut not-mapped area. ....from ppc-32 code. */
231 if (ULONG_MAX - ent->addr < ent->size)
232 ent->size = ULONG_MAX - ent->addr;
233
234 /*
235 * We've already checked virt_addr_valid so we know this address
236 * is a valid pointer, therefore we can check against it to determine
237 * if we need to trim
238 */
239 if (VMALLOC_START > ent->addr) {
240 if (VMALLOC_START - ent->addr < ent->size)
241 ent->size = VMALLOC_START - ent->addr;
242 }
243
244 ent->type = KCORE_RAM;
245 list_add_tail(&ent->list, head);
246
247 if (!get_sparsemem_vmemmap_info(ent, head)) {
248 list_del(&ent->list);
249 goto free_out;
250 }
251
252 return 0;
253free_out:
254 kfree(ent);
255 return 1;
256}
257
258static int kcore_update_ram(void)
259{
260 int nid, ret;
261 unsigned long end_pfn;
262 LIST_HEAD(head);
263
264 /* Not inialized....update now */
265 /* find out "max pfn" */
266 end_pfn = 0;
267 for_each_node_state(nid, N_MEMORY) {
268 unsigned long node_end;
269 node_end = node_end_pfn(nid);
270 if (end_pfn < node_end)
271 end_pfn = node_end;
272 }
273 /* scan 0 to max_pfn */
274 ret = walk_system_ram_range(0, end_pfn, &head, kclist_add_private);
275 if (ret) {
276 free_kclist_ents(&head);
277 return -ENOMEM;
278 }
279 __kcore_update_ram(&head);
280 return ret;
281}
282#endif /* CONFIG_HIGHMEM */
283
284/*****************************************************************************/
285/*
286 * determine size of ELF note
287 */
288static int notesize(struct memelfnote *en)
289{
290 int sz;
291
292 sz = sizeof(struct elf_note);
293 sz += roundup((strlen(en->name) + 1), 4);
294 sz += roundup(en->datasz, 4);
295
296 return sz;
297} /* end notesize() */
298
299/*****************************************************************************/
300/*
301 * store a note in the header buffer
302 */
303static char *storenote(struct memelfnote *men, char *bufp)
304{
305 struct elf_note en;
306
307#define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
308
309 en.n_namesz = strlen(men->name) + 1;
310 en.n_descsz = men->datasz;
311 en.n_type = men->type;
312
313 DUMP_WRITE(&en, sizeof(en));
314 DUMP_WRITE(men->name, en.n_namesz);
315
316 /* XXX - cast from long long to long to avoid need for libgcc.a */
317 bufp = (char*) roundup((unsigned long)bufp,4);
318 DUMP_WRITE(men->data, men->datasz);
319 bufp = (char*) roundup((unsigned long)bufp,4);
320
321#undef DUMP_WRITE
322
323 return bufp;
324} /* end storenote() */
325
326/*
327 * store an ELF coredump header in the supplied buffer
328 * nphdr is the number of elf_phdr to insert
329 */
330static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
331{
332 struct elf_prstatus prstatus; /* NT_PRSTATUS */
333 struct elf_prpsinfo prpsinfo; /* NT_PRPSINFO */
334 struct elf_phdr *nhdr, *phdr;
335 struct elfhdr *elf;
336 struct memelfnote notes[3];
337 off_t offset = 0;
338 struct kcore_list *m;
339
340 /* setup ELF header */
341 elf = (struct elfhdr *) bufp;
342 bufp += sizeof(struct elfhdr);
343 offset += sizeof(struct elfhdr);
344 memcpy(elf->e_ident, ELFMAG, SELFMAG);
345 elf->e_ident[EI_CLASS] = ELF_CLASS;
346 elf->e_ident[EI_DATA] = ELF_DATA;
347 elf->e_ident[EI_VERSION]= EV_CURRENT;
348 elf->e_ident[EI_OSABI] = ELF_OSABI;
349 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
350 elf->e_type = ET_CORE;
351 elf->e_machine = ELF_ARCH;
352 elf->e_version = EV_CURRENT;
353 elf->e_entry = 0;
354 elf->e_phoff = sizeof(struct elfhdr);
355 elf->e_shoff = 0;
356 elf->e_flags = ELF_CORE_EFLAGS;
357 elf->e_ehsize = sizeof(struct elfhdr);
358 elf->e_phentsize= sizeof(struct elf_phdr);
359 elf->e_phnum = nphdr;
360 elf->e_shentsize= 0;
361 elf->e_shnum = 0;
362 elf->e_shstrndx = 0;
363
364 /* setup ELF PT_NOTE program header */
365 nhdr = (struct elf_phdr *) bufp;
366 bufp += sizeof(struct elf_phdr);
367 offset += sizeof(struct elf_phdr);
368 nhdr->p_type = PT_NOTE;
369 nhdr->p_offset = 0;
370 nhdr->p_vaddr = 0;
371 nhdr->p_paddr = 0;
372 nhdr->p_filesz = 0;
373 nhdr->p_memsz = 0;
374 nhdr->p_flags = 0;
375 nhdr->p_align = 0;
376
377 /* setup ELF PT_LOAD program header for every area */
378 list_for_each_entry(m, &kclist_head, list) {
379 phdr = (struct elf_phdr *) bufp;
380 bufp += sizeof(struct elf_phdr);
381 offset += sizeof(struct elf_phdr);
382
383 phdr->p_type = PT_LOAD;
384 phdr->p_flags = PF_R|PF_W|PF_X;
385 phdr->p_offset = kc_vaddr_to_offset(m->addr) + dataoff;
386 phdr->p_vaddr = (size_t)m->addr;
387 if (m->type == KCORE_RAM || m->type == KCORE_TEXT)
388 phdr->p_paddr = __pa(m->addr);
389 else
390 phdr->p_paddr = (elf_addr_t)-1;
391 phdr->p_filesz = phdr->p_memsz = m->size;
392 phdr->p_align = PAGE_SIZE;
393 }
394
395 /*
396 * Set up the notes in similar form to SVR4 core dumps made
397 * with info from their /proc.
398 */
399 nhdr->p_offset = offset;
400
401 /* set up the process status */
402 notes[0].name = CORE_STR;
403 notes[0].type = NT_PRSTATUS;
404 notes[0].datasz = sizeof(struct elf_prstatus);
405 notes[0].data = &prstatus;
406
407 memset(&prstatus, 0, sizeof(struct elf_prstatus));
408
409 nhdr->p_filesz = notesize(¬es[0]);
410 bufp = storenote(¬es[0], bufp);
411
412 /* set up the process info */
413 notes[1].name = CORE_STR;
414 notes[1].type = NT_PRPSINFO;
415 notes[1].datasz = sizeof(struct elf_prpsinfo);
416 notes[1].data = &prpsinfo;
417
418 memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
419 prpsinfo.pr_state = 0;
420 prpsinfo.pr_sname = 'R';
421 prpsinfo.pr_zomb = 0;
422
423 strcpy(prpsinfo.pr_fname, "vmlinux");
424 strlcpy(prpsinfo.pr_psargs, saved_command_line, sizeof(prpsinfo.pr_psargs));
425
426 nhdr->p_filesz += notesize(¬es[1]);
427 bufp = storenote(¬es[1], bufp);
428
429 /* set up the task structure */
430 notes[2].name = CORE_STR;
431 notes[2].type = NT_TASKSTRUCT;
432 notes[2].datasz = arch_task_struct_size;
433 notes[2].data = current;
434
435 nhdr->p_filesz += notesize(¬es[2]);
436 bufp = storenote(¬es[2], bufp);
437
438} /* end elf_kcore_store_hdr() */
439
440/*****************************************************************************/
441/*
442 * read from the ELF header and then kernel memory
443 */
444static ssize_t
445read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
446{
447 char *buf = file->private_data;
448 ssize_t acc = 0;
449 size_t size, tsz;
450 size_t elf_buflen;
451 int nphdr;
452 unsigned long start;
453
454 read_lock(&kclist_lock);
455 size = get_kcore_size(&nphdr, &elf_buflen);
456
457 if (buflen == 0 || *fpos >= size) {
458 read_unlock(&kclist_lock);
459 return 0;
460 }
461
462 /* trim buflen to not go beyond EOF */
463 if (buflen > size - *fpos)
464 buflen = size - *fpos;
465
466 /* construct an ELF core header if we'll need some of it */
467 if (*fpos < elf_buflen) {
468 char * elf_buf;
469
470 tsz = elf_buflen - *fpos;
471 if (buflen < tsz)
472 tsz = buflen;
473 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
474 if (!elf_buf) {
475 read_unlock(&kclist_lock);
476 return -ENOMEM;
477 }
478 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
479 read_unlock(&kclist_lock);
480 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
481 kfree(elf_buf);
482 return -EFAULT;
483 }
484 kfree(elf_buf);
485 buflen -= tsz;
486 *fpos += tsz;
487 buffer += tsz;
488 acc += tsz;
489
490 /* leave now if filled buffer already */
491 if (buflen == 0)
492 return acc;
493 } else
494 read_unlock(&kclist_lock);
495
496 /*
497 * Check to see if our file offset matches with any of
498 * the addresses in the elf_phdr on our list.
499 */
500 start = kc_offset_to_vaddr(*fpos - elf_buflen);
501 if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
502 tsz = buflen;
503
504 while (buflen) {
505 struct kcore_list *m;
506
507 read_lock(&kclist_lock);
508 list_for_each_entry(m, &kclist_head, list) {
509 if (start >= m->addr && start < (m->addr+m->size))
510 break;
511 }
512 read_unlock(&kclist_lock);
513
514 if (&m->list == &kclist_head) {
515 if (clear_user(buffer, tsz))
516 return -EFAULT;
517 } else if (m->type == KCORE_VMALLOC) {
518 vread(buf, (char *)start, tsz);
519 /* we have to zero-fill user buffer even if no read */
520 if (copy_to_user(buffer, buf, tsz))
521 return -EFAULT;
522 } else if (m->type == KCORE_USER) {
523 /* User page is handled prior to normal kernel page: */
524 if (copy_to_user(buffer, (char *)start, tsz))
525 return -EFAULT;
526 } else {
527 if (kern_addr_valid(start)) {
528 /*
529 * Using bounce buffer to bypass the
530 * hardened user copy kernel text checks.
531 */
532 if (probe_kernel_read(buf, (void *) start, tsz)) {
533 if (clear_user(buffer, tsz))
534 return -EFAULT;
535 } else {
536 if (copy_to_user(buffer, buf, tsz))
537 return -EFAULT;
538 }
539 } else {
540 if (clear_user(buffer, tsz))
541 return -EFAULT;
542 }
543 }
544 buflen -= tsz;
545 *fpos += tsz;
546 buffer += tsz;
547 acc += tsz;
548 start += tsz;
549 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
550 }
551
552 return acc;
553}
554
555
556static int open_kcore(struct inode *inode, struct file *filp)
557{
558 if (!capable(CAP_SYS_RAWIO))
559 return -EPERM;
560
561 filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
562 if (!filp->private_data)
563 return -ENOMEM;
564
565 if (kcore_need_update)
566 kcore_update_ram();
567 if (i_size_read(inode) != proc_root_kcore->size) {
568 inode_lock(inode);
569 i_size_write(inode, proc_root_kcore->size);
570 inode_unlock(inode);
571 }
572 return 0;
573}
574
575static int release_kcore(struct inode *inode, struct file *file)
576{
577 kfree(file->private_data);
578 return 0;
579}
580
581static const struct file_operations proc_kcore_operations = {
582 .read = read_kcore,
583 .open = open_kcore,
584 .release = release_kcore,
585 .llseek = default_llseek,
586};
587
588/* just remember that we have to update kcore */
589static int __meminit kcore_callback(struct notifier_block *self,
590 unsigned long action, void *arg)
591{
592 switch (action) {
593 case MEM_ONLINE:
594 case MEM_OFFLINE:
595 write_lock(&kclist_lock);
596 kcore_need_update = 1;
597 write_unlock(&kclist_lock);
598 }
599 return NOTIFY_OK;
600}
601
602static struct notifier_block kcore_callback_nb __meminitdata = {
603 .notifier_call = kcore_callback,
604 .priority = 0,
605};
606
607static struct kcore_list kcore_vmalloc;
608
609#ifdef CONFIG_ARCH_PROC_KCORE_TEXT
610static struct kcore_list kcore_text;
611/*
612 * If defined, special segment is used for mapping kernel text instead of
613 * direct-map area. We need to create special TEXT section.
614 */
615static void __init proc_kcore_text_init(void)
616{
617 kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
618}
619#else
620static void __init proc_kcore_text_init(void)
621{
622}
623#endif
624
625#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
626/*
627 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
628 */
629struct kcore_list kcore_modules;
630static void __init add_modules_range(void)
631{
632 if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
633 kclist_add(&kcore_modules, (void *)MODULES_VADDR,
634 MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
635 }
636}
637#else
638static void __init add_modules_range(void)
639{
640}
641#endif
642
643static int __init proc_kcore_init(void)
644{
645 proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
646 &proc_kcore_operations);
647 if (!proc_root_kcore) {
648 pr_err("couldn't create /proc/kcore\n");
649 return 0; /* Always returns 0. */
650 }
651 /* Store text area if it's special */
652 proc_kcore_text_init();
653 /* Store vmalloc area */
654 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
655 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
656 add_modules_range();
657 /* Store direct-map area from physical memory map */
658 kcore_update_ram();
659 register_hotmemory_notifier(&kcore_callback_nb);
660
661 return 0;
662}
663fs_initcall(proc_kcore_init);
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
36static 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 */
47struct memelfnote
48{
49 const char *name;
50 int type;
51 unsigned int datasz;
52 void *data;
53};
54
55static LIST_HEAD(kclist_head);
56static DEFINE_RWLOCK(kclist_lock);
57static int kcore_need_update = 1;
58
59void
60kclist_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
71static 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
96static 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 */
108static 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 */
139static 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 */
160int 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
192int get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
193{
194 return 1;
195}
196
197#endif
198
199static int
200kclist_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;
236free_out:
237 kfree(ent);
238 return 1;
239}
240
241static 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 */
272static 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 */
287static 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 */
314static 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(¬es[0]);
391 bufp = storenote(¬es[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(¬es[1]);
408 bufp = storenote(¬es[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(¬es[2]);
417 bufp = storenote(¬es[2], bufp);
418
419} /* end elf_kcore_store_hdr() */
420
421/*****************************************************************************/
422/*
423 * read from the ELF header and then kernel memory
424 */
425static ssize_t
426read_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
543static 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
558static const struct file_operations proc_kcore_operations = {
559 .read = read_kcore,
560 .open = open_kcore,
561 .llseek = default_llseek,
562};
563
564#ifdef CONFIG_MEMORY_HOTPLUG
565/* just remember that we have to update kcore */
566static 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
581static struct kcore_list kcore_vmalloc;
582
583#ifdef CONFIG_ARCH_PROC_KCORE_TEXT
584static 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 */
589static void __init proc_kcore_text_init(void)
590{
591 kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
592}
593#else
594static 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 */
603struct kcore_list kcore_modules;
604static 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
610static void __init add_modules_range(void)
611{
612}
613#endif
614
615static 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}
635module_init(proc_kcore_init);