?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mmc: meson-gx: factor out setting block size
[linux-2.6/btrfs-unstable.git] / fs / proc / kcore.c
blob4ee55274f155fce61b05ca704880db627103cdae
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 <linux/uaccess.h>
27 #include <asm/io.h>
28 #include <linux/list.h>
29 #include <linux/ioport.h>
30 #include <linux/memory.h>
31 #include <linux/sched/task.h>
32 #include <asm/sections.h>
33 #include "internal.h"
34
35 #define CORE_STR "CORE"
36
37 #ifndef ELF_CORE_EFLAGS
38 #define ELF_CORE_EFLAGS 0
39 #endif
40
41 static struct proc_dir_entry *proc_root_kcore;
42
43
44 #ifndef kc_vaddr_to_offset
45 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
46 #endif
47 #ifndef kc_offset_to_vaddr
48 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
49 #endif
50
51 /* An ELF note in memory */
52 struct memelfnote
53 {
54 const char *name;
55 int type;
56 unsigned int datasz;
57 void *data;
58 };
59
60 static LIST_HEAD(kclist_head);
61 static DEFINE_RWLOCK(kclist_lock);
62 static int kcore_need_update = 1;
63
64 void
65 kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
66 {
67 new->addr = (unsigned long)addr;
68 new->size = size;
69 new->type = type;
70
71 write_lock(&kclist_lock);
72 list_add_tail(&new->list, &kclist_head);
73 write_unlock(&kclist_lock);
74 }
75
76 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
77 {
78 size_t try, size;
79 struct kcore_list *m;
80
81 *nphdr = 1; /* PT_NOTE */
82 size = 0;
83
84 list_for_each_entry(m, &kclist_head, list) {
85 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
86 if (try > size)
87 size = try;
88 *nphdr = *nphdr + 1;
89 }
90 *elf_buflen = sizeof(struct elfhdr) +
91 (*nphdr + 2)*sizeof(struct elf_phdr) +
92 3 * ((sizeof(struct elf_note)) +
93 roundup(sizeof(CORE_STR), 4)) +
94 roundup(sizeof(struct elf_prstatus), 4) +
95 roundup(sizeof(struct elf_prpsinfo), 4) +
96 roundup(arch_task_struct_size, 4);
97 *elf_buflen = PAGE_ALIGN(*elf_buflen);
98 return size + *elf_buflen;
99 }
100
101 static void free_kclist_ents(struct list_head *head)
102 {
103 struct kcore_list *tmp, *pos;
104
105 list_for_each_entry_safe(pos, tmp, head, list) {
106 list_del(&pos->list);
107 kfree(pos);
108 }
109 }
110 /*
111 * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
112 */
113 static void __kcore_update_ram(struct list_head *list)
114 {
115 int nphdr;
116 size_t size;
117 struct kcore_list *tmp, *pos;
118 LIST_HEAD(garbage);
119
120 write_lock(&kclist_lock);
121 if (kcore_need_update) {
122 list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
123 if (pos->type == KCORE_RAM
124 || pos->type == KCORE_VMEMMAP)
125 list_move(&pos->list, &garbage);
126 }
127 list_splice_tail(list, &kclist_head);
128 } else
129 list_splice(list, &garbage);
130 kcore_need_update = 0;
131 proc_root_kcore->size = get_kcore_size(&nphdr, &size);
132 write_unlock(&kclist_lock);
133
134 free_kclist_ents(&garbage);
135 }
136
137
138 #ifdef CONFIG_HIGHMEM
139 /*
140 * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
141 * because memory hole is not as big as !HIGHMEM case.
142 * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
143 */
144 static int kcore_update_ram(void)
145 {
146 LIST_HEAD(head);
147 struct kcore_list *ent;
148 int ret = 0;
149
150 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
151 if (!ent)
152 return -ENOMEM;
153 ent->addr = (unsigned long)__va(0);
154 ent->size = max_low_pfn << PAGE_SHIFT;
155 ent->type = KCORE_RAM;
156 list_add(&ent->list, &head);
157 __kcore_update_ram(&head);
158 return ret;
159 }
160
161 #else /* !CONFIG_HIGHMEM */
162
163 #ifdef CONFIG_SPARSEMEM_VMEMMAP
164 /* calculate vmemmap's address from given system ram pfn and register it */
165 static int
166 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
167 {
168 unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
169 unsigned long nr_pages = ent->size >> PAGE_SHIFT;
170 unsigned long start, end;
171 struct kcore_list *vmm, *tmp;
172
173
174 start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
175 end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
176 end = PAGE_ALIGN(end);
177 /* overlap check (because we have to align page */
178 list_for_each_entry(tmp, head, list) {
179 if (tmp->type != KCORE_VMEMMAP)
180 continue;
181 if (start < tmp->addr + tmp->size)
182 if (end > tmp->addr)
183 end = tmp->addr;
184 }
185 if (start < end) {
186 vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
187 if (!vmm)
188 return 0;
189 vmm->addr = start;
190 vmm->size = end - start;
191 vmm->type = KCORE_VMEMMAP;
192 list_add_tail(&vmm->list, head);
193 }
194 return 1;
195
196 }
197 #else
198 static int
199 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
200 {
201 return 1;
202 }
203
204 #endif
205
206 static int
207 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
208 {
209 struct list_head *head = (struct list_head *)arg;
210 struct kcore_list *ent;
211
212 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
213 if (!ent)
214 return -ENOMEM;
215 ent->addr = (unsigned long)__va((pfn << PAGE_SHIFT));
216 ent->size = nr_pages << PAGE_SHIFT;
217
218 /* Sanity check: Can happen in 32bit arch...maybe */
219 if (ent->addr < (unsigned long) __va(0))
220 goto free_out;
221
222 /* cut not-mapped area. ....from ppc-32 code. */
223 if (ULONG_MAX - ent->addr < ent->size)
224 ent->size = ULONG_MAX - ent->addr;
225
226 /* cut when vmalloc() area is higher than direct-map area */
227 if (VMALLOC_START > (unsigned long)__va(0)) {
228 if (ent->addr > VMALLOC_START)
229 goto free_out;
230 if (VMALLOC_START - ent->addr < ent->size)
231 ent->size = VMALLOC_START - ent->addr;
232 }
233
234 ent->type = KCORE_RAM;
235 list_add_tail(&ent->list, head);
236
237 if (!get_sparsemem_vmemmap_info(ent, head)) {
238 list_del(&ent->list);
239 goto free_out;
240 }
241
242 return 0;
243 free_out:
244 kfree(ent);
245 return 1;
246 }
247
248 static int kcore_update_ram(void)
249 {
250 int nid, ret;
251 unsigned long end_pfn;
252 LIST_HEAD(head);
253
254 /* Not inialized....update now */
255 /* find out "max pfn" */
256 end_pfn = 0;
257 for_each_node_state(nid, N_MEMORY) {
258 unsigned long node_end;
259 node_end = node_end_pfn(nid);
260 if (end_pfn < node_end)
261 end_pfn = node_end;
262 }
263 /* scan 0 to max_pfn */
264 ret = walk_system_ram_range(0, end_pfn, &head, kclist_add_private);
265 if (ret) {
266 free_kclist_ents(&head);
267 return -ENOMEM;
268 }
269 __kcore_update_ram(&head);
270 return ret;
271 }
272 #endif /* CONFIG_HIGHMEM */
273
274 /*****************************************************************************/
275 /*
276 * determine size of ELF note
277 */
278 static int notesize(struct memelfnote *en)
279 {
280 int sz;
281
282 sz = sizeof(struct elf_note);
283 sz += roundup((strlen(en->name) + 1), 4);
284 sz += roundup(en->datasz, 4);
285
286 return sz;
287 } /* end notesize() */
288
289 /*****************************************************************************/
290 /*
291 * store a note in the header buffer
292 */
293 static char *storenote(struct memelfnote *men, char *bufp)
294 {
295 struct elf_note en;
296
297 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
298
299 en.n_namesz = strlen(men->name) + 1;
300 en.n_descsz = men->datasz;
301 en.n_type = men->type;
302
303 DUMP_WRITE(&en, sizeof(en));
304 DUMP_WRITE(men->name, en.n_namesz);
305
306 /* XXX - cast from long long to long to avoid need for libgcc.a */
307 bufp = (char*) roundup((unsigned long)bufp,4);
308 DUMP_WRITE(men->data, men->datasz);
309 bufp = (char*) roundup((unsigned long)bufp,4);
310
311 #undef DUMP_WRITE
312
313 return bufp;
314 } /* end storenote() */
315
316 /*
317 * store an ELF coredump header in the supplied buffer
318 * nphdr is the number of elf_phdr to insert
319 */
320 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
321 {
322 struct elf_prstatus prstatus; /* NT_PRSTATUS */
323 struct elf_prpsinfo prpsinfo; /* NT_PRPSINFO */
324 struct elf_phdr *nhdr, *phdr;
325 struct elfhdr *elf;
326 struct memelfnote notes[3];
327 off_t offset = 0;
328 struct kcore_list *m;
329
330 /* setup ELF header */
331 elf = (struct elfhdr *) bufp;
332 bufp += sizeof(struct elfhdr);
333 offset += sizeof(struct elfhdr);
334 memcpy(elf->e_ident, ELFMAG, SELFMAG);
335 elf->e_ident[EI_CLASS] = ELF_CLASS;
336 elf->e_ident[EI_DATA] = ELF_DATA;
337 elf->e_ident[EI_VERSION]= EV_CURRENT;
338 elf->e_ident[EI_OSABI] = ELF_OSABI;
339 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
340 elf->e_type = ET_CORE;
341 elf->e_machine = ELF_ARCH;
342 elf->e_version = EV_CURRENT;
343 elf->e_entry = 0;
344 elf->e_phoff = sizeof(struct elfhdr);
345 elf->e_shoff = 0;
346 elf->e_flags = ELF_CORE_EFLAGS;
347 elf->e_ehsize = sizeof(struct elfhdr);
348 elf->e_phentsize= sizeof(struct elf_phdr);
349 elf->e_phnum = nphdr;
350 elf->e_shentsize= 0;
351 elf->e_shnum = 0;
352 elf->e_shstrndx = 0;
353
354 /* setup ELF PT_NOTE program header */
355 nhdr = (struct elf_phdr *) bufp;
356 bufp += sizeof(struct elf_phdr);
357 offset += sizeof(struct elf_phdr);
358 nhdr->p_type = PT_NOTE;
359 nhdr->p_offset = 0;
360 nhdr->p_vaddr = 0;
361 nhdr->p_paddr = 0;
362 nhdr->p_filesz = 0;
363 nhdr->p_memsz = 0;
364 nhdr->p_flags = 0;
365 nhdr->p_align = 0;
366
367 /* setup ELF PT_LOAD program header for every area */
368 list_for_each_entry(m, &kclist_head, list) {
369 phdr = (struct elf_phdr *) bufp;
370 bufp += sizeof(struct elf_phdr);
371 offset += sizeof(struct elf_phdr);
372
373 phdr->p_type = PT_LOAD;
374 phdr->p_flags = PF_R|PF_W|PF_X;
375 phdr->p_offset = kc_vaddr_to_offset(m->addr) + dataoff;
376 phdr->p_vaddr = (size_t)m->addr;
377 if (m->type == KCORE_RAM || m->type == KCORE_TEXT)
378 phdr->p_paddr = __pa(m->addr);
379 else
380 phdr->p_paddr = (elf_addr_t)-1;
381 phdr->p_filesz = phdr->p_memsz = m->size;
382 phdr->p_align = PAGE_SIZE;
383 }
384
385 /*
386 * Set up the notes in similar form to SVR4 core dumps made
387 * with info from their /proc.
388 */
389 nhdr->p_offset = offset;
390
391 /* set up the process status */
392 notes[0].name = CORE_STR;
393 notes[0].type = NT_PRSTATUS;
394 notes[0].datasz = sizeof(struct elf_prstatus);
395 notes[0].data = &prstatus;
396
397 memset(&prstatus, 0, sizeof(struct elf_prstatus));
398
399 nhdr->p_filesz = notesize(&notes[0]);
400 bufp = storenote(&notes[0], bufp);
401
402 /* set up the process info */
403 notes[1].name = CORE_STR;
404 notes[1].type = NT_PRPSINFO;
405 notes[1].datasz = sizeof(struct elf_prpsinfo);
406 notes[1].data = &prpsinfo;
407
408 memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
409 prpsinfo.pr_state = 0;
410 prpsinfo.pr_sname = 'R';
411 prpsinfo.pr_zomb = 0;
412
413 strcpy(prpsinfo.pr_fname, "vmlinux");
414 strlcpy(prpsinfo.pr_psargs, saved_command_line, sizeof(prpsinfo.pr_psargs));
415
416 nhdr->p_filesz += notesize(&notes[1]);
417 bufp = storenote(&notes[1], bufp);
418
419 /* set up the task structure */
420 notes[2].name = CORE_STR;
421 notes[2].type = NT_TASKSTRUCT;
422 notes[2].datasz = arch_task_struct_size;
423 notes[2].data = current;
424
425 nhdr->p_filesz += notesize(&notes[2]);
426 bufp = storenote(&notes[2], bufp);
427
428 } /* end elf_kcore_store_hdr() */
429
430 /*****************************************************************************/
431 /*
432 * read from the ELF header and then kernel memory
433 */
434 static ssize_t
435 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
436 {
437 char *buf = file->private_data;
438 ssize_t acc = 0;
439 size_t size, tsz;
440 size_t elf_buflen;
441 int nphdr;
442 unsigned long start;
443
444 read_lock(&kclist_lock);
445 size = get_kcore_size(&nphdr, &elf_buflen);
446
447 if (buflen == 0 || *fpos >= size) {
448 read_unlock(&kclist_lock);
449 return 0;
450 }
451
452 /* trim buflen to not go beyond EOF */
453 if (buflen > size - *fpos)
454 buflen = size - *fpos;
455
456 /* construct an ELF core header if we'll need some of it */
457 if (*fpos < elf_buflen) {
458 char * elf_buf;
459
460 tsz = elf_buflen - *fpos;
461 if (buflen < tsz)
462 tsz = buflen;
463 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
464 if (!elf_buf) {
465 read_unlock(&kclist_lock);
466 return -ENOMEM;
467 }
468 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
469 read_unlock(&kclist_lock);
470 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
471 kfree(elf_buf);
472 return -EFAULT;
473 }
474 kfree(elf_buf);
475 buflen -= tsz;
476 *fpos += tsz;
477 buffer += tsz;
478 acc += tsz;
479
480 /* leave now if filled buffer already */
481 if (buflen == 0)
482 return acc;
483 } else
484 read_unlock(&kclist_lock);
485
486 /*
487 * Check to see if our file offset matches with any of
488 * the addresses in the elf_phdr on our list.
489 */
490 start = kc_offset_to_vaddr(*fpos - elf_buflen);
491 if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
492 tsz = buflen;
493
494 while (buflen) {
495 struct kcore_list *m;
496
497 read_lock(&kclist_lock);
498 list_for_each_entry(m, &kclist_head, list) {
499 if (start >= m->addr && start < (m->addr+m->size))
500 break;
501 }
502 read_unlock(&kclist_lock);
503
504 if (&m->list == &kclist_head) {
505 if (clear_user(buffer, tsz))
506 return -EFAULT;
507 } else if (is_vmalloc_or_module_addr((void *)start)) {
508 vread(buf, (char *)start, tsz);
509 /* we have to zero-fill user buffer even if no read */
510 if (copy_to_user(buffer, buf, tsz))
511 return -EFAULT;
512 } else {
513 if (kern_addr_valid(start)) {
514 unsigned long n;
515
516 /*
517 * Using bounce buffer to bypass the
518 * hardened user copy kernel text checks.
519 */
520 memcpy(buf, (char *) start, tsz);
521 n = copy_to_user(buffer, buf, tsz);
522 /*
523 * We cannot distinguish between fault on source
524 * and fault on destination. When this happens
525 * we clear too and hope it will trigger the
526 * EFAULT again.
527 */
528 if (n) {
529 if (clear_user(buffer + tsz - n,
530 n))
531 return -EFAULT;
532 }
533 } else {
534 if (clear_user(buffer, tsz))
535 return -EFAULT;
536 }
537 }
538 buflen -= tsz;
539 *fpos += tsz;
540 buffer += tsz;
541 acc += tsz;
542 start += tsz;
543 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
544 }
545
546 return acc;
547 }
548
549
550 static int open_kcore(struct inode *inode, struct file *filp)
551 {
552 if (!capable(CAP_SYS_RAWIO))
553 return -EPERM;
554
555 filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
556 if (!filp->private_data)
557 return -ENOMEM;
558
559 if (kcore_need_update)
560 kcore_update_ram();
561 if (i_size_read(inode) != proc_root_kcore->size) {
562 inode_lock(inode);
563 i_size_write(inode, proc_root_kcore->size);
564 inode_unlock(inode);
565 }
566 return 0;
567 }
568
569 static int release_kcore(struct inode *inode, struct file *file)
570 {
571 kfree(file->private_data);
572 return 0;
573 }
574
575 static const struct file_operations proc_kcore_operations = {
576 .read = read_kcore,
577 .open = open_kcore,
578 .release = release_kcore,
579 .llseek = default_llseek,
580 };
581
582 /* just remember that we have to update kcore */
583 static int __meminit kcore_callback(struct notifier_block *self,
584 unsigned long action, void *arg)
585 {
586 switch (action) {
587 case MEM_ONLINE:
588 case MEM_OFFLINE:
589 write_lock(&kclist_lock);
590 kcore_need_update = 1;
591 write_unlock(&kclist_lock);
592 }
593 return NOTIFY_OK;
594 }
595
596 static struct notifier_block kcore_callback_nb __meminitdata = {
597 .notifier_call = kcore_callback,
598 .priority = 0,
599 };
600
601 static struct kcore_list kcore_vmalloc;
602
603 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
604 static struct kcore_list kcore_text;
605 /*
606 * If defined, special segment is used for mapping kernel text instead of
607 * direct-map area. We need to create special TEXT section.
608 */
609 static void __init proc_kcore_text_init(void)
610 {
611 kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
612 }
613 #else
614 static void __init proc_kcore_text_init(void)
615 {
616 }
617 #endif
618
619 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
620 /*
621 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
622 */
623 struct kcore_list kcore_modules;
624 static void __init add_modules_range(void)
625 {
626 if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
627 kclist_add(&kcore_modules, (void *)MODULES_VADDR,
628 MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
629 }
630 }
631 #else
632 static void __init add_modules_range(void)
633 {
634 }
635 #endif
636
637 static int __init proc_kcore_init(void)
638 {
639 proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
640 &proc_kcore_operations);
641 if (!proc_root_kcore) {
642 pr_err("couldn't create /proc/kcore\n");
643 return 0; /* Always returns 0. */
644 }
645 /* Store text area if it's special */
646 proc_kcore_text_init();
647 /* Store vmalloc area */
648 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
649 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
650 add_modules_range();
651 /* Store direct-map area from physical memory map */
652 kcore_update_ram();
653 register_hotmemory_notifier(&kcore_callback_nb);
654
655 return 0;
656 }
657 fs_initcall(proc_kcore_init);
(deprecated) Btrfs devel tree