Merge remote-tracking branch 'net/master'
[linux-2.6/next.git] / fs / proc / kcore.c
blobd245cb23dd726206806304022d0c83a0ab784656
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/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>
30 #define CORE_STR "CORE"
32 #ifndef ELF_CORE_EFLAGS
33 #define ELF_CORE_EFLAGS 0
34 #endif
36 static struct proc_dir_entry *proc_root_kcore;
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
46 /* An ELF note in memory */
47 struct memelfnote
49 const char *name;
50 int type;
51 unsigned int datasz;
52 void *data;
55 static LIST_HEAD(kclist_head);
56 static DEFINE_RWLOCK(kclist_lock);
57 static int kcore_need_update = 1;
59 void
60 kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
62 new->addr = (unsigned long)addr;
63 new->size = size;
64 new->type = type;
66 write_lock(&kclist_lock);
67 list_add_tail(&new->list, &kclist_head);
68 write_unlock(&kclist_lock);
71 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
73 size_t try, size;
74 struct kcore_list *m;
76 *nphdr = 1; /* PT_NOTE */
77 size = 0;
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;
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;
96 static void free_kclist_ents(struct list_head *head)
98 struct kcore_list *tmp, *pos;
100 list_for_each_entry_safe(pos, tmp, head, list) {
101 list_del(&pos->list);
102 kfree(pos);
106 * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
108 static void __kcore_update_ram(struct list_head *list)
110 int nphdr;
111 size_t size;
112 struct kcore_list *tmp, *pos;
113 LIST_HEAD(garbage);
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);
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);
129 free_kclist_ents(&garbage);
133 #ifdef CONFIG_HIGHMEM
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.)
139 static int kcore_update_ram(void)
141 LIST_HEAD(head);
142 struct kcore_list *ent;
143 int ret = 0;
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;
156 #else /* !CONFIG_HIGHMEM */
158 #ifdef CONFIG_SPARSEMEM_VMEMMAP
159 /* calculate vmemmap's address from given system ram pfn and register it */
160 int get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
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;
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;
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);
188 return 1;
191 #else
192 int get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
194 return 1;
197 #endif
199 static int
200 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
202 struct list_head *head = (struct list_head *)arg;
203 struct kcore_list *ent;
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;
211 /* Sanity check: Can happen in 32bit arch...maybe */
212 if (ent->addr < (unsigned long) __va(0))
213 goto free_out;
215 /* cut not-mapped area. ....from ppc-32 code. */
216 if (ULONG_MAX - ent->addr < ent->size)
217 ent->size = ULONG_MAX - ent->addr;
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;
227 ent->type = KCORE_RAM;
228 list_add_tail(&ent->list, head);
230 if (!get_sparsemem_vmemmap_info(ent, head)) {
231 list_del(&ent->list);
232 goto free_out;
235 return 0;
236 free_out:
237 kfree(ent);
238 return 1;
241 static int kcore_update_ram(void)
243 int nid, ret;
244 unsigned long end_pfn;
245 LIST_HEAD(head);
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;
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;
263 __kcore_update_ram(&head);
264 return ret;
266 #endif /* CONFIG_HIGHMEM */
268 /*****************************************************************************/
270 * determine size of ELF note
272 static int notesize(struct memelfnote *en)
274 int sz;
276 sz = sizeof(struct elf_note);
277 sz += roundup((strlen(en->name) + 1), 4);
278 sz += roundup(en->datasz, 4);
280 return sz;
281 } /* end notesize() */
283 /*****************************************************************************/
285 * store a note in the header buffer
287 static char *storenote(struct memelfnote *men, char *bufp)
289 struct elf_note en;
291 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
293 en.n_namesz = strlen(men->name) + 1;
294 en.n_descsz = men->datasz;
295 en.n_type = men->type;
297 DUMP_WRITE(&en, sizeof(en));
298 DUMP_WRITE(men->name, en.n_namesz);
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);
305 #undef DUMP_WRITE
307 return bufp;
308 } /* end storenote() */
311 * store an ELF coredump header in the supplied buffer
312 * nphdr is the number of elf_phdr to insert
314 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
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;
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;
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;
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);
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;
377 * Set up the notes in similar form to SVR4 core dumps made
378 * with info from their /proc.
380 nhdr->p_offset = offset;
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;
388 memset(&prstatus, 0, sizeof(struct elf_prstatus));
390 nhdr->p_filesz = notesize(&notes[0]);
391 bufp = storenote(&notes[0], bufp);
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;
399 memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
400 prpsinfo.pr_state = 0;
401 prpsinfo.pr_sname = 'R';
402 prpsinfo.pr_zomb = 0;
404 strcpy(prpsinfo.pr_fname, "vmlinux");
405 strncpy(prpsinfo.pr_psargs, saved_command_line, ELF_PRARGSZ);
407 nhdr->p_filesz += notesize(&notes[1]);
408 bufp = storenote(&notes[1], bufp);
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;
416 nhdr->p_filesz += notesize(&notes[2]);
417 bufp = storenote(&notes[2], bufp);
419 } /* end elf_kcore_store_hdr() */
421 /*****************************************************************************/
423 * read from the ELF header and then kernel memory
425 static ssize_t
426 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
428 ssize_t acc = 0;
429 size_t size, tsz;
430 size_t elf_buflen;
431 int nphdr;
432 unsigned long start;
434 read_lock(&kclist_lock);
435 size = get_kcore_size(&nphdr, &elf_buflen);
437 if (buflen == 0 || *fpos >= size) {
438 read_unlock(&kclist_lock);
439 return 0;
442 /* trim buflen to not go beyond EOF */
443 if (buflen > size - *fpos)
444 buflen = size - *fpos;
446 /* construct an ELF core header if we'll need some of it */
447 if (*fpos < elf_buflen) {
448 char * elf_buf;
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;
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;
464 kfree(elf_buf);
465 buflen -= tsz;
466 *fpos += tsz;
467 buffer += tsz;
468 acc += tsz;
470 /* leave now if filled buffer already */
471 if (buflen == 0)
472 return acc;
473 } else
474 read_unlock(&kclist_lock);
477 * Check to see if our file offset matches with any of
478 * the addresses in the elf_phdr on our list.
480 start = kc_offset_to_vaddr(*fpos - elf_buflen);
481 if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
482 tsz = buflen;
484 while (buflen) {
485 struct kcore_list *m;
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;
492 read_unlock(&kclist_lock);
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;
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;
509 kfree(elf_buf);
510 } else {
511 if (kern_addr_valid(start)) {
512 unsigned long n;
514 n = copy_to_user(buffer, (char *)start, tsz);
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.
521 if (n) {
522 if (clear_user(buffer + tsz - n,
524 return -EFAULT;
526 } else {
527 if (clear_user(buffer, tsz))
528 return -EFAULT;
531 buflen -= tsz;
532 *fpos += tsz;
533 buffer += tsz;
534 acc += tsz;
535 start += tsz;
536 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
539 return acc;
543 static int open_kcore(struct inode *inode, struct file *filp)
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);
554 return 0;
558 static const struct file_operations proc_kcore_operations = {
559 .read = read_kcore,
560 .open = open_kcore,
561 .llseek = default_llseek,
564 #ifdef CONFIG_MEMORY_HOTPLUG
565 /* just remember that we have to update kcore */
566 static int __meminit kcore_callback(struct notifier_block *self,
567 unsigned long action, void *arg)
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);
576 return NOTIFY_OK;
578 #endif
581 static struct kcore_list kcore_vmalloc;
583 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
584 static struct kcore_list kcore_text;
586 * If defined, special segment is used for mapping kernel text instead of
587 * direct-map area. We need to create special TEXT section.
589 static void __init proc_kcore_text_init(void)
591 kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
593 #else
594 static void __init proc_kcore_text_init(void)
597 #endif
599 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
601 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
603 struct kcore_list kcore_modules;
604 static void __init add_modules_range(void)
606 kclist_add(&kcore_modules, (void *)MODULES_VADDR,
607 MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
609 #else
610 static void __init add_modules_range(void)
613 #endif
615 static int __init proc_kcore_init(void)
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. */
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);
633 return 0;
635 module_init(proc_kcore_init);