netns xfrm: fix "ip xfrm state|policy count" misreport
[safe/jmp/linux-2.6] / fs / proc / kcore.c
index f9327e5..a44a789 100644 (file)
 #include <linux/elfcore.h>
 #include <linux/vmalloc.h>
 #include <linux/highmem.h>
+#include <linux/bootmem.h>
 #include <linux/init.h>
 #include <asm/uaccess.h>
 #include <asm/io.h>
 #include <linux/list.h>
+#include <linux/ioport.h>
+#include <linux/memory.h>
+#include <asm/sections.h>
 
 #define CORE_STR "CORE"
 
 
 static struct proc_dir_entry *proc_root_kcore;
 
-static int open_kcore(struct inode * inode, struct file * filp)
-{
-       return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
-}
-
-static ssize_t read_kcore(struct file *, char __user *, size_t, loff_t *);
-
-static const struct file_operations proc_kcore_operations = {
-       .read           = read_kcore,
-       .open           = open_kcore,
-};
 
 #ifndef kc_vaddr_to_offset
 #define        kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
@@ -60,12 +53,14 @@ struct memelfnote
 
 static LIST_HEAD(kclist_head);
 static DEFINE_RWLOCK(kclist_lock);
+static int kcore_need_update = 1;
 
 void
-kclist_add(struct kcore_list *new, void *addr, size_t size)
+kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
 {
        new->addr = (unsigned long)addr;
        new->size = size;
+       new->type = type;
 
        write_lock(&kclist_lock);
        list_add_tail(&new->list, &kclist_head);
@@ -97,6 +92,177 @@ static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
        return size + *elf_buflen;
 }
 
+static void free_kclist_ents(struct list_head *head)
+{
+       struct kcore_list *tmp, *pos;
+
+       list_for_each_entry_safe(pos, tmp, head, list) {
+               list_del(&pos->list);
+               kfree(pos);
+       }
+}
+/*
+ * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
+ */
+static void __kcore_update_ram(struct list_head *list)
+{
+       int nphdr;
+       size_t size;
+       struct kcore_list *tmp, *pos;
+       LIST_HEAD(garbage);
+
+       write_lock(&kclist_lock);
+       if (kcore_need_update) {
+               list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
+                       if (pos->type == KCORE_RAM
+                               || pos->type == KCORE_VMEMMAP)
+                               list_move(&pos->list, &garbage);
+               }
+               list_splice_tail(list, &kclist_head);
+       } else
+               list_splice(list, &garbage);
+       kcore_need_update = 0;
+       proc_root_kcore->size = get_kcore_size(&nphdr, &size);
+       write_unlock(&kclist_lock);
+
+       free_kclist_ents(&garbage);
+}
+
+
+#ifdef CONFIG_HIGHMEM
+/*
+ * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
+ * because memory hole is not as big as !HIGHMEM case.
+ * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
+ */
+static int kcore_update_ram(void)
+{
+       LIST_HEAD(head);
+       struct kcore_list *ent;
+       int ret = 0;
+
+       ent = kmalloc(sizeof(*ent), GFP_KERNEL);
+       if (!ent)
+               return -ENOMEM;
+       ent->addr = (unsigned long)__va(0);
+       ent->size = max_low_pfn << PAGE_SHIFT;
+       ent->type = KCORE_RAM;
+       list_add(&ent->list, &head);
+       __kcore_update_ram(&head);
+       return ret;
+}
+
+#else /* !CONFIG_HIGHMEM */
+
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+/* calculate vmemmap's address from given system ram pfn and register it */
+int get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
+{
+       unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
+       unsigned long nr_pages = ent->size >> PAGE_SHIFT;
+       unsigned long start, end;
+       struct kcore_list *vmm, *tmp;
+
+
+       start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
+       end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
+       end = ALIGN(end, PAGE_SIZE);
+       /* overlap check (because we have to align page */
+       list_for_each_entry(tmp, head, list) {
+               if (tmp->type != KCORE_VMEMMAP)
+                       continue;
+               if (start < tmp->addr + tmp->size)
+                       if (end > tmp->addr)
+                               end = tmp->addr;
+       }
+       if (start < end) {
+               vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
+               if (!vmm)
+                       return 0;
+               vmm->addr = start;
+               vmm->size = end - start;
+               vmm->type = KCORE_VMEMMAP;
+               list_add_tail(&vmm->list, head);
+       }
+       return 1;
+
+}
+#else
+int get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
+{
+       return 1;
+}
+
+#endif
+
+static int
+kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
+{
+       struct list_head *head = (struct list_head *)arg;
+       struct kcore_list *ent;
+
+       ent = kmalloc(sizeof(*ent), GFP_KERNEL);
+       if (!ent)
+               return -ENOMEM;
+       ent->addr = (unsigned long)__va((pfn << PAGE_SHIFT));
+       ent->size = nr_pages << PAGE_SHIFT;
+
+       /* Sanity check: Can happen in 32bit arch...maybe */
+       if (ent->addr < (unsigned long) __va(0))
+               goto free_out;
+
+       /* cut not-mapped area. ....from ppc-32 code. */
+       if (ULONG_MAX - ent->addr < ent->size)
+               ent->size = ULONG_MAX - ent->addr;
+
+       /* cut when vmalloc() area is higher than direct-map area */
+       if (VMALLOC_START > (unsigned long)__va(0)) {
+               if (ent->addr > VMALLOC_START)
+                       goto free_out;
+               if (VMALLOC_START - ent->addr < ent->size)
+                       ent->size = VMALLOC_START - ent->addr;
+       }
+
+       ent->type = KCORE_RAM;
+       list_add_tail(&ent->list, head);
+
+       if (!get_sparsemem_vmemmap_info(ent, head)) {
+               list_del(&ent->list);
+               goto free_out;
+       }
+
+       return 0;
+free_out:
+       kfree(ent);
+       return 1;
+}
+
+static int kcore_update_ram(void)
+{
+       int nid, ret;
+       unsigned long end_pfn;
+       LIST_HEAD(head);
+
+       /* Not inialized....update now */
+       /* find out "max pfn" */
+       end_pfn = 0;
+       for_each_node_state(nid, N_HIGH_MEMORY) {
+               unsigned long node_end;
+               node_end  = NODE_DATA(nid)->node_start_pfn +
+                       NODE_DATA(nid)->node_spanned_pages;
+               if (end_pfn < node_end)
+                       end_pfn = node_end;
+       }
+       /* scan 0 to max_pfn */
+       ret = walk_system_ram_range(0, end_pfn, &head, kclist_add_private);
+       if (ret) {
+               free_kclist_ents(&head);
+               return -ENOMEM;
+       }
+       __kcore_update_ram(&head);
+       return ret;
+}
+#endif /* CONFIG_HIGHMEM */
 
 /*****************************************************************************/
 /*
@@ -265,7 +431,8 @@ read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
        unsigned long start;
 
        read_lock(&kclist_lock);
-       proc_root_kcore->size = size = get_kcore_size(&nphdr, &elf_buflen);
+       size = get_kcore_size(&nphdr, &elf_buflen);
+
        if (buflen == 0 || *fpos >= size) {
                read_unlock(&kclist_lock);
                return 0;
@@ -326,7 +493,7 @@ read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
                if (m == NULL) {
                        if (clear_user(buffer, tsz))
                                return -EFAULT;
-               } else if (is_vmalloc_addr((void *)start)) {
+               } else if (is_vmalloc_or_module_addr((void *)start)) {
                        char * elf_buf;
 
                        elf_buf = kzalloc(tsz, GFP_KERNEL);
@@ -371,9 +538,96 @@ read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
        return acc;
 }
 
+
+static int open_kcore(struct inode *inode, struct file *filp)
+{
+       if (!capable(CAP_SYS_RAWIO))
+               return -EPERM;
+       if (kcore_need_update)
+               kcore_update_ram();
+       if (i_size_read(inode) != proc_root_kcore->size) {
+               mutex_lock(&inode->i_mutex);
+               i_size_write(inode, proc_root_kcore->size);
+               mutex_unlock(&inode->i_mutex);
+       }
+       return 0;
+}
+
+
+static const struct file_operations proc_kcore_operations = {
+       .read           = read_kcore,
+       .open           = open_kcore,
+};
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+/* just remember that we have to update kcore */
+static int __meminit kcore_callback(struct notifier_block *self,
+                                   unsigned long action, void *arg)
+{
+       switch (action) {
+       case MEM_ONLINE:
+       case MEM_OFFLINE:
+               write_lock(&kclist_lock);
+               kcore_need_update = 1;
+               write_unlock(&kclist_lock);
+       }
+       return NOTIFY_OK;
+}
+#endif
+
+
+static struct kcore_list kcore_vmalloc;
+
+#ifdef CONFIG_ARCH_PROC_KCORE_TEXT
+static struct kcore_list kcore_text;
+/*
+ * If defined, special segment is used for mapping kernel text instead of
+ * direct-map area. We need to create special TEXT section.
+ */
+static void __init proc_kcore_text_init(void)
+{
+       kclist_add(&kcore_text, _stext, _end - _stext, KCORE_TEXT);
+}
+#else
+static void __init proc_kcore_text_init(void)
+{
+}
+#endif
+
+#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
+/*
+ * MODULES_VADDR has no intersection with VMALLOC_ADDR.
+ */
+struct kcore_list kcore_modules;
+static void __init add_modules_range(void)
+{
+       kclist_add(&kcore_modules, (void *)MODULES_VADDR,
+                       MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
+}
+#else
+static void __init add_modules_range(void)
+{
+}
+#endif
+
 static int __init proc_kcore_init(void)
 {
-       proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, &proc_kcore_operations);
+       proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
+                                     &proc_kcore_operations);
+       if (!proc_root_kcore) {
+               printk(KERN_ERR "couldn't create /proc/kcore\n");
+               return 0; /* Always returns 0. */
+       }
+       /* Store text area if it's special */
+       proc_kcore_text_init();
+       /* Store vmalloc area */
+       kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
+               VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
+       add_modules_range();
+       /* Store direct-map area from physical memory map */
+       kcore_update_ram();
+       hotplug_memory_notifier(kcore_callback, 0);
+
        return 0;
 }
 module_init(proc_kcore_init);