]> git.baikalelectronics.ru Git - kernel.git/commitdiff
proc/vmcore: convert oldmem_pfn_is_ram callback to more generic vmcore callbacks
authorDavid Hildenbrand <david@redhat.com>
Tue, 9 Nov 2021 02:31:48 +0000 (18:31 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 9 Nov 2021 18:02:48 +0000 (10:02 -0800)
Let's support multiple registered callbacks, making sure that
registering vmcore callbacks cannot fail.  Make the callback return a
bool instead of an int, handling how to deal with errors internally.
Drop unused HAVE_OLDMEM_PFN_IS_RAM.

We soon want to make use of this infrastructure from other drivers:
virtio-mem, registering one callback for each virtio-mem device, to
prevent reading unplugged virtio-mem memory.

Handle it via a generic vmcore_cb structure, prepared for future
extensions: for example, once we support virtio-mem on s390x where the
vmcore is completely constructed in the second kernel, we want to detect
and add plugged virtio-mem memory ranges to the vmcore in order for them
to get dumped properly.

Handle corner cases that are unexpected and shouldn't happen in sane
setups: registering a callback after the vmcore has already been opened
(warn only) and unregistering a callback after the vmcore has already been
opened (warn and essentially read only zeroes from that point on).

Link: https://lkml.kernel.org/r/20211005121430.30136-6-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Dave Young <dyoung@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/x86/kernel/aperture_64.c
arch/x86/xen/mmu_hvm.c
fs/proc/vmcore.c
include/linux/crash_dump.h

index 10562885f5fc6108d4f39a246814b5f21bbbe27f..af3ba08b684b5a0bbfbd40f6fa02f24b822b1b55 100644 (file)
@@ -73,12 +73,23 @@ static int gart_mem_pfn_is_ram(unsigned long pfn)
                      (pfn >= aperture_pfn_start + aperture_page_count));
 }
 
+#ifdef CONFIG_PROC_VMCORE
+static bool gart_oldmem_pfn_is_ram(struct vmcore_cb *cb, unsigned long pfn)
+{
+       return !!gart_mem_pfn_is_ram(pfn);
+}
+
+static struct vmcore_cb gart_vmcore_cb = {
+       .pfn_is_ram = gart_oldmem_pfn_is_ram,
+};
+#endif
+
 static void __init exclude_from_core(u64 aper_base, u32 aper_order)
 {
        aperture_pfn_start = aper_base >> PAGE_SHIFT;
        aperture_page_count = (32 * 1024 * 1024) << aper_order >> PAGE_SHIFT;
 #ifdef CONFIG_PROC_VMCORE
-       WARN_ON(register_oldmem_pfn_is_ram(&gart_mem_pfn_is_ram));
+       register_vmcore_cb(&gart_vmcore_cb);
 #endif
 #ifdef CONFIG_PROC_KCORE
        WARN_ON(register_mem_pfn_is_ram(&gart_mem_pfn_is_ram));
index 6ba8826dcdcc04fe2c43f8bf83781017a8e4a1d9..509bdee3ab9071c6fbdbbd5f86cf95890ea1a4a9 100644 (file)
  * The kdump kernel has to check whether a pfn of the crashed kernel
  * was a ballooned page. vmcore is using this function to decide
  * whether to access a pfn of the crashed kernel.
- * Returns 0 if the pfn is not backed by a RAM page, the caller may
+ * Returns "false" if the pfn is not backed by a RAM page, the caller may
  * handle the pfn special in this case.
  */
-static int xen_oldmem_pfn_is_ram(unsigned long pfn)
+static bool xen_vmcore_pfn_is_ram(struct vmcore_cb *cb, unsigned long pfn)
 {
        struct xen_hvm_get_mem_type a = {
                .domid = DOMID_SELF,
@@ -24,10 +24,13 @@ static int xen_oldmem_pfn_is_ram(unsigned long pfn)
 
        if (HYPERVISOR_hvm_op(HVMOP_get_mem_type, &a)) {
                pr_warn_once("Unexpected HVMOP_get_mem_type failure\n");
-               return -ENXIO;
+               return true;
        }
        return a.mem_type != HVMMEM_mmio_dm;
 }
+static struct vmcore_cb xen_vmcore_cb = {
+       .pfn_is_ram = xen_vmcore_pfn_is_ram,
+};
 #endif
 
 static void xen_hvm_exit_mmap(struct mm_struct *mm)
@@ -61,6 +64,6 @@ void __init xen_hvm_init_mmu_ops(void)
        if (is_pagetable_dying_supported())
                pv_ops.mmu.exit_mmap = xen_hvm_exit_mmap;
 #ifdef CONFIG_PROC_VMCORE
-       WARN_ON(register_oldmem_pfn_is_ram(&xen_oldmem_pfn_is_ram));
+       register_vmcore_cb(&xen_vmcore_cb);
 #endif
 }
index a9bd80ab670ecbf4cb7427d5d351f66678c8b3b0..7a04b2eca287d98ad0c7a31540d798cd4323e07b 100644 (file)
@@ -62,46 +62,75 @@ core_param(novmcoredd, vmcoredd_disabled, bool, 0);
 /* Device Dump Size */
 static size_t vmcoredd_orig_sz;
 
-/*
- * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
- * The called function has to take care of module refcounting.
- */
-static int (*oldmem_pfn_is_ram)(unsigned long pfn);
-
-int register_oldmem_pfn_is_ram(int (*fn)(unsigned long pfn))
+static DECLARE_RWSEM(vmcore_cb_rwsem);
+/* List of registered vmcore callbacks. */
+static LIST_HEAD(vmcore_cb_list);
+/* Whether we had a surprise unregistration of a callback. */
+static bool vmcore_cb_unstable;
+/* Whether the vmcore has been opened once. */
+static bool vmcore_opened;
+
+void register_vmcore_cb(struct vmcore_cb *cb)
 {
-       if (oldmem_pfn_is_ram)
-               return -EBUSY;
-       oldmem_pfn_is_ram = fn;
-       return 0;
+       down_write(&vmcore_cb_rwsem);
+       INIT_LIST_HEAD(&cb->next);
+       list_add_tail(&cb->next, &vmcore_cb_list);
+       /*
+        * Registering a vmcore callback after the vmcore was opened is
+        * very unusual (e.g., manual driver loading).
+        */
+       if (vmcore_opened)
+               pr_warn_once("Unexpected vmcore callback registration\n");
+       up_write(&vmcore_cb_rwsem);
 }
-EXPORT_SYMBOL_GPL(register_oldmem_pfn_is_ram);
+EXPORT_SYMBOL_GPL(register_vmcore_cb);
 
-void unregister_oldmem_pfn_is_ram(void)
+void unregister_vmcore_cb(struct vmcore_cb *cb)
 {
-       oldmem_pfn_is_ram = NULL;
-       wmb();
+       down_write(&vmcore_cb_rwsem);
+       list_del(&cb->next);
+       /*
+        * Unregistering a vmcore callback after the vmcore was opened is
+        * very unusual (e.g., forced driver removal), but we cannot stop
+        * unregistering.
+        */
+       if (vmcore_opened) {
+               pr_warn_once("Unexpected vmcore callback unregistration\n");
+               vmcore_cb_unstable = true;
+       }
+       up_write(&vmcore_cb_rwsem);
 }
-EXPORT_SYMBOL_GPL(unregister_oldmem_pfn_is_ram);
+EXPORT_SYMBOL_GPL(unregister_vmcore_cb);
 
 static bool pfn_is_ram(unsigned long pfn)
 {
-       int (*fn)(unsigned long pfn);
-       /* pfn is ram unless fn() checks pagetype */
+       struct vmcore_cb *cb;
        bool ret = true;
 
-       /*
-        * Ask hypervisor if the pfn is really ram.
-        * A ballooned page contains no data and reading from such a page
-        * will cause high load in the hypervisor.
-        */
-       fn = oldmem_pfn_is_ram;
-       if (fn)
-               ret = !!fn(pfn);
+       lockdep_assert_held_read(&vmcore_cb_rwsem);
+       if (unlikely(vmcore_cb_unstable))
+               return false;
+
+       list_for_each_entry(cb, &vmcore_cb_list, next) {
+               if (unlikely(!cb->pfn_is_ram))
+                       continue;
+               ret = cb->pfn_is_ram(cb, pfn);
+               if (!ret)
+                       break;
+       }
 
        return ret;
 }
 
+static int open_vmcore(struct inode *inode, struct file *file)
+{
+       down_read(&vmcore_cb_rwsem);
+       vmcore_opened = true;
+       up_read(&vmcore_cb_rwsem);
+
+       return 0;
+}
+
 /* Reads a page from the oldmem device from given offset. */
 ssize_t read_from_oldmem(char *buf, size_t count,
                         u64 *ppos, int userbuf,
@@ -117,6 +146,7 @@ ssize_t read_from_oldmem(char *buf, size_t count,
        offset = (unsigned long)(*ppos % PAGE_SIZE);
        pfn = (unsigned long)(*ppos / PAGE_SIZE);
 
+       down_read(&vmcore_cb_rwsem);
        do {
                if (count > (PAGE_SIZE - offset))
                        nr_bytes = PAGE_SIZE - offset;
@@ -136,8 +166,10 @@ ssize_t read_from_oldmem(char *buf, size_t count,
                                tmp = copy_oldmem_page(pfn, buf, nr_bytes,
                                                       offset, userbuf);
 
-                       if (tmp < 0)
+                       if (tmp < 0) {
+                               up_read(&vmcore_cb_rwsem);
                                return tmp;
+                       }
                }
                *ppos += nr_bytes;
                count -= nr_bytes;
@@ -147,6 +179,7 @@ ssize_t read_from_oldmem(char *buf, size_t count,
                offset = 0;
        } while (count);
 
+       up_read(&vmcore_cb_rwsem);
        return read;
 }
 
@@ -537,14 +570,19 @@ static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
                            unsigned long from, unsigned long pfn,
                            unsigned long size, pgprot_t prot)
 {
+       int ret;
+
        /*
         * Check if oldmem_pfn_is_ram was registered to avoid
         * looping over all pages without a reason.
         */
-       if (oldmem_pfn_is_ram)
-               return remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
+       down_read(&vmcore_cb_rwsem);
+       if (!list_empty(&vmcore_cb_list) || vmcore_cb_unstable)
+               ret = remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
        else
-               return remap_oldmem_pfn_range(vma, from, pfn, size, prot);
+               ret = remap_oldmem_pfn_range(vma, from, pfn, size, prot);
+       up_read(&vmcore_cb_rwsem);
+       return ret;
 }
 
 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
@@ -668,6 +706,7 @@ static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 #endif
 
 static const struct proc_ops vmcore_proc_ops = {
+       .proc_open      = open_vmcore,
        .proc_read      = read_vmcore,
        .proc_lseek     = default_llseek,
        .proc_mmap      = mmap_vmcore,
index 2618577a4d6da77e9983dd61f3c7ae8aa65c42a3..0c547d866f1e573ffc027d0094ffef5c0d6b2500 100644 (file)
@@ -91,9 +91,29 @@ static inline void vmcore_unusable(void)
                elfcorehdr_addr = ELFCORE_ADDR_ERR;
 }
 
-#define HAVE_OLDMEM_PFN_IS_RAM 1
-extern int register_oldmem_pfn_is_ram(int (*fn)(unsigned long pfn));
-extern void unregister_oldmem_pfn_is_ram(void);
+/**
+ * struct vmcore_cb - driver callbacks for /proc/vmcore handling
+ * @pfn_is_ram: check whether a PFN really is RAM and should be accessed when
+ *              reading the vmcore. Will return "true" if it is RAM or if the
+ *              callback cannot tell. If any callback returns "false", it's not
+ *              RAM and the page must not be accessed; zeroes should be
+ *              indicated in the vmcore instead. For example, a ballooned page
+ *              contains no data and reading from such a page will cause high
+ *              load in the hypervisor.
+ * @next: List head to manage registered callbacks internally; initialized by
+ *        register_vmcore_cb().
+ *
+ * vmcore callbacks allow drivers managing physical memory ranges to
+ * coordinate with vmcore handling code, for example, to prevent accessing
+ * physical memory ranges that should not be accessed when reading the vmcore,
+ * although included in the vmcore header as memory ranges to dump.
+ */
+struct vmcore_cb {
+       bool (*pfn_is_ram)(struct vmcore_cb *cb, unsigned long pfn);
+       struct list_head next;
+};
+extern void register_vmcore_cb(struct vmcore_cb *cb);
+extern void unregister_vmcore_cb(struct vmcore_cb *cb);
 
 #else /* !CONFIG_CRASH_DUMP */
 static inline bool is_kdump_kernel(void) { return 0; }