pstore/zone: Introduce common layer to manage storage zones
authorWeiXiong Liao <liaoweixiong@allwinnertech.com>
Wed, 25 Mar 2020 08:54:56 +0000 (16:54 +0800)
committerKees Cook <keescook@chromium.org>
Sat, 30 May 2020 17:34:03 +0000 (10:34 -0700)
Implement a common set of APIs needed to support pstore storage zones,
based on how ramoops is designed. This will be used by pstore/blk with
the intention of migrating pstore/ram in the future.

Signed-off-by: WeiXiong Liao <liaoweixiong@allwinnertech.com>
Link: https://lore.kernel.org/lkml/20200511233229.27745-2-keescook@chromium.org/
Co-developed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
fs/pstore/Kconfig
fs/pstore/Makefile
fs/pstore/zone.c [new file with mode: 0644]
include/linux/pstore_zone.h [new file with mode: 0644]

index 8f0369a..98d2457 100644 (file)
@@ -153,3 +153,10 @@ config PSTORE_RAM
          "ramoops.ko".
 
          For more information, see Documentation/admin-guide/ramoops.rst.
+
+config PSTORE_ZONE
+       tristate
+       depends on PSTORE
+       help
+         The common layer for pstore/blk (and pstore/ram in the future)
+         to manage storage in zones.
index 967b589..58a967c 100644 (file)
@@ -12,3 +12,6 @@ pstore-$(CONFIG_PSTORE_PMSG)  += pmsg.o
 
 ramoops-objs += ram.o ram_core.o
 obj-$(CONFIG_PSTORE_RAM)       += ramoops.o
+
+pstore_zone-objs += zone.o
+obj-$(CONFIG_PSTORE_ZONE)      += pstore_zone.o
diff --git a/fs/pstore/zone.c b/fs/pstore/zone.c
new file mode 100644 (file)
index 0000000..362d72a
--- /dev/null
@@ -0,0 +1,985 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Provide a pstore intermediate backend, organized into kernel memory
+ * allocated zones that are then mapped and flushed into a single
+ * contiguous region on a storage backend of some kind (block, mtd, etc).
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/mount.h>
+#include <linux/printk.h>
+#include <linux/fs.h>
+#include <linux/pstore_zone.h>
+#include <linux/kdev_t.h>
+#include <linux/device.h>
+#include <linux/namei.h>
+#include <linux/fcntl.h>
+#include <linux/uio.h>
+#include <linux/writeback.h>
+#include "internal.h"
+
+/**
+ * struct psz_head - header of zone to flush to storage
+ *
+ * @sig: signature to indicate header (PSZ_SIG xor PSZONE-type value)
+ * @datalen: length of data in @data
+ * @data: zone data.
+ */
+struct psz_buffer {
+#define PSZ_SIG (0x43474244) /* DBGC */
+       uint32_t sig;
+       atomic_t datalen;
+       uint8_t data[];
+};
+
+/**
+ * struct psz_kmsg_header - kmsg dump-specific header to flush to storage
+ *
+ * @magic: magic num for kmsg dump header
+ * @time: kmsg dump trigger time
+ * @compressed: whether conpressed
+ * @counter: kmsg dump counter
+ * @reason: the kmsg dump reason (e.g. oops, panic, etc)
+ * @data: pointer to log data
+ *
+ * This is a sub-header for a kmsg dump, trailing after &psz_buffer.
+ */
+struct psz_kmsg_header {
+#define PSTORE_KMSG_HEADER_MAGIC 0x4dfc3ae5 /* Just a random number */
+       uint32_t magic;
+       struct timespec64 time;
+       bool compressed;
+       uint32_t counter;
+       enum kmsg_dump_reason reason;
+       uint8_t data[];
+};
+
+/**
+ * struct pstore_zone - single stored buffer
+ *
+ * @off: zone offset of storage
+ * @type: front-end type for this zone
+ * @name: front-end name for this zone
+ * @buffer: pointer to data buffer managed by this zone
+ * @oldbuf: pointer to old data buffer
+ * @buffer_size: bytes in @buffer->data
+ * @should_recover: whether this zone should recover from storage
+ * @dirty: whether the data in @buffer dirty
+ *
+ * zone structure in memory.
+ */
+struct pstore_zone {
+       loff_t off;
+       const char *name;
+       enum pstore_type_id type;
+
+       struct psz_buffer *buffer;
+       struct psz_buffer *oldbuf;
+       size_t buffer_size;
+       bool should_recover;
+       atomic_t dirty;
+};
+
+/**
+ * struct psz_context - all about running state of pstore/zone
+ *
+ * @kpszs: kmsg dump storage zones
+ * @kmsg_max_cnt: max count of @kpszs
+ * @kmsg_read_cnt: counter of total read kmsg dumps
+ * @kmsg_write_cnt: counter of total kmsg dump writes
+ * @oops_counter: counter of oops dumps
+ * @panic_counter: counter of panic dumps
+ * @recovered: whether finished recovering data from storage
+ * @on_panic: whether panic is happening
+ * @pstore_zone_info_lock: lock to @pstore_zone_info
+ * @pstore_zone_info: information from backend
+ * @pstore: structure for pstore
+ */
+struct psz_context {
+       struct pstore_zone **kpszs;
+       unsigned int kmsg_max_cnt;
+       unsigned int kmsg_read_cnt;
+       unsigned int kmsg_write_cnt;
+       /*
+        * These counters should be calculated during recovery.
+        * It records the oops/panic times after crashes rather than boots.
+        */
+       unsigned int oops_counter;
+       unsigned int panic_counter;
+       atomic_t recovered;
+       atomic_t on_panic;
+
+       /*
+        * pstore_zone_info_lock protects this entire structure during calls
+        * to register_pstore_zone()/unregister_pstore_zone().
+        */
+       struct mutex pstore_zone_info_lock;
+       struct pstore_zone_info *pstore_zone_info;
+       struct pstore_info pstore;
+};
+static struct psz_context pstore_zone_cxt;
+
+/**
+ * enum psz_flush_mode - flush mode for psz_zone_write()
+ *
+ * @FLUSH_NONE: do not flush to storage but update data on memory
+ * @FLUSH_PART: just flush part of data including meta data to storage
+ * @FLUSH_META: just flush meta data of zone to storage
+ * @FLUSH_ALL: flush all of zone
+ */
+enum psz_flush_mode {
+       FLUSH_NONE = 0,
+       FLUSH_PART,
+       FLUSH_META,
+       FLUSH_ALL,
+};
+
+static inline int buffer_datalen(struct pstore_zone *zone)
+{
+       return atomic_read(&zone->buffer->datalen);
+}
+
+static inline bool is_on_panic(void)
+{
+       return atomic_read(&pstore_zone_cxt.on_panic);
+}
+
+static ssize_t psz_zone_read(struct pstore_zone *zone, char *buf,
+               size_t len, unsigned long off)
+{
+       if (!buf || !zone->buffer)
+               return -EINVAL;
+       if (off > zone->buffer_size)
+               return -EINVAL;
+       len = min_t(size_t, len, zone->buffer_size - off);
+       memcpy(buf, zone->buffer->data + off, len);
+       return len;
+}
+
+static int psz_zone_write(struct pstore_zone *zone,
+               enum psz_flush_mode flush_mode, const char *buf,
+               size_t len, unsigned long off)
+{
+       struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info;
+       ssize_t wcnt = 0;
+       ssize_t (*writeop)(const char *buf, size_t bytes, loff_t pos);
+       size_t wlen;
+
+       if (off > zone->buffer_size)
+               return -EINVAL;
+
+       wlen = min_t(size_t, len, zone->buffer_size - off);
+       if (buf && wlen) {
+               memcpy(zone->buffer->data + off, buf, wlen);
+               atomic_set(&zone->buffer->datalen, wlen + off);
+       }
+
+       /* avoid to damage old records */
+       if (!is_on_panic() && !atomic_read(&pstore_zone_cxt.recovered))
+               goto dirty;
+
+       writeop = is_on_panic() ? info->panic_write : info->write;
+       if (!writeop)
+               goto dirty;
+
+       switch (flush_mode) {
+       case FLUSH_NONE:
+               if (unlikely(buf && wlen))
+                       goto dirty;
+               return 0;
+       case FLUSH_PART:
+               wcnt = writeop((const char *)zone->buffer->data + off, wlen,
+                               zone->off + sizeof(*zone->buffer) + off);
+               if (wcnt != wlen)
+                       goto dirty;
+               fallthrough;
+       case FLUSH_META:
+               wlen = sizeof(struct psz_buffer);
+               wcnt = writeop((const char *)zone->buffer, wlen, zone->off);
+               if (wcnt != wlen)
+                       goto dirty;
+               break;
+       case FLUSH_ALL:
+               wlen = zone->buffer_size + sizeof(*zone->buffer);
+               wcnt = writeop((const char *)zone->buffer, wlen, zone->off);
+               if (wcnt != wlen)
+                       goto dirty;
+               break;
+       }
+
+       return 0;
+dirty:
+       atomic_set(&zone->dirty, true);
+       return -EBUSY;
+}
+
+static int psz_flush_dirty_zone(struct pstore_zone *zone)
+{
+       int ret;
+
+       if (unlikely(!zone))
+               return -EINVAL;
+
+       if (unlikely(!atomic_read(&pstore_zone_cxt.recovered)))
+               return -EBUSY;
+
+       if (!atomic_xchg(&zone->dirty, false))
+               return 0;
+
+       ret = psz_zone_write(zone, FLUSH_ALL, NULL, 0, 0);
+       if (ret)
+               atomic_set(&zone->dirty, true);
+       return ret;
+}
+
+static int psz_flush_dirty_zones(struct pstore_zone **zones, unsigned int cnt)
+{
+       int i, ret;
+       struct pstore_zone *zone;
+
+       if (!zones)
+               return -EINVAL;
+
+       for (i = 0; i < cnt; i++) {
+               zone = zones[i];
+               if (!zone)
+                       return -EINVAL;
+               ret = psz_flush_dirty_zone(zone);
+               if (ret)
+                       return ret;
+       }
+       return 0;
+}
+
+static int psz_move_zone(struct pstore_zone *old, struct pstore_zone *new)
+{
+       const char *data = (const char *)old->buffer->data;
+       int ret;
+
+       ret = psz_zone_write(new, FLUSH_ALL, data, buffer_datalen(old), 0);
+       if (ret) {
+               atomic_set(&new->buffer->datalen, 0);
+               atomic_set(&new->dirty, false);
+               return ret;
+       }
+       atomic_set(&old->buffer->datalen, 0);
+       return 0;
+}
+
+static int psz_kmsg_recover_data(struct psz_context *cxt)
+{
+       struct pstore_zone_info *info = cxt->pstore_zone_info;
+       struct pstore_zone *zone = NULL;
+       struct psz_buffer *buf;
+       unsigned long i;
+       ssize_t rcnt;
+
+       if (!info->read)
+               return -EINVAL;
+
+       for (i = 0; i < cxt->kmsg_max_cnt; i++) {
+               zone = cxt->kpszs[i];
+               if (unlikely(!zone))
+                       return -EINVAL;
+               if (atomic_read(&zone->dirty)) {
+                       unsigned int wcnt = cxt->kmsg_write_cnt;
+                       struct pstore_zone *new = cxt->kpszs[wcnt];
+                       int ret;
+
+                       ret = psz_move_zone(zone, new);
+                       if (ret) {
+                               pr_err("move zone from %lu to %d failed\n",
+                                               i, wcnt);
+                               return ret;
+                       }
+                       cxt->kmsg_write_cnt = (wcnt + 1) % cxt->kmsg_max_cnt;
+               }
+               if (!zone->should_recover)
+                       continue;
+               buf = zone->buffer;
+               rcnt = info->read((char *)buf, zone->buffer_size + sizeof(*buf),
+                               zone->off);
+               if (rcnt != zone->buffer_size + sizeof(*buf))
+                       return (int)rcnt < 0 ? (int)rcnt : -EIO;
+       }
+       return 0;
+}
+
+static int psz_kmsg_recover_meta(struct psz_context *cxt)
+{
+       struct pstore_zone_info *info = cxt->pstore_zone_info;
+       struct pstore_zone *zone;
+       size_t rcnt, len;
+       struct psz_buffer *buf;
+       struct psz_kmsg_header *hdr;
+       struct timespec64 time = { };
+       unsigned long i;
+       /*
+        * Recover may on panic, we can't allocate any memory by kmalloc.
+        * So, we use local array instead.
+        */
+       char buffer_header[sizeof(*buf) + sizeof(*hdr)] = {0};
+
+       if (!info->read)
+               return -EINVAL;
+
+       len = sizeof(*buf) + sizeof(*hdr);
+       buf = (struct psz_buffer *)buffer_header;
+       for (i = 0; i < cxt->kmsg_max_cnt; i++) {
+               zone = cxt->kpszs[i];
+               if (unlikely(!zone))
+                       return -EINVAL;
+
+               rcnt = info->read((char *)buf, len, zone->off);
+               if (rcnt != len) {
+                       pr_err("read %s with id %lu failed\n", zone->name, i);
+                       return (int)rcnt < 0 ? (int)rcnt : -EIO;
+               }
+
+               if (buf->sig != zone->buffer->sig) {
+                       pr_debug("no valid data in kmsg dump zone %lu\n", i);
+                       continue;
+               }
+
+               if (zone->buffer_size < atomic_read(&buf->datalen)) {
+                       pr_info("found overtop zone: %s: id %lu, off %lld, size %zu\n",
+                                       zone->name, i, zone->off,
+                                       zone->buffer_size);
+                       continue;
+               }
+
+               hdr = (struct psz_kmsg_header *)buf->data;
+               if (hdr->magic != PSTORE_KMSG_HEADER_MAGIC) {
+                       pr_info("found invalid zone: %s: id %lu, off %lld, size %zu\n",
+                                       zone->name, i, zone->off,
+                                       zone->buffer_size);
+                       continue;
+               }
+
+               /*
+                * we get the newest zone, and the next one must be the oldest
+                * or unused zone, because we do write one by one like a circle.
+                */
+               if (hdr->time.tv_sec >= time.tv_sec) {
+                       time.tv_sec = hdr->time.tv_sec;
+                       cxt->kmsg_write_cnt = (i + 1) % cxt->kmsg_max_cnt;
+               }
+
+               if (hdr->reason == KMSG_DUMP_OOPS)
+                       cxt->oops_counter =
+                               max(cxt->oops_counter, hdr->counter);
+               else if (hdr->reason == KMSG_DUMP_PANIC)
+                       cxt->panic_counter =
+                               max(cxt->panic_counter, hdr->counter);
+
+               if (!atomic_read(&buf->datalen)) {
+                       pr_debug("found erased zone: %s: id %lu, off %lld, size %zu, datalen %d\n",
+                                       zone->name, i, zone->off,
+                                       zone->buffer_size,
+                                       atomic_read(&buf->datalen));
+                       continue;
+               }
+
+               if (!is_on_panic())
+                       zone->should_recover = true;
+               pr_debug("found nice zone: %s: id %lu, off %lld, size %zu, datalen %d\n",
+                               zone->name, i, zone->off,
+                               zone->buffer_size, atomic_read(&buf->datalen));
+       }
+
+       return 0;
+}
+
+static int psz_kmsg_recover(struct psz_context *cxt)
+{
+       int ret;
+
+       if (!cxt->kpszs)
+               return 0;
+
+       ret = psz_kmsg_recover_meta(cxt);
+       if (ret)
+               goto recover_fail;
+
+       ret = psz_kmsg_recover_data(cxt);
+       if (ret)
+               goto recover_fail;
+
+       return 0;
+recover_fail:
+       pr_debug("psz_recover_kmsg failed\n");
+       return ret;
+}
+
+/**
+ * psz_recovery() - recover data from storage
+ * @cxt: the context of pstore/zone
+ *
+ * recovery means reading data back from storage after rebooting
+ *
+ * Return: 0 on success, others on failure.
+ */
+static inline int psz_recovery(struct psz_context *cxt)
+{
+       int ret;
+
+       if (atomic_read(&cxt->recovered))
+               return 0;
+
+       ret = psz_kmsg_recover(cxt);
+
+       if (unlikely(ret))
+               pr_err("recover failed\n");
+       else {
+               pr_debug("recover end!\n");
+               atomic_set(&cxt->recovered, 1);
+       }
+       return ret;
+}
+
+static int psz_pstore_open(struct pstore_info *psi)
+{
+       struct psz_context *cxt = psi->data;
+
+       cxt->kmsg_read_cnt = 0;
+       return 0;
+}
+
+static inline bool psz_ok(struct pstore_zone *zone)
+{
+       if (zone && zone->buffer && buffer_datalen(zone))
+               return true;
+       return false;
+}
+
+static inline int psz_kmsg_erase(struct psz_context *cxt,
+               struct pstore_zone *zone, struct pstore_record *record)
+{
+       struct psz_buffer *buffer = zone->buffer;
+       struct psz_kmsg_header *hdr =
+               (struct psz_kmsg_header *)buffer->data;
+
+       if (unlikely(!psz_ok(zone)))
+               return 0;
+       /* this zone is already updated, no need to erase */
+       if (record->count != hdr->counter)
+               return 0;
+
+       atomic_set(&zone->buffer->datalen, 0);
+       return psz_zone_write(zone, FLUSH_META, NULL, 0, 0);
+}
+
+static int psz_pstore_erase(struct pstore_record *record)
+{
+       struct psz_context *cxt = record->psi->data;
+
+       switch (record->type) {
+       case PSTORE_TYPE_DMESG:
+               if (record->id >= cxt->kmsg_max_cnt)
+                       return -EINVAL;
+               return psz_kmsg_erase(cxt, cxt->kpszs[record->id], record);
+       default:
+               return -EINVAL;
+       }
+}
+
+static void psz_write_kmsg_hdr(struct pstore_zone *zone,
+               struct pstore_record *record)
+{
+       struct psz_context *cxt = record->psi->data;
+       struct psz_buffer *buffer = zone->buffer;
+       struct psz_kmsg_header *hdr =
+               (struct psz_kmsg_header *)buffer->data;
+
+       hdr->magic = PSTORE_KMSG_HEADER_MAGIC;
+       hdr->compressed = record->compressed;
+       hdr->time.tv_sec = record->time.tv_sec;
+       hdr->time.tv_nsec = record->time.tv_nsec;
+       hdr->reason = record->reason;
+       if (hdr->reason == KMSG_DUMP_OOPS)
+               hdr->counter = ++cxt->oops_counter;
+       else if (hdr->reason == KMSG_DUMP_PANIC)
+               hdr->counter = ++cxt->panic_counter;
+       else
+               hdr->counter = 0;
+}
+
+static inline int notrace psz_kmsg_write_record(struct psz_context *cxt,
+               struct pstore_record *record)
+{
+       size_t size, hlen;
+       struct pstore_zone *zone;
+       unsigned int zonenum;
+
+       zonenum = cxt->kmsg_write_cnt;
+       zone = cxt->kpszs[zonenum];
+       if (unlikely(!zone))
+               return -ENOSPC;
+       cxt->kmsg_write_cnt = (zonenum + 1) % cxt->kmsg_max_cnt;
+
+       pr_debug("write %s to zone id %d\n", zone->name, zonenum);
+       psz_write_kmsg_hdr(zone, record);
+       hlen = sizeof(struct psz_kmsg_header);
+       size = min_t(size_t, record->size, zone->buffer_size - hlen);
+       return psz_zone_write(zone, FLUSH_ALL, record->buf, size, hlen);
+}
+
+static int notrace psz_kmsg_write(struct psz_context *cxt,
+               struct pstore_record *record)
+{
+       int ret;
+
+       /*
+        * Explicitly only take the first part of any new crash.
+        * If our buffer is larger than kmsg_bytes, this can never happen,
+        * and if our buffer is smaller than kmsg_bytes, we don't want the
+        * report split across multiple records.
+        */
+       if (record->part != 1)
+               return -ENOSPC;
+
+       if (!cxt->kpszs)
+               return -ENOSPC;
+
+       ret = psz_kmsg_write_record(cxt, record);
+       if (!ret) {
+               pr_debug("try to flush other dirty zones\n");
+               psz_flush_dirty_zones(cxt->kpszs, cxt->kmsg_max_cnt);
+       }
+
+       /* always return 0 as we had handled it on buffer */
+       return 0;
+}
+
+static int notrace psz_pstore_write(struct pstore_record *record)
+{
+       struct psz_context *cxt = record->psi->data;
+
+       if (record->type == PSTORE_TYPE_DMESG &&
+                       record->reason == KMSG_DUMP_PANIC)
+               atomic_set(&cxt->on_panic, 1);
+
+       switch (record->type) {
+       case PSTORE_TYPE_DMESG:
+               return psz_kmsg_write(cxt, record);
+       default:
+               return -EINVAL;
+       }
+}
+
+static struct pstore_zone *psz_read_next_zone(struct psz_context *cxt)
+{
+       struct pstore_zone *zone = NULL;
+
+       while (cxt->kmsg_read_cnt < cxt->kmsg_max_cnt) {
+               zone = cxt->kpszs[cxt->kmsg_read_cnt++];
+               if (psz_ok(zone))
+                       return zone;
+       }
+
+       return NULL;
+}
+
+static int psz_kmsg_read_hdr(struct pstore_zone *zone,
+               struct pstore_record *record)
+{
+       struct psz_buffer *buffer = zone->buffer;
+       struct psz_kmsg_header *hdr =
+               (struct psz_kmsg_header *)buffer->data;
+
+       if (hdr->magic != PSTORE_KMSG_HEADER_MAGIC)
+               return -EINVAL;
+       record->compressed = hdr->compressed;
+       record->time.tv_sec = hdr->time.tv_sec;
+       record->time.tv_nsec = hdr->time.tv_nsec;
+       record->reason = hdr->reason;
+       record->count = hdr->counter;
+       return 0;
+}
+
+static ssize_t psz_kmsg_read(struct pstore_zone *zone,
+               struct pstore_record *record)
+{
+       ssize_t size, hlen = 0;
+
+       size = buffer_datalen(zone);
+       /* Clear and skip this kmsg dump record if it has no valid header */
+       if (psz_kmsg_read_hdr(zone, record)) {
+               atomic_set(&zone->buffer->datalen, 0);
+               atomic_set(&zone->dirty, 0);
+               return -ENOMSG;
+       }
+       size -= sizeof(struct psz_kmsg_header);
+
+       if (!record->compressed) {
+               char *buf = kasprintf(GFP_KERNEL, "%s: Total %d times\n",
+                                     kmsg_dump_reason_str(record->reason),
+                                     record->count);
+               hlen = strlen(buf);
+               record->buf = krealloc(buf, hlen + size, GFP_KERNEL);
+               if (!record->buf) {
+                       kfree(buf);
+                       return -ENOMEM;
+               }
+       } else {
+               record->buf = kmalloc(size, GFP_KERNEL);
+               if (!record->buf)
+                       return -ENOMEM;
+       }
+
+       size = psz_zone_read(zone, record->buf + hlen, size,
+                       sizeof(struct psz_kmsg_header));
+       if (unlikely(size < 0)) {
+               kfree(record->buf);
+               return -ENOMSG;
+       }
+
+       return size + hlen;
+}
+
+static ssize_t psz_pstore_read(struct pstore_record *record)
+{
+       struct psz_context *cxt = record->psi->data;
+       ssize_t (*readop)(struct pstore_zone *zone,
+                       struct pstore_record *record);
+       struct pstore_zone *zone;
+       ssize_t ret;
+
+       /* before read, we must recover from storage */
+       ret = psz_recovery(cxt);
+       if (ret)
+               return ret;
+
+next_zone:
+       zone = psz_read_next_zone(cxt);
+       if (!zone)
+               return 0;
+
+       record->type = zone->type;
+       switch (record->type) {
+       case PSTORE_TYPE_DMESG:
+               readop = psz_kmsg_read;
+               record->id = cxt->kmsg_read_cnt - 1;
+               break;
+       default:
+               goto next_zone;
+       }
+
+       ret = readop(zone, record);
+       if (ret == -ENOMSG)
+               goto next_zone;
+       return ret;
+}
+
+static struct psz_context pstore_zone_cxt = {
+       .pstore_zone_info_lock =
+               __MUTEX_INITIALIZER(pstore_zone_cxt.pstore_zone_info_lock),
+       .recovered = ATOMIC_INIT(0),
+       .on_panic = ATOMIC_INIT(0),
+       .pstore = {
+               .owner = THIS_MODULE,
+               .open = psz_pstore_open,
+               .read = psz_pstore_read,
+               .write = psz_pstore_write,
+               .erase = psz_pstore_erase,
+       },
+};
+
+static void psz_free_zone(struct pstore_zone **pszone)
+{
+       struct pstore_zone *zone = *pszone;
+
+       if (!zone)
+               return;
+
+       kfree(zone->buffer);
+       kfree(zone);
+       *pszone = NULL;
+}
+
+static void psz_free_zones(struct pstore_zone ***pszones, unsigned int *cnt)
+{
+       struct pstore_zone **zones = *pszones;
+
+       if (!zones)
+               return;
+
+       while (*cnt > 0) {
+               (*cnt)--;
+               psz_free_zone(&(zones[*cnt]));
+       }
+       kfree(zones);
+       *pszones = NULL;
+}
+
+static void psz_free_all_zones(struct psz_context *cxt)
+{
+       if (cxt->kpszs)
+               psz_free_zones(&cxt->kpszs, &cxt->kmsg_max_cnt);
+}
+
+static struct pstore_zone *psz_init_zone(enum pstore_type_id type,
+               loff_t *off, size_t size)
+{
+       struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info;
+       struct pstore_zone *zone;
+       const char *name = pstore_type_to_name(type);
+
+       if (!size)
+               return NULL;
+
+       if (*off + size > info->total_size) {
+               pr_err("no room for %s (0x%zx@0x%llx over 0x%lx)\n",
+                       name, size, *off, info->total_size);
+               return ERR_PTR(-ENOMEM);
+       }
+
+       zone = kzalloc(sizeof(struct pstore_zone), GFP_KERNEL);
+       if (!zone)
+               return ERR_PTR(-ENOMEM);
+
+       zone->buffer = kmalloc(size, GFP_KERNEL);
+       if (!zone->buffer) {
+               kfree(zone);
+               return ERR_PTR(-ENOMEM);
+       }
+       memset(zone->buffer, 0xFF, size);
+       zone->off = *off;
+       zone->name = name;
+       zone->type = type;
+       zone->buffer_size = size - sizeof(struct psz_buffer);
+       zone->buffer->sig = type ^ PSZ_SIG;
+       atomic_set(&zone->dirty, 0);
+       atomic_set(&zone->buffer->datalen, 0);
+
+       *off += size;
+
+       pr_debug("pszone %s: off 0x%llx, %zu header, %zu data\n", zone->name,
+                       zone->off, sizeof(*zone->buffer), zone->buffer_size);
+       return zone;
+}
+
+static struct pstore_zone **psz_init_zones(enum pstore_type_id type,
+       loff_t *off, size_t total_size, ssize_t record_size,
+       unsigned int *cnt)
+{
+       struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info;
+       struct pstore_zone **zones, *zone;
+       const char *name = pstore_type_to_name(type);
+       int c, i;
+
+       *cnt = 0;
+       if (!total_size || !record_size)
+               return NULL;
+
+       if (*off + total_size > info->total_size) {
+               pr_err("no room for zones %s (0x%zx@0x%llx over 0x%lx)\n",
+                       name, total_size, *off, info->total_size);
+               return ERR_PTR(-ENOMEM);
+       }
+
+       c = total_size / record_size;
+       zones = kcalloc(c, sizeof(*zones), GFP_KERNEL);
+       if (!zones) {
+               pr_err("allocate for zones %s failed\n", name);
+               return ERR_PTR(-ENOMEM);
+       }
+       memset(zones, 0, c * sizeof(*zones));
+
+       for (i = 0; i < c; i++) {
+               zone = psz_init_zone(type, off, record_size);
+               if (!zone || IS_ERR(zone)) {
+                       pr_err("initialize zones %s failed\n", name);
+                       psz_free_zones(&zones, &i);
+                       return (void *)zone;
+               }
+               zones[i] = zone;
+       }
+
+       *cnt = c;
+       return zones;
+}
+
+static int psz_alloc_zones(struct psz_context *cxt)
+{
+       struct pstore_zone_info *info = cxt->pstore_zone_info;
+       loff_t off = 0;
+       int err;
+       size_t size;
+
+       size = info->total_size;
+       cxt->kpszs = psz_init_zones(PSTORE_TYPE_DMESG, &off, size,
+                       info->kmsg_size, &cxt->kmsg_max_cnt);
+       if (IS_ERR(cxt->kpszs)) {
+               err = PTR_ERR(cxt->kpszs);
+               cxt->kpszs = NULL;
+               goto fail_out;
+       }
+
+       return 0;
+fail_out:
+       return err;
+}
+
+/**
+ * register_pstore_zone() - register to pstore/zone
+ *
+ * @info: back-end driver information. See &struct pstore_zone_info.
+ *
+ * Only one back-end at one time.
+ *
+ * Return: 0 on success, others on failure.
+ */
+int register_pstore_zone(struct pstore_zone_info *info)
+{
+       int err = -EINVAL;
+       struct psz_context *cxt = &pstore_zone_cxt;
+
+       if (info->total_size < 4096) {
+               pr_warn("total_size must be >= 4096\n");
+               return -EINVAL;
+       }
+
+       if (!info->kmsg_size) {
+               pr_warn("at least one record size must be non-zero\n");
+               return -EINVAL;
+       }
+
+       if (!info->name || !info->name[0])
+               return -EINVAL;
+
+#define check_size(name, size) {                                       \
+               if (info->name > 0 && info->name < (size)) {            \
+                       pr_err(#name " must be over %d\n", (size));     \
+                       return -EINVAL;                                 \
+               }                                                       \
+               if (info->name & (size - 1)) {                          \
+                       pr_err(#name " must be a multiple of %d\n",     \
+                                       (size));                        \
+                       return -EINVAL;                                 \
+               }                                                       \
+       }
+
+       check_size(total_size, 4096);
+       check_size(kmsg_size, SECTOR_SIZE);
+
+#undef check_size
+
+       /*
+        * the @read and @write must be applied.
+        * if no @read, pstore may mount failed.
+        * if no @write, pstore do not support to remove record file.
+        */
+       if (!info->read || !info->write) {
+               pr_err("no valid general read/write interface\n");
+               return -EINVAL;
+       }
+
+       mutex_lock(&cxt->pstore_zone_info_lock);
+       if (cxt->pstore_zone_info) {
+               pr_warn("'%s' already loaded: ignoring '%s'\n",
+                               cxt->pstore_zone_info->name, info->name);
+               mutex_unlock(&cxt->pstore_zone_info_lock);
+               return -EBUSY;
+       }
+       cxt->pstore_zone_info = info;
+
+       pr_debug("register %s with properties:\n", info->name);
+       pr_debug("\ttotal size : %ld Bytes\n", info->total_size);
+       pr_debug("\tkmsg size : %ld Bytes\n", info->kmsg_size);
+
+       err = psz_alloc_zones(cxt);
+       if (err) {
+               pr_err("alloc zones failed\n");
+               goto fail_out;
+       }
+
+       if (info->kmsg_size) {
+               cxt->pstore.bufsize = cxt->kpszs[0]->buffer_size -
+                       sizeof(struct psz_kmsg_header);
+               cxt->pstore.buf = kzalloc(cxt->pstore.bufsize, GFP_KERNEL);
+               if (!cxt->pstore.buf) {
+                       err = -ENOMEM;
+                       goto fail_free;
+               }
+       }
+       cxt->pstore.data = cxt;
+
+       pr_info("registered %s as backend for", info->name);
+       cxt->pstore.max_reason = info->max_reason;
+       cxt->pstore.name = info->name;
+       if (info->kmsg_size) {
+               cxt->pstore.flags |= PSTORE_FLAGS_DMESG;
+               pr_cont(" kmsg(%s",
+                       kmsg_dump_reason_str(cxt->pstore.max_reason));
+               if (cxt->pstore_zone_info->panic_write)
+                       pr_cont(",panic_write");
+               pr_cont(")");
+       }
+       pr_cont("\n");
+
+       err = pstore_register(&cxt->pstore);
+       if (err) {
+               pr_err("registering with pstore failed\n");
+               goto fail_free;
+       }
+       mutex_unlock(&pstore_zone_cxt.pstore_zone_info_lock);
+
+       return 0;
+
+fail_free:
+       kfree(cxt->pstore.buf);
+       cxt->pstore.buf = NULL;
+       cxt->pstore.bufsize = 0;
+       psz_free_all_zones(cxt);
+fail_out:
+       pstore_zone_cxt.pstore_zone_info = NULL;
+       mutex_unlock(&pstore_zone_cxt.pstore_zone_info_lock);
+       return err;
+}
+EXPORT_SYMBOL_GPL(register_pstore_zone);
+
+/**
+ * unregister_pstore_zone() - unregister to pstore/zone
+ *
+ * @info: back-end driver information. See struct pstore_zone_info.
+ */
+void unregister_pstore_zone(struct pstore_zone_info *info)
+{
+       struct psz_context *cxt = &pstore_zone_cxt;
+
+       mutex_lock(&cxt->pstore_zone_info_lock);
+       if (!cxt->pstore_zone_info) {
+               mutex_unlock(&cxt->pstore_zone_info_lock);
+               return;
+       }
+
+       /* Stop incoming writes from pstore. */
+       pstore_unregister(&cxt->pstore);
+
+       /* Clean up allocations. */
+       kfree(cxt->pstore.buf);
+       cxt->pstore.buf = NULL;
+       cxt->pstore.bufsize = 0;
+       cxt->pstore_zone_info = NULL;
+
+       psz_free_all_zones(cxt);
+
+       /* Clear counters and zone state. */
+       cxt->oops_counter = 0;
+       cxt->panic_counter = 0;
+       atomic_set(&cxt->recovered, 0);
+       atomic_set(&cxt->on_panic, 0);
+
+       mutex_unlock(&cxt->pstore_zone_info_lock);
+}
+EXPORT_SYMBOL_GPL(unregister_pstore_zone);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("WeiXiong Liao <liaoweixiong@allwinnertech.com>");
+MODULE_AUTHOR("Kees Cook <keescook@chromium.org>");
+MODULE_DESCRIPTION("Storage Manager for pstore/blk");
diff --git a/include/linux/pstore_zone.h b/include/linux/pstore_zone.h
new file mode 100644 (file)
index 0000000..eb005d9
--- /dev/null
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#ifndef __PSTORE_ZONE_H_
+#define __PSTORE_ZONE_H_
+
+#include <linux/types.h>
+
+typedef ssize_t (*pstore_zone_read_op)(char *, size_t, loff_t);
+typedef ssize_t (*pstore_zone_write_op)(const char *, size_t, loff_t);
+/**
+ * struct pstore_zone_info - pstore/zone back-end driver structure
+ *
+ * @owner:     Module which is responsible for this back-end driver.
+ * @name:      Name of the back-end driver.
+ * @total_size: The total size in bytes pstore/zone can use. It must be greater
+ *             than 4096 and be multiple of 4096.
+ * @kmsg_size: The size of oops/panic zone. Zero means disabled, otherwise,
+ *             it must be multiple of SECTOR_SIZE(512 Bytes).
+ * @max_reason: Maximum kmsg dump reason to store.
+ * @read:      The general read operation. Both of the function parameters
+ *             @size and @offset are relative value to storage.
+ *             On success, the number of bytes should be returned, others
+ *             means error.
+ * @write:     The same as @read.
+ * @panic_write:The write operation only used for panic case. It's optional
+ *             if you do not care panic log. The parameters and return value
+ *             are the same as @read.
+ */
+struct pstore_zone_info {
+       struct module *owner;
+       const char *name;
+
+       unsigned long total_size;
+       unsigned long kmsg_size;
+       int max_reason;
+       pstore_zone_read_op read;
+       pstore_zone_write_op write;
+       pstore_zone_write_op panic_write;
+};
+
+extern int register_pstore_zone(struct pstore_zone_info *info);
+extern void unregister_pstore_zone(struct pstore_zone_info *info);
+
+#endif