From: Adrian Hunter <adrian.hun...@intel.com> blkoops is a pstore back end to write panic / oops logs to a block device. It is initially intended for use with eMMC as an alternative to using a crash kernel.
Signed-off-by: Adrian Hunter <adrian.hun...@intel.com> Signed-off-by: Irina Tirdea <irina.tir...@intel.com> --- Documentation/blockdev/00-INDEX | 2 + Documentation/blockdev/blkoops.txt | 104 +++ drivers/block/Kconfig | 10 + drivers/block/Makefile | 1 + drivers/block/blkoops.c | 1569 ++++++++++++++++++++++++++++++++++++ 5 files changed, 1686 insertions(+) create mode 100644 Documentation/blockdev/blkoops.txt create mode 100644 drivers/block/blkoops.c diff --git a/Documentation/blockdev/00-INDEX b/Documentation/blockdev/00-INDEX index c08df56..c45cef8 100644 --- a/Documentation/blockdev/00-INDEX +++ b/Documentation/blockdev/00-INDEX @@ -2,6 +2,8 @@ - this file README.DAC960 - info on Mylex DAC960/DAC1100 PCI RAID Controller Driver for Linux. +blkoops.txt + - info on block device oops / panic logger cciss.txt - info, major/minor #'s for Compaq's SMART Array Controllers. cpqarray.txt diff --git a/Documentation/blockdev/blkoops.txt b/Documentation/blockdev/blkoops.txt new file mode 100644 index 0000000..fb08664 --- /dev/null +++ b/Documentation/blockdev/blkoops.txt @@ -0,0 +1,104 @@ +Block device oops / panic logger +-------------------------------- + +Contents: + + 1) Overview + 2) Format + 3) Parameters + 4) blkoops and pstore + 5) debugfs + +1) Overview +----------- + + blkoops is a pstore back end to write panic / oops logs to a block + device. It is initially intended for use with eMMC as an alternative to + using a crash kernel. + +2) Format +--------- + + Data is written in chunks called nodes which are preceded by a + header. The header is always aligned to a block boundary. Nodes are + written sequentially starting at the second block. The first block + contains a special node that fulfils 2 purposes: 1) the blkoops magic + number must be present or blkoops will not attach to the block device, + and 2) erase information is recorded there. Nodes can be arbitrarily + long. + + Nodes are identified by session number, file number and part number. + A session may have up to 2^32 - 1 files each with up to 2^32 - 1 parts. + + A new session begins when blkoops attaches to a block device and ends + when it detaches or there is a reboot. A new session overwrites the + previous session. Once the media is full no more nodes are written. + +3) Parameters +------------- + +devname + + Canonical block device name or number + + devname may be set on the kernel command line e.g. + + blkoops.devname=/dev/mmcblk0p7 + + or by writing to sysfs e.g. + + echo /dev/mmcblk0p1 > /sys/module/blkoops/parameters/devname + + devname is NOT the name of a file system object. e.g. /dev/mmcblk0p7 + does NOT mean the block device special file mmcblk0p7 in the /dev + directory. Instead it means partition 7 of the device named mmcblk0. + For more information see name_to_dev_t comment in init/do_mounts.c + + When devname is changed, the old devname (if any) is detached from + blkoops and the new devname (if any) is attached. + + blkoops will reject a block device that does not have the blkoops magic + number written on the 1st sector. For example, to prepare + /dev/mmcblk0p7 for blkoops: + + sudo bash -c "echo -e -n '\0034\0327\0130\0350' \ + | dd count=1 conv=sync \ + > /dev/mmcblk0p7" + +dump_oops + + set to 1 to dump oopses, 0 to dump only panics (default 1) + +4) blkoops and pstore +--------------------- + + pstore creates file names from pstore type code, back end name and + pstore 64-bit id. blkoops records the pstore type code, uses back end + name "blkoops", and creates the pstore 64-bit id from session number and + file number (session << 32 | file). blkoops concatenates all parts + together and presents them as one file. + + pstore noramally reads back end data entirely into memory when mounting. + However if a blkoops file is too big it will be read from media as + needed instead. + + blkoops suppreses pstore heading lines from dumped data. + + blkoops increases pstore default kmsg_bytes to ULONG_MAX. + +5) debugfs +---------- + +blkoops/type + + pstore type code to use when dumping data via blkoops/data + +blkoops/reason + + kmsg dump reason code to use when dumping data via blkoops/data + +blkoops/data + + Data written to blkoops/data is dumped to the block device + using blkoops/type. blkoops/reason must be the numberical value of + KMSG_DUMP_PANIC or (if dump_oops is 1) KMSG_DUMP_OOPS. diff --git a/drivers/block/Kconfig b/drivers/block/Kconfig index 824e09c..af5b325 100644 --- a/drivers/block/Kconfig +++ b/drivers/block/Kconfig @@ -544,4 +544,14 @@ config BLK_DEV_RBD If unsure, say N. +config BLK_DEV_OOPS + bool "Block Oops / Panic Logger" + select BLK_DEV_PANIC_WRITE + select PSTORE + default n + help + This enables panic and oops messages to be logged to a block device. + + See <file:Documentation/blockdev/blkoops.txt> for more information. + endif # BLK_DEV diff --git a/drivers/block/Makefile b/drivers/block/Makefile index 17e82df..db44850 100644 --- a/drivers/block/Makefile +++ b/drivers/block/Makefile @@ -25,6 +25,7 @@ obj-$(CONFIG_MG_DISK) += mg_disk.o obj-$(CONFIG_SUNVDC) += sunvdc.o obj-$(CONFIG_BLK_DEV_NVME) += nvme.o obj-$(CONFIG_BLK_DEV_OSD) += osdblk.o +obj-$(CONFIG_BLK_DEV_OOPS) += blkoops.o obj-$(CONFIG_BLK_DEV_UMEM) += umem.o obj-$(CONFIG_BLK_DEV_NBD) += nbd.o diff --git a/drivers/block/blkoops.c b/drivers/block/blkoops.c new file mode 100644 index 0000000..bafe17e --- /dev/null +++ b/drivers/block/blkoops.c @@ -0,0 +1,1569 @@ +/* + * Block Oops / Panic Logger + * + * Copyright (C) 2012 Intel Corporation + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#include <linux/module.h> +#include <linux/swap.h> /* For nr_free_buffer_pages() */ +#include <linux/debugfs.h> +#include <linux/completion.h> +#include <linux/blkdev.h> +#include <linux/fs.h> +#include <linux/mount.h> +#include <linux/pstore.h> + +#undef DEBUG + +#ifdef DEBUG +#define DBG_PRINTK printk +#else +#define DBG_PRINTK no_printk +#endif + +#define DBG(fmt, ...) \ + DBG_PRINTK(KERN_DEBUG "%s: " fmt "\n", __func__, ##__VA_ARGS__) + +#define BLKOOPS_BUFSIZE (32 * 1024) + +#define BLKOOPS_MAGIC 0xe858d71c + +#define BLKOOPS_VERSION 1 + +#define BLKOOPS_HDR_SZ (sizeof(struct blkoops_header)) + +#define BLKOOPS_MODE (FMODE_READ | FMODE_WRITE | FMODE_EXCL) + +#define BLKOOPS_DEVNAME_SZ 256 + +static char devname[BLKOOPS_DEVNAME_SZ]; +static int dump_oops; +static int init_done; + +static DEFINE_MUTEX(blkoops_mutex); +static DEFINE_SPINLOCK(blkoops_lock); + +/** + * struct blkoops - blkoops private data. + * @psi: pstore information + * @present: non-zero if blkoops is attached to a block device + * @bdev: block device to which that blkoops is attached + * @devid: block device (major, minor) number + * @blksize: block device block size + * @sects_per_blk: block device sectors per block + * @nr_sects: block device size in sectors + * @size: block device size in bytes + * @bdev_name: block device name + * @buf: panic write buffer + * @bufsize: panic write buffer size + * @last_session: last session number on media + * @next_session: next session number to write + * @next_file: next file number to write + * @next_sect: next sector to write + * @cache_invalid: page cache is invalid because + * @root: rb-tree of all blkoops nodes sorted by session / file / part numbers + * @file_cnt: number of (non-erased) files on media + * @max_file: maximum file number on media + * @used_sects: number of sectors used on media by the last session + * @erased_session: session number of a completely or partly erased session + * @erased_file: maximum erased file number of @erased_session + * @erased_sects: number of sectors used on media by @erased_session + * @read_anew: start reading from the first file + * @read_file: last file number read + * @dbg_root: blkoops debugfs root directory + * @dbg_type: pstore type code to use when writing + * @dbg_reason: kmsg dump reason code to use when writing + * @dbg_part: next part number to use when writing + * @dbg_lock: with @dbg_open, prevent debugfs 'data' file from having more than + 1 user, and also cause the call to 'blkoops_write()' to be in an + atomic context + * @dbg_open: non-zero if debugfs 'data' file is open + * @dbg_buf: debug write buffer + * @dbg_bufsize: debug write buffer size + * @dbg_used: number of bytes used in @dbg_buf + */ +struct blkoops { + struct pstore_info psi; + + int present; + + struct block_device *bdev; + dev_t devid; + unsigned int blksize; + unsigned int sects_per_blk; + sector_t nr_sects; + loff_t size; + char bdev_name[BDEVNAME_SIZE]; + + char *buf; + size_t bufsize; + + unsigned int last_session; + + unsigned int next_session; + unsigned int next_file; + sector_t next_sect; + + int flush_needed; + int cache_invalid; + + struct rb_root root; + unsigned int file_cnt; + unsigned int max_file; + sector_t used_sects; + + unsigned int erased_session; + unsigned int erased_file; + sector_t erased_sects; + + int read_anew; + unsigned int read_file; + +#ifdef CONFIG_DEBUG_FS + struct dentry *dbg_root; + u32 dbg_type; + u32 dbg_reason; + u32 dbg_part; + spinlock_t dbg_lock; /* debug serialization */ + int dbg_open; + void *dbg_buf; + size_t dbg_bufsize; + size_t dbg_used; +#endif +}; + +/** + * struct blkoops_node - blkoops node. + * @node: rb-tree of all blkoops nodes + * @session: session number + * @file: file number + * @part: part number + * @len: length in bytes (excluding header) + * @nr_sects: length in sectors including header + * @type: pstore type code + * @timestamp: time the node was written to media + * @offs: offset on block device + * @file_offs: offset within file + * @tot_len: total length of file (only recorded on the first part) + */ +struct blkoops_node { + struct rb_node node; + unsigned int session; + unsigned int file; + unsigned int part; + loff_t len; + sector_t nr_sects; + enum pstore_type_id type; + struct timespec timestamp; + loff_t offs; + loff_t file_offs; + loff_t tot_len; +}; + +/** + * struct blkoops_header - blkoops on-media node header. + * @magic: blkoops magic number + * @version: blkoops media format version + * @session: session number + * @file: file number + * @part: part number + * @type: pstore type code + * @timestamp: time the node was written to media + * @len: length in bytes (excluding header) + * @nr_sects: length in sectors including header + * @padding: reserved for future, zeroes + * + * Data is written in chunks called nodes which are preceded by this header. + * The header is always aligned to a block boundary. Nodes are written + * sequentially starting at the second block. The first block contains a + * special node that fulfils 2 purposes: 1) the blkoops magic number must be + * present or blkoops will not attach to the block device, and 2) erase + * information is recorded there. Nodes can be arbitrarily long. + * + * Nodes are identified by session number, file number and part number. A + * session may have up to 2^32 - 1 files each with up to 2^32 - 1 parts. + * + * A new session begins when blkoops attaches to a block device and ends when it + * detaches or there is a reboot. A new session overwrites the previous session. + * Once the media is full no more nodes are written. + */ +struct blkoops_header { + __le32 magic; + __le32 version; + __le32 session; + __le32 file; + __le32 part; + __le32 type; + __le64 timestamp; + __le64 len; + __le64 nr_sects; + __u8 padding[16]; +} __packed; + +static int blkoops_lt(struct blkoops_node *bn1, struct blkoops_node *bn2) +{ + if (bn1->session == bn2->session) { + if (bn1->file == bn2->file) + return bn1->part < bn2->part; + return bn1->file < bn2->file; + } + + return bn1->session < bn2->session; +} + +static void blkoops_add_node(struct blkoops *c, struct blkoops_node *bn) +{ + struct rb_node **p = &c->root.rb_node; + struct rb_node *parent = NULL; + struct blkoops_node *t; + + while (*p) { + parent = *p; + t = rb_entry(parent, struct blkoops_node, node); + + if (blkoops_lt(bn, t)) + p = &(*p)->rb_left; + else + p = &(*p)->rb_right; + } + + rb_link_node(&bn->node, parent, p); + rb_insert_color(&bn->node, &c->root); +} + +static int blkoops_lt_file(unsigned int session, unsigned int file, + struct blkoops_node *bn) +{ + if (session == bn->session) + return file < bn->file; + return session < bn->session; +} + +static struct blkoops_node *blkoops_lookup_next(struct blkoops *c, + unsigned int session, + unsigned int file) +{ + struct rb_node *node = c->root.rb_node; + struct blkoops_node *bn, *next = NULL; + + while (node) { + bn = rb_entry(node, struct blkoops_node, node); + + if (blkoops_lt_file(session, file, bn)) { + node = node->rb_left; + next = bn; + } else { + node = node->rb_right; + } + } + return next; +} + +static int blkoops_cmp_offs(unsigned int session, unsigned int file, loff_t pos, + struct blkoops_node *bn) +{ + if (session == bn->session) { + if (file == bn->file) { + if (pos < bn->file_offs) + return -1; + else if (pos >= bn->file_offs + bn->len) + return 1; + return 0; + } + return file < bn->file ? -1 : 1; + } + return session < bn->session ? -1 : 1; +} + +static struct blkoops_node *blkoops_lookup_pos(struct blkoops *c, + unsigned int session, + unsigned int file, loff_t pos) +{ + struct rb_node *node = c->root.rb_node; + struct blkoops_node *bn; + int cmp; + + while (node) { + bn = rb_entry(node, struct blkoops_node, node); + cmp = blkoops_cmp_offs(session, file, pos, bn); + if (cmp == -1) + node = node->rb_left; + else if (cmp == 1) + node = node->rb_right; + else + return bn; + } + return NULL; +} + +static struct page *blkoops_read_page(struct blkoops *c, loff_t pos) +{ + pgoff_t index = pos >> PAGE_CACHE_SHIFT; + + return read_mapping_page(c->bdev->bd_inode->i_mapping, index, NULL); +} + +static int blkoops_validate_node(struct blkoops *c, struct blkoops_node *bn) +{ + sector_t nr_sects; + loff_t len; + + if (bn->offs + bn->len > c->size) { + DBG("bad node size (offs %lld + len %lld > size %lld)", + bn->offs, bn->len, c->size); + return -EINVAL; + } + len = bn->len + BLKOOPS_HDR_SZ; + nr_sects = (len >> 9) + (len & 511 ? 1 : 0); + if (bn->nr_sects < nr_sects) { + DBG("bad node nr sects %llu vs len %lld", + (u64)bn->nr_sects, bn->len); + return -EINVAL; + } + if (bn->offs - BLKOOPS_HDR_SZ + ((loff_t)bn->nr_sects << 9) > c->size) { + DBG("bad node nr sects %llu vs size %lld", + (u64)bn->nr_sects, c->size); + return -EINVAL; + } + return 0; +} + +static struct blkoops_node *blkoops_alloc_node(struct blkoops *c, + struct blkoops_header *hdr, + loff_t pos) +{ + struct blkoops_node *bn; + + if (__le32_to_cpu(hdr->magic) != BLKOOPS_MAGIC) { + DBG("bad magic at pos %lld", pos); + return ERR_PTR(-EINVAL); + } + + bn = kzalloc(sizeof(struct blkoops_node), GFP_KERNEL); + if (!bn) + return ERR_PTR(-ENOMEM); + + RB_CLEAR_NODE(&bn->node); + + bn->session = __le32_to_cpu(hdr->session); + bn->file = __le32_to_cpu(hdr->file); + bn->part = __le32_to_cpu(hdr->part); + bn->type = __le32_to_cpu(hdr->type); + bn->timestamp.tv_sec = __le64_to_cpu(hdr->timestamp); + bn->len = __le64_to_cpu(hdr->len); + bn->nr_sects = __le64_to_cpu(hdr->nr_sects); + bn->offs = pos + BLKOOPS_HDR_SZ; + + return bn; +} + +static struct blkoops_node *blkoops_read_node(struct blkoops *c, loff_t pos) +{ + struct blkoops_header *hdr; + struct blkoops_node *bn; + struct page *page; + + if (pos >= c->size || pos & 511) { + DBG("bad pos %lld", pos); + return ERR_PTR(-EINVAL); + } + + page = blkoops_read_page(c, pos); + if (IS_ERR(page)) { + pr_err("blkoops: bad page at pos %lld", pos); + return ERR_CAST(page); + } + /* + * Whole header must be within page because the header is sector aligned + * and smaller than 1 sector. + */ + hdr = page_address(page) + (pos & (PAGE_CACHE_SIZE - 1)); + + bn = blkoops_alloc_node(c, hdr, pos); + + page_cache_release(page); + return bn; +} + +static inline loff_t blkoops_scan_start(struct blkoops *c) +{ + return c->blksize; +} + +static loff_t blkoops_do_scan(struct blkoops *c, loff_t pos) +{ + struct blkoops_node *bn; + int err; + + c->used_sects = pos >> 9; + + if (pos >= c->size) { + DBG("end of scan at pos %lld (size %lld)", pos, c->size); + return 0; + } + + bn = blkoops_read_node(c, pos); + if (IS_ERR(bn)) + return PTR_ERR(bn); + + err = blkoops_validate_node(c, bn); + if (err) { + DBG("bad node at pos %lld", pos); + goto out_free; + } + + DBG("node at %lld sess %u file %u part %u len %lld nr sects %llu", + pos, bn->session, bn->file, bn->part, bn->len, (u64)bn->nr_sects); + + /* A new oops may have written an updated session number */ + if (pos == blkoops_scan_start(c)) + c->last_session = bn->session; + + /* Only scan the last session */ + if (bn->session != c->last_session) { + DBG("wrong session %u (expected %u) at pos %lld", + bn->session, c->last_session, pos); + err = 0; + goto out_free; + } + + /* Skip erased files */ + if (c->erased_file && bn->session == c->erased_session && + bn->file <= c->erased_file) { + loff_t next_pos, new_pos; + + /* + * Try to skip to next non-erased node, but only if it makes + * sense. + */ + next_pos = pos + ((loff_t)bn->nr_sects << 9); + new_pos = (loff_t)c->erased_sects << 9; + if (new_pos < next_pos || new_pos > c->size) + new_pos = next_pos; + DBG("skipping erased node at pos %lld continuing from pos %lld", + pos, new_pos); + pos = new_pos; + } else { + blkoops_add_node(c, bn); + pos += (loff_t)bn->nr_sects << 9; + } + + return pos; + +out_free: + kfree(bn); + return err; +} + +static void blkoops_free_tree(struct blkoops *c) +{ + struct rb_node *node, *next; + struct blkoops_node *bn; + + node = rb_first(&c->root); + while (node) { + bn = rb_entry(node, struct blkoops_node, node); + next = rb_next(node); + rb_erase(node, &c->root); + kfree(bn); + node = next; + } + + c->file_cnt = 0; +} + +static void blkoops_count_files(struct blkoops *c) +{ + struct rb_node *node; + struct blkoops_node *bn, *last = NULL; + + c->file_cnt = 0; + c->max_file = 0; + + for (node = rb_first(&c->root); node; node = rb_next(node)) { + bn = rb_entry(node, struct blkoops_node, node); + if (last && bn->session == last->session && + bn->file == last->file) { + bn->file_offs = last->tot_len; + last->tot_len += bn->len; + continue; + } + last = bn; + last->file_offs = 0; + last->tot_len = bn->len; + c->file_cnt += 1; + if (bn->file > c->max_file) + c->max_file = bn->file; + } + DBG("file cnt %u max file %u used sects %llu", + c->file_cnt, c->max_file, (u64)c->used_sects); +} + +static int blkoops_scan(struct blkoops *c) +{ + loff_t pos = blkoops_scan_start(c); + + blkoops_free_tree(c); + + if (c->cache_invalid) { + DBG("cache_invalid"); + c->cache_invalid = 0; + invalidate_mapping_pages(c->bdev->bd_inode->i_mapping, 0, -1); + } + + do { + pos = blkoops_do_scan(c, pos); + } while (pos > 0); + + blkoops_count_files(c); + + return pos; +} + +static int blkoops_open(struct pstore_info *psi) +{ + struct blkoops *c = psi->data; + + mutex_lock(&blkoops_mutex); + if (!c->present) + goto out; + blkoops_scan(c); + c->read_anew = 1; +out: + mutex_unlock(&blkoops_mutex); + return 0; +} + +static struct blkoops_node *blkoops_next_node(struct blkoops_node *bn) +{ + struct rb_node *node; + struct blkoops_node *t; + + node = rb_next(&bn->node); + if (!node) + return NULL; + t = rb_entry(node, struct blkoops_node, node); + if (t->session == bn->session && t->file == bn->file) + return t; + return NULL; +} + +static int blkoops_read_to_buf(struct blkoops *c, char *buf, loff_t pos, + loff_t len) +{ + struct page *page; + loff_t offs; + size_t n; + int err = 0; + + while (len) { + page = blkoops_read_page(c, pos); + + offs = pos & (PAGE_CACHE_SIZE - 1); + n = PAGE_CACHE_SIZE - offs; + if (n > len) + n = len; + + if (IS_ERR(page)) { + if (!err) + err = PTR_ERR(page); + memset(buf, 0, n); + } else { + memcpy(buf, page_address(page) + offs, n); + } + + pos += n; + buf += n; + len -= n; + + page_cache_release(page); + } + + return err; +} + +static int blkoops_fill_buf(struct blkoops *c, char *buf, + struct blkoops_node *bn) +{ + loff_t tot_len = bn->tot_len; + int err = 0, err1; + + while (bn && tot_len) { + err1 = blkoops_read_to_buf(c, buf, bn->offs, bn->len); + if (!err) + err = err1; + tot_len -= bn->len; + buf += bn->len; + bn = blkoops_next_node(bn); + } + return err; +} + +static struct blkoops_node *blkoops_read_next(struct blkoops *c) +{ + struct blkoops_node *bn = NULL; + struct rb_node *node; + + if (c->read_anew) { + c->read_anew = 0; + node = rb_first(&c->root); + if (node) + bn = rb_entry(node, struct blkoops_node, node); + } else { + bn = blkoops_lookup_next(c, c->last_session, c->read_file); + } + if (bn) + c->read_file = bn->file; + return bn; +} + +static ssize_t blkoops_read_to_userbuf(struct blkoops *c, char __user *userbuf, + loff_t pos, size_t len) +{ + struct page *page; + ssize_t ret = 0; + loff_t offs; + size_t n, rn; + int err; + + while (len) { + page = blkoops_read_page(c, pos); + + offs = pos & (PAGE_CACHE_SIZE - 1); + n = PAGE_CACHE_SIZE - offs; + if (n > len) + n = len; + + if (IS_ERR(page)) { + err = PTR_ERR(page); + goto out_err; + } + + rn = copy_to_user(userbuf, page_address(page) + offs, n); + if (rn) { + page_cache_release(page); + ret += n - rn; + err = -EFAULT; + goto out_err; + } + + pos += n; + userbuf += n; + len -= n; + ret += n; + + page_cache_release(page); + } + + return ret; + +out_err: + if (!ret) + ret = err; + return ret; +} + +static inline unsigned int blkoops_session(u64 id) +{ + return id >> 32; +} + +static inline unsigned int blkoops_file(u64 id) +{ + return id; +} + +static ssize_t blkoops_file_read(u64 id, enum pstore_type_id type, + char __user *userbuf, size_t count, + loff_t *ppos, struct pstore_info *psi) +{ + struct blkoops *c = psi->data; + struct blkoops_node *bn; + unsigned int session, file; + ssize_t done, res = 0; + loff_t noffs, dpos; + size_t len; + int err; + + if (*ppos < 0) + return -EINVAL; + + session = blkoops_session(id); + file = blkoops_file(id); + + mutex_lock(&blkoops_mutex); + if (!c->present) { + err = -ENODEV; + goto out_err; + } + + DBG("sess %u file %u pos %lld count %zu", session, file, *ppos, count); + + while (count) { + bn = blkoops_lookup_pos(c, session, file, *ppos); + if (!bn) + break; + noffs = *ppos - bn->file_offs; + len = min_t(loff_t, count, bn->len - noffs); + dpos = bn->offs + noffs; + done = blkoops_read_to_userbuf(c, userbuf, dpos, len); + if (done < 0) { + err = done; + goto out_err; + } + res += done; + if (done < len) + break; + *ppos += len; + userbuf += len; + count -= len; + } +out: + DBG("sess %u file %u pos %lld res %zd", session, file, *ppos, res); + mutex_unlock(&blkoops_mutex); + return res; + +out_err: + if (!res) + res = err; + goto out; +} + +static inline u64 blkoops_id(u32 session, u32 file) +{ + return (u64)session << 32 | file; +} + +static int blkoops_read(u64 *id, enum pstore_type_id *type, + struct timespec *time, char **buf, loff_t *size, + struct pstore_info *psi) +{ + struct blkoops *c = psi->data; + struct blkoops_node *bn; + unsigned long limit; + u64 bn_id; + int err; + + mutex_lock(&blkoops_mutex); + if (!c->present) { + err = -ENODEV; + goto out; + } + + bn = blkoops_read_next(c); + if (!bn) { + err = -ENOENT; + goto out; + } + + bn_id = blkoops_id(bn->session, bn->file); + + DBG("node at pos %lld sess %u file %u part %u len %lld nr sects %llu", + bn->offs, bn->session, bn->file, bn->part, bn->len, + (u64)bn->nr_sects); + + *type = bn->type; + *id = bn_id; + *time = bn->timestamp; + *size = bn->tot_len; + + limit = nr_free_buffer_pages() << (PAGE_SHIFT - 3); + limit /= c->file_cnt; + + if (bn->tot_len > limit) { + DBG("file size %lld over limit %lu", bn->tot_len, limit); + err = -EFBIG; + goto out; + } + + *buf = kmalloc(bn->tot_len, GFP_KERNEL | __GFP_NOWARN); + if (!*buf) { + DBG("failed to allocate %lld bytes", bn->tot_len); + err = -EFBIG; + goto out; + } + + err = blkoops_fill_buf(c, *buf, bn); + if (err) { + pr_err("blkoops: read failed, file id %lld, error %d\n", + bn_id, err); + err = 0; + } +out: + mutex_unlock(&blkoops_mutex); + return err; +} + +static int blkoops_write(enum pstore_type_id type, enum kmsg_dump_reason reason, + u64 *id, unsigned int part, size_t size, + struct pstore_info *psi) +{ + struct blkoops *c = psi->data; + struct blkoops_header *hdr; + size_t nr, sz, rsz, len; + u32 partno; + int err; + + /* If blkoops_lock is locked then there is no back end, so give up */ + if (!spin_trylock(&blkoops_lock)) + return -ENODEV; + + if (!c->present) { + err = -ENODEV; + goto out_unlock; + } + + if (reason != KMSG_DUMP_OOPS && + reason != KMSG_DUMP_PANIC) { + err = -EINVAL; + goto out_unlock; + } + + if (reason == KMSG_DUMP_OOPS && !dump_oops) { + err = -EINVAL; + goto out_unlock; + } + + if (c->next_sect + c->sects_per_blk > c->nr_sects) { + err = -ENOSPC; + goto out_unlock; + } + + if (part == 1) + c->next_file += 1; + + /* + * Special case: dmesg is written backwards so reverse the order of the + * part numbers. + */ + if (type == PSTORE_TYPE_DMESG) + partno = -part; + else + partno = part; + + *id = blkoops_id(c->next_session, c->next_file); + + /* Round up the size to block size and pad with zeroes */ + sz = size + BLKOOPS_HDR_SZ; + rsz = roundup(sz, c->blksize); + nr = rsz >> 9; + memset(c->psi.buf + size, 0, rsz - sz); + + /* + * Truncate the node to fit the remaining space. Note, we have + * already checked that there is enough space for at least 1 block. + */ + len = size; + while (c->next_sect + nr > c->nr_sects) { + nr -= c->sects_per_blk; + len = (nr << 9) - BLKOOPS_HDR_SZ; + } + + hdr = (void *)c->buf; + memset(hdr, 0, BLKOOPS_HDR_SZ); + hdr->magic = __cpu_to_le32(BLKOOPS_MAGIC); + hdr->version = __cpu_to_le32(BLKOOPS_VERSION); + hdr->session = __cpu_to_le32(c->next_session); + hdr->file = __cpu_to_le32(c->next_file); + hdr->part = __cpu_to_le32(partno); + hdr->type = __cpu_to_le32(type); + hdr->timestamp = __cpu_to_le64(get_seconds()); + hdr->len = __cpu_to_le64(len); + hdr->nr_sects = __cpu_to_le64(nr); + + c->cache_invalid = 1; + c->flush_needed = 1; + + err = blk_panic_write(c->bdev, c->next_sect, c->buf, nr << 9); + if (err) + goto out_unlock; + + c->next_sect += nr; + +out_unlock: + spin_unlock(&blkoops_lock); + return err; +} + +struct blkoops_bio_batch { + atomic_t done; + unsigned long flags; + struct completion *wait; +}; + +static void blkoops_end_io(struct bio *bio, int err) +{ + struct blkoops_bio_batch *bb = bio->bi_private; + + if (err) + clear_bit(BIO_UPTODATE, &bb->flags); + if (atomic_dec_and_test(&bb->done)) + complete(bb->wait); + bio_put(bio); +} + +static int blkoops_direct_io(int type, struct block_device *bdev, sector_t sect, + unsigned long nr, void *buf) +{ + DECLARE_COMPLETION_ONSTACK(wait); + struct blkoops_bio_batch bb; + struct bio *bio; + int ret = 0; + unsigned long len = nr << 9; + + DBG("type %#x sect %llu nr %lu", type, (u64)sect, nr); + + atomic_set(&bb.done, 1); + bb.flags = 1 << BIO_UPTODATE; + bb.wait = &wait; + + while (len) { + bio = bio_alloc(GFP_KERNEL, 1); + if (!bio) { + ret = -ENOMEM; + break; + } + + bio->bi_sector = sect; + bio->bi_end_io = blkoops_end_io; + bio->bi_bdev = bdev; + bio->bi_private = &bb; + + while (len) { + unsigned int offs = offset_in_page(buf); + unsigned int n = PAGE_SIZE - offs; + int bytes; + + if (n > len) + n = len; + bytes = bio_add_page(bio, virt_to_page(buf), n, offs); + if (bytes <= 0) + break; + len -= bytes; + buf += bytes; + } + sect += bio->bi_size >> 9; + + atomic_inc(&bb.done); + submit_bio(type, bio); + } + + if (!atomic_dec_and_test(&bb.done)) + wait_for_completion(&wait); + + if (!test_bit(BIO_UPTODATE, &bb.flags)) + ret = -EIO; + + if (ret) + DBG("I/O error %d", ret); + return ret; +} + +static void blkoops_invalidate_range(struct block_device *bdev, sector_t sect, + unsigned int nr) +{ + pgoff_t start, end; + + start = sect >> (PAGE_CACHE_SHIFT - 9); + end = (sect + nr - 1) >> (PAGE_CACHE_SHIFT - 9); + + invalidate_mapping_pages(bdev->bd_inode->i_mapping, start, end); +} + +static int blkoops_write_meta(struct block_device *bdev, sector_t sect, + unsigned long nr, void *buf) +{ + blkoops_invalidate_range(bdev, sect, nr); + + return blkoops_direct_io(REQ_WRITE | REQ_META, bdev, sect, nr, buf); +} + +static int blkoops_mark_erased(struct blkoops *c) +{ + struct blkoops_header *hdr; + int err; + + DBG("session %u file %u sects %llu", + c->erased_session, c->erased_file, (u64)c->erased_sects); + + hdr = kzalloc(c->blksize, GFP_KERNEL); + if (!hdr) + return -ENOMEM; + + hdr->magic = __cpu_to_le32(BLKOOPS_MAGIC); + hdr->version = __cpu_to_le32(BLKOOPS_VERSION); + hdr->session = __cpu_to_le32(c->erased_session); + hdr->file = __cpu_to_le32(c->erased_file); + hdr->timestamp = __cpu_to_le64(get_seconds()); + hdr->nr_sects = __cpu_to_le64(c->erased_sects); + + err = blkoops_write_meta(c->bdev, 0, c->sects_per_blk, hdr); + + kfree(hdr); + + return err; +} + +static int blkoops_erase(enum pstore_type_id type, u64 id, + struct pstore_info *psi) +{ + struct blkoops *c = psi->data; + int err = 0; + + mutex_lock(&blkoops_mutex); + if (!c->present) { + err = -ENODEV; + goto out; + } + DBG("type %d id %lld", type, id); + /* + * Do nothing until all files are erased and then mark the range as + * erased. + */ + if (c->file_cnt && !--c->file_cnt && !c->cache_invalid) { + c->erased_session = c->last_session; + c->erased_file = c->max_file; + c->erased_sects = c->used_sects; + err = blkoops_mark_erased(c); + } +out: + mutex_unlock(&blkoops_mutex); + return err; +} + +static int blkoops_flush(struct pstore_info *psi) +{ + struct blkoops *c = psi->data; + int err; + + /* If blkoops_lock is locked then there is no back end, so give up */ + if (!spin_trylock(&blkoops_lock)) + return -ENODEV; + + if (!c->present) { + err = -ENODEV; + goto out_unlock; + } + + if (!c->flush_needed) { + err = 0; + goto out_unlock; + } + + err = blk_panic_flush(c->bdev); + +out_unlock: + spin_unlock(&blkoops_lock); + return err; +} + +#ifdef CONFIG_DEBUG_FS + +static int blkoops_dbg_write_buf(struct blkoops *c, void *buf, size_t len) +{ + unsigned long flags; + size_t n; + int ret; + u64 id; + + while (len) { + n = min(len, c->psi.bufsize); + memcpy(c->psi.buf, buf, n); + spin_lock_irqsave(&c->dbg_lock, flags); + ret = blkoops_write(c->dbg_type, c->dbg_reason, &id, + c->dbg_part++, n, &c->psi); + spin_unlock_irqrestore(&c->dbg_lock, flags); + if (ret) { + pr_err("blkoops: debug write failed, error %d\n", ret); + return ret; + } + buf += n; + len -= n; + } + return 0; +} + +static int blkoops_dbg_flush_buf(struct blkoops *c) +{ + int err; + + err = blkoops_dbg_write_buf(c, c->dbg_buf, c->dbg_used); + c->dbg_used = 0; + return err; +} + +static int blkoops_dbg_drain_buf(struct blkoops *c) +{ + size_t written = 0; + int err = 0; + + while (c->dbg_used >= c->psi.bufsize) { + err = blkoops_dbg_write_buf(c, c->dbg_buf + written, + c->psi.bufsize); + c->dbg_used -= c->psi.bufsize; + written += c->psi.bufsize; + } + memmove(c->dbg_buf, c->dbg_buf + written, c->dbg_used); + return err; +} + +static ssize_t blkoops_dbg_fill_buf(struct blkoops *c, const char __user *buf, + size_t len) +{ + size_t remains = c->dbg_bufsize - c->dbg_used, n = min(len, remains); + void *addr = c->dbg_buf + c->dbg_used; + + if (copy_from_user(addr, buf, n)) + return -EFAULT; + c->dbg_used += n; + return n; +} + +static ssize_t blkoops_dbg_write(struct file *file, const char __user *buf, + size_t len, loff_t *ppos) +{ + struct blkoops *c = file->private_data; + ssize_t n, res = len; + + while (len) { + n = blkoops_dbg_fill_buf(c, buf, len); + if (n < 0) + return n; + buf += n; + len -= n; + if (c->dbg_used == c->dbg_bufsize) + blkoops_dbg_drain_buf(c); + } + return res; +} + +static int blkoops_dbg_file_open(struct inode *inode, struct file *file) +{ + struct blkoops *c = inode->i_private; + unsigned long flags; + int err = 0; + + spin_lock_irqsave(&c->dbg_lock, flags); + if (c->dbg_open) + err = -EBUSY; + else + c->dbg_open = 1; + spin_unlock_irqrestore(&c->dbg_lock, flags); + if (err) + return err; + + c->dbg_bufsize = BLKOOPS_BUFSIZE; + c->dbg_buf = kmalloc(c->dbg_bufsize, GFP_KERNEL); + if (!c->dbg_buf) { + err = -ENOMEM; + goto out_err; + } + + err = nonseekable_open(inode, file); + if (err) + goto out_err; + + file->private_data = c; + + c->dbg_part = 1; + c->dbg_used = 0; + + return 0; + +out_err: + spin_lock_irqsave(&c->dbg_lock, flags); + c->dbg_open = 0; + spin_unlock_irqrestore(&c->dbg_lock, flags); + return err; +} + +static int blkoops_dbg_file_release(struct inode *inode, struct file *file) +{ + struct blkoops *c = inode->i_private; + unsigned long flags; + int err, err2; + + err = blkoops_dbg_flush_buf(c); + + spin_lock_irqsave(&c->dbg_lock, flags); + err2 = blkoops_flush(&c->psi); + spin_unlock_irqrestore(&c->dbg_lock, flags); + if (err2 && !err) + err = err2; + + kfree(c->dbg_buf); + + spin_lock_irqsave(&c->dbg_lock, flags); + c->dbg_open = 0; + spin_unlock_irqrestore(&c->dbg_lock, flags); + + return err; +} + +static const struct file_operations dbg_data_fops = { + .owner = THIS_MODULE, + .open = blkoops_dbg_file_open, + .release = blkoops_dbg_file_release, + .write = blkoops_dbg_write, + .llseek = no_llseek, +}; + +static void blkoops_init_debugfs(struct blkoops *c) +{ + umode_t rw = S_IRUSR | S_IWUSR, wo = S_IWUSR; + + c->dbg_root = debugfs_create_dir("blkoops", NULL); + + debugfs_create_u32("type", rw, c->dbg_root, &c->dbg_type); + debugfs_create_u32("reason", rw, c->dbg_root, &c->dbg_reason); + debugfs_create_file("data", wo, c->dbg_root, c, &dbg_data_fops); +} + +static void blkoops_remove_debugfs(struct blkoops *c) +{ + debugfs_remove_recursive(c->dbg_root); +} + +#else + +static inline void blkoops_init_debugfs(struct blkoops *c) +{ +} + +static void blkoops_remove_debugfs(struct blkoops *c) +{ +} + +#endif + +static void *blkoops_alloc_buf(unsigned int blksize, loff_t max_size, + unsigned int *rsz) +{ + unsigned int size; + void *addr; + gfp_t flgs; + + size = BLKOOPS_BUFSIZE; + if (size > max_size) + size = max_size; + + while (1) { + if (size < blksize) + *rsz = blksize; + else + *rsz = roundup(size, blksize); + flgs = *rsz == blksize ? GFP_KERNEL : GFP_KERNEL | __GFP_NOWARN; + addr = kzalloc(*rsz, flgs); + if (addr || *rsz == blksize) + break; + size >>= 1; + } + return addr; +} + +static int blkoops_read_session(struct blkoops *c) +{ + struct blkoops_node *bn; + + bn = blkoops_read_node(c, 0); + if (IS_ERR(bn)) { + if (PTR_ERR(bn) == -EINVAL) + pr_err("blkoops: %s: bad magic\n", c->bdev_name); + return PTR_ERR(bn); + } + + c->last_session = bn->session; + c->next_session = c->last_session + 1; + c->next_file = 0; + + c->erased_session = bn->session; + c->erased_file = bn->file; + c->erased_sects = bn->nr_sects; + + kfree(bn); + + bn = blkoops_read_node(c, blkoops_scan_start(c)); + if (IS_ERR(bn)) + goto out; + + if (!blkoops_validate_node(c, bn)) { + c->last_session = bn->session; + c->next_session = c->last_session + 1; + } + + kfree(bn); +out: + DBG("sess %u size %lld blksz %u", c->last_session, c->size, c->blksize); + if (c->erased_file) { + DBG("erased sess %u file %u sects %llu", + c->erased_session, c->erased_file, (u64)c->erased_sects); + } + + return 0; +} + +static int blkoops_get_bdev_size(struct blkoops *c, struct block_device *bdev) +{ + unsigned int blksize; + loff_t size; + + blksize = bdev_logical_block_size(bdev); + if (blksize < 512 || blksize & 511) + return -EINVAL; + + size = i_size_read(bdev->bd_inode); + if (size < 2 * blksize) + return -EINVAL; + + c->blksize = blksize; + c->sects_per_blk = blksize >> 9; + c->size = size; + c->nr_sects = size >> 9; + + return 0; +} + +static struct blkoops *blkoops_alloc(void) +{ + struct blkoops *c; + + c = kzalloc(sizeof(struct blkoops), GFP_KERNEL); + if (!c) + return NULL; + + c->psi.owner = THIS_MODULE, + c->psi.name = "blkoops", + c->psi.flags = PSTORE_NO_HEADINGS | PSTORE_MAX_KMSG_BYTES, + c->psi.open = blkoops_open, + c->psi.read = blkoops_read, + c->psi.file_read = blkoops_file_read, + c->psi.write = blkoops_write, + c->psi.erase = blkoops_erase, + c->psi.flush = blkoops_flush, + c->psi.data = c; + + spin_lock_init(&c->psi.buf_lock); +#ifdef CONFIG_DEBUG_FS + spin_lock_init(&c->dbg_lock); +#endif + return c; +} + +static struct blkoops *blkoops; + +static int blkoops_do_add(struct blkoops *c) +{ + int err; + + if (c->present) + return -EINVAL; + + if (!*devname) + return -EINVAL; + + *c->bdev_name = '\0'; + + c->devid = name_to_dev_t(devname); + if (!c->devid) { + err = -ENODEV; + goto out_err; + } + + c->bdev = blkdev_get_by_dev(c->devid, BLKOOPS_MODE, blkoops_do_add); + if (IS_ERR(c->bdev)) { + err = PTR_ERR(c->bdev); + goto out_err; + } + + bdevname(c->bdev, c->bdev_name); + + err = blk_panic_init(c->bdev); + if (err) + goto out_put; + + err = blkoops_get_bdev_size(c, c->bdev); + if (err) + goto out_cleanup; + + err = blkoops_read_session(c); + if (err) + goto out_cleanup; + + if (!c->buf) { + unsigned int rsz; + void *addr; + + addr = blkoops_alloc_buf(c->blksize, c->size, &rsz); + if (!addr) { + err = -ENOMEM; + goto out_cleanup; + } + c->buf = addr; + c->bufsize = rsz; + c->root = RB_ROOT; + } + + c->next_sect = blkoops_scan_start(c) >> 9; + + spin_lock(&blkoops_lock); + c->present = 1; + spin_unlock(&blkoops_lock); + + if (!c->psi.buf) { + c->psi.buf = c->buf + BLKOOPS_HDR_SZ; + c->psi.bufsize = c->bufsize - BLKOOPS_HDR_SZ; + err = pstore_register(&c->psi); + if (err) + goto out_no_pstore; + } + + blkoops_init_debugfs(c); + + pr_info("blkoops initialized on %s\n", c->bdev_name); + + return 0; + +out_no_pstore: + c->psi.buf = NULL; + c->present = 0; + kfree(c->buf); + c->buf = NULL; +out_cleanup: + blk_panic_cleanup(c->bdev); +out_put: + blkdev_put(c->bdev, BLKOOPS_MODE); +out_err: + pr_err("blkoops initialization failed on %s, error %d\n", + *c->bdev_name ? c->bdev_name : devname, err); + return err; +} + +static void blkoops_do_remove(struct blkoops *c) +{ + if (!c->present) + return; + + blkoops_remove_debugfs(c); + + spin_lock(&blkoops_lock); + c->present = 0; + spin_unlock(&blkoops_lock); + + blkoops_free_tree(c); + + blk_panic_cleanup(c->bdev); + + blkdev_put(c->bdev, BLKOOPS_MODE); + + pr_info("blkoops detached from %s\n", c->bdev_name); +} + +static int blkoops_add(void) +{ + int err; + + /* + * Headers are sector aligned and less than 1 sector in size so that + * only whole headers are read. + */ + BUILD_BUG_ON(BLKOOPS_HDR_SZ > 512); + + if (!blkoops) + blkoops = blkoops_alloc(); + if (!blkoops) + return -ENOMEM; + + err = blkoops_do_add(blkoops); + + if (!blkoops->psi.buf) { + /* + * 'blkoops' can be freed if the registration with pstore + * failed, otherwise 'blkoops' is never freed. + */ + kfree(blkoops); + blkoops = NULL; + } + + return err; +} + +static void blkoops_remove(void) +{ + if (blkoops) + blkoops_do_remove(blkoops); +} + +static int __init blkoops_init(void) +{ + int err = 0; + dev_t devid; + + mutex_lock(&blkoops_mutex); + init_done = 1; + if (!*devname) + goto out; + devid = name_to_dev_t(devname); + if (!devid) + wait_for_device_probe(); + err = blkoops_add(); +out: + mutex_unlock(&blkoops_mutex); + return err; +} + +late_initcall(blkoops_init); + +static int param_set_devname(const char *val, const struct kernel_param *kp) +{ + int err = 0; + + if (strlen(val) >= BLKOOPS_DEVNAME_SZ) { + pr_err("blkoops: devname parameter too long\n"); + return -ENOSPC; + } + + mutex_lock(&blkoops_mutex); + if (init_done) + blkoops_remove(); + if (sscanf(val, "%s", devname) != 1) + *devname = '\0'; + if (init_done && *devname) + err = blkoops_add(); + mutex_unlock(&blkoops_mutex); + + return err; +} + +static int param_get_devname(char *buffer, const struct kernel_param *kp) +{ + return snprintf(buffer, BLKOOPS_DEVNAME_SZ, "%s", devname); +} + +static struct kernel_param_ops param_ops_devname = { + .set = param_set_devname, + .get = param_get_devname, +}; + +module_param_cb(devname, ¶m_ops_devname, &devname, S_IRUSR | S_IWUSR); +MODULE_PARM_DESC(devname, "Canonical block device name or number"); + +module_param(dump_oops, int, S_IRUSR | S_IWUSR); +MODULE_PARM_DESC(dump_oops, + "set to 1 to dump oopses, 0 to dump only panics (default 0)"); + +MODULE_LICENSE("GPL v2"); +MODULE_AUTHOR("Adrian Hunter"); +MODULE_DESCRIPTION("Block Oops / Panic Logger"); -- 1.7.9.5 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/