[Cluster-devel] [PATCH 1/2] gfs2_edit: Add a savemeta file metadata header
Steven Whitehouse
swhiteho at redhat.com
Mon May 19 12:17:54 UTC 2014
Hi,
On 19/05/14 13:15, Andrew Price wrote:
> On 19/05/14 13:03, Steven Whitehouse wrote:
>> Hi,
>>
>> On 19/05/14 12:54, Andrew Price wrote:
>>> Previously 'gfs2_edit savemeta' just saved blocks into the file and any
>>> characterisation of the contents had to be calculated by reading the
>>> gfs2 structures out of it. Add a metadata header to savemeta output
>>> files which initially holds the original fs size and a file creation
>>> time, leaving some room to spare for future fields to be added.
>>>
>>> Signed-off-by: Andrew Price <anprice at redhat.com>
>>> ---
>>> gfs2/edit/savemeta.c | 95
>>> +++++++++++++++++++++++++++++++++++++++++++++++-----
>>> 1 file changed, 86 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/gfs2/edit/savemeta.c b/gfs2/edit/savemeta.c
>>> index ab7f86f..a10da19 100644
>>> --- a/gfs2/edit/savemeta.c
>>> +++ b/gfs2/edit/savemeta.c
>>> @@ -19,6 +19,7 @@
>>> #include <sys/time.h>
>>> #include <linux/gfs2_ondisk.h>
>>> #include <zlib.h>
>>> +#include <time.h>
>>> #include <logging.h>
>>> #include "osi_list.h"
>>> @@ -30,6 +31,17 @@
>>> #define DFT_SAVE_FILE "/tmp/gfsmeta.XXXXXX"
>>> #define MAX_JOURNALS_SAVED 256
>>> +/* Header for the savemeta output file */
>>> +struct savemeta_header {
>>> +#define SAVEMETA_MAGIC (0x01171970)
>>> + uint32_t sh_magic;
>>> +#define SAVEMETA_FORMAT (1)
>>> + uint32_t sh_format; /* In case we want to change the layout */
>>> + uint64_t sh_time; /* When savemeta was run */
>>> + uint64_t sh_fs_bytes; /* Size of the fs */
>>> + uint8_t __reserved[104];
>>> +};
>>> +
>> I assume that the intent is that this new header will be block sized and
>> that its going to look like a fs block in the saved data stream? In that
>> case we don't need the reserved field, since there is nothing else that
>> will follow it (within that block)
>
> Well not exactly. The header size is fairly arbitrary besides choosing
> a 64-bit aligned size which is less than 256 bytes (see below). The
> file starts with the header and then the saved blocks begin. The
> savemeta file format doesn't have a fixed block size as the 'siglen'
> field of struct saved_metablock dictates the length of a block's body.
> That's why patch 2/2 removes the fixed-sized buffer from struct
> saved_metablock.
>
>> What happens if you feed a saved meta file with this header to an older
>> toolset that doesn't understand it? Will it just be ignored?
>
> Older versions actually have some code which searches for the saved
> superblock in the first 256 bytes of the file. The header will get
> ignored in that case and old tools will be able to restore new-format
> metadata dumps. That is, unless the header includes the superblock
> signature, which is very unlikely.
>
> Andy
>
Well it sounds a bit odd, but it seems to make sense to me :-) I think
that should do the trick at least,
Steve.
>>
>> Steve.
>>
>>> struct saved_metablock {
>>> uint64_t blk;
>>> uint16_t siglen; /* significant data length */
>>> @@ -670,6 +682,51 @@ static void save_allocated(struct rgrp_tree *rgd,
>>> struct metafd *mfd)
>>> free(ibuf);
>>> }
>>> +static int save_header(struct metafd *mfd, uint64_t fsbytes)
>>> +{
>>> + struct savemeta_header smh = {
>>> + .sh_magic = cpu_to_be32(SAVEMETA_MAGIC),
>>> + .sh_format = cpu_to_be32(SAVEMETA_FORMAT),
>>> + .sh_time = cpu_to_be64(time(NULL)),
>>> + .sh_fs_bytes = cpu_to_be64(fsbytes)
>>> + };
>>> +
>>> + if (savemetawrite(mfd, (char *)(&smh), sizeof(smh)) !=
>>> sizeof(smh))
>>> + return -1;
>>> + return 0;
>>> +}
>>> +
>>> +static int read_header(gzFile gzin_fd, struct savemeta_header *smh)
>>> +{
>>> + size_t rs;
>>> + struct savemeta_header smh_be = {0};
>>> +
>>> + rs = gzread(gzin_fd, &smh_be, sizeof(smh_be));
>>> + if (rs == -1) {
>>> + perror("Failed to read savemeta file header");
>>> + return -1;
>>> + }
>>> + if (rs != sizeof(smh_be))
>>> + return 1;
>>> +
>>> + smh->sh_magic = be32_to_cpu(smh_be.sh_magic);
>>> + smh->sh_format = be32_to_cpu(smh_be.sh_format);
>>> + smh->sh_time = be64_to_cpu(smh_be.sh_time);
>>> + smh->sh_fs_bytes = be64_to_cpu(smh_be.sh_fs_bytes);
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +static int check_header(struct savemeta_header *smh)
>>> +{
>>> + if (smh->sh_magic != SAVEMETA_MAGIC || smh->sh_format >
>>> SAVEMETA_FORMAT)
>>> + return -1;
>>> + printf("Savemeta file format %"PRIu32"\n", smh->sh_format);
>>> + printf("Created %s\n", ctime((time_t *)&smh->sh_time));
>>> + printf("File system size %s\n",
>>> anthropomorphize(smh->sh_fs_bytes));
>>> + return 0;
>>> +}
>>> +
>>> void savemeta(char *out_fn, int saveoption, int gziplevel)
>>> {
>>> int rgcount;
>>> @@ -678,6 +735,7 @@ void savemeta(char *out_fn, int saveoption, int
>>> gziplevel)
>>> struct metafd mfd;
>>> int sane;
>>> struct osi_node *n, *next = NULL;
>>> + int err = 0;
>>> sbd.md.journals = 1;
>>> @@ -746,8 +804,14 @@ void savemeta(char *out_fn, int saveoption, int
>>> gziplevel)
>>> get_journal_inode_blocks();
>>> + /* Write the savemeta file header */
>>> + err = save_header(&mfd, sbd.fssize * sbd.bsize);
>>> + if (err) {
>>> + perror("Failed to write metadata file header");
>>> + exit(1);
>>> + }
>>> /* Save off the superblock */
>>> - save_block(sbd.device_fd, &mfd, 0x10 * (4096 / sbd.bsize));
>>> + save_block(sbd.device_fd, &mfd, GFS2_SB_ADDR * GFS2_BASIC_BLOCK /
>>> sbd.bsize);
>>> /* If this is gfs1, save off the rindex because it's not
>>> part of the file system as it is in gfs2. */
>>> if (sbd.gfs1) {
>>> @@ -809,7 +873,7 @@ void savemeta(char *out_fn, int saveoption, int
>>> gziplevel)
>>> }
>>> static int restore_data(int fd, gzFile gzin_fd, int printblocksonly,
>>> - int find_highblk)
>>> + int find_highblk, const int startpos)
>>> {
>>> size_t rs;
>>> uint64_t buf64, writes = 0, highest_valid_block = 0;
>>> @@ -821,21 +885,21 @@ static int restore_data(int fd, gzFile gzin_fd,
>>> int printblocksonly,
>>> if (!printblocksonly)
>>> lseek(fd, 0, SEEK_SET);
>>> - gzseek(gzin_fd, 0, SEEK_SET);
>>> + gzseek(gzin_fd, startpos, SEEK_SET);
>>> rs = gzread(gzin_fd, rdbuf, sizeof(rdbuf));
>>> if (rs != sizeof(rdbuf)) {
>>> fprintf(stderr, "Error: File is too small.\n");
>>> return -1;
>>> }
>>> - for (pos = 0; pos < sizeof(rdbuf) - sizeof(uint64_t) -
>>> sizeof(uint16_t);
>>> + for (pos = startpos; pos < startpos + sizeof(rdbuf) -
>>> sizeof(uint64_t) - sizeof(uint16_t);
>>> pos++) {
>>> if (!memcmp(&rdbuf[pos + sizeof(uint64_t) +
>>> sizeof(uint16_t)],
>>> gfs_superblock_id, sizeof(gfs_superblock_id))) {
>>> break;
>>> }
>>> }
>>> - if (pos == sizeof(rdbuf) - sizeof(uint64_t) - sizeof(uint16_t))
>>> - pos = 0;
>>> + if (pos == startpos + sizeof(rdbuf) - sizeof(uint64_t) -
>>> sizeof(uint16_t))
>>> + pos = startpos;
>>> if (gzseek(gzin_fd, pos, SEEK_SET) != pos) {
>>> fprintf(stderr, "bad seek: %s from %s:%d: "
>>> "offset %lld (0x%llx)\n", strerror(errno),
>>> @@ -969,7 +1033,7 @@ static int restore_data(int fd, gzFile gzin_fd,
>>> int printblocksonly,
>>> }
>>> if (!printblocksonly && !find_highblk)
>>> warm_fuzzy_stuff(sbd.fssize, TRUE);
>>> - if (find_highblk) {
>>> + if (find_highblk && startpos == 0) {
>>> printf("File system size: %lld (0x%llx) blocks, aka %sB\n",
>>> (unsigned long long)highest_valid_block,
>>> (unsigned long long)highest_valid_block,
>>> @@ -991,6 +1055,8 @@ void restoremeta(const char *in_fn, const char
>>> *out_device,
>>> {
>>> int error;
>>> gzFile gzfd;
>>> + int startpos = 0;
>>> + struct savemeta_header smh = {0};
>>> termlines = 0;
>>> if (!in_fn)
>>> @@ -1014,9 +1080,20 @@ void restoremeta(const char *in_fn, const char
>>> *out_device,
>>> if (!savedata)
>>> die("Can't allocate memory for the restore operation.\n");
>>> + gzseek(gzfd, 0, SEEK_SET);
>>> + error = read_header(gzfd, &smh);
>>> + if (error < 0) {
>>> + exit(1);
>>> + } else if (check_header(&smh) != 0) {
>>> + printf("No valid file header found. Falling back to old
>>> format...\n");
>>> + gzseek(gzfd, 0, SEEK_SET);
>>> + } else if (error == 0) {
>>> + startpos = sizeof(smh);
>>> + }
>>> +
>>> blks_saved = 0;
>>> - restore_data(sbd.device_fd, gzfd, printblocksonly, 1);
>>> - error = restore_data(sbd.device_fd, gzfd, printblocksonly, 0);
>>> + restore_data(sbd.device_fd, gzfd, printblocksonly, 1, startpos);
>>> + error = restore_data(sbd.device_fd, gzfd, printblocksonly, 0,
>>> startpos);
>>> printf("File %s %s %s.\n", in_fn,
>>> (printblocksonly ? "print" : "restore"),
>>> (error ? "error" : "successful"));
>>
>
More information about the Cluster-devel
mailing list