/* compute object size given "stripeno" and the ost size */
obd_size lov_stripe_size(struct lov_stripe_md *lsm, obd_size ost_size,
int stripeno)
{
unsigned long ssize = lsm->lsm_stripe_size;
unsigned long stripe_size;
obd_off swidth;
obd_size lov_size;
int magic = lsm->lsm_magic;
ENTRY;
if (ost_size == 0)
RETURN(0);
LASSERT(lsm_op_find(magic) != NULL);
lsm_op_find(magic)->lsm_stripe_by_index(lsm, &stripeno, NULL, &swidth);
/* lov_do_div64(a, b) returns a % b, and a = a / b */
stripe_size = lov_do_div64(ost_size, ssize);
if (stripe_size)
lov_size = ost_size * swidth + stripeno * ssize + stripe_size;
else
lov_size = (ost_size - 1) * swidth + (stripeno + 1) * ssize;
RETURN(lov_size);
}
static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
{
int rc;
if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
char *buffer;
int sz;
CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
sz = lmm_bytes * 2 + 1;
OBD_ALLOC_LARGE(buffer, sz);
if (buffer != NULL) {
int i;
for (i = 0; i < lmm_bytes; i++)
sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
buffer[sz - 1] = '\0';
CERROR("%s\n", buffer);
OBD_FREE_LARGE(buffer, sz);
}
return -EINVAL;
}
rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
lmm_bytes, stripe_count);
return rc;
}
/* Unpack LOV object metadata from disk storage. It is packed in LE byte
* order and is opaque to the networking layer.
*/
struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, struct lov_mds_md *lmm,
size_t lmm_size)
{
struct lov_stripe_md *lsm;
u16 stripe_count;
u32 pattern;
u32 magic;
int rc;
rc = lov_verify_lmm(lmm, lmm_size, &stripe_count);
if (rc)
return ERR_PTR(rc);
magic = le32_to_cpu(lmm->lmm_magic);
pattern = le32_to_cpu(lmm->lmm_pattern);
lsm = lov_lsm_alloc(stripe_count, pattern, magic);
if (IS_ERR(lsm))
return lsm;
LASSERT(lsm_op_find(magic));
rc = lsm_op_find(magic)->lsm_unpackmd(lov, lsm, lmm);
if (rc) {
lov_free_memmd(&lsm);
return ERR_PTR(rc);
}
return lsm;
}
void lov_free_memmd(struct lov_stripe_md **lsmp)
{
struct lov_stripe_md *lsm = *lsmp;
LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
*lsmp = NULL;
}
/* Unpack LOV object metadata from disk storage. It is packed in LE byte
* order and is opaque to the networking layer.
*/
int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
struct lov_mds_md *lmm, int lmm_bytes)
{
struct obd_device *obd = class_exp2obd(exp);
struct lov_obd *lov = &obd->u.lov;
int rc = 0, lsm_size;
__u16 stripe_count;
__u32 magic;
__u32 pattern;
ENTRY;
/* If passed an MDS struct use values from there, otherwise defaults */
if (lmm) {
rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
if (rc)
RETURN(rc);
magic = le32_to_cpu(lmm->lmm_magic);
pattern = le32_to_cpu(lmm->lmm_pattern);
} else {
magic = LOV_MAGIC;
stripe_count = lov_get_stripecnt(lov, magic, 0);
pattern = LOV_PATTERN_RAID0;
}
/* If we aren't passed an lsmp struct, we just want the size */
if (!lsmp) {
/* XXX LOV STACKING call into osc for sizes */
LBUG();
RETURN(lov_stripe_md_size(stripe_count));
}
/* If we are passed an allocated struct but nothing to unpack, free */
if (*lsmp && !lmm) {
lov_free_memmd(lsmp);
RETURN(0);
}
lsm_size = lov_alloc_memmd(lsmp, stripe_count, pattern, magic);
if (lsm_size < 0)
RETURN(lsm_size);
/* If we are passed a pointer but nothing to unpack, we only alloc */
if (!lmm)
RETURN(lsm_size);
LASSERT(lsm_op_find(magic) != NULL);
rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
if (rc) {
lov_free_memmd(lsmp);
RETURN(rc);
}
RETURN(lsm_size);
}
int lov_free_memmd(struct lov_stripe_md **lsmp)
{
struct lov_stripe_md *lsm = *lsmp;
int refc;
*lsmp = NULL;
LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 0);
if ((refc = cfs_atomic_dec_return(&lsm->lsm_refc)) == 0) {
LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
}
return refc;
}
static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
{
int rc;
if (!lsm_op_find(le32_to_cpu(*(__u32 *)lmm))) {
CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
CERROR("%*phN\n", lmm_bytes, lmm);
return -EINVAL;
}
rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
lmm_bytes,
stripe_count);
return rc;
}
/* compute which stripe number "lov_off" will be written into */
int lov_stripe_number(struct lov_stripe_md *lsm, obd_off lov_off)
{
unsigned long ssize = lsm->lsm_stripe_size;
obd_off stripe_off, swidth;
int magic = lsm->lsm_magic;
LASSERT(lsm_op_find(magic) != NULL);
lsm_op_find(magic)->lsm_stripe_by_offset(lsm, NULL, &lov_off, &swidth);
stripe_off = lov_do_div64(lov_off, swidth);
/* Puts stripe_off/ssize result into stripe_off */
lov_do_div64(stripe_off, ssize);
return stripe_off;
}
int lov_free_memmd(struct lov_stripe_md **lsmp)
{
struct lov_stripe_md *lsm = *lsmp;
int refc;
*lsmp = NULL;
LASSERT(atomic_read(&lsm->lsm_refc) > 0);
refc = atomic_dec_return(&lsm->lsm_refc);
if (refc == 0)
lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
return refc;
}
/* Given a whole-file size and a stripe number, give the file size which
* corresponds to the individual object of that stripe.
*
* This behaves basically in the same was as lov_stripe_offset, except that
* file sizes falling before the beginning of a stripe are clamped to the end
* of the previous stripe, not the beginning of the next:
*
* S
* ---------------------------------------------------------------------
* | 0 | 1 | 2 | 0 | 1 | 2 |
* ---------------------------------------------------------------------
*
* if clamped to stripe 2 becomes:
*
* S
* ---------------------------------------------------------------------
* | 0 | 1 | 2 | 0 | 1 | 2 |
* ---------------------------------------------------------------------
*/
loff_t lov_size_to_stripe(struct lov_stripe_md *lsm, u64 file_size,
int stripeno)
{
unsigned long ssize = lsm->lsm_stripe_size;
loff_t stripe_off;
loff_t this_stripe;
loff_t swidth;
u32 magic = lsm->lsm_magic;
if (file_size == OBD_OBJECT_EOF)
return OBD_OBJECT_EOF;
LASSERT(lsm_op_find(magic) != NULL);
lsm_op_find(magic)->lsm_stripe_by_index(lsm, &stripeno, &file_size,
&swidth);
/* lov_do_div64(a, b) returns a % b, and a = a / b */
stripe_off = lov_do_div64(file_size, swidth);
this_stripe = (loff_t)stripeno * ssize;
if (stripe_off < this_stripe) {
/* Move to end of previous stripe, or zero */
if (file_size > 0) {
file_size--;
stripe_off = ssize;
} else {
stripe_off = 0;
}
} else {
stripe_off -= this_stripe;
if (stripe_off >= ssize) {
/* Clamp to end of this stripe */
stripe_off = ssize;
}
}
return (file_size * ssize + stripe_off);
}
/* we have an offset in file backed by an lov and want to find out where
* that offset lands in our given stripe of the file. for the easy
* case where the offset is within the stripe, we just have to scale the
* offset down to make it relative to the stripe instead of the lov.
*
* the harder case is what to do when the offset doesn't intersect the
* stripe. callers will want start offsets clamped ahead to the start
* of the nearest stripe in the file. end offsets similarly clamped to the
* nearest ending byte of a stripe in the file:
*
* all this function does is move offsets to the nearest region of the
* stripe, and it does its work "mod" the full length of all the stripes.
* consider a file with 3 stripes:
*
* S E
* ---------------------------------------------------------------------
* | 0 | 1 | 2 | 0 | 1 | 2 |
* ---------------------------------------------------------------------
*
* to find stripe 1's offsets for S and E, it divides by the full stripe
* width and does its math in the context of a single set of stripes:
*
* S E
* -----------------------------------
* | 0 | 1 | 2 |
* -----------------------------------
*
* it'll notice that E is outside stripe 1 and clamp it to the end of the
* stripe, then multiply it back out by lov_off to give the real offsets in
* the stripe:
*
* S E
* ---------------------------------------------------------------------
* | 1 | 1 | 1 | 1 | 1 | 1 |
* ---------------------------------------------------------------------
*
* it would have done similarly and pulled S forward to the start of a 1
* stripe if, say, S had landed in a 0 stripe.
*
* this rounding isn't always correct. consider an E lov offset that lands
* on a 0 stripe, the "mod stripe width" math will pull it forward to the
* start of a 1 stripe, when in fact it wanted to be rounded back to the end
* of a previous 1 stripe. this logic is handled by callers and this is why:
*
* this function returns < 0 when the offset was "before" the stripe and
* was moved forward to the start of the stripe in question; 0 when it
* falls in the stripe and no shifting was done; > 0 when the offset
* was outside the stripe and was pulled back to its final byte. */
int lov_stripe_offset(struct lov_stripe_md *lsm, loff_t lov_off, int stripeno,
loff_t *obdoff)
{
unsigned long ssize = lsm->lsm_stripe_size;
loff_t stripe_off;
loff_t this_stripe;
loff_t swidth;
u32 magic = lsm->lsm_magic;
int ret = 0;
if (lov_off == OBD_OBJECT_EOF) {
*obdoff = OBD_OBJECT_EOF;
return 0;
}
LASSERT(lsm_op_find(magic) != NULL);
lsm_op_find(magic)->lsm_stripe_by_index(lsm, &stripeno, &lov_off,
&swidth);
/* lov_do_div64(a, b) returns a % b, and a = a / b */
stripe_off = lov_do_div64(lov_off, swidth);
this_stripe = (loff_t)stripeno * ssize;
if (stripe_off < this_stripe) {
stripe_off = 0;
ret = -1;
} else {
stripe_off -= this_stripe;
if (stripe_off >= ssize) {
stripe_off = ssize;
ret = 1;
}
}
*obdoff = lov_off * ssize + stripe_off;
return ret;
}
/* Unpack LOV object metadata from disk storage. It is packed in LE byte
* order and is opaque to the networking layer.
*/
struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
size_t buf_size)
{
const struct lsm_operations *op;
struct lov_stripe_md *lsm;
u32 magic;
ENTRY;
if (buf_size < sizeof(magic))
RETURN(ERR_PTR(-EINVAL));
magic = le32_to_cpu(*(u32 *)buf);
op = lsm_op_find(magic);
if (op == NULL)
RETURN(ERR_PTR(-EINVAL));
lsm = op->lsm_unpackmd(lov, buf, buf_size);
RETURN(lsm);
}
