/**
* _slm_fcmh_endow - "Endow" or apply inheritance to a new directory
* entry from its parent directory replica layout.
* Note: the bulk of this is empty until we have a place to store such
* info in the SLASH2 metafile.
*/
int
_slm_fcmh_endow(int vfsid, struct fidc_membh *p, struct fidc_membh *c,
int wr)
{
sl_replica_t repls[SL_MAX_REPLICAS];
int nr, rc = 0;
uint32_t pol;
FCMH_LOCK(p);
pol = fcmh_2_ino(p)->ino_replpol;
nr = fcmh_2_nrepls(p);
memcpy(repls, fcmh_2_ino(p)->ino_repls, sizeof(repls[0]) *
SL_DEF_REPLICAS);
if (nr > SL_DEF_REPLICAS) {
mds_inox_ensure_loaded(fcmh_2_inoh(p));
memcpy(&repls[SL_DEF_REPLICAS],
fcmh_2_inox(p)->inox_repls, sizeof(repls[0]) *
SL_INOX_NREPLICAS);
}
FCMH_ULOCK(p);
FCMH_WAIT_BUSY(c);
fcmh_2_replpol(c) = pol;
fcmh_2_ino(c)->ino_nrepls = nr;
memcpy(fcmh_2_ino(c)->ino_repls, repls, sizeof(repls[0]) *
SL_DEF_REPLICAS);
if (nr > SL_DEF_REPLICAS) {
mds_inox_ensure_loaded(fcmh_2_inoh(c));
memcpy(fcmh_2_inox(c)->inox_repls,
&repls[SL_DEF_REPLICAS], sizeof(repls[0]) *
SL_INOX_NREPLICAS);
}
if (wr)
mds_inodes_odsync(vfsid, c, mdslog_ino_repls);
FCMH_UNBUSY(c);
return (rc);
}
void
slm_pack_inode(struct fidc_membh *f, struct srt_inode *in)
{
struct slash_inode_handle *ih;
int rc;
ih = fcmh_2_inoh(f);
in->newreplpol = ih->inoh_ino.ino_replpol;
in->nrepls = ih->inoh_ino.ino_nrepls;
memcpy(in->reptbl, &ih->inoh_ino.ino_repls,
sizeof(ih->inoh_ino.ino_repls));
if (in->nrepls > SL_DEF_REPLICAS) {
rc = mds_inox_ensure_loaded(ih);
if (!rc)
memcpy(&in->reptbl[SL_DEF_REPLICAS],
&ih->inoh_extras->inox_repls,
sizeof(ih->inoh_extras->inox_repls));
}
}
void
slm_repl_upd_write(struct bmap *b, int rel)
{
struct {
sl_replica_t iosv[SL_MAX_REPLICAS];
char *stat[SL_MAX_REPLICAS];
unsigned nios;
} add, del, chg;
int off, vold, vnew, sprio, uprio, rc;
struct sl_mds_iosinfo *si;
struct bmap_mds_info *bmi;
struct fidc_membh *f;
struct sl_resource *r;
sl_ios_id_t resid;
unsigned n, nrepls;
bmi = bmap_2_bmi(b);
f = b->bcm_fcmh;
sprio = bmi->bmi_sys_prio;
uprio = bmi->bmi_usr_prio;
add.nios = 0;
del.nios = 0;
chg.nios = 0;
nrepls = fcmh_2_nrepls(f);
for (n = 0, off = 0; n < nrepls; n++, off += SL_BITS_PER_REPLICA) {
if (n == SL_DEF_REPLICAS)
mds_inox_ensure_loaded(fcmh_2_inoh(f));
resid = fcmh_2_repl(f, n);
vold = SL_REPL_GET_BMAP_IOS_STAT(bmi->bmi_orepls, off);
vnew = SL_REPL_GET_BMAP_IOS_STAT(bmi->bmi_repls, off);
r = libsl_id2res(resid);
si = r ? res2iosinfo(r) : &slm_null_iosinfo;
if (vold == vnew)
;
/* Work was added. */
else if ((vold != BREPLST_REPL_SCHED &&
vold != BREPLST_GARBAGE_QUEUED &&
vold != BREPLST_GARBAGE_SCHED &&
vnew == BREPLST_REPL_QUEUED) ||
(vold != BREPLST_GARBAGE_SCHED &&
vnew == BREPLST_GARBAGE_QUEUED &&
(si->si_flags & SIF_PRECLAIM_NOTSUP) == 0)) {
OPSTAT_INCR("repl-work-add");
PUSH_IOS(b, &add, resid, NULL);
}
/* Work has finished. */
else if ((vold == BREPLST_REPL_QUEUED ||
vold == BREPLST_REPL_SCHED ||
vold == BREPLST_TRUNC_SCHED ||
vold == BREPLST_TRUNC_QUEUED ||
vold == BREPLST_GARBAGE_SCHED ||
vold == BREPLST_VALID) &&
(((si->si_flags & SIF_PRECLAIM_NOTSUP) &&
vnew == BREPLST_GARBAGE_QUEUED) ||
vnew == BREPLST_VALID ||
vnew == BREPLST_INVALID)) {
OPSTAT_INCR("repl-work-del");
PUSH_IOS(b, &del, resid, NULL);
}
/*
* Work that was previously scheduled failed so
* requeue it.
*/
else if (vold == BREPLST_REPL_SCHED ||
vold == BREPLST_GARBAGE_SCHED ||
vold == BREPLST_TRUNC_SCHED)
PUSH_IOS(b, &chg, resid, "Q");
/* Work was scheduled. */
else if (vnew == BREPLST_REPL_SCHED ||
vnew == BREPLST_GARBAGE_SCHED ||
vnew == BREPLST_TRUNC_SCHED)
PUSH_IOS(b, &chg, resid, "S");
/* Work was reprioritized. */
else if (sprio != -1 || uprio != -1)
PUSH_IOS(b, &chg, resid, NULL);
}
for (n = 0; n < add.nios; n++) {
rc = slm_upsch_insert(b, add.iosv[n].bs_id, sprio,
uprio);
if (!rc)
continue;
psclog_warnx("upsch insert failed: bno = %d, "
"fid=%"PRId64", ios= %d, rc = %d",
b->bcm_bmapno, bmap_2_fid(b),
add.iosv[n].bs_id, rc);
}
for (n = 0; n < del.nios; n++) {
spinlock(&slm_upsch_lock);
dbdo(NULL, NULL,
" DELETE FROM upsch"
" WHERE resid = ?"
" AND fid = ?"
" AND bno = ?",
SQLITE_INTEGER, del.iosv[n].bs_id,
SQLITE_INTEGER64, bmap_2_fid(b),
SQLITE_INTEGER, b->bcm_bmapno);
freelock(&slm_upsch_lock);
}
for (n = 0; n < chg.nios; n++) {
spinlock(&slm_upsch_lock);
dbdo(NULL, NULL,
" UPDATE upsch"
" SET status = IFNULL(?, status),"
" sys_prio = IFNULL(?, sys_prio),"
" usr_prio = IFNULL(?, usr_prio)"
" WHERE resid = ?"
" AND fid = ?"
" AND bno = ?",
chg.stat[n] ? SQLITE_TEXT : SQLITE_NULL,
chg.stat[n] ? chg.stat[n] : 0,
sprio == -1 ? SQLITE_NULL : SQLITE_INTEGER,
sprio == -1 ? 0 : sprio,
uprio == -1 ? SQLITE_NULL : SQLITE_INTEGER,
uprio == -1 ? 0 : uprio,
SQLITE_INTEGER, chg.iosv[n].bs_id,
SQLITE_INTEGER64, bmap_2_fid(b),
SQLITE_INTEGER, b->bcm_bmapno);
freelock(&slm_upsch_lock);
}
bmap_2_bmi(b)->bmi_sys_prio = -1;
bmap_2_bmi(b)->bmi_usr_prio = -1;
if (rel) {
BMAP_LOCK(b);
b->bcm_flags &= ~BMAPF_REPLMODWR;
bmap_wake_locked(b);
bmap_op_done_type(b, BMAP_OPCNT_WORK);
}
}
/*
* Return the index of the given IOS ID or a negative error code on failure.
*/
int
_mds_repl_ios_lookup(int vfsid, struct slash_inode_handle *ih,
sl_ios_id_t ios, int flag)
{
int locked, rc;
struct slm_inox_od *ix = NULL;
struct sl_resource *res;
struct fidc_membh *f;
sl_replica_t *repl;
uint32_t i, j, nr;
char buf[LINE_MAX];
switch (flag) {
case IOSV_LOOKUPF_ADD:
OPSTAT_INCR("replicate-add");
break;
case IOSV_LOOKUPF_DEL:
OPSTAT_INCR("replicate-del");
break;
case IOSV_LOOKUPF_LOOKUP:
OPSTAT_INCR("replicate-lookup");
break;
default:
psc_fatalx("Invalid IOS lookup flag %d", flag);
}
/*
* Can I assume that IOS ID are non-zeros. If so, I can use
* zero to mark a free slot. See sl_global_id_build().
*/
f = inoh_2_fcmh(ih);
nr = ih->inoh_ino.ino_nrepls;
repl = ih->inoh_ino.ino_repls;
locked = INOH_RLOCK(ih);
psc_assert(nr <= SL_MAX_REPLICAS);
if (nr == SL_MAX_REPLICAS && flag == IOSV_LOOKUPF_ADD) {
DEBUG_INOH(PLL_WARN, ih, buf, "too many replicas");
PFL_GOTOERR(out, rc = -ENOSPC);
}
res = libsl_id2res(ios);
if (res == NULL || !RES_ISFS(res))
PFL_GOTOERR(out, rc = -SLERR_RES_BADTYPE);
/*
* 09/29/2016: Hit SLERR_SHORTIO in the function. Need more investigation.
*/
/*
* Return ENOENT by default for IOSV_LOOKUPF_DEL & IOSV_LOOKUPF_LOOKUP.
*/
rc = -ENOENT;
/*
* Search the existing replicas to see if the given IOS is
* already there.
*
* The following code can step through zero IOS IDs just fine.
*
*/
for (i = 0, j = 0; i < nr; i++, j++) {
if (i == SL_DEF_REPLICAS) {
/*
* The first few replicas are in the inode
* itself, the rest are in the extra inode
* block.
*/
rc = mds_inox_ensure_loaded(ih);
if (rc)
goto out;
ix = ih->inoh_extras;
repl = ix->inox_repls;
j = 0;
}
DEBUG_INOH(PLL_DEBUG, ih, buf, "is rep[%u](=%u) == %u ?",
j, repl[j].bs_id, ios);
if (repl[j].bs_id == ios) {
/*
* Luckily, this code is only called by mds_repl_delrq()
* for directories.
*
* Make sure that the logic works for at least the following
* edge cases:
*
* (1) There is only one item in the basic array.
* (2) There is only one item in the extra array.
* (3) The number of items is SL_DEF_REPLICAS.
* (4) The number of items is SL_MAX_REPLICAS.
*/
if (flag == IOSV_LOOKUPF_DEL) {
/*
* Compact the array if the IOS is not the last
* one. The last one will be either overwritten
* or zeroed. Note that we might move extra
* garbage at the end if the total number is less
* than SL_DEF_REPLICAS.
*/
if (i < SL_DEF_REPLICAS - 1) {
memmove(&repl[j], &repl[j + 1],
(SL_DEF_REPLICAS - j - 1) *
sizeof(*repl));
}
/*
* All items in the basic array, zero the last
* one and we are done.
*/
if (nr <= SL_DEF_REPLICAS) {
repl[nr-1].bs_id = 0;
goto syncit;
}
/*
* Now we know we have more than SL_DEF_REPLICAS
* items. However, if we are in the basic array,
* we have not read the extra array yet. In this
* case, we should also move the first item from
* the extra array to the last one in the basic
* array (overwrite).
*/
if (i < SL_DEF_REPLICAS) {
rc = mds_inox_ensure_loaded(ih);
if (rc)
goto out;
ix = ih->inoh_extras;
repl[SL_DEF_REPLICAS - 1].bs_id =
ix->inox_repls[0].bs_id;
repl = ix->inox_repls;
j = 0;
}
/*
* Compact the extra array unless the IOS is
* the last one, which will be zeroed.
*/
if (i < SL_MAX_REPLICAS - 1) {
memmove(&repl[j], &repl[j + 1],
(SL_INOX_NREPLICAS - j - 1) *
sizeof(*repl));
}
repl[nr-SL_DEF_REPLICAS-1].bs_id = 0;
syncit:
ih->inoh_ino.ino_nrepls = nr - 1;
rc = mds_inodes_odsync(vfsid, f, mdslog_ino_repls);
if (rc)
goto out;
}
/* XXX EEXIST for IOSV_LOOKUPF_ADD? */
rc = i;
goto out;
}
}
/* It doesn't exist; add to inode replica table if requested. */
if (flag == IOSV_LOOKUPF_ADD) {
/* paranoid */
psc_assert(i == nr);
if (nr >= SL_DEF_REPLICAS) {
/* be careful with the case of nr = SL_DEF_REPLICAS */
rc = mds_inox_ensure_loaded(ih);
if (rc)
goto out;
repl = ih->inoh_extras->inox_repls;
j = i - SL_DEF_REPLICAS;
} else {
repl = ih->inoh_ino.ino_repls;
j = i;
}
repl[j].bs_id = ios;
DEBUG_INOH(PLL_DIAG, ih, buf, "add IOS(%u) at idx %d", ios, i);
ih->inoh_ino.ino_nrepls = nr + 1;
rc = mds_inodes_odsync(vfsid, f, mdslog_ino_repls);
if (!rc)
rc = i;
}
out:
INOH_URLOCK(ih, locked);
return (rc);
}
int
mds_inode_update_interrupted(int vfsid, struct slash_inode_handle *ih,
int *rc)
{
char fn[NAME_MAX + 1];
struct srt_stat sstb;
struct iovec iovs[2];
uint64_t crc, od_crc;
void *h = NULL, *th;
mdsio_fid_t inum;
int exists = 0;
size_t nb;
th = inoh_2_mfh(ih);
snprintf(fn, sizeof(fn), "%016"PRIx64".update",
inoh_2_fid(ih));
*rc = mdsio_lookup(vfsid, mds_tmpdir_inum[vfsid], fn, &inum,
&rootcreds, NULL);
if (*rc)
PFL_GOTOERR(out, *rc);
*rc = mdsio_opencreatef(vfsid, inum, &rootcreds, O_RDONLY,
MDSIO_OPENCRF_NOLINK, 0644, NULL, NULL, NULL, &h, NULL,
NULL, 0);
if (*rc)
PFL_GOTOERR(out, *rc);
iovs[0].iov_base = &ih->inoh_ino;
iovs[0].iov_len = sizeof(ih->inoh_ino);
iovs[1].iov_base = &od_crc;
iovs[1].iov_len = sizeof(od_crc);
*rc = mdsio_preadv(vfsid, &rootcreds, iovs, nitems(iovs), &nb, 0, h);
if (*rc)
PFL_GOTOERR(out, *rc);
psc_crc64_calc(&crc, &ih->inoh_ino, sizeof(ih->inoh_ino));
if (crc != od_crc) {
*rc = PFLERR_BADCRC;
PFL_GOTOERR(out, *rc);
}
exists = 1;
psc_assert(ih->inoh_extras == NULL);
ih->inoh_extras = PSCALLOC(INOX_SZ);
inoh_2_mfh(ih) = h;
*rc = mds_inox_ensure_loaded(ih);
if (*rc)
PFL_GOTOERR(out, *rc);
inoh_2_mfh(ih) = th;
memset(&sstb, 0, sizeof(sstb));
*rc = mdsio_setattr(vfsid, 0, &sstb, SL_SETATTRF_METASIZE,
&rootcreds, NULL, th, NULL);
if (*rc)
PFL_GOTOERR(out, *rc);
*rc = mds_inode_dump(vfsid, NULL, ih, h);
if (*rc)
PFL_GOTOERR(out, *rc);
mdsio_unlink(vfsid, mds_tmpdir_inum[vfsid], NULL, fn,
&rootcreds, NULL, NULL);
out:
if (h)
mdsio_release(vfsid, &rootcreds, h);
if (*rc)
mdsio_unlink(vfsid, mds_tmpdir_inum[vfsid], NULL, fn,
&rootcreds, NULL, NULL);
inoh_2_mfh(ih) = th;
return (exists);
}