void
slm_fcmh_dtor(struct fidc_membh *f)
{
struct fcmh_mds_info *fmi;
int rc, vfsid;
fmi = fcmh_2_fmi(f);
if (fcmh_isreg(f)) {
psc_assert(psc_dynarray_len(&fmi->fmi_ptrunc_clients) == 0);
psc_dynarray_free(&fmi->fmi_ptrunc_clients);
}
if (fcmh_isreg(f) || fcmh_isdir(f)) {
/* XXX Need to worry about other modes here */
if (!fmi->fmi_ctor_rc) {
slfid_to_vfsid(fcmh_2_fid(f), &vfsid);
rc = mdsio_release(vfsid, &rootcreds,
fcmh_2_mfh(f));
psc_assert(rc == 0);
}
}
if (fcmh_isdir(f)) {
slfid_to_vfsid(fcmh_2_fid(f), &vfsid);
rc = mdsio_release(vfsid, &rootcreds,
fcmh_2_dino_mfh(f));
psc_assert(rc == 0);
}
if (fmi->fmi_inodeh.inoh_extras)
PSCFREE(fmi->fmi_inodeh.inoh_extras);
}
int
slm_fcmh_ctor(struct fidc_membh *f, __unusedx int flags)
{
struct fcmh_mds_info *fmi;
struct mio_fh *ino_mfh;
struct slm_inoh *ih;
mio_fid_t ino_mfid;
int rc, vfsid;
DEBUG_FCMH(PLL_DIAG, f, "ctor");
rc = slfid_to_vfsid(fcmh_2_fid(f), &vfsid);
if (rc) {
DEBUG_FCMH(PLL_WARN, f, "invalid file system ID; "
"rc=%d", rc);
return (rc);
}
fmi = fcmh_2_fmi(f);
memset(fmi, 0, sizeof(*fmi));
rc = mdsio_lookup_slfid(vfsid, fcmh_2_fid(f), &rootcreds,
&f->fcmh_sstb, &fcmh_2_mfid(f));
if (rc) {
fmi->fmi_ctor_rc = rc;
DEBUG_FCMH(PLL_WARN, f, "mdsio_lookup_slfid failed; "
"fid="SLPRI_FID" rc=%d",
fcmh_2_fid(f), rc);
return (rc);
}
ih = &fmi->fmi_inodeh;
ih->inoh_flags = INOH_INO_NOTLOADED;
ino_mfid = fcmh_2_mfid(f);
ino_mfh = fcmh_2_mfhp(f);
if (fcmh_isdir(f)) {
mio_fid_t pmfid;
char fn[24];
rc = mdsio_opendir(vfsid, fcmh_2_mfid(f), &rootcreds,
NULL, &fcmh_2_mfh(f));
if (rc) {
DEBUG_FCMH(PLL_WARN, f, "mdsio_opendir failed; "
"mio_fid=%"PRIx64" rc=%d", fcmh_2_mfid(f),
rc);
return (rc);
}
snprintf(fn, sizeof(fn), "%016"PRIx64".ino",
fcmh_2_fid(f));
pmfid = mdsio_getfidlinkdir(fcmh_2_fid(f));
rc = mdsio_lookup(vfsid, pmfid, fn,
&fcmh_2_dino_mfid(f), &rootcreds, NULL);
if (rc == ENOENT) {
struct slm_inox_od inox;
rc = mdsio_opencreatef(vfsid, pmfid, &rootcreds,
O_CREAT | O_EXCL | O_RDWR,
MDSIO_OPENCRF_NOLINK, 0644, fn,
&fcmh_2_dino_mfid(f), NULL,
&fcmh_2_dino_mfh(f), NULL, NULL, 0);
psc_assert(rc == 0);
INOH_LOCK(ih);
rc = mds_inode_write(vfsid, ih, NULL, NULL);
psc_assert(rc == 0);
memset(&inox, 0, sizeof(inox));
ih->inoh_extras = &inox;
rc = mds_inox_write(vfsid, ih, NULL, NULL);
ih->inoh_extras = NULL;
INOH_ULOCK(ih);
psc_assert(rc == 0);
mdsio_release(vfsid, &rootcreds,
fcmh_2_dino_mfh(f));
} else if (rc) {
fmi->fmi_ctor_rc = rc;
DEBUG_FCMH(PLL_WARN, f,
"mdsio_lookup failed; rc=%d", rc);
return (rc);
}
ino_mfid = fcmh_2_dino_mfid(f);
ino_mfh = fcmh_2_dino_mfhp(f);
}
if (fcmh_isreg(f))
psc_dynarray_init(&fmi->fmi_ptrunc_clients);
if (fcmh_isdir(f) || fcmh_isreg(f)) {
/*
* We shouldn't need O_LARGEFILE because SLASH2
* metafiles are small.
*
* I created a file with size of 8070450532247928832
* using dd by seeking to a large offset and writing one
* byte. Somehow, the ZFS size becomes 5119601018368.
* Without O_LARGEFILE, I got EOVERFLOW (75) here. The
* SLASH2 size is correct though.
*/
rc = mdsio_opencreate(vfsid, ino_mfid, &rootcreds,
O_RDWR, 0, NULL, NULL, NULL, &ino_mfh->fh, NULL,
NULL, 0);
if (rc == 0) {
rc = mds_inode_read(&fmi->fmi_inodeh);
if (rc)
DEBUG_FCMH(PLL_WARN, f,
"could not load inode; "
"mfid=%"PRIx64" rc=%d",
ino_mfid, rc);
} else {
fmi->fmi_ctor_rc = rc;
DEBUG_FCMH(PLL_WARN, f,
"mdsio_opencreate failed; "
"mfid=%"PRIx64" rc=%d",
ino_mfid, rc);
}
} else
DEBUG_FCMH(PLL_DIAG, f, "special file, no zfs obj");
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);
}
/*
* Handle a NAMESPACE_FORWARD request from another MDS.
*/
int
slm_rmm_handle_namespace_forward(struct pscrpc_request *rq)
{
char *from, *to, *name, *linkname;
struct fidc_membh *p = NULL, *op = NULL, *np = NULL;
struct srm_forward_req *mq;
struct srm_forward_rep *mp;
struct slash_creds cr;
struct srt_stat sstb;
void *mfh;
int vfsid;
p = op = np = NULL;
SL_RSX_ALLOCREP(rq, mq, mp);
if (mq->op != SLM_FORWARD_MKDIR &&
mq->op != SLM_FORWARD_RMDIR &&
mq->op != SLM_FORWARD_CREATE &&
mq->op != SLM_FORWARD_UNLINK &&
mq->op != SLM_FORWARD_SYMLINK &&
mq->op != SLM_FORWARD_RENAME &&
mq->op != SLM_FORWARD_SETATTR) {
mp->rc = -EINVAL;
return (0);
}
psclog_info("op=%d, name=%s", mq->op, mq->req.name);
mp->rc = slfid_to_vfsid(mq->fg.fg_fid, &vfsid);
if (mp->rc)
return (0);
if (current_vfsid != vfsid) {
mp->rc = -EINVAL;
return (0);
}
cr.scr_uid = mq->creds.scr_uid;
cr.scr_gid = mq->creds.scr_gid;
mds_reserve_slot(2);
switch (mq->op) {
case SLM_FORWARD_MKDIR:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
sstb.sst_mode = mq->mode;
sstb.sst_uid = mq->creds.scr_uid;
sstb.sst_gid = mq->creds.scr_gid;
mp->rc = -mdsio_mkdir(vfsid, fcmh_2_mfid(p),
mq->req.name, &sstb, 0, 0, &mp->attr, NULL,
mdslog_namespace, slm_get_next_slashfid, 0);
break;
case SLM_FORWARD_CREATE:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
mp->rc = mdsio_opencreate(vfsid, fcmh_2_mfid(p),
&cr, O_CREAT | O_EXCL | O_RDWR, mq->mode,
mq->req.name, NULL, &mp->attr, &mfh,
mdslog_namespace, slm_get_next_slashfid, 0);
if (!mp->rc)
mdsio_release(vfsid, &rootcreds, mfh);
break;
case SLM_FORWARD_RMDIR:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
mp->rc = mdsio_rmdir(vfsid, fcmh_2_mfid(p), NULL,
mq->req.name, &rootcreds, mdslog_namespace);
break;
case SLM_FORWARD_UNLINK:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
mp->rc = -mdsio_unlink(vfsid, fcmh_2_mfid(p), NULL,
mq->req.name, &rootcreds, mdslog_namespace,
&mp->attr);
break;
case SLM_FORWARD_RENAME:
mp->rc = slm_fcmh_get(&mq->fg, &op);
if (mp->rc)
break;
mp->rc = slm_fcmh_get(&mq->nfg, &np);
if (mp->rc)
break;
from = mq->req.name;
to = mq->req.name + strlen(mq->req.name) + 1;
mp->rc = mdsio_rename(vfsid, fcmh_2_mfid(op), from,
fcmh_2_mfid(np), to, &rootcreds,
mdslog_namespace, &mp->attr);
break;
case SLM_FORWARD_SETATTR:
/*
* This is tough because we have some logic at the fcmh
* layer dealing with (partial) truncates. It is not a
* pure namespace operation.
*/
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
mp->rc = -mdsio_setattr(vfsid, fcmh_2_mfid(p),
&mq->req.sstb, mq->to_set, &rootcreds, &mp->attr,
fcmh_2_mfh(p), mdslog_namespace);
break;
case SLM_FORWARD_SYMLINK:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
name = mq->req.name;
linkname = mq->req.name + strlen(mq->req.name) + 1;
mp->rc = mdsio_symlink(vfsid, linkname,
fcmh_2_mfid(p), name, &cr, &mp->attr, NULL,
NULL, slm_get_next_slashfid, 0);
break;
}
mds_unreserve_slot(2);
if (p)
fcmh_op_done(p);
if (op)
fcmh_op_done(op);
if (np)
fcmh_op_done(np);
return (0);
}
int
mds_inode_update(int vfsid, struct slash_inode_handle *ih,
int old_version)
{
char fn[NAME_MAX + 1];
struct sl_ino_compat *sic;
struct fidc_membh *f;
struct srt_stat sstb;
void *h = NULL, *th;
int rc;
sic = &sl_ino_compat_table[old_version];
rc = sic->sic_read_ino(ih);
if (rc)
return (rc);
DEBUG_INOH(PLL_INFO, ih, "updating old inode (v %d)",
old_version);
f = inoh_2_fcmh(ih);
snprintf(fn, sizeof(fn), "%016"PRIx64".update", fcmh_2_fid(f));
rc = mdsio_opencreatef(vfsid, mds_tmpdir_inum[vfsid],
&rootcreds, O_RDWR | O_CREAT | O_TRUNC,
MDSIO_OPENCRF_NOLINK, 0644, fn, NULL, NULL, &h, NULL, NULL,
0);
if (rc)
PFL_GOTOERR(out, rc);
psc_assert(ih->inoh_extras == NULL);
ih->inoh_extras = PSCALLOC(INOX_SZ);
/* convert old structures into new into temp file */
rc = sic->sic_read_inox(ih);
if (rc)
PFL_GOTOERR(out, rc);
th = inoh_2_mfhp(ih)->fh;
inoh_2_mfhp(ih)->fh = h;
rc = mds_inode_dump(vfsid, sic, ih, th);
inoh_2_mfhp(ih)->fh = th;
if (rc)
PFL_GOTOERR(out, rc);
/* move new structures to inode meta file */
memset(&sstb, 0, sizeof(sstb));
rc = mdsio_setattr(vfsid, 0, &sstb, SL_SETATTRF_METASIZE,
&rootcreds, NULL, th, NULL);
if (rc)
PFL_GOTOERR(out, rc);
// mdsio_rename(mds_tmpdir_inum, NULL, fn, &rootcreds, NULL);
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);
DEBUG_INOH(PLL_ERROR, ih, "error updating old inode "
"rc=%d", rc);
}
return (rc);
}
