/* Copyright (c) 2010 Gluster, Inc. This file is part of GlusterFS. GlusterFS is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. GlusterFS 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, see . */ #ifndef _CONFIG_H #define _CONFIG_H #include "config.h" #endif #include "rpcsvc.h" #include "dict.h" #include "xlator.h" #include "mount3.h" #include "xdr-nfs3.h" #include "msg-nfs3.h" #include "iobuf.h" #include "nfs3.h" #include "mem-pool.h" #include "logging.h" #include "nfs-fops.h" #include "nfs-inodes.h" #include "nfs-generics.h" #include "nfs3-helpers.h" #include "nfs-mem-types.h" #include #include #include #include #define nfs3_validate_strlen_or_goto(str, len, label, status, retval) \ do { \ if ((str)) { \ if (strlen ((str)) > (len)) { \ status = NFS3ERR_NAMETOOLONG; \ retval = -ENAMETOOLONG; \ goto label; \ } \ } \ } while (0); \ #define nfs3_validate_nfs3_state(request, state, status, label, retval) \ do { \ state = nfs_rpcsvc_request_program_private (request); \ if (!nfs3) { \ gf_log (GF_NFS3, GF_LOG_ERROR, "NFSv3 state " \ "missing from RPC request"); \ status = NFS3ERR_SERVERFAULT; \ ret = -EFAULT; \ goto label; \ } \ } while (0); \ #define nfs3_export_access(nfs3state, xlid) ((nfs3state)->exports[xlid]).access #define nfs3_check_rw_volaccess(nfs3state, xlid, status, label) \ do { \ if (nfs3_export_access (nfs3state,xlid)!=GF_NFS3_VOLACCESS_RW){\ gf_log (GF_NFS3, GF_LOG_TRACE, "No read-write access");\ status = NFS3ERR_ROFS; \ goto label; \ } \ } while (0) \ #define nfs3_map_fh_to_volume(nfs3state, handle, rqst, volume, status, label) \ do { \ volume = nfs3_fh_to_xlator ((nfs3state)->exportslist, handle); \ if (!volume) { \ gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to map " \ "FH to vol"); \ status = NFS3ERR_STALE; \ goto label; \ } else { \ gf_log (GF_NFS3, GF_LOG_TRACE, "FH to Volume: %s"\ ,volume->name); \ nfs_rpcsvc_request_set_private (req, volume); \ } \ } while (0); \ #define nfs3_validate_gluster_fh(handle, status, errlabel) \ do { \ if ((handle)) { \ if (!nfs3_fh_validate (handle)) { \ status = NFS3ERR_BADHANDLE; \ goto errlabel; \ } \ } \ } while (0) \ #define nfs3_check_fh_resolve_status(cst, nfstat, erlabl) \ do { \ if ((cst)->resolve_ret < 0) { \ nfstat = nfs3_errno_to_nfsstat3 (cst->resolve_errno);\ goto erlabl; \ } \ } while (0) \ #define nfs3_check_new_fh_resolve_status(cst, nfstat, erlabl) \ do { \ if (((cst)->resolve_ret < 0) && \ ((cst)->resolve_errno != ENOENT)) { \ nfstat = nfs3_errno_to_nfsstat3 (cs->resolve_errno);\ goto erlabl; \ } \ } while (0) \ #define nfs3_funge_solaris_zerolen_fh(nfs3st, fhd, enam, nfsst, erl) \ do { \ xlator_t *fungexl = NULL; \ fungexl =nfs_mntpath_to_xlator ((nfs3st)->exportslist,enam);\ if (!fungexl) { \ (nfsst) = NFS3ERR_NOENT; \ goto erl; \ } \ \ (fhd)->xlatorid = nfs_xlator_to_xlid ((nfs3st)->exportslist, \ fungexl); \ (fhd)->gen = 0; \ (fhd)->ino = 1; \ (enam) = NULL; \ } while (0) \ #define nfs3_export_sync_trusted(nf3stt, xlid) ((nf3stt)->exports[xlid]).trusted_sync #define nfs3_export_write_trusted(nf3stt, xlid) ((nf3stt)->exports[xlid]).trusted_write int nfs3_solaris_zerolen_fh (struct nfs3_fh *fh, int fhlen) { if (!fh) return 0; if (nfs3_fh_validate (fh)) return 0; if (fhlen == 0) return 1; return 0; } /* Function pointer that represents the generic prototypes of functions used * to serialize NFS3 message structures into the XDR format. * For usage, see the nfs3svc_XXX_cbk functions. */ typedef ssize_t (*nfs3_serializer) (struct iovec outmsg, void *args); nfs3_call_state_t * nfs3_call_state_init (struct nfs3_state *s, rpcsvc_request_t *req, xlator_t *v) { nfs3_call_state_t *cs = NULL; if ((!s) || (!req) || (!v)) return NULL; cs = (nfs3_call_state_t *) mem_get (s->localpool); if (!cs) return NULL; memset (cs, 0, sizeof (*cs)); INIT_LIST_HEAD (&cs->entries.list); INIT_LIST_HEAD (&cs->openwait_q); cs->operrno = EINVAL; cs->req = req; cs->vol = v; cs->nfsx = s->nfsx; cs->nfs3state = s; return cs; } void nfs3_call_state_wipe (nfs3_call_state_t *cs) { struct nfs3_state *nfs3 = NULL; if (!cs) return; nfs3 = cs->nfs3state; if (cs->fd) { gf_log (GF_NFS3, GF_LOG_TRACE, "fd 0x%lx ref: %d", (long)cs->fd, cs->fd->refcount); fd_unref (cs->fd); } if (cs->resolventry) GF_FREE (cs->resolventry); if (cs->pathname) GF_FREE (cs->pathname); if (!list_empty (&cs->entries.list)) gf_dirent_free (&cs->entries); nfs_loc_wipe (&cs->oploc); nfs_loc_wipe (&cs->resolvedloc); if (cs->iob) iobuf_unref (cs->iob); memset (cs, 0, sizeof (*cs)); mem_put (nfs3->localpool, cs); /* Already refd by fd_lookup, so no need to ref again. */ } #define nfs3_handle_call_state_init(nfs3state, calls, rq, vl ,opstat, errlabel)\ do { \ calls = nfs3_call_state_init ((nfs3state), (rq), (vl)); \ if (!calls) { \ gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to " \ "init call state"); \ opstat = NFS3ERR_SERVERFAULT; \ goto errlabel; \ } \ } while (0) \ struct iobuf * nfs3_serialize_reply (rpcsvc_request_t *req, void *arg, nfs3_serializer sfunc, struct iovec *outmsg) { struct nfs3_state *nfs3 = NULL; struct iobuf *iob = NULL; ssize_t retlen = -1; nfs3 = (struct nfs3_state *)nfs_rpcsvc_request_program_private (req); if (!nfs3) { gf_log (GF_NFS3, GF_LOG_ERROR, "NFSv3 state not found in RPC" " request"); goto ret; } /* First, get the io buffer into which the reply in arg will * be serialized. */ iob = iobuf_get (nfs3->iobpool); if (!iob) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to get iobuf"); goto ret; } iobuf_to_iovec (iob, outmsg); /* Use the given serializer to translate the give C structure in arg * to XDR format which will be written into the buffer in outmsg. */ /* retlen is used to received the error since size_t is unsigned and we * need -1 for error notification during encoding. */ retlen = sfunc (*outmsg, arg); if (retlen == -1) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to encode message"); goto ret; } outmsg->iov_len = retlen; ret: if (retlen == -1) { iobuf_unref (iob); iob = NULL; } return iob; } /* Generic reply function for NFSv3 specific replies. */ int nfs3svc_submit_reply (rpcsvc_request_t *req, void *arg, nfs3_serializer sfunc) { struct iovec outmsg = {0, }; struct iobuf *iob = NULL; int ret = -1; if (!req) return -1; iob = nfs3_serialize_reply (req, arg, sfunc, &outmsg); if (!iob) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to serialize reply"); goto ret; } /* Then, submit the message for transmission. */ ret = nfs_rpcsvc_submit_message (req, outmsg, iob); /* Now that we've done our job of handing the message to the RPC layer * we can safely unref the iob in the hope that RPC layer must have * ref'ed the iob on receiving into the txlist. */ iobuf_unref (iob); if (ret == -1) { gf_log (GF_NFS3, GF_LOG_ERROR, "Reply submission failed"); goto ret; } ret = 0; ret: return ret; } int nfs3svc_submit_vector_reply (rpcsvc_request_t *req, void *arg, nfs3_serializer sfunc, struct iovec *payload, int vcount, struct iobref *piobref) { struct iovec outmsg = {0, }; struct iobuf *iob = NULL; int ret = -1; if (!req) return -1; iob = nfs3_serialize_reply (req, arg, sfunc, &outmsg); if (!iob) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to serialize reply"); goto err; } ret = nfs_rpcsvc_request_attach_vector (req, outmsg, iob, NULL, 0); iobuf_unref (iob); if (piobref) ret = nfs_rpcsvc_request_attach_vectors (req, payload, vcount, piobref); if (ret == -1) goto err; ret = nfs_rpcsvc_submit_vectors (req); err: return ret; } uint16_t nfs3_request_xlator_id (rpcsvc_request_t *rq) { struct nfs3_state *nfs3 = NULL; xlator_t *xl = NULL; if (!rq) return 0; xl = nfs_rpcsvc_request_private (rq); nfs3 = nfs_rpcsvc_request_program_private (rq); return nfs_xlator_to_xlid (nfs3->exportslist, xl); } int nfs3svc_null (rpcsvc_request_t *req) { struct iovec dummyvec = {0, }; if (!req) return RPCSVC_ACTOR_ERROR; nfs_rpcsvc_submit_generic (req, dummyvec, NULL); return RPCSVC_ACTOR_SUCCESS; } int nfs3_getattr_reply (rpcsvc_request_t *req, nfsstat3 status, struct iatt *buf) { getattr3res res; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_getattr3res (&res, status, buf, xlid); nfs3svc_submit_reply (req, &res, (nfs3_serializer)xdr_serialize_getattr3res); return 0; } int32_t nfs3svc_getattr_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xattr, struct iatt *postparent) { nfsstat3 status = NFS3_OK; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) status = nfs3_errno_to_nfsstat3 (op_errno); nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "GETATTR", status, op_errno); nfs3_getattr_reply (cs->req, status, buf); nfs3_call_state_wipe (cs); return 0; } int32_t nfs3svc_getattr_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf) { nfsstat3 status = NFS3_OK; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) status = nfs3_errno_to_nfsstat3 (op_errno); nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "GETATTR", status, op_errno); nfs3_getattr_reply (cs->req, status, buf); nfs3_call_state_wipe (cs); return 0; } int nfs3_getattr_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); /* If inode which is to be getattr'd is the root, we need to do a * lookup instead because after a server reboot, it is not necessary * for the root to have been looked up when the getattr on the root is * sent. AND, this causes a problem for stat-prefetch in that it * expects even the root inode to have been looked up. */ if (cs->resolvedloc.inode->ino == 1) ret = nfs_lookup (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_getattr_lookup_cbk, cs); else ret = nfs_stat (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_getattr_stat_cbk, cs); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Stat fop failed: %s: %s", cs->oploc.path, strerror (-ret)); stat = nfs3_errno_to_nfsstat3 (-ret); } nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "GETATTR", stat, -ret); nfs3_getattr_reply (cs->req, stat, NULL); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3_getattr (rpcsvc_request_t *req, struct nfs3_fh *fh) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cstate = NULL; if ((!req) || (!fh)) return -1; nfs3_log_common_call (nfs_rpcsvc_request_xid (req), "GETATTR", fh); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cstate, req, vol, stat, nfs3err); ret = nfs3_fh_resolve_and_resume (cstate, fh, NULL,nfs3_getattr_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "GETATTR", stat, -ret); nfs3_getattr_reply (req, stat, NULL); ret = 0; nfs3_call_state_wipe (cstate); } return ret; } int nfs3svc_getattr (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; getattr3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_getattr3args (&args, &fh); if (xdr_to_getattr3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_getattr (req, &fh); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "GETATTR procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_setattr_reply (rpcsvc_request_t *req, nfsstat3 stat, struct iatt *preop, struct iatt *postop) { setattr3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_setattr3res (&res, stat, preop, postop, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer) xdr_serialize_setattr3res); return 0; } int32_t nfs3svc_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf) { nfsstat3 stat = NFS3ERR_SERVERFAULT; struct iatt *prestat = NULL; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } /* If the first stat was got from the guarded setattr callback, or * from an earlier setattr call then we'll need to use that stat * instead of the preop returned here. */ if (cs->preparent.ia_ino != 0) prestat = &cs->preparent; else prestat = prebuf; stat = NFS3_OK; nfs3err: nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "SETATTR", stat, op_errno); nfs3_setattr_reply (cs->req, stat, prestat, postbuf); nfs3_call_state_wipe (cs); return 0; } int32_t nfs3svc_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preop, struct iatt *postop) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -1; struct iatt *prebuf = NULL; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } /* If the first stat was got from the guarded setattr callback, then * we'll need to use that stat instead of the preop returned here. */ if (cs->preparent.ia_ino != 0) prebuf = &cs->preparent; else { prebuf = preop; /* Store the current preop in case we need to send a truncate, * in which case the preop to be returned will be this one. */ cs->preparent = *preop; } /* Only truncate if the size is not already same as the requested * truncation and also only if this is not a directory. */ if ((gf_attr_size_set (cs->setattr_valid)) && (!IA_ISDIR (postop->ia_type))) { nfs_request_user_init (&nfu, cs->req); ret = nfs_truncate (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, cs->stbuf.ia_size, nfs3svc_truncate_cbk,cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); } else { ret = -1; /* Force a reply in the branch below. */ stat = NFS3_OK; } nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "SETATTR", stat, op_errno); nfs3_setattr_reply (cs->req, stat, prebuf, postop); nfs3_call_state_wipe (cs); } return 0; } int32_t nfs3svc_setattr_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf) { int ret = -EFAULT; nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } if (buf->ia_ctime != cs->timestamp.seconds) { gf_log (GF_NFS3, GF_LOG_TRACE, "Timestamps not in sync"); stat = NFS3ERR_NOT_SYNC; goto nfs3err; } /* Not a clean way but no motivation to add a new member to local. */ cs->preparent = *buf; nfs_request_user_init (&nfu, cs->req); ret = nfs_setattr (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc,&cs->stbuf, cs->setattr_valid, nfs3svc_setattr_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "SETATTR", stat, op_errno); nfs3_setattr_reply (cs->req, stat, NULL, NULL); nfs3_call_state_wipe (cs); } return 0; } int nfs3_setattr_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); /* If no ctime check is required, head straight to setting the attrs. */ if (cs->sattrguardcheck) ret = nfs_stat (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_setattr_stat_cbk, cs); else ret = nfs_setattr (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, &cs->stbuf, cs->setattr_valid, nfs3svc_setattr_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "SETATTR", stat, -ret); nfs3_setattr_reply (cs->req, stat, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_setattr (rpcsvc_request_t *req, struct nfs3_fh *fh, sattr3 *sattr, sattrguard3 *guard) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh) || (!sattr) || (!guard)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_common_call (nfs_rpcsvc_request_xid (req), "SETATTR", fh); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, fh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->setattr_valid = nfs3_sattr3_to_setattr_valid (sattr, &cs->stbuf, NULL); if (guard->check) { gf_log (GF_NFS3, GF_LOG_TRACE, "Guard check required"); cs->timestamp = guard->sattrguard3_u.obj_ctime; cs->sattrguardcheck = 1; } else { gf_log (GF_NFS3, GF_LOG_TRACE, "Guard check not required"); cs->sattrguardcheck = 0; } if (!cs->setattr_valid) { ret = -EINVAL; /* Force a reply */ stat = NFS3_OK; goto nfs3err; } ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_setattr_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "SETATTR", stat, -ret); nfs3_setattr_reply (req, stat, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_setattr (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; setattr3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_setattr3args (&args, &fh); if (xdr_to_setattr3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_setattr (req, &fh, &args.new_attributes, &args.guard); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "SETATTR procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_lookup_reply (rpcsvc_request_t *req, nfsstat3 stat, struct nfs3_fh *newfh, struct iatt *stbuf, struct iatt *postparent) { lookup3res res = {0, }; nfs3_fill_lookup3res (&res, stat, newfh, stbuf, postparent); return nfs3svc_submit_reply (req, &res, (nfs3_serializer)xdr_serialize_lookup3res); } int nfs3svc_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xattr, struct iatt *postparent) { struct nfs3_fh newfh = {{0}, }; nfsstat3 status = NFS3_OK; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { status = nfs3_errno_to_nfsstat3 (op_errno); goto xmit_res; } nfs3_fh_build_child_fh (&cs->parent, buf, &newfh); xmit_res: nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "LOOKUP", status, op_errno, &newfh); nfs3_lookup_reply (cs->req, status, &newfh, buf, postparent); nfs3_call_state_wipe (cs); return 0; } int nfs3svc_lookup_parentdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xattr, struct iatt *postparent) { struct nfs3_fh newfh = {{0}, }; nfsstat3 status = NFS3_OK; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { status = nfs3_errno_to_nfsstat3 (op_errno); goto xmit_res; } /* If the buf inode shows that this is a root dir's buf, then the file * handle needs to be specially crafted, in all other cases, we'll just * create the handle normally using the buffer of the parent dir. */ if (buf->ia_ino != 1) nfs3_fh_build_parent_fh (&cs->fh, buf, &newfh); else newfh = nfs3_fh_build_root_fh (cs->nfs3state->exportslist, cs->vol); xmit_res: nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "LOOKUP", status, op_errno, &newfh); nfs3_lookup_reply (cs->req, status, &newfh, buf, postparent); nfs3_call_state_wipe (cs); return 0; } int nfs3_lookup_parentdir_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; inode_t *parent = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); /* At this point now, the loc in cs is for the directory file handle * sent by the client. This loc needs to be transformed into a loc that * represents the parent dir of cs->resolvedloc.inode. * * EXCEPT in the case where the .. is a parent of the root directory. * In this case we'll be returning the file handle and attributes of the * root itself. */ nfs_request_user_init (&nfu, cs->req); /* Save the file handle from the LOOKUP request. We'll use this to * build the file handle of the parent directory in case the parent is * not root dir. */ cs->fh = cs->resolvefh; /* If fh is that of the root, the resolvedloc will already contain * the loc for root. After that, we'll send lookup for the root dir * itself since we cannot send the lookup on the parent of root. * * For all other cases, we'll send the lookup on the parent of the * given directory file handle. */ if (!nfs3_fh_is_root_fh (&cs->fh)) { parent = inode_ref (cs->resolvedloc.parent); nfs_loc_wipe (&cs->resolvedloc); ret = nfs_inode_loc_fill (parent, &cs->resolvedloc); if (ret < 0) goto errtostat; } ret = nfs_lookup (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_lookup_parentdir_cbk, cs); errtostat: if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "LOOKUP", stat, -ret); nfs3_lookup_reply (cs->req, stat, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } if (parent) inode_unref (parent); return ret; } int nfs3_lookup_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); cs->parent = cs->resolvefh; ret = nfs_lookup (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_lookup_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "LOOKUP", stat, -ret); nfs3_lookup_reply (cs->req, stat, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_lookup (rpcsvc_request_t *req, struct nfs3_fh *fh, int fhlen, char *name) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh) || (!name)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_fh_entry_call (nfs_rpcsvc_request_xid (req), "LOOKUP", fh, name); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); if (nfs3_solaris_zerolen_fh (fh, fhlen)) nfs3_funge_solaris_zerolen_fh (nfs3, fh, name, stat, nfs3err); else nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_strlen_or_goto (name, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); if (!nfs3_is_parentdir_entry (name)) ret = nfs3_fh_resolve_and_resume (cs, fh, name, nfs3_lookup_resume); else ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_lookup_parentdir_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "LOOKUP", stat, -ret); nfs3_lookup_reply (req, stat, NULL, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_lookup (rpcsvc_request_t *req) { char name[NFS_PATH_MAX]; struct nfs3_fh fh = {{0}, }; lookup3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_lookup3args (&args, &fh, name); if (xdr_to_lookup3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_lookup (req, &fh, args.what.dir.data.data_len, name); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "LOOKUP procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_access_reply (rpcsvc_request_t *req, nfsstat3 status, struct iatt *buf, uint32_t accbits) { access3res res; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_access3res (&res, status, buf, accbits, nfs_rpcsvc_request_uid (req), nfs_rpcsvc_request_gid (req) , xlid); nfs3svc_submit_reply (req, &res, (nfs3_serializer)xdr_serialize_access3res); return 0; } int32_t nfs3svc_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf) { nfsstat3 status = NFS3_OK; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) status = nfs3_errno_to_nfsstat3 (op_errno); nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "ACCESS", status, op_errno); nfs3_access_reply (cs->req, status, buf, cs->accessbits); nfs3_call_state_wipe (cs); return 0; } int nfs3_access_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); cs->fh = cs->resolvefh; nfs_request_user_init (&nfu, cs->req); ret = nfs_stat (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_access_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "ACCESS", stat, -ret); nfs3_access_reply (cs->req, stat, NULL, 0); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3_access (rpcsvc_request_t *req, struct nfs3_fh *fh, uint32_t accbits) { xlator_t *vol = NULL; struct nfs3_state *nfs3 = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh)) return -1; nfs3_log_common_call (nfs_rpcsvc_request_xid (req), "ACCESS", fh); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->accessbits = accbits; ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_access_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "ACCESS", stat, -ret); nfs3_access_reply (req, stat, NULL, 0); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3svc_access (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; access3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_access3args (&args, &fh); if (xdr_to_access3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_access (req, &fh, args.access); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "ACCESS procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_readlink_reply (rpcsvc_request_t *req, nfsstat3 stat, char *path, struct iatt *buf) { readlink3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_readlink3res (&res, stat, path, buf, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_readlink3res); return 0; } int32_t nfs3svc_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, const char *path, struct iatt *buf) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } stat = NFS3_OK; nfs3err: nfs3_log_readlink_res (nfs_rpcsvc_request_xid (cs->req), stat, op_errno, (char *)path); nfs3_readlink_reply (cs->req, stat, (char *)path, buf); nfs3_call_state_wipe (cs); return 0; } int nfs3_readlink_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; nfs_user_t nfu = {0, }; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); ret = nfs_readlink (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_readlink_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READLINK", stat, -ret); nfs3_readlink_reply (cs->req, stat, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_readlink (rpcsvc_request_t *req, struct nfs3_fh *fh) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_common_call (nfs_rpcsvc_request_xid (req), "READLINK", fh); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_readlink_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "READLINK", stat, -ret); nfs3_readlink_reply (req, stat, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_readlink (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; readlink3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_readlink3args (&args, &fh); if (xdr_to_readlink3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_readlink (req, &fh); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "READLINK procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_read_reply (rpcsvc_request_t *req, nfsstat3 stat, count3 count, struct iovec *vec, int vcount, struct iobref *iobref, struct iatt *poststat, int is_eof) { read3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_read3res (&res, stat, count, poststat, is_eof, xlid); if (stat == NFS3_OK) { nfs_xdr_vector_round_up (vec, vcount, count); /* iob can be zero if the file size was zero. If so, op_ret * would be 0 and count = 0. */ if (count != 0) { nfs3svc_submit_vector_reply (req, (void *)&res, (nfs3_serializer) xdr_serialize_read3res_nocopy, vec, vcount, iobref); } else nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer) xdr_serialize_read3res_nocopy); } else nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer) xdr_serialize_read3res_nocopy); return 0; } int32_t nfs3svc_read_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iovec *vector, int32_t count, struct iatt *stbuf, struct iobref *iobref) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int is_eof = 0; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto err; } else stat = NFS3_OK; if (op_errno == ENOENT) is_eof = 1; err: nfs3_log_read_res (nfs_rpcsvc_request_xid (cs->req), stat, op_errno, op_ret, is_eof, vector, count); nfs3_read_reply (cs->req, stat, op_ret, vector, count, iobref, stbuf, is_eof); nfs3_call_state_wipe (cs); return 0; } int nfs3_read_fd_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); ret = nfs_read (cs->nfsx, cs->vol, &nfu, cs->fd, cs->datacount, cs->dataoffset, nfs3svc_read_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READ", stat, -ret); nfs3_read_reply (cs->req, stat, 0, NULL, 0, NULL, NULL, 0); nfs3_call_state_wipe (cs); } return ret; } int nfs3_read_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); ret = nfs3_file_open_and_resume (cs, nfs3_read_fd_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READ", stat, -ret); nfs3_read_reply (cs->req, stat, 0, NULL,0, NULL, NULL, 0); nfs3_call_state_wipe (cs); } return ret; } int nfs3_read (rpcsvc_request_t *req, struct nfs3_fh *fh, offset3 offset, count3 count) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_rw_call (nfs_rpcsvc_request_xid (req), "READ", fh, offset, count, -1); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->datacount = count; cs->dataoffset = offset; ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_read_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "READ", stat, -ret); nfs3_read_reply (req, stat, 0, NULL,0, NULL, NULL, 0); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3svc_read (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; read3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_read3args (&args, &fh); if (xdr_to_read3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_read (req, &fh, args.offset, args.count); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "READ procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_write_reply (rpcsvc_request_t *req, nfsstat3 stat, count3 count, stable_how stable, uint64_t wverf, struct iatt *prestat, struct iatt *poststat) { write3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_write3res (&res, stat, count, stable, wverf, prestat, poststat, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_write3res); return 0; } int32_t nfs3svc_write_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf) { struct nfs3_state *nfs3 = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; nfs3 = nfs_rpcsvc_request_program_private (cs->req); if (op_ret == -1) stat = nfs3_errno_to_nfsstat3 (op_errno); else stat = NFS3_OK; nfs3_log_write_res (nfs_rpcsvc_request_xid (cs->req), stat, op_errno, cs->maxcount, cs->writetype, nfs3->serverstart); nfs3_write_reply (cs->req, stat, cs->maxcount, cs->writetype, nfs3->serverstart, &cs->stbuf, postbuf); nfs3_call_state_wipe (cs); return 0; } /* * If this logic determines that the write should return a reply to the client * after this function, the return value is -1 and the writetype is reset to * the type of write we want to signify to the client. * * In case the write should continue to serve the request according to the type * of stable write, a 0 is returned and writetype is left as it is. */ int nfs3_write_how (int *writetype, int write_trusted, int sync_trusted) { int ret = -1; if (*writetype == UNSTABLE) { /* On an UNSTABLE write, only return STABLE when trusted-write * is set. TW is also set when trusted-sync is set. */ if (write_trusted) *writetype = FILE_SYNC; goto err; } else if ((*writetype == DATA_SYNC) || (*writetype == FILE_SYNC)) { /* On a STABLE write, if sync-trusted is on, only then, return * without syncing. */ if (sync_trusted) goto err; } ret = 0; err: return ret; } /* * Before going into the write reply logic, here is a matrix that shows the * requirements for a write reply as given by RFC1813. * * Requested Write Type || Possible Returns * ============================================== * FILE_SYNC || FILE_SYNC * DATA_SYNC || DATA_SYNC or FILE_SYNC * UNSTABLE || DATA_SYNC or FILE_SYNC or UNSTABLE * * Write types other than UNSTABLE are together called STABLE. * RS - Return Stable * RU - Return Unstable * WS - Write Stable * WU - Write Unstable * *+============================================+ *| Vol Opts -> || trusted-write| trusted-sync | *| Write Type || | | *|-------------||--------------|--------------| *| STABLE || WS | WU | *| || RS | RS | *|-------------||--------------|--------------| *| UNSTABLE || WU | WU | *| || RS | RS | *|-------------||--------------|--------------| *| COMMIT || fsync | getattr | *+============================================+ * * */ int32_t nfs3svc_write_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; struct nfs3_state *nfs3 = NULL; int write_trusted = 0; int sync_trusted = 0; cs = frame->local; nfs3 = nfs_rpcsvc_request_program_private (cs->req); if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto err; } stat = NFS3_OK; cs->maxcount = op_ret; write_trusted = nfs3_export_write_trusted (cs->nfs3state, cs->resolvefh.xlatorid); sync_trusted = nfs3_export_sync_trusted (cs->nfs3state, cs->resolvefh.xlatorid); ret = nfs3_write_how (&cs->writetype, write_trusted, sync_trusted); if (ret == -1) goto err; nfs_request_user_init (&nfu, cs->req); /* Store the current preattr so that this can be used as the pre attr * when fsync returns. We dont want to use the preattr in fsync because * the write fop happened before the fsync. */ cs->stbuf = *prebuf; ret = nfs_fsync (cs->nfsx, cs->vol, &nfu, cs->fd, 0, nfs3svc_write_fsync_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); err: if (ret < 0) { nfs3_log_write_res (nfs_rpcsvc_request_xid (cs->req), stat, op_errno, cs->maxcount, cs->writetype, nfs3->serverstart); nfs3_write_reply (cs->req, stat, cs->maxcount, cs->writetype, nfs3->serverstart, prebuf, postbuf); nfs3_call_state_wipe (cs); } return 0; } int __nfs3_write_resume (nfs3_call_state_t *cs) { int ret = -EFAULT; nfs_user_t nfu = {0, }; if (!cs) return ret; nfs_request_user_init (&nfu, cs->req); /* It is possible that the RPC record contains more bytes than * than the size of write requested in this request. This means, * that in the RPC message buffer, there could be more bytes * beyind the @count bytes. Since @payload is referring to the write * data directly inside the RPC request buffer(..since we performed a * no-copy deXDRing..), we might end up writing more data than * requested, because till now payload.iov_len accounts for all the * bytes not just the write request bytes. These extra bytes are present * as a requirement of the XDR encoding to round up the all string and * opaque data buffers to multiples of 4 bytes. */ cs->datavec.iov_len = cs->datacount; ret = nfs_write (cs->nfsx, cs->vol, &nfu, cs->fd, cs->iob, &cs->datavec, 1, cs->dataoffset, nfs3svc_write_cbk, cs); return ret; } int nfs3_write_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); ret = __nfs3_write_resume (cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "WRITE", stat, -ret); nfs3_write_reply (cs->req, stat, 0, cs->writetype, 0, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_write_open_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); ret = nfs3_file_open_and_resume (cs, nfs3_write_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "WRITE", stat, -ret); nfs3_write_reply (cs->req, stat, 0, cs->writetype, 0, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_write (rpcsvc_request_t *req, struct nfs3_fh *fh, offset3 offset, count3 count, stable_how stable, struct iovec payload, struct iobuf *iob) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh) || (!payload.iov_base)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_rw_call (nfs_rpcsvc_request_xid (req), "WRITE", fh, offset, count, stable); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, fh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->datacount = count; cs->dataoffset = offset; cs->writetype = stable; cs->iob = iob; cs->datavec = payload; ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_write_open_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "WRITE", stat, -ret); nfs3_write_reply (req, stat, 0, stable, 0, NULL, NULL); nfs3_call_state_wipe (cs); ret = 0; } return ret; } #define NFS3_VECWRITE_READFHLEN 1 #define NFS3_VECWRITE_READFH 2 #define NFS3_VECWRITE_READREST 3 #define NFS3_WRITE_POSTFH_SIZE 20 int nfs3svc_write_vecsizer (rpcsvc_request_t *req, ssize_t *readsize, int *newbuf) { ssize_t ret = RPCSVC_ACTOR_ERROR; int state = 0; uint32_t fhlen = 0; uint32_t fhlen_n = 0; write3args *args = NULL; if (!req) return ret; state = (long)nfs_rpcsvc_request_private (req); *newbuf = 0; if (state == 0) { nfs_rpcsvc_request_set_private (req, NFS3_VECWRITE_READFHLEN); *readsize = 4; ret = 0; } else if (state == NFS3_VECWRITE_READFHLEN) { fhlen_n = *(uint32_t *)req->msg.iov_base; fhlen = ntohl (fhlen_n); *readsize = nfs_xdr_length_round_up (fhlen, NFS3_FHSIZE); nfs_rpcsvc_request_set_private (req, NFS3_VECWRITE_READFH); ret = 0; } else if (state == NFS3_VECWRITE_READFH) { *readsize = NFS3_WRITE_POSTFH_SIZE; nfs_rpcsvc_request_set_private (req, NFS3_VECWRITE_READREST); ret = 0; } else if (state == NFS3_VECWRITE_READREST) { args = GF_CALLOC (1, sizeof (*args), gf_nfs_mt_write3args); if (!args) goto rpcerr; if (xdr_to_write3args_nocopy (req->msg, args, NULL) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } nfs_rpcsvc_request_set_private (req, args); ret = nfs_xdr_length_round_up (args->data.data_len, 1048576); *readsize = ret; *newbuf = 1; ret = 0; } rpcerr: return ret; } int nfs3svc_write_vec (rpcsvc_request_t *req, struct iobuf *iob) { write3args *args = NULL; int ret = RPCSVC_ACTOR_ERROR; struct iovec payload = {0, }; if ((!req) || (!iob)) return ret; args = nfs_rpcsvc_request_private (req); iobuf_to_iovec (iob, &payload); iobuf_ref (iob); ret = nfs3_write (req, (struct nfs3_fh *)args->file.data.data_val, args->offset, args->count, args->stable, payload,iob); xdr_free_write3args_nocopy (args); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "WRITE procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } return ret; } int nfs3svc_write (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; write3args args; int ret = RPCSVC_ACTOR_ERROR; struct iovec payload = {0, }; if (!req) return ret; nfs3_prep_write3args (&args, &fh); if (xdr_to_write3args_nocopy (req->msg, &args, &payload) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } /* To ensure that the iobuf for the current record does not * get returned to the iobpool, we need to keep a reference for * ourselves because the RPC call handler who called us will unref its * own ref of the record's iobuf when it is done handling the request. */ nfs_rpcsvc_request_record_ref (req); ret = nfs3_write (req, &fh, args.offset, args.count, args.stable, payload, nfs_rpcsvc_request_record_iob (req)); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "WRITE procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_create_reply (rpcsvc_request_t *req, nfsstat3 stat, struct nfs3_fh *newfh, struct iatt *newbuf, struct iatt *preparent, struct iatt *postparent) { create3res res = {0, }; nfs3_fill_create3res (&res, stat, newfh, newbuf, preparent, postparent); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_create3res); return 0; } int32_t nfs3svc_create_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preop, struct iatt *postop) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } stat = NFS3_OK; nfs3err: nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "CREATE", stat, op_errno, &cs->fh); nfs3_create_reply (cs->req, stat, &cs->fh, postop, &cs->preparent, &cs->postparent); nfs3_call_state_wipe (cs); return 0; } int32_t nfs3svc_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } nfs3_fh_build_child_fh (&cs->parent, buf, &cs->fh); /* Means no attributes were required to be set. */ if (!cs->setattr_valid) { stat = NFS3_OK; ret = -1; goto nfs3err; } cs->preparent = *preparent; cs->postparent = *postparent; nfs_request_user_init (&nfu, cs->req); ret = nfs_setattr (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc,&cs->stbuf, cs->setattr_valid, nfs3svc_create_setattr_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "CREATE", stat, op_errno, &cs->fh); nfs3_create_reply (cs->req, stat, &cs->fh, buf, preparent, postparent); nfs3_call_state_wipe (cs); } return 0; } int nfs3_create_common (nfs3_call_state_t *cs) { int ret = -EFAULT; int flags = 0; nfs_user_t nfu = {0, }; if (!cs) return ret; if ((cs->createmode == UNCHECKED) || (cs->createmode = EXCLUSIVE)) flags = O_RDWR; else if (cs->createmode == GUARDED) flags = (O_RDWR | O_EXCL); nfs_request_user_init (&nfu, cs->req); /* We can avoid sending the setattr call later if only the mode is * required to be set. This is possible because the create fop allows * us to specify a mode arg. */ if (cs->setattr_valid & GF_SET_ATTR_MODE) { cs->setattr_valid &= ~GF_SET_ATTR_MODE; ret = nfs_create (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, flags, cs->mode, nfs3svc_create_cbk, cs); } else ret = nfs_create (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, flags, NFS_DEFAULT_CREATE_MODE, nfs3svc_create_cbk, cs); return ret; } int32_t nfs3svc_create_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf) { int ret = -EFAULT; nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; cs = frame->local; nfs_request_user_init (&nfu, cs->req); if (op_ret == -1) { ret = -op_errno; stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } if (cs->preparent.ia_mtime == buf->ia_mtime) stat = NFS3_OK; else stat = NFS3ERR_EXIST; nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "CREATE", stat, op_errno); nfs3_create_reply (cs->req, stat, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return 0; } int nfs3_create_exclusive (nfs3_call_state_t *cs) { int ret = -EFAULT; nfs_user_t nfu = {0, }; if (!cs) return ret; /* HACK warning. */ cs->preparent.ia_mtime = cs->cookieverf; cs->preparent.ia_atime = 9669; nfs_request_user_init (&nfu, cs->req); /* If the file already existed we need to get that attributes so we can * compare and check whether a previous create operation was * interrupted due to server failure or dropped packets. */ if ((cs->resolve_ret == 0) || ((cs->resolve_ret == -1) && (cs->resolve_errno != ENOENT))) { ret = nfs_stat (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_create_stat_cbk, cs); goto nfs3err; } if (cs->setattr_valid & GF_SET_ATTR_MODE) { cs->setattr_valid &= ~GF_SET_ATTR_MODE; ret = nfs_create (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, O_RDWR, cs->mode, nfs3svc_create_cbk, cs); } else ret = nfs_create (cs->nfsx, cs->vol, &nfu, &cs->oploc, O_RDWR, NFS_DEFAULT_CREATE_MODE, nfs3svc_create_cbk, cs); nfs3err: return ret; } int nfs3_create_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_new_fh_resolve_status (cs, stat, nfs3err); if (cs->createmode == EXCLUSIVE) ret = nfs3_create_exclusive (cs); else ret = nfs3_create_common (cs); /* Handle a failure return from either of the create functions above. */ if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "CREATE", stat, -ret); nfs3_create_reply (cs->req, stat, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_create (rpcsvc_request_t *req, struct nfs3_fh *dirfh, char *name, createmode3 mode, sattr3 *sattr, uint64_t cverf) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!dirfh) || (!name) || (!sattr)) return -1; nfs3_log_create_call (nfs_rpcsvc_request_xid (req), dirfh, name, mode); nfs3_validate_gluster_fh (dirfh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_validate_strlen_or_goto (name, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, dirfh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, dirfh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->cookieverf = cverf; cs->setattr_valid = nfs3_sattr3_to_setattr_valid (sattr, NULL, &cs->mode); cs->createmode = mode; cs->parent = *dirfh; ret = nfs3_fh_resolve_and_resume (cs, dirfh, name, nfs3_create_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "CREATE", stat, -ret); nfs3_create_reply (req, stat, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3svc_create (rpcsvc_request_t *req) { char name[NFS_PATH_MAX]; struct nfs3_fh dirfh = {{0}, }; create3args args; int ret = RPCSVC_ACTOR_ERROR; uint64_t cverf = 0; if (!req) return ret; nfs3_prep_create3args (&args, &dirfh, name); if (xdr_to_create3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } cverf = *(uint64_t *)args.how.createhow3_u.verf; ret = nfs3_create (req, &dirfh, name, args.how.mode, &args.how.createhow3_u.obj_attributes, cverf); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "CREATE procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_mkdir_reply (rpcsvc_request_t *req, nfsstat3 stat, struct nfs3_fh *fh, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { mkdir3res res = {0, }; nfs3_fill_mkdir3res (&res, stat, fh, buf, preparent, postparent); nfs3svc_submit_reply (req, &res, (nfs3_serializer)xdr_serialize_mkdir3res); return 0; } int32_t nfs3svc_mkdir_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preop, struct iatt *postop) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } stat = NFS3_OK; nfs3err: nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "MKDIR", stat, op_errno, &cs->fh); nfs3_mkdir_reply (cs->req, stat, &cs->fh, postop, &cs->preparent, &cs->postparent); nfs3_call_state_wipe (cs); return 0; } int32_t nfs3svc_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } nfs3_fh_build_child_fh (&cs->parent, buf, &cs->fh); /* Means no attributes were required to be set. */ if (!cs->setattr_valid) { stat = NFS3_OK; goto nfs3err; } cs->preparent = *preparent; cs->postparent = *postparent; nfs_request_user_init (&nfu, cs->req); ret = nfs_setattr (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc,&cs->stbuf, cs->setattr_valid, nfs3svc_mkdir_setattr_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "MKDIR", stat, op_errno, &cs->fh); nfs3_mkdir_reply (cs->req, stat, &cs->fh, buf, preparent, postparent); nfs3_call_state_wipe (cs); } return 0; } int nfs3_mkdir_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_new_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); if (gf_attr_mode_set (cs->setattr_valid)) { cs->setattr_valid &= ~GF_SET_ATTR_MODE; ret = nfs_mkdir (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, cs->mode, nfs3svc_mkdir_cbk, cs); } else ret = nfs_mkdir (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, cs->mode, nfs3svc_mkdir_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "MKDIR", stat, -ret); nfs3_mkdir_reply (cs->req, stat, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return 0; } int nfs3_mkdir (rpcsvc_request_t *req, struct nfs3_fh *dirfh, char *name, sattr3 *sattr) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!dirfh) || (!name) || (!sattr)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_fh_entry_call (nfs_rpcsvc_request_xid (req), "MKDIR", dirfh, name); nfs3_validate_gluster_fh (dirfh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_validate_strlen_or_goto (name, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, dirfh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, dirfh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->parent = *dirfh; cs->setattr_valid = nfs3_sattr3_to_setattr_valid (sattr, NULL, &cs->mode); ret = nfs3_fh_resolve_and_resume (cs, dirfh, name, nfs3_mkdir_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "MKDIR", stat, -ret); nfs3_mkdir_reply (req, stat, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3svc_mkdir (rpcsvc_request_t *req) { char name[NFS_PATH_MAX]; struct nfs3_fh dirfh = {{0}, }; mkdir3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_mkdir3args (&args, &dirfh, name); if (xdr_to_mkdir3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_mkdir (req, &dirfh, name, &args.attributes); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "MKDIR procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_symlink_reply (rpcsvc_request_t *req, nfsstat3 stat, struct nfs3_fh *fh, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { symlink3res res = {0, }; nfs3_fill_symlink3res (&res, stat, fh, buf, preparent, postparent); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_symlink3res); return 0; } int32_t nfs3svc_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } nfs3_fh_build_child_fh (&cs->parent, buf, &cs->fh); stat = NFS3_OK; nfs3err: nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "SYMLINK", stat, op_errno, &cs->fh); nfs3_symlink_reply (cs->req, stat, &cs->fh, buf, preparent, postparent); nfs3_call_state_wipe (cs); return 0; } int nfs3_symlink_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; nfs_user_t nfu = {0, }; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_new_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); ret = nfs_symlink (cs->nfsx, cs->vol, &nfu, cs->pathname, &cs->resolvedloc, nfs3svc_symlink_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "SYMLINK", stat, -ret); nfs3_symlink_reply (cs->req, stat, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_symlink (rpcsvc_request_t *req, struct nfs3_fh *dirfh, char *name, char *target, sattr3 *sattr) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!dirfh) || (!name) || (!target) || (!sattr)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_symlink_call (nfs_rpcsvc_request_xid (req), dirfh, name, target); nfs3_validate_gluster_fh (dirfh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_validate_strlen_or_goto (name, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, dirfh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, dirfh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->parent = *dirfh; cs->pathname = gf_strdup (target); if (!cs->pathname) { ret = -1; stat = NFS3ERR_SERVERFAULT; goto nfs3err; } ret = nfs3_fh_resolve_and_resume (cs, dirfh, name, nfs3_symlink_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "SYMLINK", stat, -ret); nfs3_symlink_reply (req, stat, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_symlink (rpcsvc_request_t *req) { char name[NFS_PATH_MAX]; struct nfs3_fh dirfh = {{0}, }; char target[NFS_PATH_MAX]; symlink3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_symlink3args (&args, &dirfh, name, target); if (xdr_to_symlink3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_symlink (req, &dirfh, name, target, &args.symlink.symlink_attributes); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "SYMLINK procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_mknod_reply (rpcsvc_request_t *req, nfsstat3 stat, struct nfs3_fh *fh, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { mknod3res res = {0, }; nfs3_fill_mknod3res (&res, stat, fh, buf, preparent, postparent); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_mknod3res); return 0; } int32_t nfs3svc_mknod_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preop, struct iatt *postop) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } stat = NFS3_OK; nfs3err: nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "MKNOD", stat, op_errno, &cs->fh); nfs3_mknod_reply (cs->req, stat, &cs->fh, postop, &cs->preparent, &cs->postparent); nfs3_call_state_wipe (cs); return 0; } int32_t nfs3svc_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -1; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } nfs3_fh_build_child_fh (&cs->parent, buf, &cs->fh); /* Means no attributes were required to be set. */ if (!cs->setattr_valid) { stat = NFS3_OK; ret = -1; goto nfs3err; } cs->preparent = *preparent; cs->postparent = *postparent; nfs_request_user_init (&nfu, cs->req); ret = nfs_setattr (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc,&cs->stbuf, cs->setattr_valid, nfs3svc_mknod_setattr_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_newfh_res (nfs_rpcsvc_request_xid (cs->req), "MKNOD", stat, op_errno, &cs->fh); nfs3_mknod_reply (cs->req, stat, &cs->fh, buf, preparent, postparent); nfs3_call_state_wipe (cs); } return 0; } int nfs3_mknod_device (nfs3_call_state_t *cs) { int ret = -EFAULT; dev_t devnum = 0; mode_t mode = 0; nfs_user_t nfu = {0, }; if (!cs) return ret; devnum = makedev (cs->devnums.specdata1, cs->devnums.specdata2); if (cs->mknodtype == NF3CHR) mode = S_IFCHR; else mode = S_IFBLK; nfs_request_user_init (&nfu, cs->req); if (gf_attr_mode_set (cs->setattr_valid)) { cs->setattr_valid &= ~GF_SET_ATTR_MODE; mode |= cs->mode; ret = nfs_mknod (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, mode, devnum, nfs3svc_mknod_cbk, cs); } else ret = nfs_mknod (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, mode, devnum, nfs3svc_mknod_cbk, cs); return ret; } int nfs3_mknod_fifo (nfs3_call_state_t *cs) { int ret = -EFAULT; nfs_user_t nfu = {0, }; mode_t mode = S_IFIFO; if (!cs) return ret; nfs_request_user_init (&nfu, cs->req); if (gf_attr_mode_set (cs->setattr_valid)) { cs->setattr_valid &= ~GF_SET_ATTR_MODE; mode |= cs->mode; ret = nfs_mknod (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, mode, 0, nfs3svc_mknod_cbk, cs); } else ret = nfs_mknod (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, mode, 0, nfs3svc_mknod_cbk, cs); return ret; } int nfs3_mknod_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_new_fh_resolve_status (cs, stat, nfs3err); switch (cs->mknodtype) { case NF3CHR: case NF3BLK: ret = nfs3_mknod_device (cs); break; case NF3SOCK: case NF3FIFO: ret = nfs3_mknod_fifo (cs); break; default: ret = -EBADF; break; } if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "MKNOD", stat, -ret); nfs3_mknod_reply (cs->req, stat, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_mknod (rpcsvc_request_t *req, struct nfs3_fh *fh, char *name, mknoddata3 *nodedata) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; sattr3 *sattr = NULL; if ((!req) || (!fh) || (!name) || (!nodedata)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_mknod_call (nfs_rpcsvc_request_xid (req), fh, name, nodedata->type); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_validate_strlen_or_goto (name, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, fh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->mknodtype = nodedata->type; switch (nodedata->type) { case NF3CHR: case NF3BLK: cs->devnums = nodedata->mknoddata3_u.device.spec; sattr = &nodedata->mknoddata3_u.device.dev_attributes; cs->setattr_valid = nfs3_sattr3_to_setattr_valid (sattr, NULL, &cs->mode); break; case NF3SOCK: case NF3FIFO: sattr = &nodedata->mknoddata3_u.pipe_attributes; cs->setattr_valid = nfs3_sattr3_to_setattr_valid (sattr, NULL, &cs->mode); break; default: ret = -EBADF; break; } cs->parent = *fh; ret = nfs3_fh_resolve_and_resume (cs, fh, name, nfs3_mknod_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "MKNOD", stat, -ret); nfs3_mknod_reply (req, stat, NULL, NULL, NULL, NULL); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3svc_mknod (rpcsvc_request_t *req) { char name[NFS_PATH_MAX]; struct nfs3_fh fh = {{0}, }; mknod3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_mknod3args (&args, &fh, name); if (xdr_to_mknod3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_mknod (req, &fh, name, &args.what); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "MKNOD procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_remove_reply (rpcsvc_request_t *req, nfsstat3 stat, struct iatt *preparent , struct iatt *postparent) { remove3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_remove3res (&res, stat, preparent, postparent, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_remove3res); return 0; } int32_t nfs3svc_remove_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preparent, struct iatt *postparent) { nfsstat3 stat = NFS3ERR_SERVERFAULT; fd_t *openfd = NULL; nfs3_call_state_t *cs = NULL; struct nfs3_state *nfs3 = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto do_not_unref_cached_fd; } stat = NFS3_OK; /* Close any cached fds so that when any currently active write * finishes, the file is finally removed. */ openfd = fd_lookup (cs->resolvedloc.inode, 0); nfs3 = nfs_rpcsvc_request_program_private (cs->req); if (openfd) { fd_unref (openfd); nfs3_fdcache_remove (nfs3, openfd); } /* This is the unref equivalent of the ref done when the inode was * created on a lookup or a create request. * The inode is finally unrefed in call state wipe. */ inode_unref (cs->resolvedloc.inode); do_not_unref_cached_fd: nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "REMOVE", stat, op_errno); nfs3_remove_reply (cs->req, stat, preparent, postparent); nfs3_call_state_wipe (cs); return 0; } int __nfs3_remove (nfs3_call_state_t *cs) { int ret = -EFAULT; nfs_user_t nfu = {0, }; ia_type_t type = 0; if (!cs) return ret; type = cs->resolvedloc.inode->ia_type; nfs_request_user_init (&nfu, cs->req); if (IA_ISDIR (type)) ret = nfs_rmdir (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_remove_cbk, cs); else ret = nfs_unlink (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_remove_cbk, cs); return ret; } int nfs3_remove_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); ret = __nfs3_remove (cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "REMOVE", stat, -ret); nfs3_remove_reply (cs->req, stat, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_remove (rpcsvc_request_t *req, struct nfs3_fh *fh, char *name) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh) || (!name)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_fh_entry_call (nfs_rpcsvc_request_xid (req), "REMOVE", fh, name); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_validate_strlen_or_goto (name, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, fh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); ret = nfs3_fh_resolve_and_resume (cs, fh, name, nfs3_remove_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "REMOVE", stat, -ret); nfs3_remove_reply (req, stat, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_remove (rpcsvc_request_t *req) { char name[NFS_PATH_MAX]; struct nfs3_fh fh = {{0}, }; remove3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_remove3args (&args, &fh, name); if (xdr_to_remove3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_remove (req, &fh, name); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "REMOVE procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_rmdir_reply (rpcsvc_request_t *req, nfsstat3 stat, struct iatt *preparent, struct iatt *postparent) { rmdir3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_rmdir3res (&res, stat, preparent, postparent, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_rmdir3res); return 0; } int32_t nfs3svc_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preparent, struct iatt *postparent) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) stat = nfs3_errno_to_nfsstat3 (op_errno); else { inode_unref (cs->resolvedloc.inode); stat = NFS3_OK; } nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "RMDIR", stat, op_errno); nfs3_rmdir_reply (cs->req, stat, preparent, postparent); nfs3_call_state_wipe (cs); return 0; } int nfs3_rmdir_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; nfs_user_t nfu = {0, }; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); ret = nfs_rmdir (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_rmdir_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "RMDIR", stat, -ret); nfs3_rmdir_reply (cs->req, stat, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_rmdir (rpcsvc_request_t *req, struct nfs3_fh *fh, char *name) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh) || (!name)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_fh_entry_call (nfs_rpcsvc_request_xid (req), "RMDIR", fh, name); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_validate_strlen_or_goto (name, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, fh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); ret = nfs3_fh_resolve_and_resume (cs, fh, name, nfs3_rmdir_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "RMDIR", stat, -ret); nfs3_rmdir_reply (req, stat, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_rmdir (rpcsvc_request_t *req) { char name[NFS_PATH_MAX]; struct nfs3_fh fh = {{0}, }; rmdir3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_rmdir3args (&args, &fh, name); if (xdr_to_rmdir3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_rmdir (req, &fh, name); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "RMDIR procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_rename_reply (rpcsvc_request_t *req, nfsstat3 stat, struct iatt *buf, struct iatt *preoldparent, struct iatt *postoldparent, struct iatt *prenewparent, struct iatt *postnewparent) { rename3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_rename3res (&res, stat, buf, preoldparent, postoldparent, prenewparent, postnewparent, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer) xdr_serialize_rename3res); return 0; } int32_t nfs3svc_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf, struct iatt *preoldparent, struct iatt *postoldparent, struct iatt *prenewparent, struct iatt *postnewparent) { int ret = -EFAULT; nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; fd_t *openfd = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } stat = NFS3_OK; /* Close any cached fds so that when any currently active writes to the * dst finish, the file is finally removed. * * This logic works because we know resolvedloc contains dst, which is * resolved after src */ openfd = fd_lookup (cs->resolvedloc.inode, 0); if (openfd) { fd_unref (openfd); nfs3_fdcache_remove (cs->nfs3state, openfd); } /* This is the unref equivalent of the ref done when the inode was * created on a lookup or a create request. * The inode is finally unrefed in call state wipe. */ inode_unref (cs->resolvedloc.inode); nfs3err: nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "RENAME", stat, -ret); nfs3_rename_reply (cs->req, stat, buf, preoldparent, postoldparent, prenewparent, postnewparent); nfs3_call_state_wipe (cs); return 0; } int nfs3_rename_resume_dst (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; nfs_user_t nfu = {0, }; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_new_fh_resolve_status (cs, stat, nfs3err); cs->parent = cs->resolvefh; nfs_request_user_init (&nfu, cs->req); ret = nfs_rename (cs->nfsx, cs->vol, &nfu, &cs->oploc, &cs->resolvedloc, nfs3svc_rename_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "RENAME", stat, -ret); nfs3_rename_reply (cs->req, stat, NULL, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_rename_resume_src (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); /* Copy the resolved loc for the source file into another loc * for safekeeping till we resolve the dest loc. */ nfs_loc_copy (&cs->oploc, &cs->resolvedloc); nfs_loc_wipe (&cs->resolvedloc); ret = nfs3_fh_resolve_and_resume (cs, &cs->fh, cs->pathname, nfs3_rename_resume_dst); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "RENAME", stat, -ret); nfs3_rename_reply (cs->req, stat, NULL, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_rename (rpcsvc_request_t *req, struct nfs3_fh *olddirfh, char *oldname, struct nfs3_fh *newdirfh, char *newname) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!olddirfh) || (!oldname) || (!newdirfh) || (!newname)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_rename_call (nfs_rpcsvc_request_xid (req), olddirfh, oldname, newdirfh, newname); nfs3_validate_gluster_fh (olddirfh, stat, nfs3err); nfs3_validate_gluster_fh (newdirfh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_validate_strlen_or_goto(oldname, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_validate_strlen_or_goto(newname, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, olddirfh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, olddirfh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); /* While we resolve the source (fh, name) pair, we need to keep a copy * of the dest (fh,name) pair. */ cs->fh = *newdirfh; cs->pathname = gf_strdup (newname); if (!cs->pathname) { stat = NFS3ERR_SERVERFAULT; ret = -1; goto nfs3err; } ret = nfs3_fh_resolve_and_resume (cs, olddirfh, oldname, nfs3_rename_resume_src); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "RENAME", stat, -ret); nfs3_rename_reply (req, stat, NULL, NULL, NULL, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_rename (rpcsvc_request_t *req) { char newname[NFS_PATH_MAX]; char oldname[NFS_PATH_MAX]; struct nfs3_fh olddirfh = {{0}, }; struct nfs3_fh newdirfh = {{0}, }; rename3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_rename3args (&args, &olddirfh, oldname, &newdirfh, newname); if (xdr_to_rename3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_rename (req, &olddirfh, oldname, &newdirfh, newname); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "RENAME procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_link_reply (rpcsvc_request_t *req, nfsstat3 stat, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { link3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_link3res (&res, stat, buf, preparent, postparent, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_link3res); return 0; } int32_t nfs3svc_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) stat = nfs3_errno_to_nfsstat3 (op_errno); else stat = NFS3_OK; nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "LINK", stat, op_errno); nfs3_link_reply (cs->req, stat, buf, preparent, postparent); nfs3_call_state_wipe (cs); return 0; } int nfs3_link_resume_lnk (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; nfs_user_t nfu = {0, }; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_new_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); ret = nfs_link (cs->nfsx, cs->vol, &nfu, &cs->oploc, &cs->resolvedloc, nfs3svc_link_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "LINK", stat, -ret); nfs3_link_reply (cs->req, stat, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_link_resume_tgt (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_loc_copy (&cs->oploc, &cs->resolvedloc); nfs_loc_wipe (&cs->resolvedloc); ret = nfs3_fh_resolve_and_resume (cs, &cs->fh, cs->pathname, nfs3_link_resume_lnk); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "LINK", stat, -ret); nfs3_link_reply (cs->req, stat, NULL, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_link (rpcsvc_request_t *req, struct nfs3_fh *targetfh, struct nfs3_fh *dirfh, char *newname) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!targetfh) || (!dirfh) || (!newname)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_validate_gluster_fh (dirfh, stat, nfs3err); nfs3_validate_gluster_fh (targetfh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_validate_strlen_or_goto(newname, NFS_NAME_MAX, nfs3err, stat, ret); nfs3_map_fh_to_volume (nfs3, dirfh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, dirfh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->fh = *dirfh; cs->pathname = gf_strdup (newname); if (!cs->pathname) { stat = NFS3ERR_SERVERFAULT; ret = -1; goto nfs3err; } ret = nfs3_fh_resolve_and_resume (cs, targetfh, NULL, nfs3_link_resume_tgt); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "LINK", stat, -ret); nfs3_link_reply (req, stat, NULL, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_link (rpcsvc_request_t *req) { char newpath[NFS_PATH_MAX]; struct nfs3_fh dirfh = {{0}, }; struct nfs3_fh targetfh = {{0}, }; link3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_link3args (&args, &targetfh, &dirfh, newpath); if (xdr_to_link3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_link (req, &targetfh, &dirfh, newpath); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "LINK procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_readdirp_reply (rpcsvc_request_t *req, nfsstat3 stat,struct nfs3_fh *dirfh, uint64_t cverf, struct iatt *dirstat, gf_dirent_t *entries, count3 dircount, count3 maxcount, int is_eof) { readdirp3res res = {0, }; nfs3_fill_readdirp3res (&res, stat, dirfh, cverf, dirstat, entries, dircount, maxcount, is_eof); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer) xdr_serialize_readdirp3res); nfs3_free_readdirp3res (&res); return 0; } int nfs3_readdir_reply (rpcsvc_request_t *req, nfsstat3 stat, struct nfs3_fh *dirfh, uint64_t cverf, struct iatt *dirstat, gf_dirent_t *entries, count3 count, int is_eof) { readdir3res res = {0, }; nfs3_fill_readdir3res (&res, stat, dirfh, cverf, dirstat, entries, count , is_eof); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer) xdr_serialize_readdir3res); nfs3_free_readdir3res (&res); return 0; } int32_t nfs3svc_readdir_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int is_eof = 0; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto nfs3err; } /* Check whether we encountered a end of directory stream while * readdir'ing. */ if (cs->operrno == ENOENT) { gf_log (GF_NFS3, GF_LOG_TRACE, "Reached end-of-directory"); is_eof = 1; } stat = NFS3_OK; nfs3err: if (cs->maxcount == 0) { nfs3_log_readdir_res (nfs_rpcsvc_request_xid (cs->req), stat, op_errno, (uintptr_t)cs->fd, cs->dircount, is_eof); nfs3_readdir_reply (cs->req, stat, &cs->parent, (uintptr_t)cs->fd, buf, &cs->entries, cs->dircount, is_eof); } else { nfs3_log_readdirp_res (nfs_rpcsvc_request_xid (cs->req), stat, op_errno, (uintptr_t)cs->fd, cs->dircount, cs->maxcount, is_eof); nfs3_readdirp_reply (cs->req, stat, &cs->parent, (uintptr_t)cs->fd, buf, &cs->entries, cs->dircount, cs->maxcount, is_eof); } if (is_eof) { gf_log (GF_NFS3, GF_LOG_TRACE, "EOF REF: %d", cs->fd->refcount); fd_unref (cs->fd); } gf_log (GF_NFS3, GF_LOG_TRACE, "CS WIPE REF: %d", cs->fd->refcount); nfs3_call_state_wipe (cs); return 0; } int32_t nfs3svc_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, gf_dirent_t *entries) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { stat = nfs3_errno_to_nfsstat3 (op_errno); goto err; } cs->operrno = op_errno; list_splice_init (&entries->list, &cs->entries.list); nfs_request_user_init (&nfu, cs->req); ret = nfs_fstat (cs->nfsx, cs->vol, &nfu, cs->fd, nfs3svc_readdir_fstat_cbk, cs); if (ret < 0) { op_ret = -1; stat = nfs3_errno_to_nfsstat3 (-ret); op_errno = -ret; } err: if (op_ret >= 0) goto ret; if (cs->maxcount == 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READDIR", stat, op_errno); nfs3_readdir_reply (cs->req, stat, NULL, 0, NULL, NULL, 0, 0); } else { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READDIRP", stat, op_errno); nfs3_readdirp_reply (cs->req, stat, NULL, 0, NULL, NULL, 0, 0, 0); } /* For directories, we force a purge from the fd cache on close * so that next time the dir is read, we'll get any changed directory * entries. */ fd_unref (cs->fd); nfs3_call_state_wipe (cs); ret: return 0; } int nfs3_readdir_process (nfs3_call_state_t *cs) { int ret = -EFAULT; nfs_user_t nfu = {0, }; if (!cs) return ret; nfs_request_user_init (&nfu, cs->req); ret = nfs_readdirp (cs->nfsx, cs->vol, &nfu, cs->fd, cs->dircount, cs->cookie, nfs3svc_readdir_cbk, cs); return ret; } int nfs3_readdir_read_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; struct nfs3_state *nfs3 = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs3 = nfs_rpcsvc_request_program_private (cs->req); ret = nfs3_verify_dircookie (nfs3, cs->fd, cs->cookie, cs->cookieverf, &stat); if (ret < 0) /* Stat already set by verifier function above. */ goto nfs3err; ret = nfs3_readdir_process (cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { if (cs->maxcount == 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READDIR", stat, -ret); nfs3_readdir_reply (cs->req, stat, NULL, 0, NULL, NULL, 0, 0); } else { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READDIRP", stat, -ret); nfs3_readdirp_reply (cs->req, stat, NULL, 0, NULL, NULL, 0, 0, 0); } nfs3_call_state_wipe (cs); } return 0; } int nfs3_readdir_open_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); ret = nfs3_dir_open_and_resume (cs, nfs3_readdir_read_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { if (cs->maxcount == 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READDIR", stat, -ret); nfs3_readdir_reply (cs->req, stat, NULL, 0, NULL, NULL, 0, 0); } else { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "READDIRP", stat, -ret); nfs3_readdirp_reply (cs->req, stat, NULL, 0, NULL, NULL, 0, 0, 0); } nfs3_call_state_wipe (cs); } return ret; } int nfs3_readdir (rpcsvc_request_t *req, struct nfs3_fh *fh, cookie3 cookie, uint64_t cverf, count3 dircount, count3 maxcount) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_readdir_call (nfs_rpcsvc_request_xid (req), fh, dircount, maxcount); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->cookieverf = cverf; cs->dircount = dircount; cs->maxcount = maxcount; cs->cookie = cookie; cs->parent = *fh; ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_readdir_open_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { if (maxcount == 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "READDIR", stat, -ret); nfs3_readdir_reply (req, stat, NULL, 0, NULL, NULL, 0, 0); } else { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "READDIRP", stat, -ret); nfs3_readdirp_reply (req, stat, NULL, 0, NULL, NULL, 0, 0, 0); } /* Ret must be NULL after this so that the caller does not * also send an RPC reply. */ ret = 0; nfs3_call_state_wipe (cs); } return ret; } int nfs3svc_readdir (rpcsvc_request_t *req) { readdir3args ra; struct nfs3_fh fh = {{0},}; int ret = RPCSVC_ACTOR_ERROR; uint64_t verf = 0; if (!req) return ret; nfs3_prep_readdir3args (&ra, &fh); if (xdr_to_readdir3args (req->msg, &ra) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } verf = *(uint64_t *)ra.cookieverf; ret = nfs3_readdir (req, &fh, ra.cookie, verf, ra.count, 0); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "READDIR procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3svc_readdirp (rpcsvc_request_t *req) { readdirp3args ra; struct nfs3_fh fh = {{0},}; int ret = RPCSVC_ACTOR_ERROR; uint64_t cverf = 0; if (!req) return ret; nfs3_prep_readdirp3args (&ra, &fh); if (xdr_to_readdirp3args (req->msg, &ra) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } cverf = *(uint64_t *)ra.cookieverf; ret = nfs3_readdir (req, &fh, ra.cookie, cverf, ra.dircount, ra.maxcount); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "READDIRP procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_fsstat_reply (rpcsvc_request_t *req, nfsstat3 stat, struct statvfs *fsbuf, struct iatt *postbuf) { fsstat3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_fsstat3res (&res, stat, fsbuf, postbuf, xlid); return nfs3svc_submit_reply (req, &res, (nfs3_serializer)xdr_serialize_fsstat3res); } int32_t nfs3_fsstat_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) stat = nfs3_errno_to_nfsstat3 (op_errno); else stat = NFS3_OK; nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "FSTAT", stat, op_errno); nfs3_fsstat_reply (cs->req, stat, &cs->fsstat, buf); nfs3_call_state_wipe (cs); return 0; } int32_t nfs3_fsstat_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct statvfs *buf) { nfs_user_t nfu = {0, }; int ret = -EFAULT; nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) { ret = -op_errno; stat = nfs3_errno_to_nfsstat3 (op_errno); goto err; } /* Then get the stat for the fs root in order to fill in the * post_op_attr. */ cs->fsstat = *buf; nfs_request_user_init (&nfu, cs->req); ret = nfs_stat (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3_fsstat_stat_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "FSTAT", stat, -ret); nfs3_fsstat_reply (cs->req, stat, NULL, NULL); nfs3_call_state_wipe (cs); } return 0; } int nfs3_fsstat_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; nfs_user_t nfu = {0, }; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); /* First, we need to get the statvfs for the subvol */ ret = nfs_statfs (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3_fsstat_statfs_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "FSTAT", stat, -ret); nfs3_fsstat_reply (cs->req, stat, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_fsstat (rpcsvc_request_t *req, struct nfs3_fh *fh) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_common_call (nfs_rpcsvc_request_xid (req), "FSSTAT", fh); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_fsstat_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "FSTAT", stat, -ret); nfs3_fsstat_reply (req, stat, NULL, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_fsstat (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; fsstat3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_fsstat3args (&args, &fh); if (xdr_to_fsstat3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_fsstat (req, &fh); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "FSTAT procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_fsinfo_reply (rpcsvc_request_t *req, nfsstat3 status, struct iatt *fsroot) { fsinfo3res res; struct nfs3_state *nfs3 = NULL; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3 = nfs_rpcsvc_request_program_private (req); nfs3_fill_fsinfo3res (nfs3, &res, status, fsroot, xlid); nfs3svc_submit_reply (req, &res, (nfs3_serializer)xdr_serialize_fsinfo3res); return 0; } int32_t nfs3svc_fsinfo_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf) { nfsstat3 status = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; cs = frame->local; if (op_ret == -1) status = nfs3_errno_to_nfsstat3 (op_errno); else status = NFS3_OK; nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "FSINFO", status, op_errno); nfs3_fsinfo_reply (cs->req, status, buf); nfs3_call_state_wipe (cs); return 0; } int nfs3_fsinfo_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); ret = nfs_stat (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_fsinfo_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "FSINFO", stat, -ret); nfs3_fsinfo_reply (cs->req, stat, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_fsinfo (rpcsvc_request_t *req, struct nfs3_fh *fh) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_common_call (nfs_rpcsvc_request_xid (req), "FSINFO", fh); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_fsinfo_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "FSINFO", stat, -ret); nfs3_fsinfo_reply (req, stat, NULL); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3svc_fsinfo (rpcsvc_request_t *req) { int ret = RPCSVC_ACTOR_ERROR; fsinfo3args args; struct nfs3_fh root = {{0}, }; if (!req) return ret; nfs3_prep_fsinfo3args (&args, &root); if (xdr_to_fsinfo3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding arguments"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_fsinfo (req, &root); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "FSINFO procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_pathconf_reply (rpcsvc_request_t *req, nfsstat3 stat, struct iatt *buf) { pathconf3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_pathconf3res (&res, stat, buf, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_pathconf3res); return 0; } int32_t nfs3svc_pathconf_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *buf) { struct iatt *sbuf = NULL; nfs3_call_state_t *cs = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; cs = frame->local; if (op_ret == -1) stat = nfs3_errno_to_nfsstat3 (op_errno); else { /* If stat fop failed, we can still send the other components * in a pathconf reply. */ sbuf = buf; stat = NFS3_OK; } nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "PATHCONF", stat, op_errno); nfs3_pathconf_reply (cs->req, stat, sbuf); nfs3_call_state_wipe (cs); return 0; } int nfs3_pathconf_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); nfs_request_user_init (&nfu, cs->req); ret = nfs_stat (cs->nfsx, cs->vol, &nfu, &cs->resolvedloc, nfs3svc_pathconf_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "PATHCONF", stat, -ret); nfs3_pathconf_reply (cs->req, stat, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_pathconf (rpcsvc_request_t *req, struct nfs3_fh *fh) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_common_call (nfs_rpcsvc_request_xid (req), "PATHCONF", fh); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_pathconf_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "PATHCONF", stat, -ret); nfs3_pathconf_reply (req, stat, NULL); nfs3_call_state_wipe (cs); /* Ret must be 0 after this so that the caller does not * also send an RPC reply. */ ret = 0; } return ret; } int nfs3svc_pathconf (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; pathconf3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_pathconf3args (&args, &fh); if (xdr_to_pathconf3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_pathconf (req, &fh); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "PATHCONF procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } int nfs3_commit_reply (rpcsvc_request_t *req, nfsstat3 stat, uint64_t wverf, struct iatt *prestat, struct iatt *poststat) { commit3res res = {0, }; uint16_t xlid = 0; xlid = nfs3_request_xlator_id (req); nfs3_fill_commit3res (&res, stat, wverf, prestat, poststat, xlid); nfs3svc_submit_reply (req, (void *)&res, (nfs3_serializer)xdr_serialize_commit3res); return 0; } int32_t nfs3svc_commit_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf) { nfsstat3 stat = NFS3ERR_SERVERFAULT; nfs3_call_state_t *cs = NULL; struct nfs3_state *nfs3 = NULL; cs = frame->local; if (op_ret == -1) stat = nfs3_errno_to_nfsstat3 (op_errno); else stat = NFS3_OK; nfs3 = nfs_rpcsvc_request_program_private (cs->req); nfs3_log_commit_res (nfs_rpcsvc_request_xid (cs->req), stat, op_errno, nfs3->serverstart); nfs3_commit_reply (cs->req, stat, nfs3->serverstart, prebuf, postbuf); nfs3_call_state_wipe (cs); return 0; } int nfs3_commit_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs_user_t nfu = {0, }; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); if (nfs3_export_sync_trusted (cs->nfs3state, cs->resolvefh.xlatorid)) { ret = -1; stat = NFS3_OK; goto nfs3err; } nfs_request_user_init (&nfu, cs->req); ret = nfs_fsync (cs->nfsx, cs->vol, &nfu, cs->fd, 0, nfs3svc_commit_cbk, cs); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "COMMIT", stat, -ret); nfs3_commit_reply (cs->req, stat, cs->nfs3state->serverstart, NULL, NULL); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3_commit_open_resume (void *carg) { nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; nfs3_call_state_t *cs = NULL; if (!carg) return ret; cs = (nfs3_call_state_t *)carg; nfs3_check_fh_resolve_status (cs, stat, nfs3err); ret = nfs3_file_open_and_resume (cs, nfs3_commit_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (cs->req), "COMMIT", stat, -ret); nfs3_commit_reply (cs->req, stat, 0, NULL, NULL); nfs3_call_state_wipe (cs); } return ret; } int nfs3_commit (rpcsvc_request_t *req, struct nfs3_fh *fh, offset3 offset, count3 count) { xlator_t *vol = NULL; nfsstat3 stat = NFS3ERR_SERVERFAULT; int ret = -EFAULT; struct nfs3_state *nfs3 = NULL; nfs3_call_state_t *cs = NULL; if ((!req) || (!fh)) { gf_log (GF_NFS3, GF_LOG_ERROR, "Bad arguments"); return -1; } nfs3_log_rw_call (nfs_rpcsvc_request_xid (req), "COMMIT", fh, offset, count, -1); nfs3_validate_gluster_fh (fh, stat, nfs3err); nfs3_validate_nfs3_state (req, nfs3, stat, nfs3err, ret); nfs3_map_fh_to_volume (nfs3, fh, req, vol, stat, nfs3err); nfs3_check_rw_volaccess (nfs3, fh->xlatorid, stat, nfs3err); nfs3_handle_call_state_init (nfs3, cs, req, vol, stat, nfs3err); cs->datacount = count; cs->dataoffset = offset; ret = nfs3_fh_resolve_and_resume (cs, fh, NULL, nfs3_commit_open_resume); if (ret < 0) stat = nfs3_errno_to_nfsstat3 (-ret); nfs3err: if (ret < 0) { nfs3_log_common_res (nfs_rpcsvc_request_xid (req), "COMMIT", stat, -ret); nfs3_commit_reply (req, stat, 0, NULL, NULL); nfs3_call_state_wipe (cs); ret = 0; } return ret; } int nfs3svc_commit (rpcsvc_request_t *req) { struct nfs3_fh fh = {{0}, }; commit3args args; int ret = RPCSVC_ACTOR_ERROR; if (!req) return ret; nfs3_prep_commit3args (&args, &fh); if (xdr_to_commit3args (req->msg, &args) <= 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Error decoding args"); nfs_rpcsvc_request_seterr (req, GARBAGE_ARGS); goto rpcerr; } ret = nfs3_commit (req, &fh, args.offset, args.count); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "COMMIT procedure failed"); nfs_rpcsvc_request_seterr (req, SYSTEM_ERR); ret = RPCSVC_ACTOR_ERROR; } rpcerr: return ret; } rpcsvc_actor_t nfs3svc_actors[NFS3_PROC_COUNT] = { {"NULL", NFS3_NULL, nfs3svc_null, NULL, NULL}, {"GETATTR", NFS3_GETATTR, nfs3svc_getattr,NULL, NULL}, {"SETATTR", NFS3_SETATTR, nfs3svc_setattr,NULL, NULL}, {"LOOKUP", NFS3_LOOKUP, nfs3svc_lookup, NULL, NULL}, {"ACCESS", NFS3_ACCESS, nfs3svc_access, NULL, NULL}, {"READLINK", NFS3_READLINK, nfs3svc_readlink,NULL, NULL}, {"READ", NFS3_READ, nfs3svc_read, NULL, NULL}, {"WRITE", NFS3_WRITE, nfs3svc_write, nfs3svc_write_vec, nfs3svc_write_vecsizer}, {"CREATE", NFS3_CREATE, nfs3svc_create, NULL, NULL}, {"MKDIR", NFS3_MKDIR, nfs3svc_mkdir, NULL, NULL}, {"SYMLINK", NFS3_SYMLINK, nfs3svc_symlink,NULL, NULL}, {"MKNOD", NFS3_MKNOD, nfs3svc_mknod, NULL, NULL}, {"REMOVE", NFS3_REMOVE, nfs3svc_remove, NULL, NULL}, {"RMDIR", NFS3_RMDIR, nfs3svc_rmdir, NULL, NULL}, {"RENAME", NFS3_RENAME, nfs3svc_rename, NULL, NULL}, {"LINK", NFS3_LINK, nfs3svc_link, NULL, NULL}, {"READDIR", NFS3_READDIR, nfs3svc_readdir,NULL, NULL}, {"READDIRPLUS", NFS3_READDIRP, nfs3svc_readdirp,NULL, NULL}, {"FSSTAT", NFS3_FSSTAT, nfs3svc_fsstat, NULL, NULL}, {"FSINFO", NFS3_FSINFO, nfs3svc_fsinfo, NULL, NULL}, {"PATHCONF", NFS3_PATHCONF, nfs3svc_pathconf,NULL, NULL}, {"COMMIT", NFS3_COMMIT, nfs3svc_commit, NULL, NULL} }; rpcsvc_program_t nfs3prog = { .progname = "NFS3", .prognum = NFS_PROGRAM, .progver = NFS_V3, .progport = GF_NFS3_PORT, .progaddrfamily = AF_INET, .proghost = NULL, .actors = nfs3svc_actors, .numactors = NFS3_PROC_COUNT, .conn_destroy = NULL, .conn_init = NULL, /* Requests like FSINFO are sent before an auth scheme * is inited by client. See RFC 2623, Section 2.3.2. */ .min_auth = AUTH_NULL, }; int nfs3_init_options (struct nfs3_state *nfs3, xlator_t *nfsx) { int ret = -1; char *optstr = NULL; uint64_t size64 = 0; if ((!nfs3) || (!nfsx)) return -1; /* nfs3.read-size */ nfs3->readsize = GF_NFS3_RTPREF; if (dict_get (nfsx->options, "nfs3.read-size")) { ret = dict_get_str (nfsx->options, "nfs3.read-size", &optstr); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to read " " option: nfs3.read-size"); ret = -1; goto err; } ret = gf_string2bytesize (optstr, &size64); nfs3->readsize = size64; if (ret == -1) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to format" " option: nfs3.read-size"); ret = -1; goto err; } } /* nfs3.write-size */ nfs3->writesize = GF_NFS3_WTPREF; if (dict_get (nfsx->options, "nfs3.write-size")) { ret = dict_get_str (nfsx->options, "nfs3.write-size", &optstr); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to read " " option: nfs3.write-size"); ret = -1; goto err; } ret = gf_string2bytesize (optstr, &size64); nfs3->writesize = size64; if (ret == -1) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to format" " option: nfs3.write-size"); ret = -1; goto err; } } /* nfs3.readdir.size */ nfs3->readdirsize = GF_NFS3_DTPREF; if (dict_get (nfsx->options, "nfs3.readdir-size")) { ret = dict_get_str (nfsx->options,"nfs3.readdir-size", &optstr); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to read" " option: nfs3.readdir-size"); ret = -1; goto err; } ret = gf_string2bytesize (optstr, &size64); nfs3->readdirsize = size64; if (ret == -1) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to format" " option: nfs3.readdir-size"); ret = -1; goto err; } } /* We want to use the size of the biggest param for the io buffer size. */ nfs3->iobsize = nfs3->readsize; if (nfs3->iobsize < nfs3->writesize) nfs3->iobsize = nfs3->writesize; if (nfs3->iobsize < nfs3->readdirsize) nfs3->iobsize = nfs3->readdirsize; /* But this is the true size of each iobuf. We need this size to * accomodate the NFS headers also in the same buffer. */ nfs3->iobsize = nfs3->iobsize * 2; /* mem-factor */ nfs3->memfactor = GF_NFS3_DEFAULT_MEMFACTOR; ret = 0; err: return ret; } int nfs3_init_subvolume_options (struct nfs3_export *exp, dict_t *options) { int ret = -1; char *optstr = NULL; char searchkey[1024]; char *name = NULL; gf_boolean_t boolt = _gf_false; if ((!exp) || (!options)) return -1; /* Volume Access */ name = exp->subvol->name; ret = snprintf (searchkey, 1024, "nfs3.%s.volume-access", name); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "snprintf failed"); ret = -1; goto err; } exp->access = GF_NFS3_DEFAULT_VOLACCESS; if (dict_get (options, searchkey)) { ret = dict_get_str (options, searchkey, &optstr); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to read " " option: %s", searchkey); ret = -1; goto err; } if (strcmp (optstr, "read-only") == 0) exp->access = GF_NFS3_VOLACCESS_RO; } ret = snprintf (searchkey, 1024, "rpc-auth.%s.unix", name); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "snprintf failed"); ret = -1; goto err; } if (dict_get (options, searchkey)) { ret = dict_get_str (options, searchkey, &optstr); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to read " " option: %s", searchkey); ret = -1; goto err; } } exp->trusted_sync = 0; ret = snprintf (searchkey, 1024, "nfs3.%s.trusted-sync", name); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "snprintf failed"); ret = -1; goto err; } if (dict_get (options, searchkey)) { ret = dict_get_str (options, searchkey, &optstr); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to read " " option: %s", searchkey); ret = -1; goto err; } ret = gf_string2boolean (optstr, &boolt); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to convert str " "to gf_boolean_t"); ret = -1; goto err; } if (boolt == _gf_true) exp->trusted_sync = 1; } exp->trusted_write = 0; ret = snprintf (searchkey, 1024, "nfs3.%s.trusted-write", name); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "snprintf failed"); ret = -1; goto err; } if (dict_get (options, searchkey)) { ret = dict_get_str (options, searchkey, &optstr); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to read " " option: %s", searchkey); ret = -1; goto err; } ret = gf_string2boolean (optstr, &boolt); if (ret < 0) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to convert str " "to gf_boolean_t"); ret = -1; goto err; } if (boolt == _gf_true) exp->trusted_write = 1; } /* If trusted-sync is on, then we also switch on trusted-write because * tw is included in ts. In write logic, we're then only checking for * tw. */ if (exp->trusted_sync) exp->trusted_write = 1; gf_log (GF_NFS3, GF_LOG_TRACE, "%s: %s, %s, %s", exp->subvol->name, (exp->access == GF_NFS3_VOLACCESS_RO)?"read-only":"read-write", (exp->trusted_sync == 0)?"no trusted_sync":"trusted_sync", (exp->trusted_write == 0)?"no trusted_write":"trusted_write"); ret = 0; err: return ret; } int nfs3_init_subvolume (struct nfs3_state *nfs3, xlator_t *nfsx, xlator_t *subvol, int xlid) { int ret = -1; struct nfs3_export *exp = NULL; if ((!nfs3) || (!nfsx) || (!subvol)) return -1; exp = &nfs3->exports[xlid]; exp->subvol = subvol; gf_log (GF_NFS3, GF_LOG_TRACE, "Initing state: %s", exp->subvol->name); ret = nfs3_init_subvolume_options (exp, nfsx->options); if (ret == -1) gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to init subvol"); return ret; } int nfs3_init_subvolumes (struct nfs3_state *nfs3, xlator_t *nfsx) { int xl_count = 0; int ret = -1; struct xlator_list *xl_list = NULL; if ((!nfs3) || (!nfsx)) return -1; xl_list = nfsx->children; while (xl_list) { ++xl_count; xl_list = xl_list->next; } nfs3->exports = GF_CALLOC (xl_count, sizeof (struct nfs3_export), gf_nfs_mt_nfs3_export); if (!nfs3->exports) { gf_log (GF_NFS3, GF_LOG_ERROR, "Memory allocation failed"); goto err; } xl_list = nfsx->children; xl_count = 0; /* Re-using xl_count. */ while (xl_list) { ret = nfs3_init_subvolume (nfs3, nfsx, xl_list->xlator, xl_count); if (ret == -1) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to init subvol: " "%s", xl_list->xlator->name); goto err; } xl_list = xl_list->next; ++xl_count; } ret = 0; err: return ret; } struct nfs3_state * nfs3_init_state (xlator_t *nfsx) { struct nfs3_state *nfs3 = NULL; int ret = -1; unsigned int localpool = 0; if (!nfsx) return NULL; nfs3 = (struct nfs3_state *)GF_CALLOC (1, sizeof (*nfs3), gf_nfs_mt_nfs3_state); if (!nfs3) { gf_log (GF_NFS3, GF_LOG_ERROR, "Memory allocation failed"); return NULL; } ret = nfs3_init_options (nfs3, nfsx); if (ret == -1) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to init options"); goto ret; } nfs3->iobpool = nfsx->ctx->iobuf_pool; localpool = nfs3->memfactor * GF_NFS_CONCURRENT_OPS_MULT; gf_log (GF_NFS3, GF_LOG_TRACE, "local pool: %d", localpool); nfs3->localpool = mem_pool_new (nfs3_call_state_t, localpool); if (!nfs3->localpool) { gf_log (GF_NFS3, GF_LOG_ERROR, "local mempool creation failed"); ret = -1; goto ret; } nfs3->nfsx = nfsx; nfs3->exportslist = nfsx->children; ret = nfs3_init_subvolumes (nfs3, nfsx); if (ret == -1) { gf_log (GF_NFS3, GF_LOG_ERROR, "Failed to init per-subvolume " "state"); goto free_localpool; } nfs3->serverstart = (uint64_t)time (NULL); INIT_LIST_HEAD (&nfs3->fdlru); LOCK_INIT (&nfs3->fdlrulock); nfs3->fdcount = 0; ret = 0; free_localpool: if (ret == -1) mem_pool_destroy (nfs3->localpool); ret: if (ret == -1) { GF_FREE (nfs3); nfs3 = NULL; } return nfs3; } rpcsvc_program_t * nfs3svc_init (xlator_t *nfsx) { struct nfs3_state *nfs3 = NULL; if (!nfsx) return NULL; nfs3 = nfs3_init_state (nfsx); if (!nfs3) { gf_log (GF_NFS3, GF_LOG_ERROR, "NFSv3 state init failed"); return NULL; } nfs3prog.private = nfs3; return &nfs3prog; }