X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fliblustre%2Fsuper.c;h=3f3788a7461e75b03e7a5d990f2cbe32e9a0c840;hb=218322d65b7d33ea185c7186b432f684e440eada;hp=5f4c017d6b121dcce045451537a545894ad3f2a8;hpb=ff14a77c285b8ca6c1b51fb58478cd5fd4328881;p=fs%2Flustre-release.git diff --git a/lustre/liblustre/super.c b/lustre/liblustre/super.c index 5f4c017..3f3788a 100644 --- a/lustre/liblustre/super.c +++ b/lustre/liblustre/super.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -54,19 +54,6 @@ # include #endif -#include -#ifdef HAVE_XTIO_H -#include -#endif -#include -#include -#include -#ifdef HAVE_FILE_H -#include -#endif - -#undef LIST_HEAD - #include "llite_lib.h" #ifndef MAY_EXEC @@ -82,10 +69,10 @@ static int ll_permission(struct inode *inode, int mask) struct intnl_stat *st = llu_i2stat(inode); mode_t mode = st->st_mode; - if (current->fsuid == st->st_uid) - mode >>= 6; - else if (in_group_p(st->st_gid)) - mode >>= 3; + if (current->fsuid == st->st_uid) + mode >>= 6; + else if (in_group_p(st->st_gid)) + mode >>= 3; if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask) return 0; @@ -111,7 +98,7 @@ static void llu_fsop_gone(struct filesys *fs) int next = 0; ENTRY; - list_del(&sbi->ll_conn_chain); + cfs_list_del(&sbi->ll_conn_chain); cl_sb_fini(sbi); obd_disconnect(sbi->ll_dt_exp); obd_disconnect(sbi->ll_md_exp); @@ -133,7 +120,7 @@ static ldlm_mode_t llu_take_md_lock(struct inode *inode, __u64 bits, ldlm_policy_data_t policy = { .l_inodebits = {bits}}; struct lu_fid *fid; ldlm_mode_t rc; - int flags; + __u64 flags; ENTRY; fid = &llu_i2info(inode)->lli_fid; @@ -160,39 +147,33 @@ void llu_update_inode(struct inode *inode, struct lustre_md *md) st->st_mode = (st->st_mode & ~S_IFMT)|(body->mode & S_IFMT); if (lsm != NULL) { - if (lli->lli_smd == NULL) { - cl_inode_init(inode, md); - lli->lli_smd = lsm; - lli->lli_maxbytes = lsm->lsm_maxbytes; - if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES) - lli->lli_maxbytes = PAGE_CACHE_MAXBYTES; - } else { - if (lov_stripe_md_cmp(lli->lli_smd, lsm)) { - CERROR("lsm mismatch for inode %lld\n", - (long long)st->st_ino); - LBUG(); - } - } - } - - if (body->valid & OBD_MD_FLMTIME && - body->mtime > LTIME_S(st->st_mtime)) - LTIME_S(st->st_mtime) = body->mtime; - if (body->valid & OBD_MD_FLATIME && - body->atime > LTIME_S(st->st_atime)) - LTIME_S(st->st_atime) = body->atime; - - /* mtime is always updated with ctime, but can be set in past. - As write and utime(2) may happen within 1 second, and utime's - mtime has a priority over write's one, so take mtime from mds - for the same ctimes. */ - if (body->valid & OBD_MD_FLCTIME && - body->ctime >= LTIME_S(st->st_ctime)) { - LTIME_S(st->st_ctime) = body->ctime; - if (body->valid & OBD_MD_FLMTIME) + if (!lli->lli_has_smd) { + cl_file_inode_init(inode, md); + lli->lli_has_smd = true; + lli->lli_maxbytes = lsm->lsm_maxbytes; + if (lli->lli_maxbytes > MAX_LFS_FILESIZE) + lli->lli_maxbytes = MAX_LFS_FILESIZE; + } + if (md->lsm != NULL) + obd_free_memmd(llu_i2obdexp(inode), &md->lsm); + } + + if (body->valid & OBD_MD_FLATIME) { + if (body->atime > LTIME_S(st->st_atime)) + LTIME_S(st->st_atime) = body->atime; + lli->lli_lvb.lvb_atime = body->atime; + } + if (body->valid & OBD_MD_FLMTIME) { + if (body->mtime > LTIME_S(st->st_mtime)) LTIME_S(st->st_mtime) = body->mtime; + lli->lli_lvb.lvb_mtime = body->mtime; + } + if (body->valid & OBD_MD_FLCTIME) { + if (body->ctime > LTIME_S(st->st_ctime)) + LTIME_S(st->st_ctime) = body->ctime; + lli->lli_lvb.lvb_ctime = body->ctime; } - if (S_ISREG(st->st_mode)) + if (S_ISREG(st->st_mode)) st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE); else st->st_blksize = 4096; @@ -208,7 +189,7 @@ void llu_update_inode(struct inode *inode, struct lustre_md *md) lli->lli_st_flags = body->flags; if (body->valid & OBD_MD_FLSIZE) { if ((llu_i2sbi(inode)->ll_lco.lco_flags & OBD_CONNECT_SOM) && - S_ISREG(st->st_mode) && lli->lli_smd) { + S_ISREG(st->st_mode) && lli->lli_has_smd) { struct lustre_handle lockh; ldlm_mode_t mode; @@ -239,9 +220,13 @@ void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid) valid &= src->o_valid; + LASSERTF(!(valid & (OBD_MD_FLTYPE | OBD_MD_FLGENER | OBD_MD_FLFID | + OBD_MD_FLID | OBD_MD_FLGROUP)), + "object "DOSTID", valid %x\n", POSTID(&src->o_oi), valid); + if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME)) CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T - ", new %lu/%lu\n", + ", new %lu/%lu\n", src->o_valid, LTIME_S(st->st_mtime), LTIME_S(st->st_ctime), (long)src->o_mtime, (long)src->o_ctime); @@ -270,99 +255,37 @@ void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid) lli->lli_st_flags = src->o_flags; } -#define S_IRWXUGO (S_IRWXU|S_IRWXG|S_IRWXO) -#define S_IALLUGO (S_ISUID|S_ISGID|S_ISVTX|S_IRWXUGO) - -void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid) -{ - struct llu_inode_info *lli = llu_i2info(src); - struct intnl_stat *st = llu_i2stat(src); - obd_flag newvalid = 0; - - if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME)) - CDEBUG(D_INODE, "valid %x, new time "CFS_TIME_T"/"CFS_TIME_T"\n", - valid, LTIME_S(st->st_mtime), - LTIME_S(st->st_ctime)); - - if (valid & OBD_MD_FLATIME) { - dst->o_atime = LTIME_S(st->st_atime); - newvalid |= OBD_MD_FLATIME; - } - if (valid & OBD_MD_FLMTIME) { - dst->o_mtime = LTIME_S(st->st_mtime); - newvalid |= OBD_MD_FLMTIME; - } - if (valid & OBD_MD_FLCTIME) { - dst->o_ctime = LTIME_S(st->st_ctime); - newvalid |= OBD_MD_FLCTIME; - } - if (valid & OBD_MD_FLSIZE) { - dst->o_size = st->st_size; - newvalid |= OBD_MD_FLSIZE; - } - if (valid & OBD_MD_FLBLOCKS) { /* allocation of space (x512 bytes) */ - dst->o_blocks = st->st_blocks; - newvalid |= OBD_MD_FLBLOCKS; - } - if (valid & OBD_MD_FLBLKSZ) { /* optimal block size */ - dst->o_blksize = st->st_blksize; - newvalid |= OBD_MD_FLBLKSZ; - } - if (valid & OBD_MD_FLTYPE) { - dst->o_mode = (dst->o_mode & S_IALLUGO)|(st->st_mode & S_IFMT); - newvalid |= OBD_MD_FLTYPE; - } - if (valid & OBD_MD_FLMODE) { - dst->o_mode = (dst->o_mode & S_IFMT)|(st->st_mode & S_IALLUGO); - newvalid |= OBD_MD_FLMODE; - } - if (valid & OBD_MD_FLUID) { - dst->o_uid = st->st_uid; - newvalid |= OBD_MD_FLUID; - } - if (valid & OBD_MD_FLGID) { - dst->o_gid = st->st_gid; - newvalid |= OBD_MD_FLGID; - } - if (valid & OBD_MD_FLFLAGS) { - dst->o_flags = lli->lli_st_flags; - newvalid |= OBD_MD_FLFLAGS; - } - if (valid & OBD_MD_FLGENER) { - dst->o_generation = lli->lli_st_generation; - newvalid |= OBD_MD_FLGENER; - } - if (valid & OBD_MD_FLFID) { - dst->o_fid = st->st_ino; - newvalid |= OBD_MD_FLFID; - } - - dst->o_valid |= newvalid; -} - -/* - * really does the getattr on the inode and updates its fields +/** + * Performs the getattr on the inode and updates its fields. + * If @sync != 0, perform the getattr under the server-side lock. */ -int llu_inode_getattr(struct inode *inode, struct obdo *obdo) +int llu_inode_getattr(struct inode *inode, struct obdo *obdo, + __u64 ioepoch, int sync) { - struct llu_inode_info *lli = llu_i2info(inode); - struct ptlrpc_request_set *set; - struct lov_stripe_md *lsm = lli->lli_smd; - struct obd_info oinfo = { { { 0 } } }; - int rc; - ENTRY; + struct ptlrpc_request_set *set; + struct lov_stripe_md *lsm = NULL; + struct obd_info oinfo = { { { 0 } } }; + int rc; + ENTRY; + lsm = ccc_inode_lsm_get(inode); LASSERT(lsm); oinfo.oi_md = lsm; oinfo.oi_oa = obdo; - oinfo.oi_oa->o_id = lsm->lsm_object_id; - oinfo.oi_oa->o_gr = lsm->lsm_object_gr; + oinfo.oi_oa->o_oi = lsm->lsm_oi; oinfo.oi_oa->o_mode = S_IFREG; + oinfo.oi_oa->o_ioepoch = ioepoch; oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLMTIME | - OBD_MD_FLCTIME | OBD_MD_FLGROUP; + OBD_MD_FLCTIME | OBD_MD_FLGROUP | + OBD_MD_FLATIME | OBD_MD_FLEPOCH; + obdo_set_parent_fid(oinfo.oi_oa, &llu_i2info(inode)->lli_fid); + if (sync) { + oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS; + oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK; + } set = ptlrpc_prep_set(); if (set == NULL) { @@ -374,6 +297,7 @@ int llu_inode_getattr(struct inode *inode, struct obdo *obdo) rc = ptlrpc_set_wait(set); ptlrpc_set_destroy(set); } + ccc_inode_lsm_put(inode, lsm); if (rc) RETURN(rc); @@ -381,13 +305,13 @@ int llu_inode_getattr(struct inode *inode, struct obdo *obdo) OBD_MD_FLMTIME | OBD_MD_FLCTIME | OBD_MD_FLSIZE; - obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid); - CDEBUG(D_INODE, "objid "LPX64" size %Lu, blocks %Lu, " - "blksize %Lu\n", lli->lli_smd->lsm_object_id, - (long long unsigned)llu_i2stat(inode)->st_size, - (long long unsigned)llu_i2stat(inode)->st_blocks, - (long long unsigned)llu_i2stat(inode)->st_blksize); - RETURN(0); + obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid); + CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu, " + "blksize %llu\n", POSTID(&oinfo.oi_oa->o_oi), + (long long unsigned)llu_i2stat(inode)->st_size, + (long long unsigned)llu_i2stat(inode)->st_blocks, + (long long unsigned)llu_i2stat(inode)->st_blksize); + RETURN(0); } static struct inode* llu_new_inode(struct filesys *fs, @@ -417,7 +341,7 @@ static struct inode* llu_new_inode(struct filesys *fs, /* initialize lli here */ lli->lli_sbi = llu_fs2sbi(fs); - lli->lli_smd = NULL; + lli->lli_has_smd = false; lli->lli_symlink_name = NULL; lli->lli_flags = 0; lli->lli_maxbytes = (__u64)(~0UL); @@ -442,7 +366,7 @@ static int llu_have_md_lock(struct inode *inode, __u64 lockpart) struct lustre_handle lockh; ldlm_policy_data_t policy = { .l_inodebits = { lockpart } }; struct lu_fid *fid; - int flags; + __u64 flags; ENTRY; LASSERT(inode); @@ -460,7 +384,8 @@ static int llu_have_md_lock(struct inode *inode, __u64 lockpart) static int llu_inode_revalidate(struct inode *inode) { - struct lov_stripe_md *lsm = NULL; + struct llu_inode_info *lli = llu_i2info(inode); + struct intnl_stat *st = llu_i2stat(inode); ENTRY; if (!inode) { @@ -472,20 +397,25 @@ static int llu_inode_revalidate(struct inode *inode) struct lustre_md md; struct ptlrpc_request *req = NULL; struct llu_sb_info *sbi = llu_i2sbi(inode); + struct md_op_data op_data = { { 0 } }; unsigned long valid = OBD_MD_FLGETATTR; int rc, ealen = 0; /* Why don't we update all valid MDS fields here, if we're * doing an RPC anyways? -phil */ - if (S_ISREG(llu_i2stat(inode)->st_mode)) { + if (S_ISREG(st->st_mode)) { ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL); valid |= OBD_MD_FLEASIZE; } - rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), - NULL, valid, ealen, &req); + + llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen, + LUSTRE_OPC_ANY); + op_data.op_valid = valid; + + rc = md_getattr(sbi->ll_md_exp, &op_data, &req); if (rc) { CERROR("failure %d inode %llu\n", rc, - (long long)llu_i2stat(inode)->st_ino); + (long long)st->st_ino); RETURN(-abs(rc)); } rc = md_get_lustre_md(sbi->ll_md_exp, req, @@ -507,14 +437,18 @@ static int llu_inode_revalidate(struct inode *inode) llu_update_inode(inode, &md); - if (md.lsm != NULL && llu_i2info(inode)->lli_smd != md.lsm) - obd_free_memmd(sbi->ll_dt_exp, &md.lsm); - ptlrpc_req_finished(req); - } - - lsm = llu_i2info(inode)->lli_smd; - if (!lsm) /* object not yet allocated, don't validate size */ + if (md.lsm != NULL) + obd_free_memmd(sbi->ll_dt_exp, &md.lsm); + ptlrpc_req_finished(req); + } + + if (!lli->lli_has_smd) { + /* object not yet allocated, don't validate size */ + st->st_atime = lli->lli_lvb.lvb_atime; + st->st_mtime = lli->lli_lvb.lvb_mtime; + st->st_ctime = lli->lli_lvb.lvb_ctime; RETURN(0); + } /* ll_glimpse_size will prefer locally cached writes if they extend * the file */ @@ -570,8 +504,9 @@ static int null_if_equal(struct ldlm_lock *lock, void *data) void llu_clear_inode(struct inode *inode) { - struct llu_inode_info *lli = llu_i2info(inode); - struct llu_sb_info *sbi = llu_i2sbi(inode); + struct llu_inode_info *lli = llu_i2info(inode); + struct llu_sb_info *sbi = llu_i2sbi(inode); + struct lov_stripe_md *lsm; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n", @@ -579,19 +514,15 @@ void llu_clear_inode(struct inode *inode) inode); lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK; - md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode), - null_if_equal, inode); + md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode)); - if (lli->lli_smd) - obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd, - null_if_equal, inode); + lsm = ccc_inode_lsm_get(inode); + if (lsm != NULL) + obd_change_cbdata(sbi->ll_dt_exp, lsm, null_if_equal, inode); + ccc_inode_lsm_put(inode, lsm); - cl_inode_fini(inode); - - if (lli->lli_smd) { - obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd); - lli->lli_smd = NULL; - } + cl_inode_fini(inode); + lli->lli_has_smd = false; if (lli->lli_symlink_name) { OBD_FREE(lli->lli_symlink_name, @@ -624,7 +555,7 @@ static int inode_setattr(struct inode * inode, struct iattr * attr) * inode_setattr() is only ever invoked with ATTR_SIZE (by * llu_setattr_raw()) when file has no bodies. Check this. */ - LASSERT(ergo(ia_valid & ATTR_SIZE, llu_i2info(inode)->lli_smd == NULL)); + LASSERT(ergo(ia_valid & ATTR_SIZE, !llu_i2info(inode)->lli_has_smd)); if (ia_valid & ATTR_SIZE) st->st_size = attr->ia_size; @@ -638,11 +569,12 @@ static int inode_setattr(struct inode * inode, struct iattr * attr) st->st_mtime = attr->ia_mtime; if (ia_valid & ATTR_CTIME) st->st_ctime = attr->ia_ctime; - if (ia_valid & ATTR_MODE) { - st->st_mode = attr->ia_mode; - if (!in_group_p(st->st_gid) && !cfs_capable(CFS_CAP_FSETID)) - st->st_mode &= ~S_ISGID; - } + if (ia_valid & ATTR_MODE) { + st->st_mode = attr->ia_mode; + if (!in_group_p(st->st_gid) && + !cfs_capable(CFS_CAP_FSETID)) + st->st_mode &= ~S_ISGID; + } /* mark_inode_dirty(inode); */ return error; } @@ -704,13 +636,15 @@ static int llu_setattr_done_writing(struct inode *inode, CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid)); - op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE; + op_data->op_flags = MF_EPOCH_CLOSE; + llu_done_writing_attr(inode, op_data); + llu_pack_inode2opdata(inode, op_data, NULL); + rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod); if (rc == -EAGAIN) { /* MDS has instructed us to obtain Size-on-MDS attribute * from OSTs and send setattr to back to MDS. */ - rc = llu_sizeonmds_update(inode, mod, &op_data->op_handle, - op_data->op_ioepoch); + rc = llu_som_update(inode, op_data); } else if (rc) { CERROR("inode %llu mdc truncate failed: rc = %d\n", (unsigned long long)st->st_ino, rc); @@ -733,7 +667,7 @@ static int llu_setattr_done_writing(struct inode *inode, */ int llu_setattr_raw(struct inode *inode, struct iattr *attr) { - struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd; + int has_lsm = llu_i2info(inode)->lli_has_smd; struct intnl_stat *st = llu_i2stat(inode); int ia_valid = attr->ia_valid; struct md_op_data op_data = { { 0 } }; @@ -756,58 +690,45 @@ int llu_setattr_raw(struct inode *inode, struct iattr *attr) /* We mark all of the fields "set" so MDS/OST does not re-set them */ if (attr->ia_valid & ATTR_CTIME) { - attr->ia_ctime = CURRENT_TIME; + attr->ia_ctime = CFS_CURRENT_TIME; attr->ia_valid |= ATTR_CTIME_SET; } if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) { - attr->ia_atime = CURRENT_TIME; + attr->ia_atime = CFS_CURRENT_TIME; attr->ia_valid |= ATTR_ATIME_SET; } if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) { - attr->ia_mtime = CURRENT_TIME; + attr->ia_mtime = CFS_CURRENT_TIME; attr->ia_valid |= ATTR_MTIME_SET; } - if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) { - /* To avoid stale mtime on mds, obtain it from ost and send - to mds. */ - rc = cl_glimpse_size(inode); - if (rc) - RETURN(rc); - - attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME; - attr->ia_mtime = inode->i_stbuf.st_mtime; - } if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME)) CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T ", now = "CFS_TIME_T"\n", - LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime), - LTIME_S(CURRENT_TIME)); - - /* NB: ATTR_SIZE will only be set after this point if the size - * resides on the MDS, ie, this file has no objects. */ - if (lsm) - attr->ia_valid &= ~ATTR_SIZE; - - /* If only OST attributes being set on objects, don't do MDS RPC. - * In that case, we need to check permissions and update the local - * inode ourselves so we can call obdo_from_inode() always. */ - if (ia_valid & (lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) { + LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime), + LTIME_S(CFS_CURRENT_TIME)); + + /* NB: ATTR_SIZE will only be set after this point if the size + * resides on the MDS, ie, this file has no objects. */ + if (has_lsm) + attr->ia_valid &= ~ATTR_SIZE; + + /* If only OST attributes being set on objects, don't do MDS RPC. + * In that case, we need to check permissions and update the local + * inode ourselves so we can call obdo_from_inode() always. */ + if (ia_valid & (has_lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) { memcpy(&op_data.op_attr, attr, sizeof(*attr)); /* Open epoch for truncate. */ - if (ia_valid & ATTR_SIZE) + if (exp_connect_som(llu_i2mdexp(inode)) && + (ia_valid & ATTR_SIZE)) op_data.op_flags = MF_EPOCH_OPEN; rc = llu_md_setattr(inode, &op_data, &mod); if (rc) RETURN(rc); - if (op_data.op_ioepoch) - CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for " - "truncate\n", op_data.op_ioepoch, - PFID(&llu_i2info(inode)->lli_fid)); - - if (!lsm || !S_ISREG(st->st_mode)) { + llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch); + if (!has_lsm || !S_ISREG(st->st_mode)) { CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n"); GOTO(out, rc); } @@ -836,13 +757,18 @@ int llu_setattr_raw(struct inode *inode, struct iattr *attr) inode_setattr(inode, attr); } - if (ia_valid & ATTR_SIZE) { - rc = cl_setattr_do_truncate(inode, attr->ia_size, NULL); - } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) { - CDEBUG(D_INODE, "set mtime on OST inode %llu to %lu\n", - (long long unsigned)st->st_ino, LTIME_S(attr->ia_mtime)); - rc = cl_setattr_ost(inode, NULL); - } + if (ia_valid & ATTR_SIZE) + attr->ia_valid |= ATTR_SIZE; + if (ia_valid & (ATTR_SIZE | + ATTR_ATIME | ATTR_ATIME_SET | + ATTR_MTIME | ATTR_MTIME_SET)) + /* on truncate and utimes send attributes to osts, setting + * mtime/atime to past will be performed under PW 0:EOF extent + * lock (new_size:EOF for truncate) + * it may seem excessive to send mtime/atime updates to osts + * when not setting times to past, but it is necessary due to + * possible time de-synchronization */ + rc = cl_setattr_ost(inode, attr, NULL); EXIT; out: if (op_data.op_ioepoch) @@ -895,7 +821,7 @@ static int llu_iop_setattr(struct pnode *pno, } iattr.ia_valid |= ATTR_RAW | ATTR_CTIME; - iattr.ia_ctime = CURRENT_TIME; + iattr.ia_ctime = CFS_CURRENT_TIME; rc = llu_setattr_raw(ino, &iattr); liblustre_wait_idle(); @@ -939,6 +865,7 @@ static int llu_readlink_internal(struct inode *inode, struct llu_sb_info *sbi = llu_i2sbi(inode); struct mdt_body *body; struct intnl_stat *st = llu_i2stat(inode); + struct md_op_data op_data = {{ 0 }}; int rc, symlen = st->st_size + 1; ENTRY; @@ -951,8 +878,11 @@ static int llu_readlink_internal(struct inode *inode, RETURN(0); } - rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), NULL, - OBD_MD_LINKNAME, symlen, request); + llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen, + LUSTRE_OPC_ANY); + op_data.op_valid = OBD_MD_LINKNAME; + + rc = md_getattr(sbi->ll_md_exp, &op_data, request); if (rc) { CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc); RETURN(rc); @@ -1157,7 +1087,7 @@ static int llu_statfs_internal(struct llu_sb_info *sbi, int rc; ENTRY; - rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, 0); + rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, 0); if (rc) { CERROR("md_statfs fails: rc = %d\n", rc); RETURN(rc); @@ -1204,7 +1134,8 @@ static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs) /* For now we will always get up-to-date statfs values, but in the * future we may allow some amount of caching on the client (e.g. * from QOS or lprocfs updates). */ - rc = llu_statfs_internal(sbi, &osfs, cfs_time_current_64() - HZ); + rc = llu_statfs_internal(sbi, &osfs, + cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS)); if (rc) return rc; @@ -1329,20 +1260,23 @@ static int llu_file_flock(struct inode *ino, int cmd, struct file_lock *file_lock) { - struct llu_inode_info *lli = llu_i2info(ino); - struct intnl_stat *st = llu_i2stat(ino); - struct ldlm_res_id res_id = - { .name = {fid_seq(&lli->lli_fid), - fid_oid(&lli->lli_fid), - fid_ver(&lli->lli_fid), - LDLM_FLOCK} }; - struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL, - ldlm_flock_completion_ast, NULL, NULL, file_lock }; - - struct lustre_handle lockh = {0}; - ldlm_policy_data_t flock; - int flags = 0; - int rc; + struct llu_inode_info *lli = llu_i2info(ino); + struct ldlm_res_id res_id = + { .name = {fid_seq(&lli->lli_fid), + fid_oid(&lli->lli_fid), + fid_ver(&lli->lli_fid), + LDLM_FLOCK} }; + struct ldlm_enqueue_info einfo = { + .ei_type = LDLM_FLOCK, + .ei_mode = 0, + .ei_cb_cp = ldlm_flock_completion_ast, + .ei_cbdata = file_lock, + }; + struct intnl_stat *st = llu_i2stat(ino); + struct lustre_handle lockh = {0}; + ldlm_policy_data_t flock; + __u64 flags = 0; + int rc; CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n", (unsigned long long)st->st_ino, file_lock); @@ -1397,7 +1331,7 @@ static int llu_file_flock(struct inode *ino, LBUG(); } - CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#x, mode=%u, " + CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#llx, mode=%u, " "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino, flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start, flock.l_flock.end); @@ -1411,11 +1345,12 @@ static int llu_file_flock(struct inode *ino, if (lmv->desc.ld_tgt_count < 1) RETURN(rc = -ENODEV); - if (lmv->tgts[0].ltd_exp != NULL) - rc = ldlm_cli_enqueue(lmv->tgts[0].ltd_exp, NULL, &einfo, &res_id, - &flock, &flags, NULL, 0, NULL, &lockh, 0); - else - rc = -ENODEV; + if (lmv->tgts[0] != NULL && lmv->tgts[0]->ltd_exp != NULL) + rc = ldlm_cli_enqueue(lmv->tgts[0]->ltd_exp, NULL, + &einfo, &res_id, &flock, &flags, + NULL, 0, LVB_T_NONE, &lockh, 0); + else + rc = -ENODEV; } RETURN(rc); } @@ -1602,56 +1537,51 @@ static int llu_get_grouplock(struct inode *inode, unsigned long arg) { struct llu_inode_info *lli = llu_i2info(inode); struct ll_file_data *fd = lli->lli_file_data; - ldlm_policy_data_t policy = { .l_extent = { .start = 0, - .end = OBD_OBJECT_EOF}}; - struct lustre_handle lockh = { 0 }; - struct lov_stripe_md *lsm = lli->lli_smd; - ldlm_error_t err; - int flags = 0; + int rc; + struct ccc_grouplock grouplock; ENTRY; + if (fd->fd_flags & LL_FILE_IGNORE_LOCK) { + RETURN(-ENOTSUPP); + } if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { RETURN(-EINVAL); } + LASSERT(fd->fd_grouplock.cg_lock == NULL); - policy.l_extent.gid = arg; - if (lli->lli_open_flags & O_NONBLOCK) - flags = LDLM_FL_BLOCK_NOWAIT; + rc = cl_get_grouplock(cl_i2info(inode)->lli_clob, + arg, (lli->lli_open_flags & O_NONBLOCK), + &grouplock); - err = llu_extent_lock(fd, inode, lsm, LCK_GROUP, &policy, &lockh, - flags); - if (err) - RETURN(err); + if (rc) + RETURN(rc); - fd->fd_flags |= LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK; - fd->fd_gid = arg; - memcpy(&fd->fd_cwlockh, &lockh, sizeof(lockh)); + fd->fd_flags |= LL_FILE_GROUP_LOCKED; + fd->fd_grouplock = grouplock; RETURN(0); } -static int llu_put_grouplock(struct inode *inode, unsigned long arg) +int llu_put_grouplock(struct inode *inode, unsigned long arg) { struct llu_inode_info *lli = llu_i2info(inode); struct ll_file_data *fd = lli->lli_file_data; - struct lov_stripe_md *lsm = lli->lli_smd; - ldlm_error_t err; + struct ccc_grouplock grouplock; ENTRY; if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) RETURN(-EINVAL); - if (fd->fd_gid != arg) - RETURN(-EINVAL); + LASSERT(fd->fd_grouplock.cg_lock != NULL); - fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK); + if (fd->fd_grouplock.cg_gid != arg) + RETURN(-EINVAL); - err = llu_extent_unlock(fd, inode, lsm, LCK_GROUP, &fd->fd_cwlockh); - if (err) - RETURN(err); + grouplock = fd->fd_grouplock; + memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); + fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; - fd->fd_gid = 0; - memset(&fd->fd_cwlockh, 0, sizeof(fd->fd_cwlockh)); + cl_put_grouplock(&grouplock); RETURN(0); } @@ -1670,7 +1600,7 @@ static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg) LASSERT(sizeof(lum) == sizeof(*lump)); LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0])); - if (copy_from_user(&lum, lump, sizeof(lum))) + if (copy_from_user(&lum, lump, sizeof(lum))) return(-EFAULT); switch (lum.lmm_magic) { @@ -1708,45 +1638,32 @@ static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg) } static int llu_lov_setstripe_ea_info(struct inode *ino, int flags, - struct lov_user_md *lum, int lum_size) + struct lov_user_md *lum, int lum_size) { - struct llu_sb_info *sbi = llu_i2sbi(ino); - struct llu_inode_info *lli = llu_i2info(ino); - struct llu_inode_info *lli2 = NULL; - struct lov_stripe_md *lsm; - struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; - struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR, - llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL }; - - struct ptlrpc_request *req = NULL; - struct lustre_md md; - struct md_op_data data = {{ 0 }}; - struct lustre_handle lockh; - int rc = 0; - ENTRY; - - lsm = lli->lli_smd; - if (lsm) { - CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n", - PFID(&lli->lli_fid)); - return -EEXIST; - } - - OBD_ALLOC(lli2, sizeof(struct llu_inode_info)); - if (!lli2) - return -ENOMEM; - - memcpy(lli2, lli, sizeof(struct llu_inode_info)); - lli2->lli_open_count = 0; - lli2->lli_it = NULL; - lli2->lli_file_data = NULL; - lli2->lli_smd = NULL; - lli2->lli_symlink_name = NULL; - ino->i_private = lli2; + struct llu_sb_info *sbi = llu_i2sbi(ino); + struct llu_inode_info *lli = llu_i2info(ino); + struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; + struct ldlm_enqueue_info einfo = { + .ei_type = LDLM_IBITS, + .ei_mode = LCK_CR, + .ei_cb_bl = llu_md_blocking_ast, + .ei_cb_cp = ldlm_completion_ast, + }; + struct ptlrpc_request *req = NULL; + struct lustre_md md; + struct md_op_data data = {{ 0 }}; + struct lustre_handle lockh; + int rc = 0; + ENTRY; + + if (lli->lli_has_smd) { + CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n", + PFID(&lli->lli_fid)); + return -EEXIST; + } llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR, LUSTRE_OPC_ANY); - rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data, &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY); if (rc) @@ -1770,28 +1687,20 @@ static int llu_lov_setstripe_ea_info(struct inode *ino, int flags, if (rc) GOTO(out, rc); - lli->lli_smd = lli2->lli_smd; - lli2->lli_smd = NULL; - - llu_local_open(lli2, &oit); - + llu_update_inode(ino, &md); + llu_local_open(lli, &oit); /* release intent */ if (lustre_handle_is_used(&lockh)) ldlm_lock_decref(&lockh, LCK_CR); - ptlrpc_req_finished(req); req = NULL; - rc = llu_file_release(ino); - out: - ino->i_private = lli; - if (!rc) - llu_update_inode(ino, &md); - if (lli2) - OBD_FREE(lli2, sizeof(struct llu_inode_info)); + EXIT; + +out: if (req != NULL) ptlrpc_req_finished(req); - RETURN(rc); + return rc; } static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg) @@ -1803,7 +1712,7 @@ static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg) LASSERT(sizeof(lum) == sizeof(*lump)); LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0])); - if (copy_from_user(&lum, lump, sizeof(lum))) + if (copy_from_user(&lum, lump, sizeof(lum))) RETURN(-EFAULT); rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum)); @@ -1823,13 +1732,15 @@ static int llu_lov_setstripe(struct inode *ino, unsigned long arg) static int llu_lov_getstripe(struct inode *ino, unsigned long arg) { - struct lov_stripe_md *lsm = llu_i2info(ino)->lli_smd; - - if (!lsm) - RETURN(-ENODATA); - - return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm, - (void *)arg); + struct lov_stripe_md *lsm = NULL; + int rc = -ENODATA; + + lsm = ccc_inode_lsm_get(ino); + if (lsm != NULL) + rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm, + (void *)arg); + ccc_inode_lsm_put(ino, lsm); + return rc; } static int llu_iop_ioctl(struct inode *ino, unsigned long int request, @@ -1940,12 +1851,14 @@ llu_fsswop_mount(const char *source, struct lustre_md md; class_uuid_t uuid; struct config_llog_instance cfg = {0, }; - char ll_instance[sizeof(sbi) * 2 + 1]; struct lustre_profile *lprof; char *zconf_mgsnid, *zconf_profile; char *osc = NULL, *mdc = NULL; int async = 1, err = -EINVAL; struct obd_connect_data ocd = {0,}; + struct md_op_data op_data = {{0}}; + /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */ + const int instlen = sizeof(cfg.cfg_instance) * 2 + 2; ENTRY; @@ -1970,10 +1883,9 @@ llu_fsswop_mount(const char *source, /* generate a string unique to this super, let's try the address of the super itself.*/ - sprintf(ll_instance, "%p", sbi); + cfg.cfg_instance = sbi; /* retrive & parse config log */ - cfg.cfg_instance = ll_instance; cfg.cfg_uuid = sbi->ll_sb_uuid; err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1); if (err < 0) { @@ -1986,11 +1898,11 @@ llu_fsswop_mount(const char *source, CERROR("No profile found: %s\n", zconf_profile); GOTO(out_free, err = -EINVAL); } - OBD_ALLOC(osc, strlen(lprof->lp_dt) + strlen(ll_instance) + 2); - sprintf(osc, "%s-%s", lprof->lp_dt, ll_instance); + OBD_ALLOC(osc, strlen(lprof->lp_dt) + instlen + 2); + sprintf(osc, "%s-%p", lprof->lp_dt, cfg.cfg_instance); - OBD_ALLOC(mdc, strlen(lprof->lp_md) + strlen(ll_instance) + 2); - sprintf(mdc, "%s-%s", lprof->lp_md, ll_instance); + OBD_ALLOC(mdc, strlen(lprof->lp_md) + instlen + 2); + sprintf(mdc, "%s-%p", lprof->lp_md, cfg.cfg_instance); if (!osc) { CERROR("no osc\n"); @@ -2012,11 +1924,14 @@ llu_fsswop_mount(const char *source, CERROR("MDC %s: not setup or attached\n", mdc); GOTO(out_free, err = -EINVAL); } - obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC, - sizeof(async), &async, NULL); + obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC), + KEY_ASYNC, sizeof(async), &async, NULL); ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION | - OBD_CONNECT_FID | OBD_CONNECT_AT; + OBD_CONNECT_FID | OBD_CONNECT_AT | + OBD_CONNECT_VBR | OBD_CONNECT_FULL20 | + OBD_CONNECT_LVB_TYPE; + #ifdef LIBLUSTRE_POSIX_ACL ocd.ocd_connect_flags |= OBD_CONNECT_ACL; #endif @@ -2030,7 +1945,7 @@ llu_fsswop_mount(const char *source, GOTO(out_free, err); } - err = obd_statfs(obd, &osfs, 100000000, 0); + err = obd_statfs(NULL, sbi->ll_md_exp, &osfs, 100000000, 0); if (err) GOTO(out_md, err); @@ -2044,15 +1959,18 @@ llu_fsswop_mount(const char *source, CERROR("OSC %s: not setup or attached\n", osc); GOTO(out_md, err = -EINVAL); } - obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC, - sizeof(async), &async, NULL); + obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC), + KEY_ASYNC, sizeof(async), &async, NULL); obd->obd_upcall.onu_owner = &sbi->ll_lco; obd->obd_upcall.onu_upcall = cl_ocd_update; ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK | - OBD_CONNECT_FID | OBD_CONNECT_AT; + OBD_CONNECT_FID | OBD_CONNECT_AT | + OBD_CONNECT_FULL20 | OBD_CONNECT_EINPROGRESS | + OBD_CONNECT_LVB_TYPE; + ocd.ocd_version = LUSTRE_VERSION_CODE; err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL); if (err) { @@ -2075,9 +1993,10 @@ llu_fsswop_mount(const char *source, } CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid)); + op_data.op_fid1 = sbi->ll_root_fid; + op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS; /* fetch attr of root inode */ - err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, NULL, - OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS, 0, &request); + err = md_getattr(sbi->ll_md_exp, &op_data, &request); if (err) { CERROR("md_getattr failed for root: rc = %d\n", err); GOTO(out_lock_cn_cb, err); @@ -2118,9 +2037,8 @@ llu_fsswop_mount(const char *source, ptlrpc_req_finished(request); CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source); - liblustre_wait_idle(); - - return 0; + err = 0; + goto out_free; out_inode: _sysio_i_gone(root); @@ -2132,10 +2050,11 @@ out_md: obd_disconnect(sbi->ll_md_exp); out_free: if (osc) - OBD_FREE(osc, strlen(osc) + 1); + OBD_FREE(osc, strlen(lprof->lp_dt) + instlen + 2); if (mdc) - OBD_FREE(mdc, strlen(mdc) + 1); - OBD_FREE(sbi, sizeof(*sbi)); + OBD_FREE(mdc, strlen(lprof->lp_md) + instlen + 2); + if (err != 0) + OBD_FREE(sbi, sizeof(*sbi)); liblustre_wait_idle(); return err; }