X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fliblustre%2Fsuper.c;h=9980b9523e974c0668a209f1eb69178d041220e6;hp=100f985a81e63668c099b3c590a85dea956b67de;hb=8d161d44214f907a9f2d9cf5a79cd2c83de995c3;hpb=00d783456c551aa990e911c075f8cc0db0c45f96 diff --git a/lustre/liblustre/super.c b/lustre/liblustre/super.c index 100f985..9980b95 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/ @@ -40,31 +40,34 @@ #define DEBUG_SUBSYSTEM S_LLITE +#include +#include +#include +#include +#include +#include #include +#include #include -#include #include -#include +#include +#include #include -#include -#include -#ifndef __CYGWIN__ -# include -#else -# include -#endif - -#include -#ifdef HAVE_XTIO_H -#include -#endif -#include -#include -#include -#ifdef HAVE_FILE_H -#include -#endif - +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include #include "llite_lib.h" #ifndef MAY_EXEC @@ -80,10 +83,10 @@ 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 (cfs_curproc_is_in_groups(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; @@ -131,7 +134,7 @@ 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; @@ -158,39 +161,33 @@ 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 (S_ISREG(st->st_mode)) + 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)) st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE); else st->st_blksize = 4096; @@ -206,7 +203,7 @@ 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; @@ -230,114 +227,6 @@ void llu_update_inode(struct inode *inode, struct lustre_md *md) } } -void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid) -{ - struct llu_inode_info *lli = llu_i2info(dst); - struct intnl_stat *st = llu_i2stat(dst); - - valid &= src->o_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", - src->o_valid, - LTIME_S(st->st_mtime), LTIME_S(st->st_ctime), - (long)src->o_mtime, (long)src->o_ctime); - - if (valid & OBD_MD_FLATIME) - LTIME_S(st->st_atime) = src->o_atime; - if (valid & OBD_MD_FLMTIME) - LTIME_S(st->st_mtime) = src->o_mtime; - if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime)) - LTIME_S(st->st_ctime) = src->o_ctime; - if (valid & OBD_MD_FLSIZE) - st->st_size = src->o_size; - if (valid & OBD_MD_FLBLOCKS) /* allocation of space */ - st->st_blocks = src->o_blocks; - if (valid & OBD_MD_FLBLKSZ) - st->st_blksize = src->o_blksize; - if (valid & OBD_MD_FLTYPE) - st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT); - if (valid & OBD_MD_FLMODE) - st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT); - if (valid & OBD_MD_FLUID) - st->st_uid = src->o_uid; - if (valid & OBD_MD_FLGID) - st->st_gid = src->o_gid; - if (valid & OBD_MD_FLFLAGS) - 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; -} - /** * Performs the getattr on the inode and updates its fields. * If @sync != 0, perform the getattr under the server-side lock. @@ -345,19 +234,18 @@ void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid) 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 | @@ -365,6 +253,7 @@ int llu_inode_getattr(struct inode *inode, struct obdo *obdo, OBD_MD_FLBLKSZ | OBD_MD_FLMTIME | 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; @@ -380,6 +269,7 @@ 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); @@ -387,13 +277,13 @@ 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, @@ -423,7 +313,7 @@ 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); @@ -448,7 +338,7 @@ 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); @@ -466,14 +356,10 @@ 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) { - CERROR("REPORT THIS LINE TO PETER\n"); - RETURN(0); - } - if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) { struct lustre_md md; struct ptlrpc_request *req = NULL; @@ -484,7 +370,7 @@ static int llu_inode_revalidate(struct inode *inode) /* 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; } @@ -496,7 +382,7 @@ static int llu_inode_revalidate(struct inode *inode) 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, @@ -518,14 +404,18 @@ 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 */ @@ -579,10 +469,11 @@ static int null_if_equal(struct ldlm_lock *lock, void *data) return LDLM_ITER_CONTINUE; } -void llu_clear_inode(struct inode *inode) +static 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", @@ -590,19 +481,15 @@ 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, @@ -613,7 +500,7 @@ void llu_clear_inode(struct inode *inode) EXIT; } -void llu_iop_gone(struct inode *inode) +static void llu_iop_gone(struct inode *inode) { struct llu_inode_info *lli = llu_i2info(inode); ENTRY; @@ -635,7 +522,7 @@ 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; @@ -649,12 +536,12 @@ 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 (!cfs_curproc_is_in_groups(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; } @@ -747,7 +634,7 @@ 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 } }; @@ -781,32 +668,22 @@ int llu_setattr_raw(struct inode *inode, struct iattr *attr) 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(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 (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. */ @@ -818,7 +695,7 @@ int llu_setattr_raw(struct inode *inode, struct iattr *attr) RETURN(rc); llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch); - if (!lsm || !S_ISREG(st->st_mode)) { + if (!has_lsm || !S_ISREG(st->st_mode)) { CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n"); GOTO(out, rc); } @@ -847,13 +724,18 @@ 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) @@ -1172,7 +1054,7 @@ 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); @@ -1219,7 +1101,8 @@ 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() - CFS_HZ); + rc = llu_statfs_internal(sbi, &osfs, + cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS)); if (rc) return rc; @@ -1344,20 +1227,23 @@ 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); @@ -1412,11 +1298,11 @@ static int llu_file_flock(struct inode *ino, LBUG(); } - CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#x, 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); - + CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags="LPX64", " + "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); { struct lmv_obd *lmv; struct obd_device *lmv_obd; @@ -1426,11 +1312,12 @@ 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, &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); } @@ -1680,7 +1567,7 @@ 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 (cfs_copy_from_user(&lum, lump, sizeof(lum))) + if (copy_from_user(&lum, lump, sizeof(lum))) return(-EFAULT); switch (lum.lmm_magic) { @@ -1718,30 +1605,34 @@ 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 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; - - if (lli->lli_smd) { - CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n", - PFID(&lli->lli_fid)); - return -EEXIST; - } + 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); + rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &oit, &data, &lockh, + LDLM_FL_INTENT_ONLY); if (rc) GOTO(out, rc); @@ -1788,7 +1679,7 @@ 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 (cfs_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)); @@ -1808,13 +1699,15 @@ 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, @@ -1867,9 +1760,8 @@ static int llu_iop_datasync(struct inode *inode) return 0; } -struct filesys_ops llu_filesys_ops = -{ - fsop_gone: llu_fsop_gone, +static struct filesys_ops llu_filesys_ops = { + .fsop_gone = llu_fsop_gone, }; struct inode *llu_iget(struct filesys *fs, struct lustre_md *md) @@ -1925,13 +1817,14 @@ 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; @@ -1956,10 +1849,9 @@ 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) { @@ -1972,11 +1864,11 @@ 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"); @@ -1998,12 +1890,14 @@ 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_VBR | OBD_CONNECT_SOM; + OBD_CONNECT_VBR | OBD_CONNECT_FULL20 | + OBD_CONNECT_LVB_TYPE; + #ifdef LIBLUSTRE_POSIX_ACL ocd.ocd_connect_flags |= OBD_CONNECT_ACL; #endif @@ -2017,7 +1911,7 @@ 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); @@ -2031,8 +1925,8 @@ 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; @@ -2040,7 +1934,8 @@ llu_fsswop_mount(const char *source, ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_FID | OBD_CONNECT_AT | - OBD_CONNECT_SOM; + 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); @@ -2108,9 +2003,8 @@ 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); @@ -2122,43 +2016,44 @@ 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; } struct fssw_ops llu_fssw_ops = { - llu_fsswop_mount + .fsswop_mount = llu_fsswop_mount, }; static struct inode_ops llu_inode_ops = { - inop_lookup: llu_iop_lookup, - inop_getattr: llu_iop_getattr, - inop_setattr: llu_iop_setattr, - inop_filldirentries: llu_iop_filldirentries, - inop_mkdir: llu_iop_mkdir_raw, - inop_rmdir: llu_iop_rmdir_raw, - inop_symlink: llu_iop_symlink_raw, - inop_readlink: llu_iop_readlink, - inop_open: llu_iop_open, - inop_close: llu_iop_close, - inop_link: llu_iop_link_raw, - inop_unlink: llu_iop_unlink_raw, - inop_rename: llu_iop_rename_raw, - inop_pos: llu_iop_pos, - inop_read: llu_iop_read, - inop_write: llu_iop_write, - inop_iodone: llu_iop_iodone, - inop_fcntl: llu_iop_fcntl, - inop_sync: llu_iop_sync, - inop_datasync: llu_iop_datasync, - inop_ioctl: llu_iop_ioctl, - inop_mknod: llu_iop_mknod_raw, + .inop_lookup = llu_iop_lookup, + .inop_getattr = llu_iop_getattr, + .inop_setattr = llu_iop_setattr, + .inop_filldirentries = llu_iop_filldirentries, + .inop_mkdir = llu_iop_mkdir_raw, + .inop_rmdir = llu_iop_rmdir_raw, + .inop_symlink = llu_iop_symlink_raw, + .inop_readlink = llu_iop_readlink, + .inop_open = llu_iop_open, + .inop_close = llu_iop_close, + .inop_link = llu_iop_link_raw, + .inop_unlink = llu_iop_unlink_raw, + .inop_rename = llu_iop_rename_raw, + .inop_pos = llu_iop_pos, + .inop_read = llu_iop_read, + .inop_write = llu_iop_write, + .inop_iodone = llu_iop_iodone, + .inop_fcntl = llu_iop_fcntl, + .inop_sync = llu_iop_sync, + .inop_datasync = llu_iop_datasync, + .inop_ioctl = llu_iop_ioctl, + .inop_mknod = llu_iop_mknod_raw, #ifdef _HAVE_STATVFS - inop_statvfs: llu_iop_statvfs, + .inop_statvfs = llu_iop_statvfs, #endif - inop_gone: llu_iop_gone, + .inop_gone = llu_iop_gone, };