X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Ffile.c;h=14e38cf7050c35306ea957d2579feb1ce9498d90;hp=34c08cc8469725e495e7554411f4abc113d84700;hb=a7811634d45319dc3812fa652c727367fa290a24;hpb=4614c11ecf8d61925b6dd17f1e6f77a4f8450297 diff --git a/lustre/llite/file.c b/lustre/llite/file.c index 34c08cc..14e38cf 100644 --- a/lustre/llite/file.c +++ b/lustre/llite/file.c @@ -27,7 +27,7 @@ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Whamcloud, Inc. + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -81,6 +81,9 @@ void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data, if (fh) op_data->op_handle = *fh; op_data->op_capa1 = ll_mdscapa_get(inode); + + if (LLIF_DATA_MODIFIED & ll_i2info(inode)->lli_flags) + op_data->op_bias |= MDS_DATA_MODIFIED; } /** @@ -154,6 +157,17 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp, CERROR("inode %lu mdc close failed: rc = %d\n", inode->i_ino, rc); } + + /* DATA_MODIFIED flag was successfully sent on close, cancel data + * modification flag. */ + if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) { + struct ll_inode_info *lli = ll_i2info(inode); + + spin_lock(&lli->lli_lock); + lli->lli_flags &= ~LLIF_DATA_MODIFIED; + spin_unlock(&lli->lli_lock); + } + ll_finish_md_op_data(op_data); if (rc == 0) { @@ -201,15 +215,15 @@ int ll_md_real_close(struct inode *inode, int flags) och_usecount = &lli->lli_open_fd_read_count; } - cfs_mutex_lock(&lli->lli_och_mutex); + mutex_lock(&lli->lli_och_mutex); if (*och_usecount) { /* There are still users of this handle, so skip freeing it. */ - cfs_mutex_unlock(&lli->lli_och_mutex); + mutex_unlock(&lli->lli_och_mutex); RETURN(0); } och=*och_p; *och_p = NULL; - cfs_mutex_unlock(&lli->lli_och_mutex); + mutex_unlock(&lli->lli_och_mutex); if (och) { /* There might be a race and somebody have freed this och already */ @@ -241,7 +255,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, struct inode *inode = file->f_dentry->d_inode; ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}}; - cfs_mutex_lock(&lli->lli_och_mutex); + mutex_lock(&lli->lli_och_mutex); if (fd->fd_omode & FMODE_WRITE) { lockmode = LCK_CW; LASSERT(lli->lli_open_fd_write_count); @@ -255,7 +269,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, LASSERT(lli->lli_open_fd_read_count); lli->lli_open_fd_read_count--; } - cfs_mutex_unlock(&lli->lli_och_mutex); + mutex_unlock(&lli->lli_och_mutex); if (!md_lock_match(md_exp, flags, ll_inode2fid(inode), LDLM_IBITS, &policy, lockmode, @@ -397,7 +411,7 @@ static int ll_intent_file_open(struct file *file, void *lmm, GOTO(out, rc); } - rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL); + rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL, itp); if (!rc && itp->d.lustre.it_lock_mode) ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode, itp, NULL); @@ -513,14 +527,14 @@ int ll_file_open(struct inode *inode, struct file *file) fd->fd_file = file; if (S_ISDIR(inode->i_mode)) { - cfs_spin_lock(&lli->lli_sa_lock); - if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL && - lli->lli_opendir_pid == 0) { - lli->lli_opendir_key = fd; - lli->lli_opendir_pid = cfs_curproc_pid(); - opendir_set = 1; - } - cfs_spin_unlock(&lli->lli_sa_lock); + spin_lock(&lli->lli_sa_lock); + if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL && + lli->lli_opendir_pid == 0) { + lli->lli_opendir_key = fd; + lli->lli_opendir_pid = cfs_curproc_pid(); + opendir_set = 1; + } + spin_unlock(&lli->lli_sa_lock); } if (inode->i_sb->s_root == file->f_dentry) { @@ -570,14 +584,14 @@ restart: och_usecount = &lli->lli_open_fd_read_count; } - cfs_mutex_lock(&lli->lli_och_mutex); + mutex_lock(&lli->lli_och_mutex); if (*och_p) { /* Open handle is present */ if (it_disposition(it, DISP_OPEN_OPEN)) { /* Well, there's extra open request that we do not need, let's close it somehow. This will decref request. */ rc = it_open_error(DISP_OPEN_OPEN, it); if (rc) { - cfs_mutex_unlock(&lli->lli_och_mutex); + mutex_unlock(&lli->lli_och_mutex); GOTO(out_openerr, rc); } @@ -588,7 +602,7 @@ restart: rc = ll_local_open(file, it, fd, NULL); if (rc) { (*och_usecount)--; - cfs_mutex_unlock(&lli->lli_och_mutex); + mutex_unlock(&lli->lli_och_mutex); GOTO(out_openerr, rc); } } else { @@ -599,7 +613,7 @@ restart: could be cancelled, and since blocking ast handler would attempt to grab och_mutex as well, that would result in a deadlock */ - cfs_mutex_unlock(&lli->lli_och_mutex); + mutex_unlock(&lli->lli_och_mutex); it->it_create_mode |= M_CHECK_STALE; rc = ll_intent_file_open(file, NULL, 0, it); it->it_create_mode &= ~M_CHECK_STALE; @@ -629,7 +643,7 @@ restart: if (rc) GOTO(out_och_free, rc); } - cfs_mutex_unlock(&lli->lli_och_mutex); + mutex_unlock(&lli->lli_och_mutex); fd = NULL; /* Must do this outside lli_och_mutex lock to prevent deadlock where @@ -651,18 +665,13 @@ restart: GOTO(out_och_free, rc); out_och_free: - if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) { - ptlrpc_req_finished(it->d.lustre.it_data); - it_clear_disposition(it, DISP_ENQ_OPEN_REF); - } - if (rc) { if (och_p && *och_p) { OBD_FREE(*och_p, sizeof (struct obd_client_handle)); *och_p = NULL; /* OBD_FREE writes some magic there */ (*och_usecount)--; } - cfs_mutex_unlock(&lli->lli_och_mutex); + mutex_unlock(&lli->lli_och_mutex); out_openerr: if (opendir_set != 0) @@ -673,6 +682,11 @@ out_openerr: ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); } + if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) { + ptlrpc_req_finished(it->d.lustre.it_data); + it_clear_disposition(it, DISP_ENQ_OPEN_REF); + } + return rc; } @@ -691,8 +705,7 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, oinfo.oi_md = lsm; oinfo.oi_oa = obdo; - oinfo.oi_oa->o_id = lsm->lsm_object_id; - oinfo.oi_oa->o_seq = lsm->lsm_object_seq; + 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 | @@ -742,10 +755,11 @@ int ll_inode_getattr(struct inode *inode, struct obdo *obdo, capa, obdo, ioepoch, sync); capa_put(capa); if (rc == 0) { + struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi; + obdo_refresh_inode(inode, obdo, obdo->o_valid); - CDEBUG(D_INODE, - "objid "LPX64" size %llu, blocks %llu, blksize %lu\n", - lsm ? lsm->lsm_object_id : 0, i_size_read(inode), + CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu," + " blksize %lu\n", POSTID(oi), i_size_read(inode), (unsigned long long)inode->i_blocks, (unsigned long)ll_inode_blksize(inode)); } @@ -753,39 +767,47 @@ int ll_inode_getattr(struct inode *inode, struct obdo *obdo, RETURN(rc); } -int ll_merge_lvb(struct inode *inode) +int ll_merge_lvb(const struct lu_env *env, struct inode *inode) { struct ll_inode_info *lli = ll_i2info(inode); - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct lov_stripe_md *lsm; + struct cl_object *obj = lli->lli_clob; + struct cl_attr *attr = ccc_env_thread_attr(env); struct ost_lvb lvb; int rc = 0; ENTRY; - lsm = ccc_inode_lsm_get(inode); ll_inode_size_lock(inode); + /* merge timestamps the most recently obtained from mds with + timestamps obtained from osts */ + LTIME_S(inode->i_atime) = lli->lli_lvb.lvb_atime; + LTIME_S(inode->i_mtime) = lli->lli_lvb.lvb_mtime; + LTIME_S(inode->i_ctime) = lli->lli_lvb.lvb_ctime; inode_init_lvb(inode, &lvb); - /* merge timestamps the most resently obtained from mds with - timestamps obtained from osts */ - lvb.lvb_atime = lli->lli_lvb.lvb_atime; - lvb.lvb_mtime = lli->lli_lvb.lvb_mtime; - lvb.lvb_ctime = lli->lli_lvb.lvb_ctime; - if (lsm != NULL) { - rc = obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 0); - cl_isize_write_nolock(inode, lvb.lvb_size); + cl_object_attr_lock(obj); + rc = cl_object_attr_get(env, obj, attr); + cl_object_attr_unlock(obj); + + if (rc == 0) { + if (lvb.lvb_atime < attr->cat_atime) + lvb.lvb_atime = attr->cat_atime; + if (lvb.lvb_ctime < attr->cat_ctime) + lvb.lvb_ctime = attr->cat_ctime; + if (lvb.lvb_mtime < attr->cat_mtime) + lvb.lvb_mtime = attr->cat_mtime; CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n", - PFID(&lli->lli_fid), lvb.lvb_size); - inode->i_blocks = lvb.lvb_blocks; + PFID(&lli->lli_fid), attr->cat_size); + cl_isize_write_nolock(inode, attr->cat_size); + + inode->i_blocks = attr->cat_blocks; LTIME_S(inode->i_mtime) = lvb.lvb_mtime; LTIME_S(inode->i_atime) = lvb.lvb_atime; LTIME_S(inode->i_ctime) = lvb.lvb_ctime; } ll_inode_size_unlock(inode); - ccc_inode_lsm_put(inode, lsm); RETURN(rc); } @@ -814,7 +836,9 @@ void ll_io_init(struct cl_io *io, const struct file *file, int write) io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK; if (write) { io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); - io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || IS_SYNC(inode); + io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || + file->f_flags & O_DIRECT || + IS_SYNC(inode); } io->ci_obj = ll_i2info(inode)->lli_clob; io->ci_lockreq = CILR_MAYBE; @@ -837,6 +861,7 @@ ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, ssize_t result; ENTRY; +restart: io = ccc_env_thread_io(env); ll_io_init(io, file, iot == CIT_WRITE); @@ -858,12 +883,12 @@ ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, #endif if ((iot == CIT_WRITE) && !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { - if (cfs_mutex_lock_interruptible(&lli-> + if (mutex_lock_interruptible(&lli-> lli_write_mutex)) GOTO(out, result = -ERESTARTSYS); write_mutex_locked = 1; } else if (iot == CIT_READ) { - cfs_down_read(&lli->lli_trunc_sem); + down_read(&lli->lli_trunc_sem); } break; case IO_SENDFILE: @@ -880,9 +905,9 @@ ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, } result = cl_io_loop(env, io); if (write_mutex_locked) - cfs_mutex_unlock(&lli->lli_write_mutex); + mutex_unlock(&lli->lli_write_mutex); else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ) - cfs_up_read(&lli->lli_trunc_sem); + up_read(&lli->lli_trunc_sem); } else { /* cl_io_rw_init() handled IO */ result = io->ci_result; @@ -895,6 +920,16 @@ ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, GOTO(out, result); out: cl_io_fini(env, io); + /* If any bit been read/written (result != 0), we just return + * short read/write instead of restart io. */ + if (result == 0 && io->ci_need_restart) { + CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n", + iot == CIT_READ ? "read" : "write", + file->f_dentry->d_name.name, *ppos, count); + LASSERTF(io->u.ci_rw.crw_count == count, "%zd != %zd\n", + io->u.ci_rw.crw_count, count); + goto restart; + } if (iot == CIT_READ) { if (result >= 0) @@ -905,7 +940,7 @@ out: ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode), LPROC_LL_WRITE_BYTES, result); fd->fd_write_failed = false; - } else { + } else if (result != -ERESTARTSYS) { fd->fd_write_failed = true; } } @@ -1272,37 +1307,36 @@ out: static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg) { - struct ll_recreate_obj ucreat; - ENTRY; + struct ll_recreate_obj ucreat; + ENTRY; - if (!cfs_capable(CFS_CAP_SYS_ADMIN)) - RETURN(-EPERM); + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + RETURN(-EPERM); - if (cfs_copy_from_user(&ucreat, (struct ll_recreate_obj *)arg, - sizeof(struct ll_recreate_obj))) - RETURN(-EFAULT); + if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg, + sizeof(ucreat))) + RETURN(-EFAULT); - RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0, - ucreat.lrc_ost_idx)); + RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0, + ucreat.lrc_ost_idx)); } static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg) { - struct lu_fid fid; - obd_id id; - obd_count ost_idx; + struct lu_fid fid; + obd_id id; + obd_count ost_idx; ENTRY; - if (!cfs_capable(CFS_CAP_SYS_ADMIN)) - RETURN(-EPERM); + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + RETURN(-EPERM); - if (cfs_copy_from_user(&fid, (struct lu_fid *)arg, - sizeof(struct lu_fid))) - RETURN(-EFAULT); + if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid))) + RETURN(-EFAULT); - id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32); - ost_idx = (fid_seq(&fid) >> 16) & 0xffff; - RETURN(ll_lov_recreate(inode, id, 0, ost_idx)); + id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32); + ost_idx = (fid_seq(&fid) >> 16) & 0xffff; + RETURN(ll_lov_recreate(inode, id, 0, ost_idx)); } int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file, @@ -1422,58 +1456,61 @@ out: static int ll_lov_setea(struct inode *inode, struct file *file, unsigned long arg) { - int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE; - struct lov_user_md *lump; - int lum_size = sizeof(struct lov_user_md) + - sizeof(struct lov_user_ost_data); - int rc; - ENTRY; + int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE; + struct lov_user_md *lump; + int lum_size = sizeof(struct lov_user_md) + + sizeof(struct lov_user_ost_data); + int rc; + ENTRY; - if (!cfs_capable(CFS_CAP_SYS_ADMIN)) - RETURN(-EPERM); + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + RETURN(-EPERM); - OBD_ALLOC_LARGE(lump, lum_size); - if (lump == NULL) { + OBD_ALLOC_LARGE(lump, lum_size); + if (lump == NULL) RETURN(-ENOMEM); - } - if (cfs_copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) { - OBD_FREE_LARGE(lump, lum_size); - RETURN(-EFAULT); - } - rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size); + if (copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) { + OBD_FREE_LARGE(lump, lum_size); + RETURN(-EFAULT); + } - OBD_FREE_LARGE(lump, lum_size); - RETURN(rc); + rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size); + + OBD_FREE_LARGE(lump, lum_size); + RETURN(rc); } static int ll_lov_setstripe(struct inode *inode, struct file *file, - unsigned long arg) -{ - struct lov_user_md_v3 lumv3; - struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; - struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg; - struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg; - int lum_size; - int rc; - int flags = FMODE_WRITE; - ENTRY; + unsigned long arg) +{ + struct lov_user_md_v3 lumv3; + struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; + struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg; + struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg; + int lum_size, rc; + int flags = FMODE_WRITE; + ENTRY; - /* first try with v1 which is smaller than v3 */ - lum_size = sizeof(struct lov_user_md_v1); - if (cfs_copy_from_user(lumv1, lumv1p, lum_size)) - RETURN(-EFAULT); + /* first try with v1 which is smaller than v3 */ + lum_size = sizeof(struct lov_user_md_v1); + if (copy_from_user(lumv1, lumv1p, lum_size)) + RETURN(-EFAULT); - if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { - lum_size = sizeof(struct lov_user_md_v3); - if (cfs_copy_from_user(&lumv3, lumv3p, lum_size)) - RETURN(-EFAULT); - } + if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { + lum_size = sizeof(struct lov_user_md_v3); + if (copy_from_user(&lumv3, lumv3p, lum_size)) + RETURN(-EFAULT); + } - rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size); - if (rc == 0) { + rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size); + if (rc == 0) { struct lov_stripe_md *lsm; + __u32 gen; + put_user(0, &lumv1p->lmm_stripe_count); + + ll_layout_refresh(inode, &gen); lsm = ccc_inode_lsm_get(inode); rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, lsm, (void *)arg); @@ -1507,47 +1544,47 @@ int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) if (ll_file_nolock(file)) RETURN(-EOPNOTSUPP); - cfs_spin_lock(&lli->lli_lock); - if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { - CWARN("group lock already existed with gid %lu\n", - fd->fd_grouplock.cg_gid); - cfs_spin_unlock(&lli->lli_lock); - RETURN(-EINVAL); - } - LASSERT(fd->fd_grouplock.cg_lock == NULL); - cfs_spin_unlock(&lli->lli_lock); - - rc = cl_get_grouplock(cl_i2info(inode)->lli_clob, - arg, (file->f_flags & O_NONBLOCK), &grouplock); - if (rc) - RETURN(rc); - - cfs_spin_lock(&lli->lli_lock); - if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { - cfs_spin_unlock(&lli->lli_lock); - CERROR("another thread just won the race\n"); - cl_put_grouplock(&grouplock); - RETURN(-EINVAL); - } + spin_lock(&lli->lli_lock); + if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { + CWARN("group lock already existed with gid %lu\n", + fd->fd_grouplock.cg_gid); + spin_unlock(&lli->lli_lock); + RETURN(-EINVAL); + } + LASSERT(fd->fd_grouplock.cg_lock == NULL); + spin_unlock(&lli->lli_lock); + + rc = cl_get_grouplock(cl_i2info(inode)->lli_clob, + arg, (file->f_flags & O_NONBLOCK), &grouplock); + if (rc) + RETURN(rc); + + spin_lock(&lli->lli_lock); + if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { + spin_unlock(&lli->lli_lock); + CERROR("another thread just won the race\n"); + cl_put_grouplock(&grouplock); + RETURN(-EINVAL); + } - fd->fd_flags |= LL_FILE_GROUP_LOCKED; - fd->fd_grouplock = grouplock; - cfs_spin_unlock(&lli->lli_lock); + fd->fd_flags |= LL_FILE_GROUP_LOCKED; + fd->fd_grouplock = grouplock; + spin_unlock(&lli->lli_lock); - CDEBUG(D_INFO, "group lock %lu obtained\n", arg); - RETURN(0); + CDEBUG(D_INFO, "group lock %lu obtained\n", arg); + RETURN(0); } int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg) { - struct ll_inode_info *lli = ll_i2info(inode); - struct ll_file_data *fd = LUSTRE_FPRIVATE(file); - struct ccc_grouplock grouplock; - ENTRY; + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + struct ccc_grouplock grouplock; + ENTRY; - cfs_spin_lock(&lli->lli_lock); - if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { - cfs_spin_unlock(&lli->lli_lock); + spin_lock(&lli->lli_lock); + if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { + spin_unlock(&lli->lli_lock); CWARN("no group lock held\n"); RETURN(-EINVAL); } @@ -1556,18 +1593,18 @@ int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg) if (fd->fd_grouplock.cg_gid != arg) { CWARN("group lock %lu doesn't match current id %lu\n", arg, fd->fd_grouplock.cg_gid); - cfs_spin_unlock(&lli->lli_lock); - RETURN(-EINVAL); - } + spin_unlock(&lli->lli_lock); + RETURN(-EINVAL); + } - grouplock = fd->fd_grouplock; - memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); - fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; - cfs_spin_unlock(&lli->lli_lock); + grouplock = fd->fd_grouplock; + memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); + fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; + spin_unlock(&lli->lli_lock); - cl_put_grouplock(&grouplock); - CDEBUG(D_INFO, "group lock %lu released\n", arg); - RETURN(0); + cl_put_grouplock(&grouplock); + CDEBUG(D_INFO, "group lock %lu released\n", arg); + RETURN(0); } /** @@ -1654,8 +1691,7 @@ int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) GOTO(out, rc = -EOPNOTSUPP); - fm_key.oa.o_id = lsm->lsm_object_id; - fm_key.oa.o_seq = lsm->lsm_object_seq; + fm_key.oa.o_oi = lsm->lsm_oi; fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE); @@ -1680,43 +1716,44 @@ out: int ll_fid2path(struct inode *inode, void *arg) { - struct obd_export *exp = ll_i2mdexp(inode); - struct getinfo_fid2path *gfout, *gfin; - int outsize, rc; - ENTRY; + struct obd_export *exp = ll_i2mdexp(inode); + struct getinfo_fid2path *gfout, *gfin; + int outsize, rc; + ENTRY; if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) && !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH)) RETURN(-EPERM); - /* Need to get the buflen */ - OBD_ALLOC_PTR(gfin); - if (gfin == NULL) - RETURN(-ENOMEM); - if (cfs_copy_from_user(gfin, arg, sizeof(*gfin))) { - OBD_FREE_PTR(gfin); - RETURN(-EFAULT); - } + /* Need to get the buflen */ + OBD_ALLOC_PTR(gfin); + if (gfin == NULL) + RETURN(-ENOMEM); + if (copy_from_user(gfin, arg, sizeof(*gfin))) { + OBD_FREE_PTR(gfin); + RETURN(-EFAULT); + } - outsize = sizeof(*gfout) + gfin->gf_pathlen; - OBD_ALLOC(gfout, outsize); - if (gfout == NULL) { - OBD_FREE_PTR(gfin); - RETURN(-ENOMEM); - } - memcpy(gfout, gfin, sizeof(*gfout)); - OBD_FREE_PTR(gfin); + outsize = sizeof(*gfout) + gfin->gf_pathlen; + OBD_ALLOC(gfout, outsize); + if (gfout == NULL) { + OBD_FREE_PTR(gfin); + RETURN(-ENOMEM); + } + memcpy(gfout, gfin, sizeof(*gfout)); + OBD_FREE_PTR(gfin); - /* Call mdc_iocontrol */ - rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); - if (rc) - GOTO(gf_free, rc); - if (cfs_copy_to_user(arg, gfout, outsize)) - rc = -EFAULT; + /* Call mdc_iocontrol */ + rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); + if (rc) + GOTO(gf_free, rc); + + if (copy_to_user(arg, gfout, outsize)) + rc = -EFAULT; gf_free: - OBD_FREE(gfout, outsize); - RETURN(rc); + OBD_FREE(gfout, outsize); + RETURN(rc); } static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) @@ -1738,10 +1775,10 @@ static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) if (fiemap_s == NULL) RETURN(-ENOMEM); - /* get the fiemap value */ - if (copy_from_user(fiemap_s,(struct ll_user_fiemap __user *)arg, - sizeof(*fiemap_s))) - GOTO(error, rc = -EFAULT); + /* get the fiemap value */ + if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg, + sizeof(*fiemap_s))) + GOTO(error, rc = -EFAULT); /* If fm_extent_count is non-zero, read the first extent since * it is used to calculate end_offset and device from previous @@ -1763,8 +1800,8 @@ static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) ret_bytes += (fiemap_s->fm_mapped_extents * sizeof(struct ll_fiemap_extent)); - if (copy_to_user((void *)arg, fiemap_s, ret_bytes)) - rc = -EFAULT; + if (copy_to_user((void *)arg, fiemap_s, ret_bytes)) + rc = -EFAULT; error: OBD_FREE_LARGE(fiemap_s, num_bytes); @@ -1778,52 +1815,202 @@ error: * Version is computed using server side locking. * * @param extent_lock Take extent lock. Not needed if a process is already - * holding the OST object group locks. + * holding the OST object group locks. */ -static int ll_data_version(struct inode *inode, __u64 *data_version, - int extent_lock) +int ll_data_version(struct inode *inode, __u64 *data_version, + int extent_lock) { - struct lov_stripe_md *lsm = NULL; - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct obdo *obdo = NULL; - int rc; + struct lov_stripe_md *lsm = NULL; + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct obdo *obdo = NULL; + int rc; ENTRY; /* If no stripe, we consider version is 0. */ lsm = ccc_inode_lsm_get(inode); if (lsm == NULL) { - *data_version = 0; - CDEBUG(D_INODE, "No object for inode\n"); - RETURN(0); - } + *data_version = 0; + CDEBUG(D_INODE, "No object for inode\n"); + RETURN(0); + } - OBD_ALLOC_PTR(obdo); + OBD_ALLOC_PTR(obdo); if (obdo == NULL) { ccc_inode_lsm_put(inode, lsm); RETURN(-ENOMEM); } - rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock); - if (!rc) { - if (!(obdo->o_valid & OBD_MD_FLDATAVERSION)) - rc = -EOPNOTSUPP; - else - *data_version = obdo->o_data_version; - } + rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock); + if (!rc) { + if (!(obdo->o_valid & OBD_MD_FLDATAVERSION)) + rc = -EOPNOTSUPP; + else + *data_version = obdo->o_data_version; + } - OBD_FREE_PTR(obdo); + OBD_FREE_PTR(obdo); ccc_inode_lsm_put(inode, lsm); RETURN(rc); } -long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) +struct ll_swap_stack { + struct iattr ia1, ia2; + __u64 dv1, dv2; + struct inode *inode1, *inode2; + bool check_dv1, check_dv2; +}; + +static int ll_swap_layouts(struct file *file1, struct file *file2, + struct lustre_swap_layouts *lsl) { - struct inode *inode = file->f_dentry->d_inode; - struct ll_file_data *fd = LUSTRE_FPRIVATE(file); - int flags; + struct mdc_swap_layouts msl; + struct md_op_data *op_data; + __u32 gid; + __u64 dv; + struct ll_swap_stack *llss = NULL; + int rc, rc1; - ENTRY; + OBD_ALLOC_PTR(llss); + if (llss == NULL) + RETURN(-ENOMEM); + + llss->inode1 = file1->f_dentry->d_inode; + llss->inode2 = file2->f_dentry->d_inode; + + if (!S_ISREG(llss->inode2->i_mode)) + GOTO(free, rc = -EINVAL); + + if (ll_permission(llss->inode1, MAY_WRITE, NULL) || + ll_permission(llss->inode2, MAY_WRITE, NULL)) + GOTO(free, rc = -EPERM); + + if (llss->inode2->i_sb != llss->inode1->i_sb) + GOTO(free, rc = -EXDEV); + + /* we use 2 bool because it is easier to swap than 2 bits */ + if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1) + llss->check_dv1 = true; + + if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2) + llss->check_dv2 = true; + + /* we cannot use lsl->sl_dvX directly because we may swap them */ + llss->dv1 = lsl->sl_dv1; + llss->dv2 = lsl->sl_dv2; + + rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2)); + if (rc == 0) /* same file, done! */ + GOTO(free, rc = 0); + + if (rc < 0) { /* sequentialize it */ + swap(llss->inode1, llss->inode2); + swap(file1, file2); + swap(llss->dv1, llss->dv2); + swap(llss->check_dv1, llss->check_dv2); + } + + gid = lsl->sl_gid; + if (gid != 0) { /* application asks to flush dirty cache */ + rc = ll_get_grouplock(llss->inode1, file1, gid); + if (rc < 0) + GOTO(free, rc); + + rc = ll_get_grouplock(llss->inode2, file2, gid); + if (rc < 0) { + ll_put_grouplock(llss->inode1, file1, gid); + GOTO(free, rc); + } + } + + /* to be able to restore mtime and atime after swap + * we need to first save them */ + if (lsl->sl_flags & + (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) { + llss->ia1.ia_mtime = llss->inode1->i_mtime; + llss->ia1.ia_atime = llss->inode1->i_atime; + llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME; + llss->ia2.ia_mtime = llss->inode2->i_mtime; + llss->ia2.ia_atime = llss->inode2->i_atime; + llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME; + } + + /* ultimate check, before swaping the layouts we check if + * dataversion has changed (if requested) */ + if (llss->check_dv1) { + rc = ll_data_version(llss->inode1, &dv, 0); + if (rc) + GOTO(putgl, rc); + if (dv != llss->dv1) + GOTO(putgl, rc = -EAGAIN); + } + + if (llss->check_dv2) { + rc = ll_data_version(llss->inode2, &dv, 0); + if (rc) + GOTO(putgl, rc); + if (dv != llss->dv2) + GOTO(putgl, rc = -EAGAIN); + } + + /* struct md_op_data is used to send the swap args to the mdt + * only flags is missing, so we use struct mdc_swap_layouts + * through the md_op_data->op_data */ + /* flags from user space have to be converted before they are send to + * server, no flag is sent today, they are only used on the client */ + msl.msl_flags = 0; + rc = -ENOMEM; + op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0, + 0, LUSTRE_OPC_ANY, &msl); + if (op_data != NULL) { + rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, + ll_i2mdexp(llss->inode1), + sizeof(*op_data), op_data, NULL); + ll_finish_md_op_data(op_data); + } + +putgl: + if (gid != 0) { + ll_put_grouplock(llss->inode2, file2, gid); + ll_put_grouplock(llss->inode1, file1, gid); + } + + /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */ + if (rc != 0) + GOTO(free, rc); + + /* clear useless flags */ + if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) { + llss->ia1.ia_valid &= ~ATTR_MTIME; + llss->ia2.ia_valid &= ~ATTR_MTIME; + } + + if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) { + llss->ia1.ia_valid &= ~ATTR_ATIME; + llss->ia2.ia_valid &= ~ATTR_ATIME; + } + + /* update time if requested */ + rc = rc1 = 0; + if (llss->ia2.ia_valid != 0) + rc = ll_setattr(file1->f_dentry, &llss->ia2); + + if (llss->ia1.ia_valid != 0) + rc1 = ll_setattr(file2->f_dentry, &llss->ia1); + +free: + if (llss != NULL) + OBD_FREE_PTR(llss); + + RETURN(rc); +} + +long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) +{ + struct inode *inode = file->f_dentry->d_inode; + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + int flags, rc; + ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino, inode->i_generation, inode, cmd); @@ -1863,6 +2050,27 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) RETURN(ll_lov_setstripe(inode, file, arg)); case LL_IOC_LOV_SETEA: RETURN(ll_lov_setea(inode, file, arg)); + case LL_IOC_LOV_SWAP_LAYOUTS: { + struct file *file2; + struct lustre_swap_layouts lsl; + + if (cfs_copy_from_user(&lsl, (char *)arg, + sizeof(struct lustre_swap_layouts))) + RETURN(-EFAULT); + + if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */ + RETURN(-EPERM); + + file2 = fget(lsl.sl_fd); + if (file2 == NULL) + RETURN(-EBADF); + + rc = -EPERM; + if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */ + rc = ll_swap_layouts(file, file2, &lsl); + fput(file2); + RETURN(rc); + } case LL_IOC_LOV_GETSTRIPE: RETURN(ll_lov_getstripe(inode, arg)); case LL_IOC_RECREATE_OBJ: @@ -1890,33 +2098,32 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) case FSFILT_IOC_SETVERSION_OLD: case FSFILT_IOC_SETVERSION: */ - case LL_IOC_FLUSHCTX: - RETURN(ll_flush_ctx(inode)); - case LL_IOC_PATH2FID: { - if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode), - sizeof(struct lu_fid))) - RETURN(-EFAULT); + case LL_IOC_FLUSHCTX: + RETURN(ll_flush_ctx(inode)); + case LL_IOC_PATH2FID: { + if (copy_to_user((void *)arg, ll_inode2fid(inode), + sizeof(struct lu_fid))) + RETURN(-EFAULT); - RETURN(0); - } - case OBD_IOC_FID2PATH: + RETURN(0); + } + case OBD_IOC_FID2PATH: RETURN(ll_fid2path(inode, (void *)arg)); - case LL_IOC_DATA_VERSION: { - struct ioc_data_version idv; - int rc; + case LL_IOC_DATA_VERSION: { + struct ioc_data_version idv; + int rc; - if (cfs_copy_from_user(&idv, (char *)arg, sizeof(idv))) - RETURN(-EFAULT); + if (copy_from_user(&idv, (char *)arg, sizeof(idv))) + RETURN(-EFAULT); - rc = ll_data_version(inode, &idv.idv_version, - !(idv.idv_flags & LL_DV_NOFLUSH)); + rc = ll_data_version(inode, &idv.idv_version, + !(idv.idv_flags & LL_DV_NOFLUSH)); - if (rc == 0 && - cfs_copy_to_user((char *) arg, &idv, sizeof(idv))) - RETURN(-EFAULT); + if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv))) + RETURN(-EFAULT); - RETURN(rc); - } + RETURN(rc); + } case LL_IOC_GET_MDTIDX: { int mdtidx; @@ -1930,55 +2137,200 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) RETURN(0); } - case OBD_IOC_GETDTNAME: - case OBD_IOC_GETMDNAME: - RETURN(ll_get_obd_name(inode, cmd, arg)); - default: { - int err; - - if (LLIOC_STOP == - ll_iocontrol_call(inode, file, cmd, arg, &err)) - RETURN(err); - - RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, - (void *)arg)); - } - } + case OBD_IOC_GETDTNAME: + case OBD_IOC_GETMDNAME: + RETURN(ll_get_obd_name(inode, cmd, arg)); + case LL_IOC_HSM_STATE_GET: { + struct md_op_data *op_data; + struct hsm_user_state *hus; + int rc; + + OBD_ALLOC_PTR(hus); + if (hus == NULL) + RETURN(-ENOMEM); + + op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, + LUSTRE_OPC_ANY, hus); + if (op_data == NULL) { + OBD_FREE_PTR(hus); + RETURN(-ENOMEM); + } + + rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), + op_data, NULL); + + if (copy_to_user((void *)arg, hus, sizeof(*hus))) + rc = -EFAULT; + + ll_finish_md_op_data(op_data); + OBD_FREE_PTR(hus); + RETURN(rc); + } + case LL_IOC_HSM_STATE_SET: { + struct md_op_data *op_data; + struct hsm_state_set *hss; + int rc; + + OBD_ALLOC_PTR(hss); + if (hss == NULL) + RETURN(-ENOMEM); + if (copy_from_user(hss, (char *)arg, sizeof(*hss))) { + OBD_FREE_PTR(hss); + RETURN(-EFAULT); + } + + /* Non-root users are forbidden to set or clear flags which are + * NOT defined in HSM_USER_MASK. */ + if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) + && !cfs_capable(CFS_CAP_SYS_ADMIN)) { + OBD_FREE_PTR(hss); + RETURN(-EPERM); + } + + op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, + LUSTRE_OPC_ANY, hss); + if (op_data == NULL) { + OBD_FREE_PTR(hss); + RETURN(-ENOMEM); + } + + rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), + op_data, NULL); + + ll_finish_md_op_data(op_data); + + OBD_FREE_PTR(hss); + RETURN(rc); + } + case LL_IOC_HSM_ACTION: { + struct md_op_data *op_data; + struct hsm_current_action *hca; + int rc; + + OBD_ALLOC_PTR(hca); + if (hca == NULL) + RETURN(-ENOMEM); + + op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, + LUSTRE_OPC_ANY, hca); + if (op_data == NULL) { + OBD_FREE_PTR(hca); + RETURN(-ENOMEM); + } + + rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), + op_data, NULL); + + if (cfs_copy_to_user((char *)arg, hca, sizeof(*hca))) + rc = -EFAULT; + + ll_finish_md_op_data(op_data); + OBD_FREE_PTR(hca); + RETURN(rc); + } + default: { + int err; + + if (LLIOC_STOP == + ll_iocontrol_call(inode, file, cmd, arg, &err)) + RETURN(err); + + RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, + (void *)arg)); + } + } } -loff_t ll_file_seek(struct file *file, loff_t offset, int origin) +#ifndef HAVE_FILE_LLSEEK_SIZE +static inline loff_t +llseek_execute(struct file *file, loff_t offset, loff_t maxsize) { - struct inode *inode = file->f_dentry->d_inode; - loff_t retval; - ENTRY; - retval = offset + ((origin == 2) ? i_size_read(inode) : - (origin == 1) ? file->f_pos : 0); - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%llu=%#llx(%s)\n", - inode->i_ino, inode->i_generation, inode, retval, retval, - origin == 2 ? "SEEK_END": origin == 1 ? "SEEK_CUR" : "SEEK_SET"); - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); - - if (origin == 2) { /* SEEK_END */ - int rc; + if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) + return -EINVAL; + if (offset > maxsize) + return -EINVAL; - rc = ll_glimpse_size(inode); - if (rc != 0) - RETURN(rc); + if (offset != file->f_pos) { + file->f_pos = offset; + file->f_version = 0; + } + return offset; +} - offset += i_size_read(inode); - } else if (origin == 1) { /* SEEK_CUR */ - offset += file->f_pos; - } +static loff_t +generic_file_llseek_size(struct file *file, loff_t offset, int origin, + loff_t maxsize, loff_t eof) +{ + struct inode *inode = file->f_dentry->d_inode; - retval = -EINVAL; - if (offset >= 0 && offset <= ll_file_maxbytes(inode)) { - if (offset != file->f_pos) { - file->f_pos = offset; - } - retval = offset; - } + switch (origin) { + case SEEK_END: + offset += eof; + break; + case SEEK_CUR: + /* + * Here we special-case the lseek(fd, 0, SEEK_CUR) + * position-querying operation. Avoid rewriting the "same" + * f_pos value back to the file because a concurrent read(), + * write() or lseek() might have altered it + */ + if (offset == 0) + return file->f_pos; + /* + * f_lock protects against read/modify/write race with other + * SEEK_CURs. Note that parallel writes and reads behave + * like SEEK_SET. + */ + mutex_lock(&inode->i_mutex); + offset = llseek_execute(file, file->f_pos + offset, maxsize); + mutex_unlock(&inode->i_mutex); + return offset; + case SEEK_DATA: + /* + * In the generic case the entire file is data, so as long as + * offset isn't at the end of the file then the offset is data. + */ + if (offset >= eof) + return -ENXIO; + break; + case SEEK_HOLE: + /* + * There is a virtual hole at the end of the file, so as long as + * offset isn't i_size or larger, return i_size. + */ + if (offset >= eof) + return -ENXIO; + offset = eof; + break; + } + + return llseek_execute(file, offset, maxsize); +} +#endif + +loff_t ll_file_seek(struct file *file, loff_t offset, int origin) +{ + struct inode *inode = file->f_dentry->d_inode; + loff_t retval, eof = 0; + + ENTRY; + retval = offset + ((origin == SEEK_END) ? i_size_read(inode) : + (origin == SEEK_CUR) ? file->f_pos : 0); + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%llu=%#llx(%d)\n", + inode->i_ino, inode->i_generation, inode, retval, retval, + origin); + ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); + + if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) { + retval = ll_glimpse_size(inode); + if (retval != 0) + RETURN(retval); + eof = i_size_read(inode); + } - RETURN(retval); + retval = ll_generic_file_llseek_size(file, offset, origin, + ll_file_maxbytes(inode), eof); + RETURN(retval); } int ll_flush(struct file *file, fl_owner_t id) @@ -2059,21 +2411,31 @@ int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end, RETURN(result); } +/* + * When dentry is provided (the 'else' case), *file->f_dentry may be + * null and dentry must be used directly rather than pulled from + * *file->f_dentry as is done otherwise. + */ + #ifdef HAVE_FILE_FSYNC_4ARGS int ll_fsync(struct file *file, loff_t start, loff_t end, int data) +{ + struct dentry *dentry = file->f_dentry; #elif defined(HAVE_FILE_FSYNC_2ARGS) int ll_fsync(struct file *file, int data) +{ + struct dentry *dentry = file->f_dentry; #else int ll_fsync(struct file *file, struct dentry *dentry, int data) -#endif { - struct inode *inode = file->f_dentry->d_inode; +#endif + struct inode *inode = dentry->d_inode; struct ll_inode_info *lli = ll_i2info(inode); struct ptlrpc_request *req; struct obd_capa *oc; - struct lov_stripe_md *lsm; int rc, err; ENTRY; + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, inode->i_generation, inode); ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1); @@ -2108,8 +2470,7 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data) if (!err) ptlrpc_req_finished(req); - lsm = ccc_inode_lsm_get(inode); - if (data && lsm) { + if (data) { struct ll_file_data *fd = LUSTRE_FPRIVATE(file); err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, @@ -2121,7 +2482,6 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data) else fd->fd_write_failed = false; } - ccc_inode_lsm_put(inode, lsm); #ifdef HAVE_FILE_FSYNC_4ARGS mutex_unlock(&inode->i_mutex); @@ -2141,6 +2501,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) ldlm_policy_data_t flock = {{0}}; int flags = 0; int rc; + int rc2 = 0; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu file_lock=%p\n", @@ -2236,15 +2597,22 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, op_data, &lockh, &flock, 0, NULL /* req */, flags); - ll_finish_md_op_data(op_data); - if ((file_lock->fl_flags & FL_FLOCK) && (rc == 0 || file_lock->fl_type == F_UNLCK)) - flock_lock_file_wait(file, file_lock); + rc2 = flock_lock_file_wait(file, file_lock); if ((file_lock->fl_flags & FL_POSIX) && (rc == 0 || file_lock->fl_type == F_UNLCK) && !(flags & LDLM_FL_TEST_LOCK)) - posix_lock_file_wait(file, file_lock); + rc2 = posix_lock_file_wait(file, file_lock); + + if (rc2 && file_lock->fl_type != F_UNLCK) { + einfo.ei_mode = LCK_NL; + md_enqueue(sbi->ll_md_exp, &einfo, NULL, + op_data, &lockh, &flock, 0, NULL /* req */, flags); + rc = rc2; + } + + ll_finish_md_op_data(op_data); RETURN(rc); } @@ -2308,41 +2676,39 @@ int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode) } ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits, - struct lustre_handle *lockh) + struct lustre_handle *lockh, __u64 flags) { ldlm_policy_data_t policy = { .l_inodebits = {bits}}; struct lu_fid *fid; ldlm_mode_t rc; - __u64 flags; ENTRY; fid = &ll_i2info(inode)->lli_fid; CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid)); - flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING; - rc = md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy, + rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags, + fid, LDLM_IBITS, &policy, LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh); RETURN(rc); } -static int ll_inode_revalidate_fini(struct inode *inode, int rc) { - if (rc == -ENOENT) { /* Already unlinked. Just update nlink - * and return success */ +static int ll_inode_revalidate_fini(struct inode *inode, int rc) +{ + /* Already unlinked. Just update nlink and return success */ + if (rc == -ENOENT) { clear_nlink(inode); /* This path cannot be hit for regular files unless in * case of obscure races, so no need to to validate * size. */ - if (!S_ISREG(inode->i_mode) && - !S_ISDIR(inode->i_mode)) + if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode)) return 0; + } else if (rc != 0) { + CERROR("%s: revalidate FID "DFID" error: rc = %d\n", + ll_get_fsname(inode->i_sb, NULL, 0), + PFID(ll_inode2fid(inode)), rc); } - if (rc) { - CERROR("failure %d inode %lu\n", rc, inode->i_ino); - return -abs(rc); - } - - return 0; + return rc; } int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, @@ -2354,10 +2720,7 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, int rc = 0; ENTRY; - if (!inode) { - CERROR("REPORT THIS LINE TO PETER\n"); - RETURN(0); - } + LASSERT(inode != NULL); CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n", inode->i_ino, inode->i_generation, inode, dentry->d_name.name); @@ -2367,7 +2730,7 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC. * But under CMD case, it caused some lock issues, should be fixed * with new CMD ibits lock. See bug 12718 */ - if (exp->exp_connect_flags & OBD_CONNECT_ATTRFID) { + if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) { struct lookup_intent oit = { .it_op = IT_GETATTR }; struct md_op_data *op_data; @@ -2438,7 +2801,7 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, RETURN(rc); } - rc = ll_prep_inode(&inode, req, NULL); + rc = ll_prep_inode(&inode, req, NULL, NULL); } out: ptlrpc_req_finished(req); @@ -2453,21 +2816,17 @@ int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, ENTRY; rc = __ll_inode_revalidate_it(dentry, it, ibits); - - /* if object not yet allocated, don't validate size */ - if (rc == 0 && !ll_i2info(dentry->d_inode)->lli_has_smd) { - LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime; - LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime; - LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime; - RETURN(0); - } - - /* ll_glimpse_size will prefer locally cached writes if they extend - * the file */ - - if (rc == 0) - rc = ll_glimpse_size(inode); - + if (rc != 0) + RETURN(rc); + + /* if object isn't regular file, don't validate size */ + if (!S_ISREG(inode->i_mode)) { + LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime; + LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime; + LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime; + } else { + rc = ll_glimpse_size(inode); + } RETURN(rc); } @@ -2554,10 +2913,10 @@ struct posix_acl * ll_get_acl(struct inode *inode, int type) struct posix_acl *acl = NULL; ENTRY; - cfs_spin_lock(&lli->lli_lock); + spin_lock(&lli->lli_lock); /* VFS' acl_permission_check->check_acl will release the refcount */ acl = posix_acl_dup(lli->lli_posix_acl); - cfs_spin_unlock(&lli->lli_lock); + spin_unlock(&lli->lli_lock); RETURN(acl); } @@ -2735,7 +3094,7 @@ struct inode_operations ll_file_inode_operations = { /* dynamic ioctl number support routins */ static struct llioc_ctl_data { - cfs_rw_semaphore_t ioc_sem; + struct rw_semaphore ioc_sem; cfs_list_t ioc_head; } llioc = { __RWSEM_INITIALIZER(llioc.ioc_sem), @@ -2772,9 +3131,9 @@ void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd) in_data->iocd_count = count; memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count); - cfs_down_write(&llioc.ioc_sem); + down_write(&llioc.ioc_sem); cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head); - cfs_up_write(&llioc.ioc_sem); + up_write(&llioc.ioc_sem); RETURN(in_data); } @@ -2786,19 +3145,19 @@ void ll_iocontrol_unregister(void *magic) if (magic == NULL) return; - cfs_down_write(&llioc.ioc_sem); + down_write(&llioc.ioc_sem); cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) { if (tmp == magic) { unsigned int size = tmp->iocd_size; cfs_list_del(&tmp->iocd_list); - cfs_up_write(&llioc.ioc_sem); + up_write(&llioc.ioc_sem); OBD_FREE(tmp, size); return; } } - cfs_up_write(&llioc.ioc_sem); + up_write(&llioc.ioc_sem); CWARN("didn't find iocontrol register block with magic: %p\n", magic); } @@ -2813,7 +3172,7 @@ enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, struct llioc_data *data; int rc = -EINVAL, i; - cfs_down_read(&llioc.ioc_sem); + down_read(&llioc.ioc_sem); cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) { for (i = 0; i < data->iocd_count; i++) { if (cmd != data->iocd_cmd[i]) @@ -2826,7 +3185,7 @@ enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, if (ret == LLIOC_STOP) break; } - cfs_up_read(&llioc.ioc_sem); + up_read(&llioc.ioc_sem); if (rcp) *rcp = rc; @@ -2850,10 +3209,127 @@ int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf) result = cl_conf_set(env, lli->lli_clob, conf); cl_env_nested_put(&nest, env); + + if (conf->coc_opc == OBJECT_CONF_SET) { + struct ldlm_lock *lock = conf->coc_lock; + + LASSERT(lock != NULL); + LASSERT(ldlm_has_layout(lock)); + if (result == 0) { + /* it can only be allowed to match after layout is + * applied to inode otherwise false layout would be + * seen. Applying layout shoud happen before dropping + * the intent lock. */ + ldlm_lock_allow_match(lock); + } + } RETURN(result); } /** + * Apply the layout to the inode. Layout lock is held and will be released + * in this function. + */ +static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode, + struct inode *inode, __u32 *gen, bool reconf) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct ldlm_lock *lock; + struct lustre_md md = { NULL }; + struct cl_object_conf conf; + int rc = 0; + bool lvb_ready; + ENTRY; + + LASSERT(lustre_handle_is_used(lockh)); + + lock = ldlm_handle2lock(lockh); + LASSERT(lock != NULL); + LASSERT(ldlm_has_layout(lock)); + + LDLM_DEBUG(lock, "File %p/"DFID" being reconfigured: %d.\n", + inode, PFID(&lli->lli_fid), reconf); + + lock_res_and_lock(lock); + lvb_ready = !!(lock->l_flags & LDLM_FL_LVB_READY); + unlock_res_and_lock(lock); + /* checking lvb_ready is racy but this is okay. The worst case is + * that multi processes may configure the file on the same time. */ + if (lvb_ready || !reconf) { + LDLM_LOCK_PUT(lock); + + rc = -ENODATA; + if (lvb_ready) { + /* layout_gen must be valid if layout lock is not + * cancelled and stripe has already set */ + *gen = lli->lli_layout_gen; + rc = 0; + } + ldlm_lock_decref(lockh, mode); + RETURN(rc); + } + + /* for layout lock, lmm is returned in lock's lvb. + * lvb_data is immutable if the lock is held so it's safe to access it + * without res lock. See the description in ldlm_lock_decref_internal() + * for the condition to free lvb_data of layout lock */ + if (lock->l_lvb_data != NULL) { + rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm, + lock->l_lvb_data, lock->l_lvb_len); + if (rc >= 0) { + *gen = LL_LAYOUT_GEN_EMPTY; + if (md.lsm != NULL) + *gen = md.lsm->lsm_layout_gen; + rc = 0; + } else { + CERROR("%s: file "DFID" unpackmd error: %d\n", + ll_get_fsname(inode->i_sb, NULL, 0), + PFID(&lli->lli_fid), rc); + } + } + if (rc < 0) { + LDLM_LOCK_PUT(lock); + ldlm_lock_decref(lockh, mode); + RETURN(rc); + } + + /* set layout to file. Unlikely this will fail as old layout was + * surely eliminated */ + memset(&conf, 0, sizeof conf); + conf.coc_opc = OBJECT_CONF_SET; + conf.coc_inode = inode; + conf.coc_lock = lock; + conf.u.coc_md = &md; + rc = ll_layout_conf(inode, &conf); + LDLM_LOCK_PUT(lock); + + ldlm_lock_decref(lockh, mode); + + if (md.lsm != NULL) + obd_free_memmd(sbi->ll_dt_exp, &md.lsm); + + /* wait for IO to complete if it's still being used. */ + if (rc == -EBUSY) { + CDEBUG(D_INODE, "%s: %p/"DFID" wait for layout reconf.\n", + ll_get_fsname(inode->i_sb, NULL, 0), + inode, PFID(&lli->lli_fid)); + + memset(&conf, 0, sizeof conf); + conf.coc_opc = OBJECT_CONF_WAIT; + conf.coc_inode = inode; + rc = ll_layout_conf(inode, &conf); + if (rc == 0) + rc = -EAGAIN; + + CDEBUG(D_INODE, "file: "DFID" waiting layout return: %d.\n", + PFID(&lli->lli_fid), rc); + } + + RETURN(rc); +} + +/** * This function checks if there exists a LAYOUT lock on the client side, * or enqueues it if it doesn't have one in cache. * @@ -2870,18 +3346,20 @@ int ll_layout_refresh(struct inode *inode, __u32 *gen) { struct ll_inode_info *lli = ll_i2info(inode); struct ll_sb_info *sbi = ll_i2sbi(inode); - struct md_op_data *op_data = NULL; - struct ptlrpc_request *req = NULL; - struct lookup_intent it = { .it_op = IT_LAYOUT }; + struct md_op_data *op_data; + struct lookup_intent it; struct lustre_handle lockh; ldlm_mode_t mode; - struct cl_object_conf conf = { .coc_inode = inode, - .coc_validate_only = true }; + struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS, + .ei_mode = LCK_CR, + .ei_cb_bl = ll_md_blocking_ast, + .ei_cb_cp = ldlm_completion_ast, + .ei_cbdata = NULL }; int rc; ENTRY; - *gen = 0; - if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_LAYOUT_LOCK)) + *gen = LL_LAYOUT_GEN_NONE; + if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK)) RETURN(0); /* sanity checks */ @@ -2890,83 +3368,69 @@ int ll_layout_refresh(struct inode *inode, __u32 *gen) /* mostly layout lock is caching on the local side, so try to match * it before grabbing layout lock mutex. */ - mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh); + mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0); if (mode != 0) { /* hit cached lock */ - struct lov_stripe_md *lsm; + rc = ll_layout_lock_set(&lockh, mode, inode, gen, false); + if (rc == 0) + RETURN(0); - lsm = ccc_inode_lsm_get(inode); - if (lsm != NULL) - *gen = lsm->lsm_layout_gen; - ccc_inode_lsm_put(inode, lsm); - ldlm_lock_decref(&lockh, mode); + /* better hold lli_layout_mutex to try again otherwise + * it will have starvation problem. */ + } - RETURN(0); + /* take layout lock mutex to enqueue layout lock exclusively. */ + mutex_lock(&lli->lli_layout_mutex); + +again: + /* try again. Maybe somebody else has done this. */ + mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0); + if (mode != 0) { /* hit cached lock */ + rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); + if (rc == -EAGAIN) + goto again; + + mutex_unlock(&lli->lli_layout_mutex); + RETURN(rc); } op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, - 0, 0, LUSTRE_OPC_ANY, NULL); - if (IS_ERR(op_data)) + 0, 0, LUSTRE_OPC_ANY, NULL); + if (IS_ERR(op_data)) { + mutex_unlock(&lli->lli_layout_mutex); RETURN(PTR_ERR(op_data)); + } - /* take layout lock mutex to enqueue layout lock exclusively. */ - cfs_mutex_lock(&lli->lli_layout_mutex); + /* have to enqueue one */ + memset(&it, 0, sizeof(it)); + it.it_op = IT_LAYOUT; + lockh.cookie = 0ULL; - /* make sure the old conf goes away */ - ll_layout_conf(inode, &conf); + LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file %p/"DFID".\n", + ll_get_fsname(inode->i_sb, NULL, 0), inode, + PFID(&lli->lli_fid)); - /* enqueue layout lock */ - rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, - &req, ll_md_blocking_ast, 0); - if (rc == 0) { - /* we get a new lock, so update the lock data */ - lockh.cookie = it.d.lustre.it_lock_handle; - md_set_lock_data(sbi->ll_md_exp, &lockh.cookie, inode, NULL); - - /* req == NULL is when lock was found in client cache, without - * any request to server (but lsm can be canceled just after a - * release) */ - if (req != NULL) { - struct ldlm_lock *lock = ldlm_handle2lock(&lockh); - struct lustre_md md = { NULL }; - void *lmm; - int lmmsize; - - /* for IT_LAYOUT lock, lmm is returned in lock's lvb - * data via completion callback */ - LASSERT(lock != NULL); - lmm = lock->l_lvb_data; - lmmsize = lock->l_lvb_len; - if (lmm != NULL) - rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm, - lmm, lmmsize); - if (rc == 0) { - if (md.lsm != NULL) - *gen = md.lsm->lsm_layout_gen; - - memset(&conf, 0, sizeof conf); - conf.coc_inode = inode; - conf.u.coc_md = &md; - ll_layout_conf(inode, &conf); - /* is this racy? */ - lli->lli_has_smd = md.lsm != NULL; - } - if (md.lsm != NULL) - obd_free_memmd(sbi->ll_dt_exp, &md.lsm); + rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh, + NULL, 0, NULL, 0); + if (it.d.lustre.it_data != NULL) + ptlrpc_req_finished(it.d.lustre.it_data); + it.d.lustre.it_data = NULL; - LDLM_LOCK_PUT(lock); - ptlrpc_req_finished(req); - } else { /* hit caching lock */ - struct lov_stripe_md *lsm; + ll_finish_md_op_data(op_data); - lsm = ccc_inode_lsm_get(inode); - if (lsm != NULL) - *gen = lsm->lsm_layout_gen; - ccc_inode_lsm_put(inode, lsm); - } - ll_intent_drop_lock(&it); + md_set_lock_data(sbi->ll_md_exp, &it.d.lustre.it_lock_handle, inode, NULL); + + mode = it.d.lustre.it_lock_mode; + it.d.lustre.it_lock_mode = 0; + ll_intent_drop_lock(&it); + + if (rc == 0) { + /* set lock data in case this is a new lock */ + ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); + rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); + if (rc == -EAGAIN) + goto again; } - cfs_mutex_unlock(&lli->lli_layout_mutex); - ll_finish_md_op_data(op_data); + mutex_unlock(&lli->lli_layout_mutex); RETURN(rc); }