X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fllite%2Fvvp_io.c;h=4d6e5b31288c638fbf8353e3508e4f885ba96c3c;hb=ac8e871d781fae211feb88ff7e95aabb6ae48e68;hp=5107795ae5cf300db9e950467f8b79f38418e127;hpb=1072318b4d89cddba00e9adeb939249f485f2d13;p=fs%2Flustre-release.git diff --git a/lustre/llite/vvp_io.c b/lustre/llite/vvp_io.c index 5107795..4d6e5b3 100644 --- a/lustre/llite/vvp_io.c +++ b/lustre/llite/vvp_io.c @@ -26,7 +26,7 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved. + * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. */ /* @@ -201,7 +201,6 @@ static int vvp_io_rw_lock(const struct lu_env *env, struct cl_io *io, int ast_flags = 0; LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); - LASSERT(vvp_env_io(env)->cui_oneshot == 0); ENTRY; ccc_io_update_iov(env, cio, io); @@ -252,26 +251,21 @@ static int vvp_io_write_lock(const struct lu_env *env, struct cl_io *io = ios->cis_io; loff_t start; loff_t end; - int result; - if (cl2vvp_io(env, ios)->cui_oneshot == 0) { - if (io->u.ci_wr.wr_append) { - start = 0; - end = OBD_OBJECT_EOF; - } else { - start = io->u.ci_wr.wr.crw_pos; - end = start + io->u.ci_wr.wr.crw_count - 1; - } - result = vvp_io_rw_lock(env, io, CLM_WRITE, start, end); - } else - result = 0; - return result; + if (io->u.ci_wr.wr_append) { + start = 0; + end = OBD_OBJECT_EOF; + } else { + start = io->u.ci_wr.wr.crw_pos; + end = start + io->u.ci_wr.wr.crw_count - 1; + } + return vvp_io_rw_lock(env, io, CLM_WRITE, start, end); } -static int vvp_io_trunc_iter_init(const struct lu_env *env, - const struct cl_io_slice *ios) +static int vvp_io_setattr_iter_init(const struct lu_env *env, + const struct cl_io_slice *ios) { - struct ccc_io *vio = cl2ccc_io(env, ios); + struct ccc_io *cio = ccc_env_io(env); struct inode *inode = ccc_object_inode(ios->cis_obj); /* @@ -282,61 +276,72 @@ static int vvp_io_trunc_iter_init(const struct lu_env *env, * This last one is especially bad for racing o_append users on other * nodes. */ - UNLOCK_INODE_MUTEX(inode); - UP_WRITE_I_ALLOC_SEM(inode); - vio->u.trunc.cui_locks_released = 1; + if (cl_io_is_trunc(ios->cis_io)) + UP_WRITE_I_ALLOC_SEM(inode); + cio->u.setattr.cui_locks_released = 1; return 0; } /** - * Implementation of cl_io_operations::cio_lock() method for CIT_TRUNC io. + * Implementation of cl_io_operations::cio_lock() method for CIT_SETATTR io. * * Handles "lockless io" mode when extent locking is done by server. */ -static int vvp_io_trunc_lock(const struct lu_env *env, - const struct cl_io_slice *ios) +static int vvp_io_setattr_lock(const struct lu_env *env, + const struct cl_io_slice *ios) { - struct ccc_io *vio = cl2ccc_io(env, ios); - struct cl_io *io = ios->cis_io; - size_t new_size = io->u.ci_truncate.tr_size; - __u32 enqflags = new_size == 0 ? CEF_DISCARD_DATA : 0; - int result; + struct ccc_io *cio = ccc_env_io(env); + struct cl_io *io = ios->cis_io; + size_t new_size; + __u32 enqflags = 0; + + if (cl_io_is_trunc(io)) { + new_size = io->u.ci_setattr.sa_attr.lvb_size; + if (new_size == 0) + enqflags = CEF_DISCARD_DATA; + } else { + LASSERT((io->u.ci_setattr.sa_attr.lvb_mtime < + io->u.ci_setattr.sa_attr.lvb_ctime) || + (io->u.ci_setattr.sa_attr.lvb_atime < + io->u.ci_setattr.sa_attr.lvb_ctime)); + new_size = 0; + } + cio->u.setattr.cui_local_lock = SETATTR_EXTENT_LOCK; + return ccc_io_one_lock(env, io, enqflags, CLM_WRITE, + new_size, OBD_OBJECT_EOF); +} + +static int vvp_do_vmtruncate(struct inode *inode, size_t size) +{ + int result; + /* + * Only ll_inode_size_lock is taken at this level. lov_stripe_lock() + * is grabbed by ll_truncate() only over call to obd_adjust_kms(). If + * vmtruncate returns 0, then ll_truncate dropped ll_inode_size_lock() + */ + ll_inode_size_lock(inode, 0); + result = vmtruncate(inode, size); + if (result != 0) + ll_inode_size_unlock(inode, 0); - vio->u.trunc.cui_local_lock = TRUNC_EXTENT; - result = ccc_io_one_lock(env, io, enqflags, CLM_WRITE, - new_size, OBD_OBJECT_EOF); return result; } -static int vvp_io_trunc_start(const struct lu_env *env, - const struct cl_io_slice *ios) +static int vvp_io_setattr_trunc(const struct lu_env *env, + const struct cl_io_slice *ios, + struct inode *inode, loff_t size) { - struct ccc_io *cio = cl2ccc_io(env, ios); struct vvp_io *vio = cl2vvp_io(env, ios); struct cl_io *io = ios->cis_io; - struct inode *inode = ccc_object_inode(io->ci_obj); struct cl_object *obj = ios->cis_obj; - size_t size = io->u.ci_truncate.tr_size; pgoff_t start = cl_index(obj, size); int result; - LASSERT(cio->u.trunc.cui_locks_released); - LASSERT(vio->cui_oneshot == 0); - - LOCK_INODE_MUTEX(inode); DOWN_WRITE_I_ALLOC_SEM(inode); - cio->u.trunc.cui_locks_released = 0; - /* - * Only ll_inode_size_lock is taken at this level. lov_stripe_lock() - * is grabbed by ll_truncate() only over call to obd_adjust_kms(). If - * vmtruncate returns 0, then ll_truncate dropped ll_inode_size_lock() - */ - ll_inode_size_lock(inode, 0); - result = vmtruncate(inode, size); - if (result != 0) - ll_inode_size_unlock(inode, 0); + result = vvp_do_vmtruncate(inode, size); + /* * If a page is partially truncated, keep it owned across truncate to * prevent... races. @@ -348,9 +353,9 @@ static int vvp_io_trunc_start(const struct lu_env *env, struct cl_object_header *hdr; hdr = cl_object_header(obj); - spin_lock(&hdr->coh_page_guard); + cfs_spin_lock(&hdr->coh_page_guard); vio->cui_partpage = cl_page_lookup(hdr, start); - spin_unlock(&hdr->coh_page_guard); + cfs_spin_unlock(&hdr->coh_page_guard); if (vio->cui_partpage != NULL) /* @@ -367,28 +372,84 @@ static int vvp_io_trunc_start(const struct lu_env *env, return result; } -static void vvp_io_trunc_end(const struct lu_env *env, - const struct cl_io_slice *ios) +static int vvp_io_setattr_time(const struct lu_env *env, + const struct cl_io_slice *ios) { - struct vvp_io *vio = cl2vvp_io(env, ios); + struct cl_io *io = ios->cis_io; + struct cl_object *obj = io->ci_obj; + struct cl_attr *attr = ccc_env_thread_attr(env); + int result; + unsigned valid = CAT_CTIME; + + cl_object_attr_lock(obj); + attr->cat_ctime = io->u.ci_setattr.sa_attr.lvb_ctime; + if (io->u.ci_setattr.sa_valid & ATTR_ATIME_SET) { + attr->cat_atime = io->u.ci_setattr.sa_attr.lvb_atime; + valid |= CAT_ATIME; + } + if (io->u.ci_setattr.sa_valid & ATTR_MTIME_SET) { + attr->cat_mtime = io->u.ci_setattr.sa_attr.lvb_mtime; + valid |= CAT_MTIME; + } + result = cl_object_attr_set(env, obj, attr, valid); + cl_object_attr_unlock(obj); + + return result; +} + +static int vvp_io_setattr_start(const struct lu_env *env, + const struct cl_io_slice *ios) +{ + struct ccc_io *cio = cl2ccc_io(env, ios); + struct cl_io *io = ios->cis_io; + struct inode *inode = ccc_object_inode(io->ci_obj); + + LASSERT(cio->u.setattr.cui_locks_released); + + LOCK_INODE_MUTEX(inode); + cio->u.setattr.cui_locks_released = 0; + + if (cl_io_is_trunc(io)) + return vvp_io_setattr_trunc(env, ios, inode, + io->u.ci_setattr.sa_attr.lvb_size); + else + return vvp_io_setattr_time(env, ios); +} +static void vvp_io_setattr_end(const struct lu_env *env, + const struct cl_io_slice *ios) +{ + struct vvp_io *vio = cl2vvp_io(env, ios); + struct cl_io *io = ios->cis_io; + struct inode *inode = ccc_object_inode(io->ci_obj); + + if (!cl_io_is_trunc(io)) + return; if (vio->cui_partpage != NULL) { cl_page_disown(env, ios->cis_io, vio->cui_partpage); cl_page_put(env, vio->cui_partpage); vio->cui_partpage = NULL; } + + /* + * Do vmtruncate again, to remove possible stale pages populated by + * competing read threads. bz20645. + */ + vvp_do_vmtruncate(inode, io->u.ci_setattr.sa_attr.lvb_size); } -static void vvp_io_trunc_fini(const struct lu_env *env, - const struct cl_io_slice *ios) +static void vvp_io_setattr_fini(const struct lu_env *env, + const struct cl_io_slice *ios) { struct ccc_io *cio = ccc_env_io(env); + struct cl_io *io = ios->cis_io; struct inode *inode = ccc_object_inode(ios->cis_io->ci_obj); - if (cio->u.trunc.cui_locks_released) { + if (cio->u.setattr.cui_locks_released) { LOCK_INODE_MUTEX(inode); - DOWN_WRITE_I_ALLOC_SEM(inode); - cio->u.trunc.cui_locks_released = 0; + if (cl_io_is_trunc(io)) + DOWN_WRITE_I_ALLOC_SEM(inode); + cio->u.setattr.cui_locks_released = 0; } vvp_io_fini(env, ios); } @@ -439,7 +500,6 @@ static int vvp_io_read_start(const struct lu_env *env, int exceed = 0; CLOBINVRNT(env, obj, ccc_object_invariant(obj)); - LASSERT(vio->cui_oneshot == 0); CDEBUG(D_VFSTRACE, "read: -> [%lli, %lli)\n", pos, pos + cnt); @@ -531,7 +591,7 @@ static int vvp_io_write_start(const struct lu_env *env, CDEBUG(D_VFSTRACE, "write: [%lli, %lli)\n", pos, pos + (long long)cnt); - if (cl2vvp_io(env, ios)->cui_oneshot > 0) + if (cio->cui_iov == NULL) /* from a temp io in ll_cl_init(). */ result = 0; else result = lustre_generic_file_write(file, cio, &pos); @@ -626,8 +686,6 @@ static int vvp_io_fault_start(const struct lu_env *env, loff_t size; pgoff_t last; /* last page in a file data region */ - LASSERT(vio->cui_oneshot == 0); - if (fio->ft_executable && LTIME_S(inode->i_mtime) != vio->u.fault.ft_mtime) CWARN("binary "DFID @@ -699,7 +757,6 @@ static int vvp_io_read_page(const struct lu_env *env, int rc; CLOBINVRNT(env, obj, ccc_object_invariant(obj)); - LASSERT(cl2vvp_io(env, ios)->cui_oneshot == 0); LASSERT(slice->cpl_obj == obj); ENTRY; @@ -747,7 +804,7 @@ static int vvp_page_sync_io(const struct lu_env *env, struct cl_io *io, cl_2queue_init_page(queue, page); cl_page_clip(env, page, 0, to); - + result = cl_io_submit_sync(env, io, crt, queue, CRP_NORMAL, 0); LASSERT(cl_page_is_owned(page, io)); cl_page_clip(env, page, 0, CFS_PAGE_SIZE); @@ -925,12 +982,20 @@ static int vvp_io_commit_write(const struct lu_env *env, size = cl_offset(obj, pg->cp_index) + to; + ll_inode_size_lock(inode, 0); if (result == 0) { - if (size > i_size_read(inode)) - i_size_write(inode, size); + if (size > i_size_read(inode)) { + cl_isize_write_nolock(inode, size); + CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n", + PFID(lu_object_fid(&obj->co_lu)), size); + } cl_page_export(env, pg, 1); - } else if (size > i_size_read(inode)) - cl_page_discard(env, io, pg); + } else { + if (size > i_size_read(inode)) + cl_page_discard(env, io, pg); + } + ll_inode_size_unlock(inode, 0); + RETURN(result); } @@ -948,12 +1013,12 @@ static const struct cl_io_operations vvp_io_ops = { .cio_start = vvp_io_write_start, .cio_advance = ccc_io_advance }, - [CIT_TRUNC] = { - .cio_fini = vvp_io_trunc_fini, - .cio_iter_init = vvp_io_trunc_iter_init, - .cio_lock = vvp_io_trunc_lock, - .cio_start = vvp_io_trunc_start, - .cio_end = vvp_io_trunc_end + [CIT_SETATTR] = { + .cio_fini = vvp_io_setattr_fini, + .cio_iter_init = vvp_io_setattr_iter_init, + .cio_lock = vvp_io_setattr_lock, + .cio_start = vvp_io_setattr_start, + .cio_end = vvp_io_setattr_end }, [CIT_FAULT] = { .cio_fini = vvp_io_fault_fini, @@ -985,7 +1050,6 @@ int vvp_io_init(const struct lu_env *env, struct cl_object *obj, CL_IO_SLICE_CLEAN(cio, cui_cl); cl_io_slice_add(io, &cio->cui_cl, obj, &vvp_io_ops); - vio->cui_oneshot = 0; vio->cui_ra_window_set = 0; result = 0; if (io->ci_type == CIT_READ || io->ci_type == CIT_WRITE) { @@ -1004,9 +1068,12 @@ int vvp_io_init(const struct lu_env *env, struct cl_object *obj, cio->cui_tot_nrsegs = 0; ll_stats_ops_tally(sbi, op, count); } - } else if (io->ci_type == CIT_TRUNC) { - /* lockless truncate? */ - ll_stats_ops_tally(sbi, LPROC_LL_TRUNC, 1); + } else if (io->ci_type == CIT_SETATTR) { + if (cl_io_is_trunc(io)) + /* lockless truncate? */ + ll_stats_ops_tally(sbi, LPROC_LL_TRUNC, 1); + else + io->ci_lockreq = CILR_MANDATORY; } RETURN(result); }