X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fllite%2Fvvp_io.c;h=a076e5a447a198b9c2ee5518cd860daf738c82af;hb=d62f58c32074bc54055801ed0f919bfd5c277f6d;hp=9a464f2864d862192212b025963b85b66799181c;hpb=adb0b75acfb2c924d15eb9b2c8f9f8f277fa5b6f;p=fs%2Flustre-release.git diff --git a/lustre/llite/vvp_io.c b/lustre/llite/vvp_io.c index 9a464f2..a076e5a 100644 --- a/lustre/llite/vvp_io.c +++ b/lustre/llite/vvp_io.c @@ -27,7 +27,7 @@ * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2013, Intel Corporation. + * Copyright (c) 2011, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,28 +41,32 @@ #define DEBUG_SUBSYSTEM S_LLITE -#ifndef __KERNEL__ -# error This file is kernel only. -#endif #include -#include - +#include "llite_internal.h" #include "vvp_internal.h" static struct vvp_io *cl2vvp_io(const struct lu_env *env, - const struct cl_io_slice *slice); + const struct cl_io_slice *slice) +{ + struct vvp_io *vio; + + vio = container_of(slice, struct vvp_io, vui_cl); + LASSERT(vio == vvp_env_io(env)); + + return vio; +} /** * True, if \a io is a normal io, False for splice_{read,write} */ -int cl_is_normalio(const struct lu_env *env, const struct cl_io *io) +static int cl_is_normalio(const struct lu_env *env, const struct cl_io *io) { - struct vvp_io *vio = vvp_env_io(env); + struct vvp_io *vio = vvp_env_io(env); - LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); + LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); - return vio->cui_io_subtype == IO_NORMAL; + return vio->vui_io_subtype == IO_NORMAL; } /** @@ -75,7 +79,7 @@ static bool can_populate_pages(const struct lu_env *env, struct cl_io *io, struct inode *inode) { struct ll_inode_info *lli = ll_i2info(inode); - struct ccc_io *cio = ccc_env_io(env); + struct vvp_io *vio = vvp_env_io(env); bool rc = true; switch (io->ci_type) { @@ -83,9 +87,10 @@ static bool can_populate_pages(const struct lu_env *env, struct cl_io *io, case CIT_WRITE: /* don't need lock here to check lli_layout_gen as we have held * extent lock and GROUP lock has to hold to swap layout */ - if (ll_layout_version_get(lli) != cio->cui_layout_gen) { + if (ll_layout_version_get(lli) != vio->vui_layout_gen || + OBD_FAIL_CHECK_RESET(OBD_FAIL_LLITE_LOST_LAYOUT, 0)) { io->ci_need_restart = 1; - /* this will return application a short read/write */ + /* this will cause a short read/write */ io->ci_continue = 0; rc = false; } @@ -98,21 +103,175 @@ static bool can_populate_pages(const struct lu_env *env, struct cl_io *io, return rc; } +static void vvp_object_size_lock(struct cl_object *obj) +{ + struct inode *inode = vvp_object_inode(obj); + + ll_inode_size_lock(inode); + cl_object_attr_lock(obj); +} + +static void vvp_object_size_unlock(struct cl_object *obj) +{ + struct inode *inode = vvp_object_inode(obj); + + cl_object_attr_unlock(obj); + ll_inode_size_unlock(inode); +} + +/** + * Helper function that if necessary adjusts file size (inode->i_size), when + * position at the offset \a pos is accessed. File size can be arbitrary stale + * on a Lustre client, but client at least knows KMS. If accessed area is + * inside [0, KMS], set file size to KMS, otherwise glimpse file size. + * + * Locking: i_size_lock is used to serialize changes to inode size and to + * protect consistency between inode size and cl_object + * attributes. cl_object_size_lock() protects consistency between cl_attr's of + * top-object and sub-objects. + */ +static int vvp_prep_size(const struct lu_env *env, struct cl_object *obj, + struct cl_io *io, loff_t start, size_t count, + int *exceed) +{ + struct cl_attr *attr = vvp_env_thread_attr(env); + struct inode *inode = vvp_object_inode(obj); + loff_t pos = start + count - 1; + loff_t kms; + int result; + + /* + * Consistency guarantees: following possibilities exist for the + * relation between region being accessed and real file size at this + * moment: + * + * (A): the region is completely inside of the file; + * + * (B-x): x bytes of region are inside of the file, the rest is + * outside; + * + * (C): the region is completely outside of the file. + * + * This classification is stable under DLM lock already acquired by + * the caller, because to change the class, other client has to take + * DLM lock conflicting with our lock. Also, any updates to ->i_size + * by other threads on this client are serialized by + * ll_inode_size_lock(). This guarantees that short reads are handled + * correctly in the face of concurrent writes and truncates. + */ + vvp_object_size_lock(obj); + result = cl_object_attr_get(env, obj, attr); + if (result == 0) { + kms = attr->cat_kms; + if (pos > kms) { + /* + * A glimpse is necessary to determine whether we + * return a short read (B) or some zeroes at the end + * of the buffer (C) + */ + vvp_object_size_unlock(obj); + result = cl_glimpse_lock(env, io, inode, obj, 0); + if (result == 0 && exceed != NULL) { + /* If objective page index exceed end-of-file + * page index, return directly. Do not expect + * kernel will check such case correctly. + * linux-2.6.18-128.1.1 miss to do that. + * --bug 17336 */ + loff_t size = i_size_read(inode); + unsigned long cur_index = start >> + PAGE_CACHE_SHIFT; + + if ((size == 0 && cur_index != 0) || + (((size - 1) >> PAGE_CACHE_SHIFT) < + cur_index)) + *exceed = 1; + } + + return result; + } else { + /* + * region is within kms and, hence, within real file + * size (A). We need to increase i_size to cover the + * read region so that generic_file_read() will do its + * job, but that doesn't mean the kms size is + * _correct_, it is only the _minimum_ size. If + * someone does a stat they will get the correct size + * which will always be >= the kms value here. + * b=11081 + */ + if (i_size_read(inode) < kms) { + i_size_write(inode, kms); + CDEBUG(D_VFSTRACE, + DFID" updating i_size "LPU64"\n", + PFID(lu_object_fid(&obj->co_lu)), + (__u64)i_size_read(inode)); + } + } + } + + vvp_object_size_unlock(obj); + + return result; +} + /***************************************************************************** * * io operations. * */ +static int vvp_io_one_lock_index(const struct lu_env *env, struct cl_io *io, + __u32 enqflags, enum cl_lock_mode mode, + pgoff_t start, pgoff_t end) +{ + struct vvp_io *vio = vvp_env_io(env); + struct cl_lock_descr *descr = &vio->vui_link.cill_descr; + struct cl_object *obj = io->ci_obj; + + CLOBINVRNT(env, obj, vvp_object_invariant(obj)); + ENTRY; + + CDEBUG(D_VFSTRACE, "lock: %d [%lu, %lu]\n", mode, start, end); + + memset(&vio->vui_link, 0, sizeof vio->vui_link); + + if (vio->vui_fd && (vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { + descr->cld_mode = CLM_GROUP; + descr->cld_gid = vio->vui_fd->fd_grouplock.lg_gid; + enqflags |= CEF_LOCK_MATCH; + } else { + descr->cld_mode = mode; + } + + descr->cld_obj = obj; + descr->cld_start = start; + descr->cld_end = end; + descr->cld_enq_flags = enqflags; + + cl_io_lock_add(env, io, &vio->vui_link); + + RETURN(0); +} + +static int vvp_io_one_lock(const struct lu_env *env, struct cl_io *io, + __u32 enqflags, enum cl_lock_mode mode, + loff_t start, loff_t end) +{ + struct cl_object *obj = io->ci_obj; + + return vvp_io_one_lock_index(env, io, enqflags, mode, + cl_index(obj, start), cl_index(obj, end)); +} + static int vvp_io_write_iter_init(const struct lu_env *env, const struct cl_io_slice *ios) { - struct ccc_io *cio = cl2ccc_io(env, ios); + struct vvp_io *vio = cl2vvp_io(env, ios); - cl_page_list_init(&cio->u.write.cui_queue); - cio->u.write.cui_written = 0; - cio->u.write.cui_from = 0; - cio->u.write.cui_to = PAGE_SIZE; + cl_page_list_init(&vio->u.write.vui_queue); + vio->u.write.vui_written = 0; + vio->u.write.vui_from = 0; + vio->u.write.vui_to = PAGE_SIZE; return 0; } @@ -120,39 +279,39 @@ static int vvp_io_write_iter_init(const struct lu_env *env, static void vvp_io_write_iter_fini(const struct lu_env *env, const struct cl_io_slice *ios) { - struct ccc_io *cio = cl2ccc_io(env, ios); + struct vvp_io *vio = cl2vvp_io(env, ios); - LASSERT(cio->u.write.cui_queue.pl_nr == 0); + LASSERT(vio->u.write.vui_queue.pl_nr == 0); } static int vvp_io_fault_iter_init(const struct lu_env *env, const struct cl_io_slice *ios) { - struct vvp_io *vio = cl2vvp_io(env, ios); - struct inode *inode = ccc_object_inode(ios->cis_obj); + struct vvp_io *vio = cl2vvp_io(env, ios); + struct inode *inode = vvp_object_inode(ios->cis_obj); + + LASSERT(inode == vio->vui_fd->fd_file->f_path.dentry->d_inode); + vio->u.fault.ft_mtime = LTIME_S(inode->i_mtime); - LASSERT(inode == - cl2ccc_io(env, ios)->cui_fd->fd_file->f_dentry->d_inode); - vio->u.fault.ft_mtime = LTIME_S(inode->i_mtime); - return 0; + return 0; } static void vvp_io_fini(const struct lu_env *env, const struct cl_io_slice *ios) { struct cl_io *io = ios->cis_io; struct cl_object *obj = io->ci_obj; - struct ccc_io *cio = cl2ccc_io(env, ios); - struct inode *inode = ccc_object_inode(obj); + struct vvp_io *vio = cl2vvp_io(env, ios); + struct inode *inode = vvp_object_inode(obj); - CLOBINVRNT(env, obj, ccc_object_invariant(obj)); + CLOBINVRNT(env, obj, vvp_object_invariant(obj)); CDEBUG(D_VFSTRACE, DFID" ignore/verify layout %d/%d, layout version %d " "restore needed %d\n", PFID(lu_object_fid(&obj->co_lu)), io->ci_ignore_layout, io->ci_verify_layout, - cio->cui_layout_gen, io->ci_restore_needed); + vio->vui_layout_gen, io->ci_restore_needed); - if (io->ci_restore_needed == 1) { + if (io->ci_restore_needed) { int rc; /* file was detected release, we need to restore it @@ -162,7 +321,7 @@ static void vvp_io_fini(const struct lu_env *env, const struct cl_io_slice *ios) /* if restore registration failed, no restart, * we will return -ENODATA */ /* The layout will change after restore, so we need to - * block on layout lock hold by the MDT + * block on layout lock held by the MDT * as MDT will not send new layout in lvb (see LU-3124) * we have to explicitly fetch it, all this will be done * by ll_layout_refresh() @@ -184,17 +343,17 @@ static void vvp_io_fini(const struct lu_env *env, const struct cl_io_slice *ios) /* check layout version */ ll_layout_refresh(inode, &gen); - io->ci_need_restart = cio->cui_layout_gen != gen; + io->ci_need_restart = vio->vui_layout_gen != gen; if (io->ci_need_restart) { CDEBUG(D_VFSTRACE, DFID" layout changed from %d to %d.\n", PFID(lu_object_fid(&obj->co_lu)), - cio->cui_layout_gen, gen); + vio->vui_layout_gen, gen); /* today successful restore is the only possible * case */ /* restore was done, clear restoring state */ - ll_i2info(ccc_object_inode(obj))->lli_flags &= - ~LLIF_FILE_RESTORING; + ll_file_clear_flag(ll_i2info(vvp_object_inode(obj)), + LLIF_FILE_RESTORING); } } } @@ -205,7 +364,7 @@ static void vvp_io_fault_fini(const struct lu_env *env, struct cl_io *io = ios->cis_io; struct cl_page *page = io->u.ci_fault.ft_page; - CLOBINVRNT(env, io->ci_obj, ccc_object_invariant(io->ci_obj)); + CLOBINVRNT(env, io->ci_obj, vvp_object_invariant(io->ci_obj)); if (page != NULL) { lu_ref_del(&page->cp_reference, "fault", io); @@ -228,54 +387,63 @@ static enum cl_lock_mode vvp_mode_from_vma(struct vm_area_struct *vma) } static int vvp_mmap_locks(const struct lu_env *env, - struct ccc_io *vio, struct cl_io *io) + struct vvp_io *vio, struct cl_io *io) { - struct ccc_thread_info *cti = ccc_env_info(env); - struct mm_struct *mm = current->mm; - struct vm_area_struct *vma; - struct cl_lock_descr *descr = &cti->cti_descr; - ldlm_policy_data_t policy; - unsigned long addr; - unsigned long seg; - ssize_t count; - int result; - ENTRY; - - LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); - - if (!cl_is_normalio(env, io)) - RETURN(0); - - if (vio->cui_iov == NULL) /* nfs or loop back device write */ - RETURN(0); - - /* No MM (e.g. NFS)? No vmas too. */ - if (mm == NULL) - RETURN(0); - - for (seg = 0; seg < vio->cui_nrsegs; seg++) { - const struct iovec *iv = &vio->cui_iov[seg]; - - addr = (unsigned long)iv->iov_base; - count = iv->iov_len; + struct vvp_thread_info *vti = vvp_env_info(env); + struct mm_struct *mm = current->mm; + struct vm_area_struct *vma; + struct cl_lock_descr *descr = &vti->vti_descr; + union ldlm_policy_data policy; +#ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER + struct iovec iov; + struct iov_iter i; +#else + unsigned long seg; +#endif + int result = 0; + ENTRY; + + LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); + + if (!cl_is_normalio(env, io)) + RETURN(0); + + /* nfs or loop back device write */ + if (vio->vui_iter == NULL) + RETURN(0); + + /* No MM (e.g. NFS)? No vmas too. */ + if (mm == NULL) + RETURN(0); + +#ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER + iov_for_each(iov, i, *(vio->vui_iter)) { +#else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */ + for (seg = 0; seg < vio->vui_iter->nr_segs; seg++) { + const struct iovec iov = vio->vui_iter->iov[seg]; +#endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */ + unsigned long addr = (unsigned long)iov.iov_base; + size_t count = iov.iov_len; + if (count == 0) continue; - count += addr & (~CFS_PAGE_MASK); - addr &= CFS_PAGE_MASK; + count += addr & ~PAGE_MASK; + addr &= PAGE_MASK; down_read(&mm->mmap_sem); while((vma = our_vma(mm, addr, count)) != NULL) { - struct inode *inode = vma->vm_file->f_dentry->d_inode; + struct dentry *de = vma->vm_file->f_path.dentry; + struct inode *inode = de->d_inode; int flags = CEF_MUST; - if (ll_file_nolock(vma->vm_file)) { - /* - * For no lock case, a lockless lock will be - * generated. - */ - flags = CEF_NEVER; - } + if (ll_file_nolock(vma->vm_file)) { + /* + * For no lock case is not allowed for mmap + */ + result = -EINVAL; + break; + } /* * XXX: Required lock mode can be weakened: CIT_WRITE @@ -296,57 +464,149 @@ static int vvp_mmap_locks(const struct lu_env *env, descr->cld_mode, descr->cld_start, descr->cld_end); - if (result < 0) - RETURN(result); + if (result < 0) + break; - if (vma->vm_end - addr >= count) - break; + if (vma->vm_end - addr >= count) + break; - count -= vma->vm_end - addr; - addr = vma->vm_end; - } - up_read(&mm->mmap_sem); - } - RETURN(0); + count -= vma->vm_end - addr; + addr = vma->vm_end; + } + up_read(&mm->mmap_sem); + if (result < 0) + break; + } + RETURN(result); +} + +static void vvp_io_advance(const struct lu_env *env, + const struct cl_io_slice *ios, + size_t nob) +{ + struct vvp_io *vio = cl2vvp_io(env, ios); + struct cl_io *io = ios->cis_io; + struct cl_object *obj = ios->cis_io->ci_obj; +#ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER + struct iovec *iov; +#endif + CLOBINVRNT(env, obj, vvp_object_invariant(obj)); + + if (!cl_is_normalio(env, io)) + return; + +#ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER + vio->vui_tot_count -= nob; + iov_iter_reexpand(vio->vui_iter, vio->vui_tot_count); +#else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */ + LASSERT(vio->vui_tot_nrsegs >= vio->vui_iter->nr_segs); + LASSERT(vio->vui_tot_count >= nob); + + /* Restore the iov changed in vvp_io_update_iov() */ + if (vio->vui_iov_olen > 0) { + unsigned long idx = vio->vui_iter->nr_segs - 1; + + /* In the latest kernels iov is const so that + * changes are done using iter helpers. In older + * kernels those helpers don't exist we lustre + * has to do some of the management of the iter + * itself. */ + iov = (struct iovec *)&vio->vui_iter->iov[idx]; + iov->iov_len = vio->vui_iov_olen; + vio->vui_iov_olen = 0; + } + + /* In the latest kernels special helpers exist to help + * advance the iov but we don't have that in older kernels + * so we need to do the book keeping ourselves. */ + iov = (struct iovec *)vio->vui_iter->iov; + while (nob > 0) { + if (iov->iov_len > nob) { + iov->iov_len -= nob; + iov->iov_base += nob; + break; + } + + nob -= iov->iov_len; + iov++; + vio->vui_tot_nrsegs--; + } + + vio->vui_iter->iov = iov; + vio->vui_tot_count -= nob; +#endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */ +} + +static void vvp_io_update_iov(const struct lu_env *env, + struct vvp_io *vio, struct cl_io *io) +{ + size_t size = io->u.ci_rw.crw_count; +#ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER + if (!cl_is_normalio(env, io) || vio->vui_iter == NULL) + return; + + iov_iter_truncate(vio->vui_iter, size); +#else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */ + unsigned long i; + + vio->vui_iov_olen = 0; + if (!cl_is_normalio(env, io) || vio->vui_tot_nrsegs == 0) + return; + + for (i = 0; i < vio->vui_tot_nrsegs; i++) { + struct iovec *iv = (struct iovec *) &vio->vui_iter->iov[i]; + + if (iv->iov_len < size) { + size -= iv->iov_len; + } else { + if (iv->iov_len > size) { + vio->vui_iov_olen = iv->iov_len; + iv->iov_len = size; + } + break; + } + } + + vio->vui_iter->nr_segs = i + 1; + LASSERTF(vio->vui_tot_nrsegs >= vio->vui_iter->nr_segs, + "tot_nrsegs: %lu, nrsegs: %lu\n", + vio->vui_tot_nrsegs, vio->vui_iter->nr_segs); +#endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */ } static int vvp_io_rw_lock(const struct lu_env *env, struct cl_io *io, enum cl_lock_mode mode, loff_t start, loff_t end) { - struct ccc_io *cio = ccc_env_io(env); - int result; - int ast_flags = 0; + struct vvp_io *vio = vvp_env_io(env); + int result; + int ast_flags = 0; + + LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); + ENTRY; - LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); - ENTRY; + vvp_io_update_iov(env, vio, io); - ccc_io_update_iov(env, cio, io); + if (io->u.ci_rw.crw_nonblock) + ast_flags |= CEF_NONBLOCK; + + result = vvp_mmap_locks(env, vio, io); + if (result == 0) + result = vvp_io_one_lock(env, io, ast_flags, mode, start, end); - if (io->u.ci_rw.crw_nonblock) - ast_flags |= CEF_NONBLOCK; - result = vvp_mmap_locks(env, cio, io); - if (result == 0) - result = ccc_io_one_lock(env, io, ast_flags, mode, start, end); RETURN(result); } static int vvp_io_read_lock(const struct lu_env *env, const struct cl_io_slice *ios) { - struct cl_io *io = ios->cis_io; - struct ll_inode_info *lli = ll_i2info(ccc_object_inode(io->ci_obj)); - int result; + struct cl_io *io = ios->cis_io; + struct cl_io_rw_common *rd = &io->u.ci_rd.rd; + int result; - ENTRY; - /* XXX: Layer violation, we shouldn't see lsm at llite level. */ - if (lli->lli_has_smd) /* lsm-less file doesn't need to lock */ - result = vvp_io_rw_lock(env, io, CLM_READ, - io->u.ci_rd.rd.crw_pos, - io->u.ci_rd.rd.crw_pos + - io->u.ci_rd.rd.crw_count - 1); - else - result = 0; - RETURN(result); + ENTRY; + result = vvp_io_rw_lock(env, io, CLM_READ, rd->crw_pos, + rd->crw_pos + rd->crw_count - 1); + RETURN(result); } static int vvp_io_fault_lock(const struct lu_env *env, @@ -357,9 +617,11 @@ static int vvp_io_fault_lock(const struct lu_env *env, /* * XXX LDLM_FL_CBPENDING */ - return ccc_io_one_lock_index - (env, io, 0, vvp_mode_from_vma(vio->u.fault.ft_vma), - io->u.ci_fault.ft_index, io->u.ci_fault.ft_index); + return vvp_io_one_lock_index(env, + io, 0, + vvp_mode_from_vma(vio->u.fault.ft_vma), + io->u.ci_fault.ft_index, + io->u.ci_fault.ft_index); } static int vvp_io_write_lock(const struct lu_env *env, @@ -393,7 +655,6 @@ static int vvp_io_setattr_iter_init(const struct lu_env *env, static int vvp_io_setattr_lock(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; __u64 new_size; __u32 enqflags = 0; @@ -403,16 +664,24 @@ static int vvp_io_setattr_lock(const struct lu_env *env, if (new_size == 0) enqflags = CEF_DISCARD_DATA; } else { - if ((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)) - return 0; - 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); + unsigned int valid = io->u.ci_setattr.sa_valid; + + if (!(valid & TIMES_SET_FLAGS)) + return 0; + + if ((!(valid & ATTR_MTIME) || + io->u.ci_setattr.sa_attr.lvb_mtime >= + io->u.ci_setattr.sa_attr.lvb_ctime) && + (!(valid & ATTR_ATIME) || + io->u.ci_setattr.sa_attr.lvb_atime >= + io->u.ci_setattr.sa_attr.lvb_ctime)) + return 0; + + new_size = 0; + } + + return vvp_io_one_lock(env, io, enqflags, CLM_WRITE, + new_size, OBD_OBJECT_EOF); } static int vvp_do_vmtruncate(struct inode *inode, size_t size) @@ -435,20 +704,12 @@ static int vvp_do_vmtruncate(struct inode *inode, size_t size) return result; } -static int vvp_io_setattr_trunc(const struct lu_env *env, - const struct cl_io_slice *ios, - struct inode *inode, loff_t size) -{ - inode_dio_wait(inode); - return 0; -} - static int vvp_io_setattr_time(const struct lu_env *env, const struct cl_io_slice *ios) { struct cl_io *io = ios->cis_io; struct cl_object *obj = io->ci_obj; - struct cl_attr *attr = ccc_env_thread_attr(env); + struct cl_attr *attr = vvp_env_thread_attr(env); int result; unsigned valid = CAT_CTIME; @@ -462,39 +723,44 @@ static int vvp_io_setattr_time(const struct lu_env *env, 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); + result = cl_object_attr_update(env, obj, attr, valid); + cl_object_attr_unlock(obj); - return result; + return result; } static int vvp_io_setattr_start(const struct lu_env *env, const struct cl_io_slice *ios) { - struct cl_io *io = ios->cis_io; - struct inode *inode = ccc_object_inode(io->ci_obj); - int result = 0; + struct cl_io *io = ios->cis_io; + struct inode *inode = vvp_object_inode(io->ci_obj); + struct ll_inode_info *lli = ll_i2info(inode); mutex_lock(&inode->i_mutex); - if (cl_io_is_trunc(io)) - result = vvp_io_setattr_trunc(env, ios, inode, - io->u.ci_setattr.sa_attr.lvb_size); - if (result == 0) - result = vvp_io_setattr_time(env, ios); - return result; + if (cl_io_is_trunc(io)) { + down_write(&lli->lli_trunc_sem); + inode_dio_wait(inode); + } + + if (io->u.ci_setattr.sa_valid & TIMES_SET_FLAGS) + return vvp_io_setattr_time(env, ios); + + return 0; } static void vvp_io_setattr_end(const struct lu_env *env, const struct cl_io_slice *ios) { - struct cl_io *io = ios->cis_io; - struct inode *inode = ccc_object_inode(io->ci_obj); + struct cl_io *io = ios->cis_io; + struct inode *inode = vvp_object_inode(io->ci_obj); + struct ll_inode_info *lli = ll_i2info(inode); if (cl_io_is_trunc(io)) { /* Truncate in memory pages - they must be clean pages * because osc has already notified to destroy osc_extents. */ vvp_do_vmtruncate(inode, io->u.ci_setattr.sa_attr.lvb_size); inode_dio_write_done(inode); + up_write(&lli->lli_trunc_sem); } mutex_unlock(&inode->i_mutex); } @@ -502,83 +768,98 @@ static void vvp_io_setattr_end(const struct lu_env *env, static void vvp_io_setattr_fini(const struct lu_env *env, const struct cl_io_slice *ios) { + bool restore_needed = ios->cis_io->ci_restore_needed; + struct inode *inode = vvp_object_inode(ios->cis_obj); + vvp_io_fini(env, ios); + + if (restore_needed && !ios->cis_io->ci_restore_needed) { + /* restore finished, set data modified flag for HSM */ + ll_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED); + } } static int vvp_io_read_start(const struct lu_env *env, - const struct cl_io_slice *ios) + const struct cl_io_slice *ios) { - struct vvp_io *vio = cl2vvp_io(env, ios); - struct ccc_io *cio = cl2ccc_io(env, ios); - struct cl_io *io = ios->cis_io; - struct cl_object *obj = io->ci_obj; - struct inode *inode = ccc_object_inode(obj); - struct ll_ra_read *bead = &vio->cui_bead; - struct file *file = cio->cui_fd->fd_file; - - int result; - loff_t pos = io->u.ci_rd.rd.crw_pos; - long cnt = io->u.ci_rd.rd.crw_count; - long tot = cio->cui_tot_count; + struct vvp_io *vio = cl2vvp_io(env, ios); + struct cl_io *io = ios->cis_io; + struct cl_object *obj = io->ci_obj; + struct inode *inode = vvp_object_inode(obj); + struct ll_inode_info *lli = ll_i2info(inode); + struct file *file = vio->vui_fd->fd_file; + + int result; + loff_t pos = io->u.ci_rd.rd.crw_pos; + long cnt = io->u.ci_rd.rd.crw_count; + long tot = vio->vui_tot_count; int exceed = 0; - CLOBINVRNT(env, obj, ccc_object_invariant(obj)); + CLOBINVRNT(env, obj, vvp_object_invariant(obj)); + + CDEBUG(D_VFSTRACE, "read: -> [%lli, %lli)\n", pos, pos + cnt); - CDEBUG(D_VFSTRACE, "read: -> [%lli, %lli)\n", pos, pos + cnt); + if (vio->vui_io_subtype == IO_NORMAL) + down_read(&lli->lli_trunc_sem); if (!can_populate_pages(env, io, inode)) return 0; - result = ccc_prep_size(env, obj, io, pos, tot, &exceed); - if (result != 0) - return result; - else if (exceed != 0) - goto out; - - LU_OBJECT_HEADER(D_INODE, env, &obj->co_lu, - "Read ino %lu, %lu bytes, offset %lld, size %llu\n", - inode->i_ino, cnt, pos, i_size_read(inode)); - - /* turn off the kernel's read-ahead */ - cio->cui_fd->fd_file->f_ra.ra_pages = 0; - - /* initialize read-ahead window once per syscall */ - if (!vio->cui_ra_window_set) { - vio->cui_ra_window_set = 1; - bead->lrr_start = cl_index(obj, pos); - bead->lrr_count = cl_index(obj, tot + PAGE_CACHE_SIZE - 1); - ll_ra_read_in(file, bead); - } + result = vvp_prep_size(env, obj, io, pos, tot, &exceed); + if (result != 0) + return result; + else if (exceed != 0) + goto out; + + LU_OBJECT_HEADER(D_INODE, env, &obj->co_lu, + "Read ino %lu, %lu bytes, offset %lld, size %llu\n", + inode->i_ino, cnt, pos, i_size_read(inode)); + + /* turn off the kernel's read-ahead */ + vio->vui_fd->fd_file->f_ra.ra_pages = 0; + + /* initialize read-ahead window once per syscall */ + if (!vio->vui_ra_valid) { + vio->vui_ra_valid = true; + vio->vui_ra_start = cl_index(obj, pos); + vio->vui_ra_count = cl_index(obj, tot + PAGE_CACHE_SIZE - 1); + ll_ras_enter(file); + } - /* BUG: 5972 */ - file_accessed(file); - switch (vio->cui_io_subtype) { - case IO_NORMAL: - LASSERT(cio->cui_iocb->ki_pos == pos); - result = generic_file_aio_read(cio->cui_iocb, - cio->cui_iov, cio->cui_nrsegs, - cio->cui_iocb->ki_pos); + /* BUG: 5972 */ + file_accessed(file); + switch (vio->vui_io_subtype) { + case IO_NORMAL: + LASSERT(vio->vui_iocb->ki_pos == pos); +#ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER + result = generic_file_read_iter(vio->vui_iocb, vio->vui_iter); +#else + result = generic_file_aio_read(vio->vui_iocb, + vio->vui_iter->iov, + vio->vui_iter->nr_segs, + vio->vui_iocb->ki_pos); +#endif break; - case IO_SPLICE: - result = generic_file_splice_read(file, &pos, - vio->u.splice.cui_pipe, cnt, - vio->u.splice.cui_flags); - /* LU-1109: do splice read stripe by stripe otherwise if it - * may make nfsd stuck if this read occupied all internal pipe - * buffers. */ - io->ci_continue = 0; - break; - default: - CERROR("Wrong IO type %u\n", vio->cui_io_subtype); - LBUG(); - } + case IO_SPLICE: + result = generic_file_splice_read(file, &pos, + vio->u.splice.vui_pipe, cnt, + vio->u.splice.vui_flags); + /* LU-1109: do splice read stripe by stripe otherwise if it + * may make nfsd stuck if this read occupied all internal pipe + * buffers. */ + io->ci_continue = 0; + break; + default: + CERROR("Wrong IO type %u\n", vio->vui_io_subtype); + LBUG(); + } out: if (result >= 0) { if (result < cnt) io->ci_continue = 0; io->ci_nob += result; - ll_rw_stats_tally(ll_i2sbi(inode), current->pid, cio->cui_fd, + ll_rw_stats_tally(ll_i2sbi(inode), current->pid, vio->vui_fd, pos, result, READ); result = 0; } @@ -586,17 +867,6 @@ out: return result; } -static void vvp_io_read_fini(const struct lu_env *env, const struct cl_io_slice *ios) -{ - struct vvp_io *vio = cl2vvp_io(env, ios); - struct ccc_io *cio = cl2ccc_io(env, ios); - - if (vio->cui_ra_window_set) - ll_ra_read_ex(cio->cui_fd->fd_file, &vio->cui_bead); - - vvp_io_fini(env, ios); -} - static int vvp_io_commit_sync(const struct lu_env *env, struct cl_io *io, struct cl_page_list *plist, int from, int to) { @@ -613,11 +883,13 @@ static int vvp_io_commit_sync(const struct lu_env *env, struct cl_io *io, page = cl_page_list_first(plist); if (plist->pl_nr == 1) { cl_page_clip(env, page, from, to); - } else if (from > 0) { - cl_page_clip(env, page, from, PAGE_SIZE); } else { - page = cl_page_list_last(plist); - cl_page_clip(env, page, 0, to); + if (from > 0) + cl_page_clip(env, page, from, PAGE_SIZE); + if (to != PAGE_SIZE) { + page = cl_page_list_last(plist); + cl_page_clip(env, page, 0, to); + } } } @@ -656,20 +928,15 @@ static int vvp_io_commit_sync(const struct lu_env *env, struct cl_io *io, static void write_commit_callback(const struct lu_env *env, struct cl_io *io, struct cl_page *page) { - struct ccc_page *cp; struct page *vmpage = page->cp_vmpage; - struct cl_object *clob = cl_io_top(io)->ci_obj; SetPageUptodate(vmpage); set_page_dirty(vmpage); - cp = cl2ccc_page(cl_object_page_slice(clob, page)); - vvp_write_pending(cl2ccc(clob), cp); - cl_page_disown(env, io, page); /* held in ll_cl_init() */ - lu_ref_del(&page->cp_reference, "cl_io", io); + lu_ref_del(&page->cp_reference, "cl_io", cl_io_top(io)); cl_page_put(env, page); } @@ -681,15 +948,15 @@ static bool page_list_sanity_check(struct cl_object *obj, pgoff_t index = CL_PAGE_EOF; cl_page_list_for_each(page, plist) { - struct ccc_page *cp = cl_object_page_slice(obj, page); + struct vvp_page *vpg = cl_object_page_slice(obj, page); if (index == CL_PAGE_EOF) { - index = ccc_index(cp); + index = vvp_index(vpg); continue; } ++index; - if (index == ccc_index(cp)) + if (index == vvp_index(vpg)) continue; return false; @@ -701,26 +968,26 @@ static bool page_list_sanity_check(struct cl_object *obj, int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io) { struct cl_object *obj = io->ci_obj; - struct inode *inode = ccc_object_inode(obj); - struct ccc_io *cio = ccc_env_io(env); - struct cl_page_list *queue = &cio->u.write.cui_queue; + struct inode *inode = vvp_object_inode(obj); + struct vvp_io *vio = vvp_env_io(env); + struct cl_page_list *queue = &vio->u.write.vui_queue; struct cl_page *page; int rc = 0; int bytes = 0; - unsigned int npages = cio->u.write.cui_queue.pl_nr; + unsigned int npages = vio->u.write.vui_queue.pl_nr; ENTRY; if (npages == 0) RETURN(0); CDEBUG(D_VFSTRACE, "commit async pages: %d, from %d, to %d\n", - npages, cio->u.write.cui_from, cio->u.write.cui_to); + npages, vio->u.write.vui_from, vio->u.write.vui_to); LASSERT(page_list_sanity_check(obj, queue)); /* submit IO with async write */ rc = cl_io_commit_async(env, io, queue, - cio->u.write.cui_from, cio->u.write.cui_to, + vio->u.write.vui_from, vio->u.write.vui_to, write_commit_callback); npages -= queue->pl_nr; /* already committed pages */ if (npages > 0) { @@ -728,18 +995,18 @@ int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io) bytes = npages << PAGE_SHIFT; /* first page */ - bytes -= cio->u.write.cui_from; + bytes -= vio->u.write.vui_from; if (queue->pl_nr == 0) /* last page */ - bytes -= PAGE_SIZE - cio->u.write.cui_to; + bytes -= PAGE_SIZE - vio->u.write.vui_to; LASSERTF(bytes > 0, "bytes = %d, pages = %d\n", bytes, npages); - cio->u.write.cui_written += bytes; + vio->u.write.vui_written += bytes; CDEBUG(D_VFSTRACE, "Committed %d pages %d bytes, tot: %ld\n", - npages, bytes, cio->u.write.cui_written); + npages, bytes, vio->u.write.vui_written); /* the first page must have been written. */ - cio->u.write.cui_from = 0; + vio->u.write.vui_from = 0; } LASSERT(page_list_sanity_check(obj, queue)); LASSERT(ergo(rc == 0, queue->pl_nr == 0)); @@ -747,16 +1014,16 @@ int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io) /* out of quota, try sync write */ if (rc == -EDQUOT && !cl_io_is_mkwrite(io)) { rc = vvp_io_commit_sync(env, io, queue, - cio->u.write.cui_from, - cio->u.write.cui_to); + vio->u.write.vui_from, + vio->u.write.vui_to); if (rc > 0) { - cio->u.write.cui_written += rc; + vio->u.write.vui_written += rc; rc = 0; } } /* update inode size */ - ll_merge_lvb(env, inode); + ll_merge_attr(env, inode); /* Now the pages in queue were failed to commit, discard them * unless they were dirtied before. */ @@ -781,42 +1048,83 @@ int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io) static int vvp_io_write_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 cl_object *obj = io->ci_obj; - struct inode *inode = ccc_object_inode(obj); - ssize_t result = 0; - loff_t pos = io->u.ci_wr.wr.crw_pos; - size_t cnt = io->u.ci_wr.wr.crw_count; + struct vvp_io *vio = cl2vvp_io(env, ios); + struct cl_io *io = ios->cis_io; + struct cl_object *obj = io->ci_obj; + struct inode *inode = vvp_object_inode(obj); + struct ll_inode_info *lli = ll_i2info(inode); + ssize_t result = 0; + loff_t pos = io->u.ci_wr.wr.crw_pos; + size_t cnt = io->u.ci_wr.wr.crw_count; - ENTRY; + ENTRY; + + if (vio->vui_io_subtype == IO_NORMAL) + down_read(&lli->lli_trunc_sem); if (!can_populate_pages(env, io, inode)) - return 0; + RETURN(0); - if (cl_io_is_append(io)) { - /* - * PARALLEL IO This has to be changed for parallel IO doing - * out-of-order writes. - */ - pos = io->u.ci_wr.wr.crw_pos = i_size_read(inode); - cio->cui_iocb->ki_pos = pos; - } else { - LASSERT(cio->cui_iocb->ki_pos == pos); + if (cl_io_is_append(io)) { + /* + * PARALLEL IO This has to be changed for parallel IO doing + * out-of-order writes. + */ + ll_merge_attr(env, inode); + pos = io->u.ci_wr.wr.crw_pos = i_size_read(inode); + vio->vui_iocb->ki_pos = pos; + } else { + LASSERT(vio->vui_iocb->ki_pos == pos); } - CDEBUG(D_VFSTRACE, "write: [%lli, %lli)\n", pos, pos + (long long)cnt); + CDEBUG(D_VFSTRACE, "write: [%lli, %lli)\n", pos, pos + (long long)cnt); + + /* The maximum Lustre file size is variable, based on the OST maximum + * object size and number of stripes. This needs another check in + * addition to the VFS checks earlier. */ + if (pos + cnt > ll_file_maxbytes(inode)) { + CDEBUG(D_INODE, + "%s: file "DFID" offset %llu > maxbytes "LPU64"\n", + ll_get_fsname(inode->i_sb, NULL, 0), + PFID(ll_inode2fid(inode)), pos + cnt, + ll_file_maxbytes(inode)); + RETURN(-EFBIG); + } - if (cio->cui_iov == NULL) /* from a temp io in ll_cl_init(). */ - result = 0; - else - result = generic_file_aio_write(cio->cui_iocb, - cio->cui_iov, cio->cui_nrsegs, - cio->cui_iocb->ki_pos); + if (vio->vui_iter == NULL) { + /* from a temp io in ll_cl_init(). */ + result = 0; + } else { + /* + * When using the locked AIO function (generic_file_aio_write()) + * testing has shown the inode mutex to be a limiting factor + * with multi-threaded single shared file performance. To get + * around this, we now use the lockless version. To maintain + * consistency, proper locking to protect against writes, + * trucates, etc. is handled in the higher layers of lustre. + */ +#ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER + result = generic_file_write_iter(vio->vui_iocb, vio->vui_iter); +#else + result = __generic_file_aio_write(vio->vui_iocb, + vio->vui_iter->iov, + vio->vui_iter->nr_segs, + &vio->vui_iocb->ki_pos); +#endif + if (result > 0 || result == -EIOCBQUEUED) { + ssize_t err; + + err = generic_write_sync(vio->vui_iocb->ki_filp, + pos, result); + if (err < 0 && result > 0) + result = err; + } + + } if (result > 0) { result = vvp_io_write_commit(env, io); - if (cio->u.write.cui_written > 0) { - result = cio->u.write.cui_written; + if (vio->u.write.vui_written > 0) { + result = vio->u.write.vui_written; io->ci_nob += result; CDEBUG(D_VFSTRACE, "write: nob %zd, result: %zd\n", @@ -824,96 +1132,97 @@ static int vvp_io_write_start(const struct lu_env *env, } } if (result > 0) { - 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_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED); if (result < cnt) io->ci_continue = 0; ll_rw_stats_tally(ll_i2sbi(inode), current->pid, - cio->cui_fd, pos, result, WRITE); + vio->vui_fd, pos, result, WRITE); result = 0; } RETURN(result); } +static void vvp_io_rw_end(const struct lu_env *env, + const struct cl_io_slice *ios) +{ + struct vvp_io *vio = cl2vvp_io(env, ios); + struct inode *inode = vvp_object_inode(ios->cis_obj); + struct ll_inode_info *lli = ll_i2info(inode); + + if (vio->vui_io_subtype == IO_NORMAL) + up_read(&lli->lli_trunc_sem); +} + static int vvp_io_kernel_fault(struct vvp_fault_io *cfio) { - struct vm_fault *vmf = cfio->fault.ft_vmf; + struct vm_fault *vmf = cfio->ft_vmf; - cfio->fault.ft_flags = filemap_fault(cfio->ft_vma, vmf); + cfio->ft_flags = filemap_fault(cfio->ft_vma, vmf); + cfio->ft_flags_valid = 1; - if (vmf->page) { - LL_CDEBUG_PAGE(D_PAGE, vmf->page, "got addr %p type NOPAGE\n", - vmf->virtual_address); - if (unlikely(!(cfio->fault.ft_flags & VM_FAULT_LOCKED))) { - lock_page(vmf->page); - cfio->fault.ft_flags &= VM_FAULT_LOCKED; - } + if (vmf->page) { + LL_CDEBUG_PAGE(D_PAGE, vmf->page, "got addr %p type NOPAGE\n", + vmf->virtual_address); + if (unlikely(!(cfio->ft_flags & VM_FAULT_LOCKED))) { + lock_page(vmf->page); + cfio->ft_flags |= VM_FAULT_LOCKED; + } - cfio->ft_vmpage = vmf->page; - return 0; - } + cfio->ft_vmpage = vmf->page; - if (cfio->fault.ft_flags & VM_FAULT_SIGBUS) { - CDEBUG(D_PAGE, "got addr %p - SIGBUS\n", vmf->virtual_address); - return -EFAULT; - } + return 0; + } - if (cfio->fault.ft_flags & VM_FAULT_OOM) { - CDEBUG(D_PAGE, "got addr %p - OOM\n", vmf->virtual_address); - return -ENOMEM; - } + if (cfio->ft_flags & VM_FAULT_SIGBUS) { + CDEBUG(D_PAGE, "got addr %p - SIGBUS\n", vmf->virtual_address); + return -EFAULT; + } + + if (cfio->ft_flags & VM_FAULT_OOM) { + CDEBUG(D_PAGE, "got addr %p - OOM\n", vmf->virtual_address); + return -ENOMEM; + } + + if (cfio->ft_flags & VM_FAULT_RETRY) + return -EAGAIN; - if (cfio->fault.ft_flags & VM_FAULT_RETRY) - return -EAGAIN; + CERROR("unknown error in page fault %d\n", cfio->ft_flags); - CERROR("unknow error in page fault %d!\n", cfio->fault.ft_flags); - return -EINVAL; + return -EINVAL; } static void mkwrite_commit_callback(const struct lu_env *env, struct cl_io *io, struct cl_page *page) { - struct ccc_page *cp; - struct cl_object *clob = cl_io_top(io)->ci_obj; - set_page_dirty(page->cp_vmpage); - - cp = cl2ccc_page(cl_object_page_slice(clob, page)); - vvp_write_pending(cl2ccc(clob), cp); } static int vvp_io_fault_start(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 inode *inode = ccc_object_inode(obj); - struct cl_fault_io *fio = &io->u.ci_fault; - struct vvp_fault_io *cfio = &vio->u.fault; - loff_t offset; - int result = 0; - struct page *vmpage = NULL; - struct cl_page *page; - loff_t size; - pgoff_t last_index; + struct vvp_io *vio = cl2vvp_io(env, ios); + struct cl_io *io = ios->cis_io; + struct cl_object *obj = io->ci_obj; + struct inode *inode = vvp_object_inode(obj); + struct ll_inode_info *lli = ll_i2info(inode); + struct cl_fault_io *fio = &io->u.ci_fault; + struct vvp_fault_io *cfio = &vio->u.fault; + loff_t offset; + int result = 0; + struct page *vmpage = NULL; + struct cl_page *page; + loff_t size; + pgoff_t last_index; ENTRY; - if (fio->ft_executable && - LTIME_S(inode->i_mtime) != vio->u.fault.ft_mtime) - CWARN("binary "DFID - " changed while waiting for the page fault lock\n", - PFID(lu_object_fid(&obj->co_lu))); + down_read(&lli->lli_trunc_sem); /* offset of the last byte on the page */ offset = cl_offset(obj, fio->ft_index + 1) - 1; LASSERT(cl_index(obj, offset) == fio->ft_index); - result = ccc_prep_size(env, obj, io, 0, offset + 1, NULL); + result = vvp_prep_size(env, obj, io, 0, offset + 1, NULL); if (result != 0) RETURN(result); @@ -984,7 +1293,7 @@ static int vvp_io_fault_start(const struct lu_env *env, wait_on_page_writeback(vmpage); if (!PageDirty(vmpage)) { struct cl_page_list *plist = &io->ci_queue.c2_qin; - struct ccc_page *cp = cl_object_page_slice(obj, page); + struct vvp_page *vpg = cl_object_page_slice(obj, page); int to = PAGE_SIZE; /* vvp_page_assume() calls wait_on_page_writeback(). */ @@ -994,8 +1303,8 @@ static int vvp_io_fault_start(const struct lu_env *env, cl_page_list_add(plist, page); /* size fixup */ - if (last_index == ccc_index(cp)) - to = size & ~CFS_PAGE_MASK; + if (last_index == vvp_index(vpg)) + to = size & ~PAGE_MASK; /* Do not set Dirty bit here so that in case IO is * started before the page is really made dirty, we @@ -1044,10 +1353,23 @@ out: /* return unlocked vmpage to avoid deadlocking */ if (vmpage != NULL) unlock_page(vmpage); - cfio->fault.ft_flags &= ~VM_FAULT_LOCKED; + + cfio->ft_flags &= ~VM_FAULT_LOCKED; + return result; } +static void vvp_io_fault_end(const struct lu_env *env, + const struct cl_io_slice *ios) +{ + struct inode *inode = vvp_object_inode(ios->cis_obj); + struct ll_inode_info *lli = ll_i2info(inode); + + CLOBINVRNT(env, ios->cis_io->ci_obj, + vvp_object_invariant(ios->cis_io->ci_obj)); + up_read(&lli->lli_trunc_sem); +} + static int vvp_io_fsync_start(const struct lu_env *env, const struct cl_io_slice *ios) { @@ -1057,59 +1379,43 @@ static int vvp_io_fsync_start(const struct lu_env *env, return 0; } -static int vvp_io_read_page(const struct lu_env *env, - const struct cl_io_slice *ios, - const struct cl_page_slice *slice) +static int vvp_io_read_ahead(const struct lu_env *env, + const struct cl_io_slice *ios, + pgoff_t start, struct cl_read_ahead *ra) { - struct cl_io *io = ios->cis_io; - struct ccc_page *cp = cl2ccc_page(slice); - struct cl_page *page = slice->cpl_page; - struct inode *inode = ccc_object_inode(slice->cpl_obj); - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ll_file_data *fd = cl2ccc_io(env, ios)->cui_fd; - struct ll_readahead_state *ras = &fd->fd_ras; - struct cl_2queue *queue = &io->ci_queue; - + int result = 0; ENTRY; - if (sbi->ll_ra_info.ra_max_pages_per_file > 0 && - sbi->ll_ra_info.ra_max_pages > 0) - ras_update(sbi, inode, ras, ccc_index(cp), - cp->cpg_defer_uptodate); + if (ios->cis_io->ci_type == CIT_READ || + ios->cis_io->ci_type == CIT_FAULT) { + struct vvp_io *vio = cl2vvp_io(env, ios); - if (cp->cpg_defer_uptodate) { - cp->cpg_ra_used = 1; - cl_page_export(env, page, 1); - } - - /* - * Add page into the queue even when it is marked uptodate above. - * this will unlock it automatically as part of cl_page_list_disown(). - */ - cl_2queue_add(queue, page); - if (sbi->ll_ra_info.ra_max_pages_per_file > 0 && - sbi->ll_ra_info.ra_max_pages > 0) - ll_readahead(env, io, &queue->c2_qin, ras, - cp->cpg_defer_uptodate); + if (unlikely(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { + ra->cra_end = CL_PAGE_EOF; + result = +1; /* no need to call down */ + } + } - RETURN(0); + RETURN(result); } static const struct cl_io_operations vvp_io_ops = { - .op = { - [CIT_READ] = { - .cio_fini = vvp_io_read_fini, - .cio_lock = vvp_io_read_lock, - .cio_start = vvp_io_read_start, - .cio_advance = ccc_io_advance - }, + .op = { + [CIT_READ] = { + .cio_fini = vvp_io_fini, + .cio_lock = vvp_io_read_lock, + .cio_start = vvp_io_read_start, + .cio_end = vvp_io_rw_end, + .cio_advance = vvp_io_advance, + }, [CIT_WRITE] = { .cio_fini = vvp_io_fini, .cio_iter_init = vvp_io_write_iter_init, .cio_iter_fini = vvp_io_write_iter_fini, .cio_lock = vvp_io_write_lock, .cio_start = vvp_io_write_start, - .cio_advance = ccc_io_advance + .cio_end = vvp_io_rw_end, + .cio_advance = vvp_io_advance, }, [CIT_SETATTR] = { .cio_fini = vvp_io_setattr_fini, @@ -1123,39 +1429,41 @@ static const struct cl_io_operations vvp_io_ops = { .cio_iter_init = vvp_io_fault_iter_init, .cio_lock = vvp_io_fault_lock, .cio_start = vvp_io_fault_start, - .cio_end = ccc_io_end + .cio_end = vvp_io_fault_end, }, [CIT_FSYNC] = { - .cio_start = vvp_io_fsync_start, - .cio_fini = vvp_io_fini + .cio_start = vvp_io_fsync_start, + .cio_fini = vvp_io_fini + }, + [CIT_MISC] = { + .cio_fini = vvp_io_fini }, - [CIT_MISC] = { - .cio_fini = vvp_io_fini - } - }, - .cio_read_page = vvp_io_read_page, + [CIT_LADVISE] = { + .cio_fini = vvp_io_fini + }, + }, + .cio_read_ahead = vvp_io_read_ahead }; int vvp_io_init(const struct lu_env *env, struct cl_object *obj, struct cl_io *io) { struct vvp_io *vio = vvp_env_io(env); - struct ccc_io *cio = ccc_env_io(env); - struct inode *inode = ccc_object_inode(obj); - int result; + struct inode *inode = vvp_object_inode(obj); + int result; - CLOBINVRNT(env, obj, ccc_object_invariant(obj)); - ENTRY; + CLOBINVRNT(env, obj, vvp_object_invariant(obj)); + ENTRY; CDEBUG(D_VFSTRACE, DFID" ignore/verify layout %d/%d, layout version %d " - "restore needed %d\n", + "restore needed %d\n", PFID(lu_object_fid(&obj->co_lu)), io->ci_ignore_layout, io->ci_verify_layout, - cio->cui_layout_gen, io->ci_restore_needed); + vio->vui_layout_gen, io->ci_restore_needed); - CL_IO_SLICE_CLEAN(cio, cui_cl); - cl_io_slice_add(io, &cio->cui_cl, obj, &vvp_io_ops); - vio->cui_ra_window_set = 0; + CL_IO_SLICE_CLEAN(vio, vui_cl); + cl_io_slice_add(io, &vio->vui_cl, obj, &vvp_io_ops); + vio->vui_ra_valid = false; result = 0; if (io->ci_type == CIT_READ || io->ci_type == CIT_WRITE) { size_t count; @@ -1166,10 +1474,12 @@ int vvp_io_init(const struct lu_env *env, struct cl_object *obj, * results." -- Single Unix Spec */ if (count == 0) result = 1; - else { - cio->cui_tot_count = count; - cio->cui_tot_nrsegs = 0; - } + else { + vio->vui_tot_count = count; +#ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER + vio->vui_tot_nrsegs = 0; +#endif + } /* for read/write, we store the jobid in the inode, and * it'll be fetched by osc when building RPC. @@ -1193,7 +1503,7 @@ int vvp_io_init(const struct lu_env *env, struct cl_object *obj, * even for operations requiring to open file, such as read and write, * because it might not grant layout lock in IT_OPEN. */ if (result == 0 && !io->ci_ignore_layout) { - result = ll_layout_refresh(inode, &cio->cui_layout_gen); + result = ll_layout_refresh(inode, &vio->vui_layout_gen); if (result == -ENOENT) /* If the inode on MDS has been removed, but the objects * on OSTs haven't been destroyed (async unlink), layout @@ -1208,11 +1518,3 @@ int vvp_io_init(const struct lu_env *env, struct cl_object *obj, RETURN(result); } - -static struct vvp_io *cl2vvp_io(const struct lu_env *env, - const struct cl_io_slice *slice) -{ - /* Caling just for assertion */ - cl2ccc_io(env, slice); - return vvp_env_io(env); -}