X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Fvvp_io.c;h=a29574c70861d967af0f4db52cf217384991b717;hp=0df098a02fecce94297096ae57b37a856a64399d;hb=d2ff62aa3509659da92c1a33bd7849b3f04f7a2d;hpb=422e5995808bee60407a78020825ab369e1449bd diff --git a/lustre/llite/vvp_io.c b/lustre/llite/vvp_io.c index 0df098a..a29574c 100644 --- a/lustre/llite/vvp_io.c +++ b/lustre/llite/vvp_io.c @@ -52,6 +52,18 @@ static struct vvp_io *cl2vvp_io(const struct lu_env *env, const struct cl_io_slice *slice); +/** + * True, if \a io is a normal io, False for sendfile() / splice_{read|write} + */ +int cl_is_normalio(const struct lu_env *env, const struct cl_io *io) +{ + struct vvp_io *vio = vvp_env_io(env); + + LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); + + return vio->cui_io_subtype == IO_NORMAL; +} + /***************************************************************************** * * io operations. @@ -76,9 +88,7 @@ static void vvp_io_fini(const struct lu_env *env, const struct cl_io_slice *ios) struct cl_object *obj = io->ci_obj; CLOBINVRNT(env, obj, ccc_object_invariant(obj)); - if (io->ci_type == CIT_WRITE) - up(&ll_i2info(ccc_object_inode(obj))->lli_write_sem); - else { + if (io->ci_type == CIT_READ) { struct vvp_io *vio = cl2vvp_io(env, ios); struct ccc_io *cio = cl2ccc_io(env, ios); @@ -123,7 +133,6 @@ static int vvp_mmap_locks(const struct lu_env *env, struct vm_area_struct *vma; struct cl_lock_descr *descr = &cti->cti_descr; ldlm_policy_data_t policy; - struct inode *inode; unsigned long addr; unsigned long seg; ssize_t count; @@ -132,7 +141,7 @@ static int vvp_mmap_locks(const struct lu_env *env, LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); - if (cl_io_is_sendfile(io)) + if (!cl_is_normalio(env, io)) RETURN(0); for (seg = 0; seg < vio->cui_nrsegs; seg++) { @@ -146,16 +155,16 @@ static int vvp_mmap_locks(const struct lu_env *env, count += addr & (~CFS_PAGE_MASK); addr &= CFS_PAGE_MASK; while((vma = our_vma(addr, count)) != NULL) { - struct file *file = vma->vm_file; - struct ll_file_data *fd; + struct inode *inode = vma->vm_file->f_dentry->d_inode; + int flags = CEF_MUST; - LASSERT(file); - fd = LUSTRE_FPRIVATE(file); - - inode = file->f_dentry->d_inode; - if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK || - ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK)) - goto cont; + if (ll_file_nolock(vma->vm_file)) { + /* + * For no lock case, a lockless lock will be + * generated. + */ + flags = CEF_NEVER; + } /* * XXX: Required lock mode can be weakened: CIT_WRITE @@ -169,13 +178,14 @@ static int vvp_mmap_locks(const struct lu_env *env, policy.l_extent.start); descr->cld_end = cl_index(descr->cld_obj, policy.l_extent.end); - result = cl_io_lock_alloc_add(env, io, descr, CEF_MUST); + descr->cld_enq_flags = flags; + result = cl_io_lock_alloc_add(env, io, descr); if (result < 0) RETURN(result); - cont: if (vma->vm_end - addr >= count) break; + count -= vma->vm_end - addr; addr = vma->vm_end; } @@ -191,7 +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); @@ -242,20 +251,15 @@ 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, @@ -280,7 +284,7 @@ static int vvp_io_trunc_iter_init(const struct lu_env *env, } /** - * Implementation of cl_io_operations::vio_lock() method for CIT_TRUNC io. + * Implementation of cl_io_operations::cio_lock() method for CIT_TRUNC io. * * Handles "lockless io" mode when extent locking is done by server. */ @@ -312,7 +316,6 @@ static int vvp_io_trunc_start(const struct lu_env *env, int result; LASSERT(cio->u.trunc.cui_locks_released); - LASSERT(vio->cui_oneshot == 0); LOCK_INODE_MUTEX(inode); DOWN_WRITE_I_ALLOC_SEM(inode); @@ -424,14 +427,13 @@ static int vvp_io_read_start(const struct lu_env *env, int result; loff_t pos = io->u.ci_rd.rd.crw_pos; - size_t cnt = io->u.ci_rd.rd.crw_count; - size_t tot = cio->cui_tot_count; + long cnt = io->u.ci_rd.rd.crw_count; + long tot = cio->cui_tot_count; int exceed = 0; CLOBINVRNT(env, obj, ccc_object_invariant(obj)); - LASSERT(vio->cui_oneshot == 0); - CDEBUG(D_VFSTRACE, "read: -> [%lli, %lli)\n", pos, pos + (long long)cnt); + CDEBUG(D_VFSTRACE, "read: -> [%lli, %lli)\n", pos, pos + cnt); result = ccc_prep_size(env, obj, io, pos, tot, 1, &exceed); if (result != 0) @@ -440,7 +442,7 @@ static int vvp_io_read_start(const struct lu_env *env, goto out; LU_OBJECT_HEADER(D_INODE, env, &obj->co_lu, - "Read ino %lu, "LPSZ" bytes, offset %lld, size %llu\n", + "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 */ @@ -459,11 +461,27 @@ static int vvp_io_read_start(const struct lu_env *env, /* BUG: 5972 */ file_accessed(file); - if (cl_io_is_sendfile(io)) { + switch (vio->cui_io_subtype) { + case IO_NORMAL: + result = lustre_generic_file_read(file, cio, &pos); + break; +#ifdef HAVE_KERNEL_SENDFILE + case IO_SENDFILE: result = generic_file_sendfile(file, &pos, cnt, - vio->u.read.cui_actor, vio->u.read.cui_target); - } else { - result = lustre_generic_file_read(file, cio, &pos); + vio->u.sendfile.cui_actor, + vio->u.sendfile.cui_target); + break; +#endif +#ifdef HAVE_KERNEL_SPLICE_READ + case IO_SPLICE: + result = generic_file_splice_read(file, &pos, + vio->u.splice.cui_pipe, cnt, + vio->u.splice.cui_flags); + break; +#endif + default: + CERROR("Wrong IO type %u\n", vio->cui_io_subtype); + LBUG(); } out: @@ -505,7 +523,7 @@ 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); @@ -521,6 +539,69 @@ static int vvp_io_write_start(const struct lu_env *env, RETURN(result); } +#ifndef HAVE_VM_OP_FAULT +static int vvp_io_kernel_fault(struct vvp_fault_io *cfio) +{ + cfs_page_t *vmpage; + + vmpage = filemap_nopage(cfio->ft_vma, cfio->nopage.ft_address, + cfio->nopage.ft_type); + + if (vmpage == NOPAGE_SIGBUS) { + CDEBUG(D_PAGE, "got addr %lu type %lx - SIGBUS\n", + cfio->nopage.ft_address,(long)cfio->nopage.ft_type); + return -EFAULT; + } else if (vmpage == NOPAGE_OOM) { + CDEBUG(D_PAGE, "got addr %lu type %lx - OOM\n", + cfio->nopage.ft_address, (long)cfio->nopage.ft_type); + return -ENOMEM; + } + + LL_CDEBUG_PAGE(D_PAGE, vmpage, "got addr %lu type %lx\n", + cfio->nopage.ft_address, (long)cfio->nopage.ft_type); + + cfio->ft_vmpage = vmpage; + + return 0; +} +#else +static int vvp_io_kernel_fault(struct vvp_fault_io *cfio) +{ + cfio->fault.ft_flags = filemap_fault(cfio->ft_vma, cfio->fault.ft_vmf); + + if (cfio->fault.ft_vmf->page) { + LL_CDEBUG_PAGE(D_PAGE, cfio->fault.ft_vmf->page, + "got addr %p type NOPAGE\n", + cfio->fault.ft_vmf->virtual_address); + /*XXX workaround to bug in CLIO - he deadlocked with + lock cancel if page locked */ + if (likely(cfio->fault.ft_flags & VM_FAULT_LOCKED)) { + unlock_page(cfio->fault.ft_vmf->page); + cfio->fault.ft_flags &= ~VM_FAULT_LOCKED; + } + + cfio->ft_vmpage = cfio->fault.ft_vmf->page; + return 0; + } + + if (unlikely (cfio->fault.ft_flags & VM_FAULT_ERROR)) { + CDEBUG(D_PAGE, "got addr %p - SIGBUS\n", + cfio->fault.ft_vmf->virtual_address); + return -EFAULT; + } + + if (unlikely (cfio->fault.ft_flags & VM_FAULT_NOPAGE)) { + CDEBUG(D_PAGE, "got addr %p - OOM\n", + cfio->fault.ft_vmf->virtual_address); + return -ENOMEM; + } + + CERROR("unknow error in page fault!\n"); + return -EINVAL; +} + +#endif + static int vvp_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios) { @@ -530,11 +611,12 @@ static int vvp_io_fault_start(const struct lu_env *env, struct inode *inode = ccc_object_inode(obj); struct cl_fault_io *fio = &io->u.ci_fault; struct vvp_fault_io *cfio = &vio->u.fault; - cfs_page_t *vmpage; loff_t offset; + int kernel_result = 0; int result = 0; - - LASSERT(vio->cui_oneshot == 0); + struct cl_page *page; + loff_t size; + pgoff_t last; /* last page in a file data region */ if (fio->ft_executable && LTIME_S(inode->i_mtime) != vio->u.fault.ft_mtime) @@ -549,55 +631,44 @@ static int vvp_io_fault_start(const struct lu_env *env, if (result != 0) return result; - vmpage = filemap_nopage(cfio->ft_vma, cfio->ft_address, cfio->ft_type); - if (vmpage != NOPAGE_SIGBUS && vmpage != NOPAGE_OOM) - LL_CDEBUG_PAGE(D_PAGE, vmpage, - "got addr %lu type %lx\n", - cfio->ft_address, (long)cfio->ft_type); - else - CDEBUG(D_PAGE, "got addr %lu type %lx - SIGBUS\n", - cfio->ft_address, (long)cfio->ft_type); - - if (vmpage == NOPAGE_SIGBUS) - result = -EFAULT; - else if (vmpage == NOPAGE_OOM) - result = -ENOMEM; - else { - struct cl_page *page; - loff_t size; - pgoff_t last; /* last page in a file data region */ - - /* Temporarily lock vmpage to keep cl_page_find() happy. */ - lock_page(vmpage); - page = cl_page_find(env, obj, fio->ft_index, vmpage, - CPT_CACHEABLE); - unlock_page(vmpage); - if (!IS_ERR(page)) { - size = i_size_read(inode); - last = cl_index(obj, size - 1); - if (fio->ft_index == last) - /* - * Last page is mapped partially. - */ - fio->ft_nob = size - cl_offset(obj, - fio->ft_index); - else - fio->ft_nob = cl_page_size(obj); - lu_ref_add(&page->cp_reference, "fault", io); - fio->ft_page = page; - /* - * Certain 2.6 kernels return not-NULL from - * filemap_nopage() when page is beyond the file size, - * on the grounds that "An external ptracer can access - * pages that normally aren't accessible.." Don't - * propagate such page fault to the lower layers to - * avoid side-effects like KMS updates. - */ - if (fio->ft_index > last) - result = +1; - } else - result = PTR_ERR(page); + /* must return locked page */ + kernel_result = vvp_io_kernel_fault(cfio); + if (kernel_result != 0) + return kernel_result; + /* Temporarily lock vmpage to keep cl_page_find() happy. */ + lock_page(cfio->ft_vmpage); + page = cl_page_find(env, obj, fio->ft_index, cfio->ft_vmpage, + CPT_CACHEABLE); + unlock_page(cfio->ft_vmpage); + if (IS_ERR(page)) { + page_cache_release(cfio->ft_vmpage); + cfio->ft_vmpage = NULL; + return PTR_ERR(page); } + + size = i_size_read(inode); + last = cl_index(obj, size - 1); + if (fio->ft_index == last) + /* + * Last page is mapped partially. + */ + fio->ft_nob = size - cl_offset(obj, fio->ft_index); + else + fio->ft_nob = cl_page_size(obj); + + lu_ref_add(&page->cp_reference, "fault", io); + fio->ft_page = page; + /* + * Certain 2.6 kernels return not-NULL from + * filemap_nopage() when page is beyond the file size, + * on the grounds that "An external ptracer can access + * pages that normally aren't accessible.." Don't + * propagate such page fault to the lower layers to + * avoid side-effects like KMS updates. + */ + if (fio->ft_index > last) + result = +1; + return result; } @@ -615,9 +686,9 @@ static int vvp_io_read_page(const struct lu_env *env, struct ll_readahead_state *ras = &fd->fd_ras; cfs_page_t *vmpage = cp->cpg_page; struct cl_2queue *queue = &io->ci_queue; + int rc; CLOBINVRNT(env, obj, ccc_object_invariant(obj)); - LASSERT(cl2vvp_io(env, ios)->cui_oneshot == 0); LASSERT(slice->cpl_obj == obj); ENTRY; @@ -627,22 +698,18 @@ static int vvp_io_read_page(const struct lu_env *env, cp->cpg_defer_uptodate); /* Sanity check whether the page is protected by a lock. */ - if (likely(!(fd->fd_flags & LL_FILE_IGNORE_LOCK))) { - int rc; - - rc = cl_page_is_under_lock(env, io, page); - if (rc != -EBUSY) { - CL_PAGE_HEADER(D_WARNING, env, page, "%s: %i\n", - rc == -ENODATA ? "without a lock" : - "match failed", rc); - if (rc != -ENODATA) - RETURN(rc); - } + rc = cl_page_is_under_lock(env, io, page); + if (rc != -EBUSY) { + CL_PAGE_HEADER(D_WARNING, env, page, "%s: %i\n", + rc == -ENODATA ? "without a lock" : + "match failed", rc); + if (rc != -ENODATA) + RETURN(rc); } if (cp->cpg_defer_uptodate) { cp->cpg_ra_used = 1; - cl_page_export(env, page); + cl_page_export(env, page, 1); } /* * Add page into the queue even when it is marked uptodate above. @@ -661,10 +728,6 @@ static int vvp_page_sync_io(const struct lu_env *env, struct cl_io *io, int to, enum cl_req_type crt) { struct cl_2queue *queue; - struct ccc_object *cobo = cl2ccc(page->cp_obj); - struct cl_sync_io *anchor = &ccc_env_info(env)->cti_sync_io; - - int writing = io->ci_type == CIT_WRITE; int result; LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE); @@ -672,19 +735,9 @@ static int vvp_page_sync_io(const struct lu_env *env, struct cl_io *io, queue = &io->ci_queue; cl_2queue_init_page(queue, page); - - if (writing) - /* Do not pass llap here as it is sync write. */ - vvp_write_pending(cobo, cp); - - cl_sync_io_init(anchor, 1); - cp->cpg_sync_io = anchor; cl_page_clip(env, page, 0, to); - result = cl_io_submit_rw(env, io, crt, queue, CRP_NORMAL); - if (result == 0) - result = cl_sync_io_wait(env, io, &queue->c2_qout, anchor); - else - cp->cpg_sync_io = NULL; + + 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); @@ -707,7 +760,7 @@ static int vvp_io_prepare_partial(const struct lu_env *env, struct cl_io *io, struct ccc_page *cp, unsigned from, unsigned to) { - struct cl_attr *attr = &ccc_env_info(env)->cti_attr; + struct cl_attr *attr = ccc_env_thread_attr(env); loff_t offset = cl_offset(obj, pg->cp_index); int result; @@ -739,7 +792,7 @@ static int vvp_io_prepare_partial(const struct lu_env *env, struct cl_io *io, * details. -jay */ if (result == 0) - cl_page_export(env, pg); + cl_page_export(env, pg, 1); } return result; } @@ -829,6 +882,9 @@ static int vvp_io_commit_write(const struct lu_env *env, tallyop = LPROC_LL_DIRTY_MISSES; vvp_write_pending(cl2ccc(obj), cp); set_page_dirty(vmpage); + /* ll_set_page_dirty() does the same for now, but + * it will not soon. */ + vvp_write_pending(cl2ccc(obj), cp); result = cl_page_cache_add(env, io, pg, CRT_WRITE); if (result == -EDQUOT) /* @@ -861,7 +917,7 @@ static int vvp_io_commit_write(const struct lu_env *env, if (result == 0) { if (size > i_size_read(inode)) i_size_write(inode, size); - cl_page_export(env, pg); + cl_page_export(env, pg, 1); } else if (size > i_size_read(inode)) cl_page_discard(env, io, pg); RETURN(result); @@ -918,7 +974,6 @@ 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) { @@ -928,8 +983,6 @@ int vvp_io_init(const struct lu_env *env, struct cl_object *obj, count = io->u.ci_rw.crw_count; op = io->ci_type == CIT_READ ? LPROC_LL_READ_BYTES : LPROC_LL_WRITE_BYTES; - if (io->ci_type == CIT_WRITE) - down(&ll_i2info(inode)->lli_write_sem); /* "If nbyte is 0, read() will return 0 and have no other * results." -- Single Unix Spec */ if (count == 0)