X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Frw26.c;h=6cbf9e0cb2f6eba279c54000a814467e7d2d1156;hp=3cfc77d33e543991dee0d92285dd6716182482a9;hb=322968acf183ab16d952cd3026f6580957b31259;hpb=b3fa6d5458e8ae1e4b695e03097c638db2738a38 diff --git a/lustre/llite/rw26.c b/lustre/llite/rw26.c index 3cfc77d..6cbf9e0 100644 --- a/lustre/llite/rw26.c +++ b/lustre/llite/rw26.c @@ -1,24 +1,44 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Lustre Lite I/O page cache routines for the 2.5/2.6 kernel version + * GPL HEADER START + * + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * Copyright (c) 2001-2003 Cluster File Systems, Inc. + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * This file is part of Lustre, http://www.lustre.org. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * - * Lustre is free software; you can redistribute it and/or - * modify it under the terms of version 2 of the GNU General Public - * License as published by the Free Software Foundation. + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. * - * Lustre is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * GPL HEADER END + */ +/* + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. + * Use is subject to license terms. + * + * Copyright (c) 2011 Whamcloud, Inc. * - * You should have received a copy of the GNU General Public License - * along with Lustre; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. + * + * lustre/lustre/llite/rw26.c + * + * Lustre Lite I/O page cache routines for the 2.5/2.6 kernel version */ #ifndef AUTOCONF_INCLUDED @@ -37,73 +57,128 @@ #include #include -#include #include #include #include -#include #include #include #include #define DEBUG_SUBSYSTEM S_LLITE -//#include #include #include "llite_internal.h" #include -static int ll_writepage_26(struct page *page, struct writeback_control *wbc) +/** + * Implements Linux VM address_space::invalidatepage() method. This method is + * called when the page is truncate from a file, either as a result of + * explicit truncate, or when inode is removed from memory (as a result of + * final iput(), umount, or memory pressure induced icache shrinking). + * + * [0, offset] bytes of the page remain valid (this is for a case of not-page + * aligned truncate). Lustre leaves partially truncated page in the cache, + * relying on struct inode::i_size to limit further accesses. + */ +static int cl_invalidatepage(struct page *vmpage, unsigned long offset) { - return ll_writepage(page); + struct inode *inode; + struct lu_env *env; + struct cl_page *page; + struct cl_object *obj; + + int result; + int refcheck; + + LASSERT(PageLocked(vmpage)); + LASSERT(!PageWriteback(vmpage)); + + /* + * It is safe to not check anything in invalidatepage/releasepage + * below because they are run with page locked and all our io is + * happening with locked page too + */ + result = 0; + if (offset == 0) { + env = cl_env_get(&refcheck); + if (!IS_ERR(env)) { + inode = vmpage->mapping->host; + obj = ll_i2info(inode)->lli_clob; + if (obj != NULL) { + page = cl_vmpage_page(vmpage, obj); + if (page != NULL) { + lu_ref_add(&page->cp_reference, + "delete", vmpage); + cl_page_delete(env, page); + result = 1; + lu_ref_del(&page->cp_reference, + "delete", vmpage); + cl_page_put(env, page); + } + } else + LASSERT(vmpage->private == 0); + cl_env_put(env, &refcheck); + } + } + return result; } -/* It is safe to not check anything in invalidatepage/releasepage below - because they are run with page locked and all our io is happening with - locked page too */ #ifdef HAVE_INVALIDATEPAGE_RETURN_INT static int ll_invalidatepage(struct page *page, unsigned long offset) { - if (offset) - return 0; - if (PagePrivate(page)) - ll_removepage(page); - return 1; + return cl_invalidatepage(page, offset); } -#else +#else /* !HAVE_INVALIDATEPAGE_RETURN_INT */ static void ll_invalidatepage(struct page *page, unsigned long offset) { - if (offset == 0 && PagePrivate(page)) - ll_removepage(page); + cl_invalidatepage(page, offset); } #endif -static int ll_releasepage(struct page *page, int gfp_mask) + +#ifdef HAVE_RELEASEPAGE_WITH_INT +#define RELEASEPAGE_ARG_TYPE int +#else +#define RELEASEPAGE_ARG_TYPE gfp_t +#endif +static int ll_releasepage(struct page *page, RELEASEPAGE_ARG_TYPE gfp_mask) { - if (PagePrivate(page)) - ll_removepage(page); + void *cookie; + + cookie = cl_env_reenter(); + ll_invalidatepage(page, 0); + cl_env_reexit(cookie); return 1; } -static int ll_set_page_dirty(struct page *page) +static int ll_set_page_dirty(struct page *vmpage) { - struct ll_async_page *llap; - ENTRY; - - llap = llap_from_page(page, LLAP_ORIGIN_UNKNOWN); - if (IS_ERR(llap)) - RETURN(PTR_ERR(llap)); - - llap_write_pending(page->mapping->host, llap); - RETURN(__set_page_dirty_nobuffers(page)); +#if 0 + struct cl_page *page = vvp_vmpage_page_transient(vmpage); + struct vvp_object *obj = cl_inode2vvp(vmpage->mapping->host); + struct vvp_page *cpg; + + /* + * XXX should page method be called here? + */ + LASSERT(&obj->co_cl == page->cp_obj); + cpg = cl2vvp_page(cl_page_at(page, &vvp_device_type)); + /* + * XXX cannot do much here, because page is possibly not locked: + * sys_munmap()->... + * ->unmap_page_range()->zap_pte_range()->set_page_dirty(). + */ + vvp_write_pending(obj, cpg); +#endif + RETURN(__set_page_dirty_nobuffers(vmpage)); } #define MAX_DIRECTIO_SIZE 2*1024*1024*1024UL static inline int ll_get_user_pages(int rw, unsigned long user_addr, - size_t size, struct page ***pages) + size_t size, struct page ***pages, + int *max_pages) { int result = -ENOMEM; - int page_count; /* set an arbitrary limit to prevent arithmetic overflow */ if (size > MAX_DIRECTIO_SIZE) { @@ -111,18 +186,18 @@ static inline int ll_get_user_pages(int rw, unsigned long user_addr, return -EFBIG; } - page_count = (user_addr + size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT; - page_count -= user_addr >> CFS_PAGE_SHIFT; + *max_pages = (user_addr + size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT; + *max_pages -= user_addr >> CFS_PAGE_SHIFT; - OBD_ALLOC_WAIT(*pages, page_count * sizeof(**pages)); + OBD_ALLOC_LARGE(*pages, *max_pages * sizeof(**pages)); if (*pages) { down_read(¤t->mm->mmap_sem); result = get_user_pages(current, current->mm, user_addr, - page_count, (rw == READ), 0, *pages, + *max_pages, (rw == READ), 0, *pages, NULL); up_read(¤t->mm->mmap_sem); - if (result < 0) - OBD_FREE(*pages, page_count * sizeof(**pages)); + if (unlikely(result <= 0)) + OBD_FREE_LARGE(*pages, *max_pages * sizeof(**pages)); } return result; @@ -135,91 +210,167 @@ static void ll_free_user_pages(struct page **pages, int npages, int do_dirty) int i; for (i = 0; i < npages; i++) { + if (pages[i] == NULL) + break; if (do_dirty) set_page_dirty_lock(pages[i]); page_cache_release(pages[i]); } - OBD_FREE(pages, npages * sizeof(*pages)); + OBD_FREE_LARGE(pages, npages * sizeof(*pages)); } -static ssize_t ll_direct_IO_26_seg(int rw, struct inode *inode, - struct address_space *mapping, - struct lov_stripe_md *lsm, - size_t size, loff_t file_offset, - struct page **pages, int page_count) +ssize_t ll_direct_rw_pages(const struct lu_env *env, struct cl_io *io, + int rw, struct inode *inode, + struct ll_dio_pages *pv) { - struct brw_page *pga; - struct obdo oa; - int opc, i, rc = 0; - size_t length; - struct obd_capa *ocapa; - loff_t file_offset_orig = file_offset; + struct cl_page *clp; + struct cl_2queue *queue; + struct cl_object *obj = io->ci_obj; + int i; + ssize_t rc = 0; + loff_t file_offset = pv->ldp_start_offset; + long size = pv->ldp_size; + int page_count = pv->ldp_nr; + struct page **pages = pv->ldp_pages; + long page_size = cl_page_size(obj); + bool do_io; + int io_pages = 0; ENTRY; - OBD_ALLOC(pga, sizeof(*pga) * page_count); - if (!pga) { - CDEBUG(D_VFSTRACE, "sizeof(*pga) = %u page_count = %u\n", - (int)sizeof(*pga), page_count); - RETURN(-ENOMEM); - } + queue = &io->ci_queue; + cl_2queue_init(queue); + for (i = 0; i < page_count; i++) { + if (pv->ldp_offsets) + file_offset = pv->ldp_offsets[i]; + + LASSERT(!(file_offset & (page_size - 1))); + clp = cl_page_find(env, obj, cl_index(obj, file_offset), + pv->ldp_pages[i], CPT_TRANSIENT); + if (IS_ERR(clp)) { + rc = PTR_ERR(clp); + break; + } - for (i = 0, length = size; length > 0; - length -=pga[i].count, file_offset +=pga[i].count,i++) {/*i last!*/ - pga[i].pg = pages[i]; - pga[i].off = file_offset; - /* To the end of the page, or the length, whatever is less */ - pga[i].count = min_t(int, CFS_PAGE_SIZE - - (file_offset & ~CFS_PAGE_MASK), - length); - pga[i].flag = 0; - if (rw == READ) - POISON_PAGE(pages[i], 0x0d); - } + rc = cl_page_own(env, io, clp); + if (rc) { + LASSERT(clp->cp_state == CPS_FREEING); + cl_page_put(env, clp); + break; + } - ll_inode_fill_obdo(inode, rw, &oa); - - if (rw == WRITE) { - lprocfs_counter_add(ll_i2sbi(inode)->ll_stats, - LPROC_LL_DIRECT_WRITE, size); - opc = CAPA_OPC_OSS_WRITE; - llap_write_pending(inode, NULL); - } else { - lprocfs_counter_add(ll_i2sbi(inode)->ll_stats, - LPROC_LL_DIRECT_READ, size); - opc = CAPA_OPC_OSS_RW; + do_io = true; + + /* check the page type: if the page is a host page, then do + * write directly */ + if (clp->cp_type == CPT_CACHEABLE) { + cfs_page_t *vmpage = cl_page_vmpage(env, clp); + cfs_page_t *src_page; + cfs_page_t *dst_page; + void *src; + void *dst; + + src_page = (rw == WRITE) ? pages[i] : vmpage; + dst_page = (rw == WRITE) ? vmpage : pages[i]; + + src = kmap_atomic(src_page, KM_USER0); + dst = kmap_atomic(dst_page, KM_USER1); + memcpy(dst, src, min(page_size, size)); + kunmap_atomic(dst, KM_USER1); + kunmap_atomic(src, KM_USER0); + + /* make sure page will be added to the transfer by + * cl_io_submit()->...->vvp_page_prep_write(). */ + if (rw == WRITE) + set_page_dirty(vmpage); + + if (rw == READ) { + /* do not issue the page for read, since it + * may reread a ra page which has NOT uptodate + * bit set. */ + cl_page_disown(env, io, clp); + do_io = false; + } + } + + if (likely(do_io)) { + cl_2queue_add(queue, clp); + + /* + * Set page clip to tell transfer formation engine + * that page has to be sent even if it is beyond KMS. + */ + cl_page_clip(env, clp, 0, min(size, page_size)); + + ++io_pages; + } + + /* drop the reference count for cl_page_find */ + cl_page_put(env, clp); + size -= page_size; + file_offset += page_size; } - ocapa = ll_osscapa_get(inode, opc); - rc = obd_brw_rqset(rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ, - ll_i2dtexp(inode), &oa, lsm, page_count, pga, NULL, - ocapa); - capa_put(ocapa); - if ((rc > 0) && (rw == WRITE)) { - lov_stripe_lock(lsm); - obd_adjust_kms(ll_i2dtexp(inode), lsm, file_offset_orig + rc, 0); - lov_stripe_unlock(lsm); + + if (rc == 0 && io_pages) { + rc = cl_io_submit_sync(env, io, + rw == READ ? CRT_READ : CRT_WRITE, + queue, CRP_NORMAL, 0); } + if (rc == 0) + rc = pv->ldp_size; - OBD_FREE(pga, sizeof(*pga) * page_count); + cl_2queue_discard(env, io, queue); + cl_2queue_disown(env, io, queue); + cl_2queue_fini(env, queue); RETURN(rc); } +EXPORT_SYMBOL(ll_direct_rw_pages); + +static ssize_t ll_direct_IO_26_seg(const struct lu_env *env, struct cl_io *io, + int rw, struct inode *inode, + struct address_space *mapping, + size_t size, loff_t file_offset, + struct page **pages, int page_count) +{ + struct ll_dio_pages pvec = { .ldp_pages = pages, + .ldp_nr = page_count, + .ldp_size = size, + .ldp_offsets = NULL, + .ldp_start_offset = file_offset + }; + + return ll_direct_rw_pages(env, io, rw, inode, &pvec); +} -/* This is the maximum size of a single O_DIRECT request, based on a 128kB +#ifdef KMALLOC_MAX_SIZE +#define MAX_MALLOC KMALLOC_MAX_SIZE +#else +#define MAX_MALLOC (128 * 1024) +#endif + +/* This is the maximum size of a single O_DIRECT request, based on the * kmalloc limit. We need to fit all of the brw_page structs, each one * representing PAGE_SIZE worth of user data, into a single buffer, and - * then truncate this to be a full-sized RPC. This is 22MB for 4kB pages. */ -#define MAX_DIO_SIZE ((128 * 1024 / sizeof(struct brw_page) * CFS_PAGE_SIZE) & \ + * then truncate this to be a full-sized RPC. For 4kB PAGE_SIZE this is + * up to 22MB for 128kB kmalloc and up to 682MB for 4MB kmalloc. */ +#define MAX_DIO_SIZE ((MAX_MALLOC / sizeof(struct brw_page) * CFS_PAGE_SIZE) & \ ~(PTLRPC_MAX_BRW_SIZE - 1)) static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t file_offset, unsigned long nr_segs) { + struct lu_env *env; + struct cl_io *io; struct file *file = iocb->ki_filp; struct inode *inode = file->f_mapping->host; - ssize_t count = iov_length(iov, nr_segs), tot_bytes = 0; + struct ccc_object *obj = cl_inode2ccc(inode); + long count = iov_length(iov, nr_segs); + long tot_bytes = 0, result = 0; struct ll_inode_info *lli = ll_i2info(inode); + struct lov_stripe_md *lsm = lli->lli_smd; unsigned long seg = 0; - size_t size = MAX_DIO_SIZE; + long size = MAX_DIO_SIZE; + int refcheck; ENTRY; if (!lli->lli_smd || !lli->lli_smd->lsm_object_id) @@ -229,17 +380,12 @@ static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb, if ((file_offset & ~CFS_PAGE_MASK) || (count & ~CFS_PAGE_MASK)) RETURN(-EINVAL); - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), size="LPSZ" (max %lu), " - "offset=%lld=%llx, pages "LPSZ" (max %lu)\n", + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), size=%lu (max %lu), " + "offset=%lld=%llx, pages %lu (max %lu)\n", inode->i_ino, inode->i_generation, inode, count, MAX_DIO_SIZE, file_offset, file_offset, count >> CFS_PAGE_SHIFT, MAX_DIO_SIZE >> CFS_PAGE_SHIFT); - if (rw == WRITE) - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_DIRECT_WRITE, count); - else - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_DIRECT_READ, count); - /* Check that all user buffers are aligned as well */ for (seg = 0; seg < nr_segs; seg++) { if (((unsigned long)iov[seg].iov_base & ~CFS_PAGE_MASK) || @@ -247,50 +393,68 @@ static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb, RETURN(-EINVAL); } + env = cl_env_get(&refcheck); + LASSERT(!IS_ERR(env)); + io = ccc_env_io(env)->cui_cl.cis_io; + LASSERT(io != NULL); + + /* 0. Need locking between buffered and direct access. and race with + * size changing by concurrent truncates and writes. + * 1. Need inode sem to operate transient pages. */ + if (rw == READ) + LOCK_INODE_MUTEX(inode); + + LASSERT(obj->cob_transient_pages == 0); for (seg = 0; seg < nr_segs; seg++) { - size_t iov_left = iov[seg].iov_len; + long iov_left = iov[seg].iov_len; unsigned long user_addr = (unsigned long)iov[seg].iov_base; + if (rw == READ) { + if (file_offset >= i_size_read(inode)) + break; + if (file_offset + iov_left > i_size_read(inode)) + iov_left = i_size_read(inode) - file_offset; + } + while (iov_left > 0) { struct page **pages; - int page_count; - ssize_t result; - - page_count = ll_get_user_pages(rw, user_addr, - min(size, iov_left), - &pages); - LASSERT(page_count != 0); - if (page_count > 0) { - result = ll_direct_IO_26_seg(rw, inode, + int page_count, max_pages = 0; + long bytes; + + bytes = min(size, iov_left); + page_count = ll_get_user_pages(rw, user_addr, bytes, + &pages, &max_pages); + if (likely(page_count > 0)) { + if (unlikely(page_count < max_pages)) + bytes = page_count << CFS_PAGE_SHIFT; + result = ll_direct_IO_26_seg(env, io, rw, inode, file->f_mapping, - lli->lli_smd, - min(size,iov_left), - file_offset, pages, - page_count); - ll_free_user_pages(pages, page_count, rw==READ); + bytes, file_offset, + pages, page_count); + ll_free_user_pages(pages, max_pages, rw==READ); + } else if (page_count == 0) { + GOTO(out, result = -EFAULT); } else { - result = 0; + result = page_count; } - if (page_count < 0 || result <= 0) { + if (unlikely(result <= 0)) { /* If we can't allocate a large enough buffer * for the request, shrink it to a smaller * PAGE_SIZE multiple and try again. * We should always be able to kmalloc for a * page worth of page pointers = 4MB on i386. */ - if ((page_count == -ENOMEM||result == -ENOMEM)&& + if (result == -ENOMEM && size > (CFS_PAGE_SIZE / sizeof(*pages)) * CFS_PAGE_SIZE) { size = ((((size / 2) - 1) | ~CFS_PAGE_MASK) + 1) & CFS_PAGE_MASK; - CDEBUG(D_VFSTRACE, "DIO size now %u\n", - (int)size); + CDEBUG(D_VFSTRACE,"DIO size now %lu\n", + size); continue; } - if (tot_bytes > 0) - RETURN(tot_bytes); - RETURN(page_count < 0 ? page_count : result); + GOTO(out, result); } tot_bytes += result; @@ -299,20 +463,113 @@ static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb, user_addr += result; } } - RETURN(tot_bytes); +out: + LASSERT(obj->cob_transient_pages == 0); + if (rw == READ) + UNLOCK_INODE_MUTEX(inode); + + if (tot_bytes > 0) { + if (rw == WRITE) { + lov_stripe_lock(lsm); + obd_adjust_kms(ll_i2dtexp(inode), lsm, file_offset, 0); + lov_stripe_unlock(lsm); + } + } + + cl_env_put(env, &refcheck); + RETURN(tot_bytes ? : result); +} + +#if defined(HAVE_KERNEL_WRITE_BEGIN_END) || defined(MS_HAS_NEW_AOPS) +static int ll_write_begin(struct file *file, struct address_space *mapping, + loff_t pos, unsigned len, unsigned flags, + struct page **pagep, void **fsdata) +{ + pgoff_t index = pos >> PAGE_CACHE_SHIFT; + struct page *page; + int rc; + unsigned from = pos & (PAGE_CACHE_SIZE - 1); + ENTRY; + + page = grab_cache_page_write_begin(mapping, index, flags); + if (!page) + RETURN(-ENOMEM); + + *pagep = page; + + rc = ll_prepare_write(file, page, from, from + len); + if (rc) { + unlock_page(page); + page_cache_release(page); + } + RETURN(rc); } +static int ll_write_end(struct file *file, struct address_space *mapping, + loff_t pos, unsigned len, unsigned copied, + struct page *page, void *fsdata) +{ + unsigned from = pos & (PAGE_CACHE_SIZE - 1); + int rc; + + rc = ll_commit_write(file, page, from, from + copied); + unlock_page(page); + page_cache_release(page); + + return rc ?: copied; +} +#endif + +#ifdef CONFIG_MIGRATION +int ll_migratepage(struct address_space *mapping, + struct page *newpage, struct page *page) +{ + /* Always fail page migration until we have a proper implementation */ + return -EIO; +} +#endif + +#ifndef MS_HAS_NEW_AOPS struct address_space_operations ll_aops = { .readpage = ll_readpage, // .readpages = ll_readpages, .direct_IO = ll_direct_IO_26, - .writepage = ll_writepage_26, + .writepage = ll_writepage, .writepages = generic_writepages, .set_page_dirty = ll_set_page_dirty, .sync_page = NULL, +#ifdef HAVE_KERNEL_WRITE_BEGIN_END + .write_begin = ll_write_begin, + .write_end = ll_write_end, +#else .prepare_write = ll_prepare_write, .commit_write = ll_commit_write, +#endif .invalidatepage = ll_invalidatepage, - .releasepage = ll_releasepage, + .releasepage = (void *)ll_releasepage, +#ifdef CONFIG_MIGRATION + .migratepage = ll_migratepage, +#endif .bmap = NULL }; +#else +struct address_space_operations_ext ll_aops = { + .orig_aops.readpage = ll_readpage, +// .orig_aops.readpages = ll_readpages, + .orig_aops.direct_IO = ll_direct_IO_26, + .orig_aops.writepage = ll_writepage, + .orig_aops.writepages = generic_writepages, + .orig_aops.set_page_dirty = ll_set_page_dirty, + .orig_aops.sync_page = NULL, + .orig_aops.prepare_write = ll_prepare_write, + .orig_aops.commit_write = ll_commit_write, + .orig_aops.invalidatepage = ll_invalidatepage, + .orig_aops.releasepage = ll_releasepage, +#ifdef CONFIG_MIGRATION + .orig_aops.migratepage = ll_migratepage, +#endif + .orig_aops.bmap = NULL, + .write_begin = ll_write_begin, + .write_end = ll_write_end +}; +#endif