X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fllite%2Frw26.c;h=b0eaf8efdcc0b652144ab644f3933403cb1551da;hb=ca58021173bbfe465a13cd34cc42b12fe5af31e4;hp=b0bd2d3a47d44af47d61b96200a48c26c0d72442;hpb=6869932b552ac705f411de3362f01bd50c1f6f7d;p=fs%2Flustre-release.git diff --git a/lustre/llite/rw26.c b/lustre/llite/rw26.c index b0bd2d3..b0eaf8e 100644 --- a/lustre/llite/rw26.c +++ b/lustre/llite/rw26.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -38,20 +38,19 @@ * Lustre Lite I/O page cache routines for the 2.5/2.6 kernel version */ -#ifndef AUTOCONF_INCLUDED -#include -#endif #include #include #include #include #include -#include #include -#include -#include #include +#ifdef HAVE_MIGRATE_H +#include +#elif defined(HAVE_MIGRATE_MODE_H) +#include +#endif #include #include #include @@ -60,66 +59,138 @@ #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) -{ - return ll_writepage(page); -} - -/* 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; -} +/** + * 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 void ll_invalidatepage(struct page *vmpage, +#ifdef HAVE_INVALIDATE_RANGE + unsigned int offset, unsigned int length #else -static void ll_invalidatepage(struct page *page, unsigned long offset) -{ - if (offset == 0 && PagePrivate(page)) - ll_removepage(page); -} + unsigned long offset #endif -static int ll_releasepage(struct page *page, gfp_t gfp_mask) + ) { - if (PagePrivate(page)) - ll_removepage(page); - return 1; + struct inode *inode; + struct lu_env *env; + struct cl_page *page; + struct cl_object *obj; + + 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 + */ +#ifdef HAVE_INVALIDATE_RANGE + if (offset == 0 && length == PAGE_CACHE_SIZE) { +#else + if (offset == 0) { +#endif + 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) { + cl_page_delete(env, page); + cl_page_put(env, page); + } + } else + LASSERT(vmpage->private == 0); + cl_env_put(env, &refcheck); + } + } } -static int ll_set_page_dirty(struct page *page) +#ifdef HAVE_RELEASEPAGE_WITH_INT +#define RELEASEPAGE_ARG_TYPE int +#else +#define RELEASEPAGE_ARG_TYPE gfp_t +#endif +static int ll_releasepage(struct page *vmpage, RELEASEPAGE_ARG_TYPE gfp_mask) { - 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)); + struct lu_env *env; + void *cookie; + struct cl_object *obj; + struct cl_page *page; + struct address_space *mapping; + int result = 0; + + LASSERT(PageLocked(vmpage)); + if (PageWriteback(vmpage) || PageDirty(vmpage)) + return 0; + + mapping = vmpage->mapping; + if (mapping == NULL) + return 1; + + obj = ll_i2info(mapping->host)->lli_clob; + if (obj == NULL) + return 1; + + /* 1 for caller, 1 for cl_page and 1 for page cache */ + if (page_count(vmpage) > 3) + return 0; + + page = cl_vmpage_page(vmpage, obj); + if (page == NULL) + return 1; + + cookie = cl_env_reenter(); + env = cl_env_percpu_get(); + LASSERT(!IS_ERR(env)); + + if (!cl_page_in_use(page)) { + result = 1; + cl_page_delete(env, page); + } + + /* To use percpu env array, the call path can not be rescheduled; + * otherwise percpu array will be messed if ll_releaspage() called + * again on the same CPU. + * + * If this page holds the last refc of cl_object, the following + * call path may cause reschedule: + * cl_page_put -> cl_page_free -> cl_object_put -> + * lu_object_put -> lu_object_free -> lov_delete_raid0. + * + * However, the kernel can't get rid of this inode until all pages have + * been cleaned up. Now that we hold page lock here, it's pretty safe + * that we won't get into object delete path. + */ + LASSERT(cl_object_refc(obj) > 1); + cl_page_put(env, page); + + cl_env_percpu_put(env); + cl_env_reexit(cookie); + return result; } #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) { @@ -127,18 +198,19 @@ 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 + PAGE_CACHE_SIZE - 1) >> + PAGE_CACHE_SHIFT; + *max_pages -= user_addr >> PAGE_CACHE_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; @@ -151,100 +223,179 @@ 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 obd_info *oinfo, - struct ptlrpc_request_set *set, - 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; - int i, rc = 0; - size_t length; - ENTRY; + 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; + size_t size = pv->ldp_size; + int page_count = pv->ldp_nr; + struct page **pages = pv->ldp_pages; + size_t page_size = cl_page_size(obj); + bool do_io; + int io_pages = 0; + ENTRY; + + 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; + } - 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); - } + rc = cl_page_own(env, io, clp); + if (rc) { + LASSERT(clp->cp_state == CPS_FREEING); + cl_page_put(env, clp); + break; + } + + do_io = true; + + /* check the page type: if the page is a host page, then do + * write directly */ + if (clp->cp_type == CPT_CACHEABLE) { + struct page *vmpage = cl_page_vmpage(clp); + struct page *src_page; + struct page *dst_page; + void *src; + void *dst; + + src_page = (rw == WRITE) ? pages[i] : vmpage; + dst_page = (rw == WRITE) ? vmpage : pages[i]; + + src = ll_kmap_atomic(src_page, KM_USER0); + dst = ll_kmap_atomic(dst_page, KM_USER1); + memcpy(dst, src, min(page_size, size)); + ll_kunmap_atomic(dst, KM_USER1); + ll_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); - 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); + /* + * 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; } - rc = obd_brw_async(rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ, - ll_i2dtexp(inode), oinfo, page_count, - pga, NULL, set); + if (rc == 0 && io_pages) { + rc = cl_io_submit_sync(env, io, + rw == READ ? CRT_READ : CRT_WRITE, + queue, 0); + } if (rc == 0) - rc = size; + 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); -/* This is the maximum size of a single O_DIRECT request, based on a 128kB +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); +} + +#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) & \ - ~(PTLRPC_MAX_BRW_SIZE - 1)) + * 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) * PAGE_CACHE_SIZE) & \ + ~(DT_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; + 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; - struct ptlrpc_request_set *set; - struct obd_info oinfo; - struct obdo oa; unsigned long seg = 0; - size_t size = MAX_DIO_SIZE; - int opc; + long size = MAX_DIO_SIZE; + int refcheck; ENTRY; - if (!lli->lli_smd || !lli->lli_smd->lsm_object_id) + if (!lli->lli_has_smd) RETURN(-EBADF); /* FIXME: io smaller than PAGE_SIZE is broken on ia64 ??? */ 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", - 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); - opc = CAPA_OPC_OSS_WRITE; - llap_write_pending(inode, NULL); - } else { - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_DIRECT_READ, count); - opc = CAPA_OPC_OSS_RW; - } + CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), size=%lu (max %lu), " + "offset=%lld=%llx, pages %lu (max %lu)\n", + PFID(ll_inode2fid(inode)), inode, count, MAX_DIO_SIZE, + file_offset, file_offset, count >> PAGE_CACHE_SHIFT, + MAX_DIO_SIZE >> PAGE_CACHE_SHIFT); /* Check that all user buffers are aligned as well */ for (seg = 0; seg < nr_segs; seg++) { @@ -253,71 +404,61 @@ static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb, RETURN(-EINVAL); } - set = ptlrpc_prep_set(); - if (set == NULL) - RETURN(-ENOMEM); - - ll_inode_fill_obdo(inode, rw, &oa); - oinfo.oi_oa = &oa; - oinfo.oi_md = lsm; - oinfo.oi_capa = ll_osscapa_get(inode, opc); - - /* need locking between buffered and direct access. and race with - *size changing by concurrent truncates and writes. */ - if (rw == READ) - LOCK_INODE_MUTEX(inode); + env = cl_env_get(&refcheck); + LASSERT(!IS_ERR(env)); + io = ccc_env_io(env)->cui_cl.cis_io; + LASSERT(io != NULL); 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 >= inode->i_size) + if (file_offset >= i_size_read(inode)) break; - if (file_offset + iov_left > inode->i_size) - iov_left = inode->i_size - file_offset; + 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 << PAGE_CACHE_SHIFT; + result = ll_direct_IO_26_seg(env, io, rw, inode, file->f_mapping, - &oinfo, set, - 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)&& - size > (CFS_PAGE_SIZE / sizeof(*pages)) * - CFS_PAGE_SIZE) { + if (result == -ENOMEM && + size > (PAGE_CACHE_SIZE / sizeof(*pages)) * + PAGE_CACHE_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) - tot_bytes = page_count < 0 ? page_count : result; - GOTO(out, tot_bytes); + GOTO(out, result); } tot_bytes += result; @@ -327,38 +468,255 @@ static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb, } } out: - if (rw == READ) - UNLOCK_INODE_MUTEX(inode); - if (tot_bytes > 0) { - int rc; - - rc = ptlrpc_set_wait(set); - if (rc) { - tot_bytes = rc; - } else if (rw == WRITE) { - lov_stripe_lock(lsm); - obd_adjust_kms(ll_i2dtexp(inode), lsm, file_offset, 0); - lov_stripe_unlock(lsm); - } - } + struct ccc_io *cio = ccc_env_io(env); + + /* no commit async for direct IO */ + cio->u.write.cui_written += tot_bytes; + } + + cl_env_put(env, &refcheck); + RETURN(tot_bytes ? tot_bytes : result); +} - capa_put(oinfo.oi_capa); - ptlrpc_set_destroy(set); - RETURN(tot_bytes); +/** + * Prepare partially written-to page for a write. + */ +static int ll_prepare_partial_page(const struct lu_env *env, struct cl_io *io, + struct cl_page *pg) +{ + struct cl_attr *attr = ccc_env_thread_attr(env); + struct cl_object *obj = io->ci_obj; + struct ccc_page *cp = cl_object_page_slice(obj, pg); + loff_t offset = cl_offset(obj, ccc_index(cp)); + int result; + + cl_object_attr_lock(obj); + result = cl_object_attr_get(env, obj, attr); + cl_object_attr_unlock(obj); + if (result == 0) { + /* + * If are writing to a new page, no need to read old data. + * The extent locking will have updated the KMS, and for our + * purposes here we can treat it like i_size. + */ + if (attr->cat_kms <= offset) { + char *kaddr = ll_kmap_atomic(cp->cpg_page, KM_USER0); + + memset(kaddr, 0, cl_page_size(obj)); + ll_kunmap_atomic(kaddr, KM_USER0); + } else if (cp->cpg_defer_uptodate) + cp->cpg_ra_used = 1; + else + result = ll_page_sync_io(env, io, pg, CRT_READ); + } + return result; +} + +static int ll_write_begin(struct file *file, struct address_space *mapping, + loff_t pos, unsigned len, unsigned flags, + struct page **pagep, void **fsdata) +{ + struct ll_cl_context *lcc; + const struct lu_env *env; + struct cl_io *io; + struct cl_page *page; + + struct cl_object *clob = ll_i2info(mapping->host)->lli_clob; + pgoff_t index = pos >> PAGE_CACHE_SHIFT; + struct page *vmpage = NULL; + unsigned from = pos & (PAGE_CACHE_SIZE - 1); + unsigned to = from + len; + int result = 0; + ENTRY; + + CDEBUG(D_VFSTRACE, "Writing %lu of %d to %d bytes\n", index, from, len); + + lcc = ll_cl_find(file); + if (lcc == NULL) + GOTO(out, result = -EIO); + + env = lcc->lcc_env; + io = lcc->lcc_io; + + /* To avoid deadlock, try to lock page first. */ + vmpage = grab_cache_page_nowait(mapping, index); + + if (unlikely(vmpage == NULL || + PageDirty(vmpage) || PageWriteback(vmpage))) { + struct ccc_io *cio = ccc_env_io(env); + struct cl_page_list *plist = &cio->u.write.cui_queue; + + /* if the page is already in dirty cache, we have to commit + * the pages right now; otherwise, it may cause deadlock + * because it holds page lock of a dirty page and request for + * more grants. It's okay for the dirty page to be the first + * one in commit page list, though. */ + if (vmpage != NULL && plist->pl_nr > 0) { + unlock_page(vmpage); + page_cache_release(vmpage); + vmpage = NULL; + } + + /* commit pages and then wait for page lock */ + result = vvp_io_write_commit(env, io); + if (result < 0) + GOTO(out, result); + + if (vmpage == NULL) { + vmpage = grab_cache_page_write_begin(mapping, index, + flags); + if (vmpage == NULL) + GOTO(out, result = -ENOMEM); + } + } + + page = cl_page_find(env, clob, vmpage->index, vmpage, CPT_CACHEABLE); + if (IS_ERR(page)) + GOTO(out, result = PTR_ERR(page)); + + lcc->lcc_page = page; + lu_ref_add(&page->cp_reference, "cl_io", io); + + cl_page_assume(env, io, page); + if (!PageUptodate(vmpage)) { + /* + * We're completely overwriting an existing page, + * so _don't_ set it up to date until commit_write + */ + if (from == 0 && to == PAGE_SIZE) { + CL_PAGE_HEADER(D_PAGE, env, page, "full page write\n"); + POISON_PAGE(vmpage, 0x11); + } else { + /* TODO: can be optimized at OSC layer to check if it + * is a lockless IO. In that case, it's not necessary + * to read the data. */ + result = ll_prepare_partial_page(env, io, page); + if (result == 0) + SetPageUptodate(vmpage); + } + } + if (result < 0) + cl_page_unassume(env, io, page); + EXIT; +out: + if (result < 0) { + if (vmpage != NULL) { + unlock_page(vmpage); + page_cache_release(vmpage); + } + } else { + *pagep = vmpage; + *fsdata = lcc; + } + RETURN(result); } -struct address_space_operations ll_aops = { - .readpage = ll_readpage, -// .readpages = ll_readpages, +static int ll_write_end(struct file *file, struct address_space *mapping, + loff_t pos, unsigned len, unsigned copied, + struct page *vmpage, void *fsdata) +{ + struct ll_cl_context *lcc = fsdata; + const struct lu_env *env; + struct cl_io *io; + struct ccc_io *cio; + struct cl_page *page; + unsigned from = pos & (PAGE_CACHE_SIZE - 1); + bool unplug = false; + int result = 0; + ENTRY; + + page_cache_release(vmpage); + + LASSERT(lcc != NULL); + env = lcc->lcc_env; + page = lcc->lcc_page; + io = lcc->lcc_io; + cio = ccc_env_io(env); + + LASSERT(cl_page_is_owned(page, io)); + if (copied > 0) { + struct cl_page_list *plist = &cio->u.write.cui_queue; + + lcc->lcc_page = NULL; /* page will be queued */ + + /* Add it into write queue */ + cl_page_list_add(plist, page); + if (plist->pl_nr == 1) /* first page */ + cio->u.write.cui_from = from; + else + LASSERT(from == 0); + cio->u.write.cui_to = from + copied; + + /* To address the deadlock in balance_dirty_pages() where + * this dirty page may be written back in the same thread. */ + if (PageDirty(vmpage)) + unplug = true; + + /* We may have one full RPC, commit it soon */ + if (plist->pl_nr >= PTLRPC_MAX_BRW_PAGES) + unplug = true; + + CL_PAGE_DEBUG(D_VFSTRACE, env, page, + "queued page: %d.\n", plist->pl_nr); + } else { + cl_page_disown(env, io, page); + + lcc->lcc_page = NULL; + lu_ref_del(&page->cp_reference, "cl_io", io); + cl_page_put(env, page); + + /* page list is not contiguous now, commit it now */ + unplug = true; + } + if (unplug || + file->f_flags & O_SYNC || IS_SYNC(file->f_dentry->d_inode)) + result = vvp_io_write_commit(env, io); + + RETURN(result >= 0 ? copied : result); +} + +#ifdef CONFIG_MIGRATION +static int ll_migratepage(struct address_space *mapping, + struct page *newpage, struct page *page +#ifdef HAVE_MIGRATEPAGE_4ARGS + , enum migrate_mode mode +#endif + ) +{ + /* Always fail page migration until we have a proper implementation */ + return -EIO; +} +#endif + +#ifndef MS_HAS_NEW_AOPS +const struct address_space_operations ll_aops = { + .readpage = ll_readpage, .direct_IO = ll_direct_IO_26, - .writepage = ll_writepage_26, - .writepages = generic_writepages, - .set_page_dirty = ll_set_page_dirty, - .sync_page = NULL, - .prepare_write = ll_prepare_write, - .commit_write = ll_commit_write, + .writepage = ll_writepage, + .writepages = ll_writepages, + .set_page_dirty = __set_page_dirty_nobuffers, + .write_begin = ll_write_begin, + .write_end = ll_write_end, .invalidatepage = ll_invalidatepage, - .releasepage = ll_releasepage, - .bmap = NULL + .releasepage = (void *)ll_releasepage, +#ifdef CONFIG_MIGRATION + .migratepage = ll_migratepage, +#endif }; +#else +const struct address_space_operations_ext ll_aops = { + .orig_aops.readpage = ll_readpage, + .orig_aops.direct_IO = ll_direct_IO_26, + .orig_aops.writepage = ll_writepage, + .orig_aops.writepages = ll_writepages, + .orig_aops.set_page_dirty = __set_page_dirty_nobuffers, + .orig_aops.invalidatepage = ll_invalidatepage, + .orig_aops.releasepage = ll_releasepage, +#ifdef CONFIG_MIGRATION + .orig_aops.migratepage = ll_migratepage, +#endif + .write_begin = ll_write_begin, + .write_end = ll_write_end +}; +#endif