11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 181da177e4SLinus Torvalds #include <linux/nfs_fs.h> 191da177e4SLinus Torvalds #include <linux/nfs_mount.h> 201da177e4SLinus Torvalds #include <linux/nfs_page.h> 213fcfab16SAndrew Morton #include <linux/backing-dev.h> 223fcfab16SAndrew Morton 231da177e4SLinus Torvalds #include <asm/uaccess.h> 241da177e4SLinus Torvalds 251da177e4SLinus Torvalds #include "delegation.h" 2649a70f27STrond Myklebust #include "internal.h" 2791d5b470SChuck Lever #include "iostat.h" 281da177e4SLinus Torvalds 291da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 321da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds /* 351da177e4SLinus Torvalds * Local function declarations 361da177e4SLinus Torvalds */ 371da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context*, 381da177e4SLinus Torvalds struct page *, 391da177e4SLinus Torvalds unsigned int, unsigned int); 40c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, 41c63c7b05STrond Myklebust struct inode *inode, int ioflags); 42f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 43788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 44788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 461da177e4SLinus Torvalds 47e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 483feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 491da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 501da177e4SLinus Torvalds 51e9f7bee1STrond Myklebust struct nfs_write_data *nfs_commit_alloc(void) 521da177e4SLinus Torvalds { 53e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5440859d7eSChuck Lever 551da177e4SLinus Torvalds if (p) { 561da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 571da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 581da177e4SLinus Torvalds } 591da177e4SLinus Torvalds return p; 601da177e4SLinus Torvalds } 611da177e4SLinus Torvalds 625e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 631da177e4SLinus Torvalds { 6440859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6540859d7eSChuck Lever kfree(p->pagevec); 661da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 671da177e4SLinus Torvalds } 681da177e4SLinus Torvalds 698d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 703feb2d49STrond Myklebust { 71e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 723feb2d49STrond Myklebust 733feb2d49STrond Myklebust if (p) { 743feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 753feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 76e9f7bee1STrond Myklebust p->npages = pagecount; 770d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 780d0b5cb3SChuck Lever p->pagevec = p->page_array; 793feb2d49STrond Myklebust else { 800d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 810d0b5cb3SChuck Lever if (!p->pagevec) { 823feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 833feb2d49STrond Myklebust p = NULL; 843feb2d49STrond Myklebust } 853feb2d49STrond Myklebust } 863feb2d49STrond Myklebust } 873feb2d49STrond Myklebust return p; 883feb2d49STrond Myklebust } 893feb2d49STrond Myklebust 905e4424afSTrond Myklebust static void nfs_writedata_free(struct nfs_write_data *p) 913feb2d49STrond Myklebust { 923feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 933feb2d49STrond Myklebust kfree(p->pagevec); 943feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 953feb2d49STrond Myklebust } 963feb2d49STrond Myklebust 97383ba719STrond Myklebust void nfs_writedata_release(void *data) 981da177e4SLinus Torvalds { 99383ba719STrond Myklebust struct nfs_write_data *wdata = data; 100383ba719STrond Myklebust 101383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 1021da177e4SLinus Torvalds nfs_writedata_free(wdata); 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds 1057b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1067b159fc1STrond Myklebust { 1077b159fc1STrond Myklebust ctx->error = error; 1087b159fc1STrond Myklebust smp_wmb(); 1097b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1107b159fc1STrond Myklebust } 1117b159fc1STrond Myklebust 112277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 113277459d2STrond Myklebust { 114277459d2STrond Myklebust struct nfs_page *req = NULL; 115277459d2STrond Myklebust 116277459d2STrond Myklebust if (PagePrivate(page)) { 117277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 118277459d2STrond Myklebust if (req != NULL) 119c03b4024STrond Myklebust kref_get(&req->wb_kref); 120277459d2STrond Myklebust } 121277459d2STrond Myklebust return req; 122277459d2STrond Myklebust } 123277459d2STrond Myklebust 124277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 125277459d2STrond Myklebust { 126587142f8STrond Myklebust struct inode *inode = page->mapping->host; 127277459d2STrond Myklebust struct nfs_page *req = NULL; 128277459d2STrond Myklebust 129587142f8STrond Myklebust spin_lock(&inode->i_lock); 130277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 131587142f8STrond Myklebust spin_unlock(&inode->i_lock); 132277459d2STrond Myklebust return req; 133277459d2STrond Myklebust } 134277459d2STrond Myklebust 1351da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1361da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1371da177e4SLinus Torvalds { 1381da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1391da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 140ca52fec1STrond Myklebust pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1411da177e4SLinus Torvalds 1421da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1431da177e4SLinus Torvalds return; 1441da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1451da177e4SLinus Torvalds if (i_size >= end) 1461da177e4SLinus Torvalds return; 14791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1481da177e4SLinus Torvalds i_size_write(inode, end); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 152a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 153a301b777STrond Myklebust { 154a301b777STrond Myklebust SetPageError(page); 155a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 156a301b777STrond Myklebust } 157a301b777STrond Myklebust 1581da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1591da177e4SLinus Torvalds * covers the full page. 1601da177e4SLinus Torvalds */ 1611da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1621da177e4SLinus Torvalds { 1631da177e4SLinus Torvalds if (PageUptodate(page)) 1641da177e4SLinus Torvalds return; 1651da177e4SLinus Torvalds if (base != 0) 1661da177e4SLinus Torvalds return; 16749a70f27STrond Myklebust if (count != nfs_page_length(page)) 1681da177e4SLinus Torvalds return; 1691da177e4SLinus Torvalds SetPageUptodate(page); 1701da177e4SLinus Torvalds } 1711da177e4SLinus Torvalds 172e21195a7STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1731da177e4SLinus Torvalds unsigned int offset, unsigned int count) 1741da177e4SLinus Torvalds { 1751da177e4SLinus Torvalds struct nfs_page *req; 176e21195a7STrond Myklebust int ret; 1771da177e4SLinus Torvalds 178e21195a7STrond Myklebust for (;;) { 179e21195a7STrond Myklebust req = nfs_update_request(ctx, page, offset, count); 180e21195a7STrond Myklebust if (!IS_ERR(req)) 181e21195a7STrond Myklebust break; 182e21195a7STrond Myklebust ret = PTR_ERR(req); 183e21195a7STrond Myklebust if (ret != -EBUSY) 184e21195a7STrond Myklebust return ret; 185e21195a7STrond Myklebust ret = nfs_wb_page(page->mapping->host, page); 186e21195a7STrond Myklebust if (ret != 0) 187e21195a7STrond Myklebust return ret; 188e21195a7STrond Myklebust } 1891da177e4SLinus Torvalds /* Update file length */ 1901da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 191acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 192abd3e641STrond Myklebust return 0; 1931da177e4SLinus Torvalds } 1941da177e4SLinus Torvalds 1951da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1961da177e4SLinus Torvalds { 1971da177e4SLinus Torvalds if (wbc->for_reclaim) 198c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 1991da177e4SLinus Torvalds if (wbc->for_kupdate) 2001da177e4SLinus Torvalds return FLUSH_LOWPRI; 2011da177e4SLinus Torvalds return 0; 2021da177e4SLinus Torvalds } 2031da177e4SLinus Torvalds 2041da177e4SLinus Torvalds /* 20589a09141SPeter Zijlstra * NFS congestion control 20689a09141SPeter Zijlstra */ 20789a09141SPeter Zijlstra 20889a09141SPeter Zijlstra int nfs_congestion_kb; 20989a09141SPeter Zijlstra 21089a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 21189a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 21289a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 21389a09141SPeter Zijlstra 2145a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 21589a09141SPeter Zijlstra { 2165a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2175a6d41b3STrond Myklebust 2185a6d41b3STrond Myklebust if (!ret) { 21989a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 22089a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 22189a09141SPeter Zijlstra 222277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 22389a09141SPeter Zijlstra NFS_CONGESTION_ON_THRESH) 22489a09141SPeter Zijlstra set_bdi_congested(&nfss->backing_dev_info, WRITE); 22589a09141SPeter Zijlstra } 2265a6d41b3STrond Myklebust return ret; 22789a09141SPeter Zijlstra } 22889a09141SPeter Zijlstra 22989a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 23089a09141SPeter Zijlstra { 23189a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 23289a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 23389a09141SPeter Zijlstra 23489a09141SPeter Zijlstra end_page_writeback(page); 235c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 23689a09141SPeter Zijlstra clear_bdi_congested(&nfss->backing_dev_info, WRITE); 23789a09141SPeter Zijlstra } 23889a09141SPeter Zijlstra 23989a09141SPeter Zijlstra /* 240e261f51fSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 2419cccef95STrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 242e261f51fSTrond Myklebust */ 243c63c7b05STrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 244c63c7b05STrond Myklebust struct page *page) 245e261f51fSTrond Myklebust { 246587142f8STrond Myklebust struct inode *inode = page->mapping->host; 247e261f51fSTrond Myklebust struct nfs_page *req; 248e261f51fSTrond Myklebust int ret; 249e261f51fSTrond Myklebust 250587142f8STrond Myklebust spin_lock(&inode->i_lock); 251e261f51fSTrond Myklebust for(;;) { 252e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 253e261f51fSTrond Myklebust if (req == NULL) { 254587142f8STrond Myklebust spin_unlock(&inode->i_lock); 2559cccef95STrond Myklebust return 0; 256e261f51fSTrond Myklebust } 257acee478aSTrond Myklebust if (nfs_set_page_tag_locked(req)) 258e261f51fSTrond Myklebust break; 259e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 260acee478aSTrond Myklebust * then the call to nfs_set_page_tag_locked() will always 261e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 262e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 263e261f51fSTrond Myklebust */ 264587142f8STrond Myklebust spin_unlock(&inode->i_lock); 265e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 266e261f51fSTrond Myklebust nfs_release_request(req); 267e261f51fSTrond Myklebust if (ret != 0) 268e261f51fSTrond Myklebust return ret; 269587142f8STrond Myklebust spin_lock(&inode->i_lock); 270e261f51fSTrond Myklebust } 271612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 272612c9384STrond Myklebust /* This request is marked for commit */ 273587142f8STrond Myklebust spin_unlock(&inode->i_lock); 274acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 275c63c7b05STrond Myklebust nfs_pageio_complete(pgio); 2769cccef95STrond Myklebust return 0; 277612c9384STrond Myklebust } 278c63c7b05STrond Myklebust if (nfs_set_page_writeback(page) != 0) { 279587142f8STrond Myklebust spin_unlock(&inode->i_lock); 280c63c7b05STrond Myklebust BUG(); 281c63c7b05STrond Myklebust } 282587142f8STrond Myklebust spin_unlock(&inode->i_lock); 283f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 284f8512ad0SFred Isaman nfs_redirty_request(req); 285f8512ad0SFred Isaman return pgio->pg_error; 286f8512ad0SFred Isaman } 2879cccef95STrond Myklebust return 0; 288e261f51fSTrond Myklebust } 289e261f51fSTrond Myklebust 290f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 291f758c885STrond Myklebust { 292f758c885STrond Myklebust struct inode *inode = page->mapping->host; 293f758c885STrond Myklebust 294f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 295f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 296f758c885STrond Myklebust 297f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 298f758c885STrond Myklebust return nfs_page_async_flush(pgio, page); 299f758c885STrond Myklebust } 300f758c885STrond Myklebust 301e261f51fSTrond Myklebust /* 3021da177e4SLinus Torvalds * Write an mmapped page to the server. 3031da177e4SLinus Torvalds */ 3044d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3051da177e4SLinus Torvalds { 306f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 307e261f51fSTrond Myklebust int err; 3081da177e4SLinus Torvalds 309f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 310f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 311f758c885STrond Myklebust nfs_pageio_complete(&pgio); 312f758c885STrond Myklebust if (err < 0) 3134d770ccfSTrond Myklebust return err; 314f758c885STrond Myklebust if (pgio.pg_error < 0) 315f758c885STrond Myklebust return pgio.pg_error; 316f758c885STrond Myklebust return 0; 3174d770ccfSTrond Myklebust } 3184d770ccfSTrond Myklebust 3194d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3204d770ccfSTrond Myklebust { 321f758c885STrond Myklebust int ret; 3224d770ccfSTrond Myklebust 323f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3241da177e4SLinus Torvalds unlock_page(page); 325f758c885STrond Myklebust return ret; 326f758c885STrond Myklebust } 327f758c885STrond Myklebust 328f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 329f758c885STrond Myklebust { 330f758c885STrond Myklebust int ret; 331f758c885STrond Myklebust 332f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 333f758c885STrond Myklebust unlock_page(page); 334f758c885STrond Myklebust return ret; 3351da177e4SLinus Torvalds } 3361da177e4SLinus Torvalds 3371da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3381da177e4SLinus Torvalds { 3391da177e4SLinus Torvalds struct inode *inode = mapping->host; 340c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3411da177e4SLinus Torvalds int err; 3421da177e4SLinus Torvalds 34391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 34491d5b470SChuck Lever 345c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 346f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 347c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 348f758c885STrond Myklebust if (err < 0) 3491da177e4SLinus Torvalds return err; 350f758c885STrond Myklebust if (pgio.pg_error < 0) 351c63c7b05STrond Myklebust return pgio.pg_error; 352c63c7b05STrond Myklebust return 0; 3531da177e4SLinus Torvalds } 3541da177e4SLinus Torvalds 3551da177e4SLinus Torvalds /* 3561da177e4SLinus Torvalds * Insert a write request into an inode 3571da177e4SLinus Torvalds */ 35827852596SNick Piggin static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3591da177e4SLinus Torvalds { 3601da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3611da177e4SLinus Torvalds int error; 3621da177e4SLinus Torvalds 3631da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 36427852596SNick Piggin BUG_ON(error); 3651da177e4SLinus Torvalds if (!nfsi->npages) { 3661da177e4SLinus Torvalds igrab(inode); 3671da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3681da177e4SLinus Torvalds nfsi->change_attr++; 3691da177e4SLinus Torvalds } 370deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 371277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3721da177e4SLinus Torvalds nfsi->npages++; 373c03b4024STrond Myklebust kref_get(&req->wb_kref); 37427852596SNick Piggin radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, 37527852596SNick Piggin NFS_PAGE_TAG_LOCKED); 3761da177e4SLinus Torvalds } 3771da177e4SLinus Torvalds 3781da177e4SLinus Torvalds /* 37989a09141SPeter Zijlstra * Remove a write request from an inode 3801da177e4SLinus Torvalds */ 3811da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 3821da177e4SLinus Torvalds { 38388be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 3841da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3851da177e4SLinus Torvalds 3861da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 3871da177e4SLinus Torvalds 388587142f8STrond Myklebust spin_lock(&inode->i_lock); 389277459d2STrond Myklebust set_page_private(req->wb_page, 0); 390deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 3911da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 3921da177e4SLinus Torvalds nfsi->npages--; 3931da177e4SLinus Torvalds if (!nfsi->npages) { 394587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3951da177e4SLinus Torvalds iput(inode); 3961da177e4SLinus Torvalds } else 397587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3981da177e4SLinus Torvalds nfs_clear_request(req); 3991da177e4SLinus Torvalds nfs_release_request(req); 4001da177e4SLinus Torvalds } 4011da177e4SLinus Torvalds 40261822ab5STrond Myklebust static void 403*6d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 40461822ab5STrond Myklebust { 40561822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 40661822ab5STrond Myklebust } 40761822ab5STrond Myklebust 4081da177e4SLinus Torvalds /* 4091da177e4SLinus Torvalds * Check if a request is dirty 4101da177e4SLinus Torvalds */ 4111da177e4SLinus Torvalds static inline int 4121da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 4131da177e4SLinus Torvalds { 4145a6d41b3STrond Myklebust struct page *page = req->wb_page; 4155a6d41b3STrond Myklebust 416612c9384STrond Myklebust if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags)) 4175a6d41b3STrond Myklebust return 0; 4185a6d41b3STrond Myklebust return !PageWriteback(req->wb_page); 4191da177e4SLinus Torvalds } 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4221da177e4SLinus Torvalds /* 4231da177e4SLinus Torvalds * Add a request to the inode's commit list. 4241da177e4SLinus Torvalds */ 4251da177e4SLinus Torvalds static void 4261da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4271da177e4SLinus Torvalds { 42888be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4291da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4301da177e4SLinus Torvalds 431587142f8STrond Myklebust spin_lock(&inode->i_lock); 4321da177e4SLinus Torvalds nfsi->ncommit++; 433612c9384STrond Myklebust set_bit(PG_NEED_COMMIT, &(req)->wb_flags); 4345c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4355c369683STrond Myklebust req->wb_index, 4365c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 437587142f8STrond Myklebust spin_unlock(&inode->i_lock); 438fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 439c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 440a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4411da177e4SLinus Torvalds } 4428e821cadSTrond Myklebust 4438e821cadSTrond Myklebust static inline 4448e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4458e821cadSTrond Myklebust { 4468e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4478e821cadSTrond Myklebust } 4488e821cadSTrond Myklebust 4498e821cadSTrond Myklebust static inline 4508e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4518e821cadSTrond Myklebust { 452612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4538e821cadSTrond Myklebust nfs_mark_request_commit(req); 4548e821cadSTrond Myklebust return 1; 4558e821cadSTrond Myklebust } 4568e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 457*6d884e8fSFred nfs_mark_request_dirty(req); 4588e821cadSTrond Myklebust return 1; 4598e821cadSTrond Myklebust } 4608e821cadSTrond Myklebust return 0; 4618e821cadSTrond Myklebust } 4628e821cadSTrond Myklebust #else 4638e821cadSTrond Myklebust static inline void 4648e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4658e821cadSTrond Myklebust { 4668e821cadSTrond Myklebust } 4678e821cadSTrond Myklebust 4688e821cadSTrond Myklebust static inline 4698e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4708e821cadSTrond Myklebust { 4718e821cadSTrond Myklebust return 0; 4728e821cadSTrond Myklebust } 4738e821cadSTrond Myklebust 4748e821cadSTrond Myklebust static inline 4758e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4768e821cadSTrond Myklebust { 4778e821cadSTrond Myklebust return 0; 4788e821cadSTrond Myklebust } 4791da177e4SLinus Torvalds #endif 4801da177e4SLinus Torvalds 4811da177e4SLinus Torvalds /* 4821da177e4SLinus Torvalds * Wait for a request to complete. 4831da177e4SLinus Torvalds * 484150030b7SMatthew Wilcox * Interruptible by fatal signals only. 4851da177e4SLinus Torvalds */ 486ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) 4871da177e4SLinus Torvalds { 4881da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4891da177e4SLinus Torvalds struct nfs_page *req; 490ca52fec1STrond Myklebust pgoff_t idx_end, next; 4911da177e4SLinus Torvalds unsigned int res = 0; 4921da177e4SLinus Torvalds int error; 4931da177e4SLinus Torvalds 4941da177e4SLinus Torvalds if (npages == 0) 4951da177e4SLinus Torvalds idx_end = ~0; 4961da177e4SLinus Torvalds else 4971da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 4981da177e4SLinus Torvalds 4991da177e4SLinus Torvalds next = idx_start; 5009fd367f0STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { 5011da177e4SLinus Torvalds if (req->wb_index > idx_end) 5021da177e4SLinus Torvalds break; 5031da177e4SLinus Torvalds 5041da177e4SLinus Torvalds next = req->wb_index + 1; 505c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5061da177e4SLinus Torvalds 507c03b4024STrond Myklebust kref_get(&req->wb_kref); 508587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5091da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5101da177e4SLinus Torvalds nfs_release_request(req); 511587142f8STrond Myklebust spin_lock(&inode->i_lock); 5121da177e4SLinus Torvalds if (error < 0) 5131da177e4SLinus Torvalds return error; 5141da177e4SLinus Torvalds res++; 5151da177e4SLinus Torvalds } 5161da177e4SLinus Torvalds return res; 5171da177e4SLinus Torvalds } 5181da177e4SLinus Torvalds 51983715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 52083715ad5STrond Myklebust { 52183715ad5STrond Myklebust struct nfs_page *req; 52283715ad5STrond Myklebust 52383715ad5STrond Myklebust while(!list_empty(head)) { 52483715ad5STrond Myklebust req = nfs_list_entry(head->next); 525b6dff26aSTrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 526c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 527c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 52883715ad5STrond Myklebust nfs_list_remove_request(req); 529612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 53083715ad5STrond Myklebust nfs_inode_remove_request(req); 531b6dff26aSTrond Myklebust nfs_unlock_request(req); 53283715ad5STrond Myklebust } 53383715ad5STrond Myklebust } 53483715ad5STrond Myklebust 5351da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5361da177e4SLinus Torvalds /* 5371da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5381da177e4SLinus Torvalds * @inode: NFS inode to scan 5391da177e4SLinus Torvalds * @dst: destination list 5401da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5411da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5421da177e4SLinus Torvalds * 5431da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5441da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5451da177e4SLinus Torvalds */ 5461da177e4SLinus Torvalds static int 547ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5481da177e4SLinus Torvalds { 5491da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5503da28eb1STrond Myklebust int res = 0; 5513da28eb1STrond Myklebust 5523da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 5535c369683STrond Myklebust res = nfs_scan_list(nfsi, dst, idx_start, npages, 5545c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 5551da177e4SLinus Torvalds nfsi->ncommit -= res; 5563da28eb1STrond Myklebust } 5571da177e4SLinus Torvalds return res; 5581da177e4SLinus Torvalds } 559c42de9ddSTrond Myklebust #else 560ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 561c42de9ddSTrond Myklebust { 562c42de9ddSTrond Myklebust return 0; 563c42de9ddSTrond Myklebust } 5641da177e4SLinus Torvalds #endif 5651da177e4SLinus Torvalds 5661da177e4SLinus Torvalds /* 5671da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 5681da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 5691da177e4SLinus Torvalds * the calling process. 5701da177e4SLinus Torvalds * 5711da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 5721da177e4SLinus Torvalds */ 5731da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 574e21195a7STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 5751da177e4SLinus Torvalds { 57689a09141SPeter Zijlstra struct address_space *mapping = page->mapping; 57789a09141SPeter Zijlstra struct inode *inode = mapping->host; 5781da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 579ca52fec1STrond Myklebust pgoff_t rqend, end; 5801da177e4SLinus Torvalds 5811da177e4SLinus Torvalds end = offset + bytes; 5821da177e4SLinus Torvalds 5831da177e4SLinus Torvalds for (;;) { 5841da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 5851da177e4SLinus Torvalds * A request for the page we wish to update 5861da177e4SLinus Torvalds */ 58727852596SNick Piggin if (new) { 58827852596SNick Piggin if (radix_tree_preload(GFP_NOFS)) { 58927852596SNick Piggin nfs_release_request(new); 59027852596SNick Piggin return ERR_PTR(-ENOMEM); 59127852596SNick Piggin } 59227852596SNick Piggin } 59327852596SNick Piggin 594587142f8STrond Myklebust spin_lock(&inode->i_lock); 595277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 5961da177e4SLinus Torvalds if (req) { 597acee478aSTrond Myklebust if (!nfs_set_page_tag_locked(req)) { 5981da177e4SLinus Torvalds int error; 599277459d2STrond Myklebust 600587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6011da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6021da177e4SLinus Torvalds nfs_release_request(req); 6031dd594b2SNeil Brown if (error < 0) { 60427852596SNick Piggin if (new) { 60527852596SNick Piggin radix_tree_preload_end(); 6061dd594b2SNeil Brown nfs_release_request(new); 60727852596SNick Piggin } 6081da177e4SLinus Torvalds return ERR_PTR(error); 6091dd594b2SNeil Brown } 6101da177e4SLinus Torvalds continue; 6111da177e4SLinus Torvalds } 612587142f8STrond Myklebust spin_unlock(&inode->i_lock); 61327852596SNick Piggin if (new) { 61427852596SNick Piggin radix_tree_preload_end(); 6151da177e4SLinus Torvalds nfs_release_request(new); 61627852596SNick Piggin } 6171da177e4SLinus Torvalds break; 6181da177e4SLinus Torvalds } 6191da177e4SLinus Torvalds 6201da177e4SLinus Torvalds if (new) { 6211da177e4SLinus Torvalds nfs_lock_request_dontget(new); 62227852596SNick Piggin nfs_inode_add_request(inode, new); 623587142f8STrond Myklebust spin_unlock(&inode->i_lock); 62427852596SNick Piggin radix_tree_preload_end(); 62561e930a9STrond Myklebust req = new; 62661e930a9STrond Myklebust goto zero_page; 6271da177e4SLinus Torvalds } 628587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6291da177e4SLinus Torvalds 6301da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 6311da177e4SLinus Torvalds if (IS_ERR(new)) 6321da177e4SLinus Torvalds return new; 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds 6351da177e4SLinus Torvalds /* We have a request for our page. 6361da177e4SLinus Torvalds * If the creds don't match, or the 6371da177e4SLinus Torvalds * page addresses don't match, 6381da177e4SLinus Torvalds * tell the caller to wait on the conflicting 6391da177e4SLinus Torvalds * request. 6401da177e4SLinus Torvalds */ 6411da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 6421da177e4SLinus Torvalds if (req->wb_context != ctx 6431da177e4SLinus Torvalds || req->wb_page != page 6441da177e4SLinus Torvalds || !nfs_dirty_request(req) 6451da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 646acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 6471da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 6481da177e4SLinus Torvalds } 6491da177e4SLinus Torvalds 6501da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6511da177e4SLinus Torvalds if (offset < req->wb_offset) { 6521da177e4SLinus Torvalds req->wb_offset = offset; 6531da177e4SLinus Torvalds req->wb_pgbase = offset; 65461e930a9STrond Myklebust req->wb_bytes = max(end, rqend) - req->wb_offset; 65561e930a9STrond Myklebust goto zero_page; 6561da177e4SLinus Torvalds } 6571da177e4SLinus Torvalds 6581da177e4SLinus Torvalds if (end > rqend) 6591da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds return req; 66261e930a9STrond Myklebust zero_page: 66361e930a9STrond Myklebust /* If this page might potentially be marked as up to date, 66461e930a9STrond Myklebust * then we need to zero any uninitalised data. */ 66561e930a9STrond Myklebust if (req->wb_pgbase == 0 && req->wb_bytes != PAGE_CACHE_SIZE 66661e930a9STrond Myklebust && !PageUptodate(req->wb_page)) 667eebd2aa3SChristoph Lameter zero_user_segment(req->wb_page, req->wb_bytes, PAGE_CACHE_SIZE); 66861e930a9STrond Myklebust return req; 6691da177e4SLinus Torvalds } 6701da177e4SLinus Torvalds 6711da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6721da177e4SLinus Torvalds { 673cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 6741da177e4SLinus Torvalds struct nfs_page *req; 6751a54533eSTrond Myklebust int do_flush, status; 6761da177e4SLinus Torvalds /* 6771da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6781da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6791da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 6801da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 6811da177e4SLinus Torvalds * Also do the same if we find a request from an existing 6821da177e4SLinus Torvalds * dropped page. 6831da177e4SLinus Torvalds */ 6841a54533eSTrond Myklebust do { 685277459d2STrond Myklebust req = nfs_page_find_request(page); 6861a54533eSTrond Myklebust if (req == NULL) 6871a54533eSTrond Myklebust return 0; 6881a54533eSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx 689e261f51fSTrond Myklebust || !nfs_dirty_request(req); 6901da177e4SLinus Torvalds nfs_release_request(req); 6911a54533eSTrond Myklebust if (!do_flush) 6921a54533eSTrond Myklebust return 0; 693277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 6941a54533eSTrond Myklebust } while (status == 0); 6951a54533eSTrond Myklebust return status; 6961da177e4SLinus Torvalds } 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds /* 6995d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 7005d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 7015d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 7025d47a356STrond Myklebust */ 7035d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 7045d47a356STrond Myklebust { 7055d47a356STrond Myklebust return PageUptodate(page) && 7065d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 7075d47a356STrond Myklebust } 7085d47a356STrond Myklebust 7095d47a356STrond Myklebust /* 7101da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7111da177e4SLinus Torvalds * 7121da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7131da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7141da177e4SLinus Torvalds */ 7151da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7161da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7171da177e4SLinus Torvalds { 718cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7191da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7201da177e4SLinus Torvalds int status = 0; 7211da177e4SLinus Torvalds 72291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 72391d5b470SChuck Lever 7241da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 72501cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 72601cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7270bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 7281da177e4SLinus Torvalds 7291da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7305d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7315d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7325d47a356STrond Myklebust * inefficiencies. 7331da177e4SLinus Torvalds */ 7345d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7355d47a356STrond Myklebust inode->i_flock == NULL && 7364b5621f6STrond Myklebust !(file->f_flags & O_SYNC)) { 73749a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7381da177e4SLinus Torvalds offset = 0; 7391da177e4SLinus Torvalds } 7401da177e4SLinus Torvalds 741e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 742e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7431da177e4SLinus Torvalds 7441da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 7451da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7461da177e4SLinus Torvalds if (status < 0) 747a301b777STrond Myklebust nfs_set_pageerror(page); 7481da177e4SLinus Torvalds return status; 7491da177e4SLinus Torvalds } 7501da177e4SLinus Torvalds 7511da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7521da177e4SLinus Torvalds { 7538e821cadSTrond Myklebust 75444dd151dSTrond Myklebust if (PageError(req->wb_page)) { 75544dd151dSTrond Myklebust nfs_end_page_writeback(req->wb_page); 75644dd151dSTrond Myklebust nfs_inode_remove_request(req); 75744dd151dSTrond Myklebust } else if (!nfs_reschedule_unstable_write(req)) { 75844dd151dSTrond Myklebust /* Set the PG_uptodate flag */ 75944dd151dSTrond Myklebust nfs_mark_uptodate(req->wb_page, req->wb_pgbase, req->wb_bytes); 76089a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7611da177e4SLinus Torvalds nfs_inode_remove_request(req); 7628e821cadSTrond Myklebust } else 7638e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 7649fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 7651da177e4SLinus Torvalds } 7661da177e4SLinus Torvalds 7673ff7576dSTrond Myklebust static int flush_task_priority(int how) 7681da177e4SLinus Torvalds { 7691da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7701da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7711da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7721da177e4SLinus Torvalds case FLUSH_LOWPRI: 7731da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7761da177e4SLinus Torvalds } 7771da177e4SLinus Torvalds 7781da177e4SLinus Torvalds /* 7791da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 7801da177e4SLinus Torvalds */ 7811da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 7821da177e4SLinus Torvalds struct nfs_write_data *data, 783788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7841da177e4SLinus Torvalds unsigned int count, unsigned int offset, 7851da177e4SLinus Torvalds int how) 7861da177e4SLinus Torvalds { 78784115e1cSTrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 78884115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 7893ff7576dSTrond Myklebust int priority = flush_task_priority(how); 79007737691STrond Myklebust struct rpc_task *task; 791bdc7f021STrond Myklebust struct rpc_message msg = { 792bdc7f021STrond Myklebust .rpc_argp = &data->args, 793bdc7f021STrond Myklebust .rpc_resp = &data->res, 794bdc7f021STrond Myklebust .rpc_cred = req->wb_context->cred, 795bdc7f021STrond Myklebust }; 79684115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 79784115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 79807737691STrond Myklebust .task = &data->task, 799bdc7f021STrond Myklebust .rpc_message = &msg, 80084115e1cSTrond Myklebust .callback_ops = call_ops, 80184115e1cSTrond Myklebust .callback_data = data, 802101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 80384115e1cSTrond Myklebust .flags = flags, 8043ff7576dSTrond Myklebust .priority = priority, 80584115e1cSTrond Myklebust }; 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8081da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds data->req = req; 81188be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 812bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 8131da177e4SLinus Torvalds 8141da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8151da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8161da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8171da177e4SLinus Torvalds data->args.pages = data->pagevec; 8181da177e4SLinus Torvalds data->args.count = count; 819383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 820bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 821bdc7f021STrond Myklebust if (how & FLUSH_STABLE) { 822bdc7f021STrond Myklebust data->args.stable = NFS_DATA_SYNC; 823bdc7f021STrond Myklebust if (!NFS_I(inode)->ncommit) 824bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 825bdc7f021STrond Myklebust } 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8281da177e4SLinus Torvalds data->res.count = count; 8291da177e4SLinus Torvalds data->res.verf = &data->verf; 8300e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8311da177e4SLinus Torvalds 832788e7a89STrond Myklebust /* Set up the initial task struct. */ 833bdc7f021STrond Myklebust NFS_PROTO(inode)->write_setup(data, &msg); 8341da177e4SLinus Torvalds 835a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 836a3f565b1SChuck Lever "(req %s/%Ld, %u bytes @ offset %Lu)\n", 8370bbacc40SChuck Lever data->task.tk_pid, 8381da177e4SLinus Torvalds inode->i_sb->s_id, 8391da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8401da177e4SLinus Torvalds count, 8411da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8421da177e4SLinus Torvalds 84307737691STrond Myklebust task = rpc_run_task(&task_setup_data); 84407737691STrond Myklebust if (!IS_ERR(task)) 84507737691STrond Myklebust rpc_put_task(task); 8461da177e4SLinus Torvalds } 8471da177e4SLinus Torvalds 848*6d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 849*6d884e8fSFred * call this on each, which will prepare them to be retried on next 850*6d884e8fSFred * writeback using standard nfs. 851*6d884e8fSFred */ 852*6d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 853*6d884e8fSFred { 854*6d884e8fSFred nfs_mark_request_dirty(req); 855*6d884e8fSFred nfs_end_page_writeback(req->wb_page); 856*6d884e8fSFred nfs_clear_page_tag_locked(req); 857*6d884e8fSFred } 858*6d884e8fSFred 8591da177e4SLinus Torvalds /* 8601da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8611da177e4SLinus Torvalds * contiguous dirty area on one page. 8621da177e4SLinus Torvalds */ 8638d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8641da177e4SLinus Torvalds { 8651da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 8661da177e4SLinus Torvalds struct page *page = req->wb_page; 8671da177e4SLinus Torvalds struct nfs_write_data *data; 868e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 869e9f7bee1STrond Myklebust unsigned int offset; 8701da177e4SLinus Torvalds int requests = 0; 8711da177e4SLinus Torvalds LIST_HEAD(list); 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds nfs_list_remove_request(req); 8741da177e4SLinus Torvalds 875bcb71bbaSTrond Myklebust nbytes = count; 876e9f7bee1STrond Myklebust do { 877e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 878e9f7bee1STrond Myklebust 8798d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 8801da177e4SLinus Torvalds if (!data) 8811da177e4SLinus Torvalds goto out_bad; 8821da177e4SLinus Torvalds list_add(&data->pages, &list); 8831da177e4SLinus Torvalds requests++; 884e9f7bee1STrond Myklebust nbytes -= len; 885e9f7bee1STrond Myklebust } while (nbytes != 0); 8861da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 8871da177e4SLinus Torvalds 8881da177e4SLinus Torvalds ClearPageError(page); 8891da177e4SLinus Torvalds offset = 0; 890bcb71bbaSTrond Myklebust nbytes = count; 8911da177e4SLinus Torvalds do { 8921da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8931da177e4SLinus Torvalds list_del_init(&data->pages); 8941da177e4SLinus Torvalds 8951da177e4SLinus Torvalds data->pagevec[0] = page; 8961da177e4SLinus Torvalds 897bcb71bbaSTrond Myklebust if (nbytes < wsize) 898bcb71bbaSTrond Myklebust wsize = nbytes; 899788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 900788e7a89STrond Myklebust wsize, offset, how); 9011da177e4SLinus Torvalds offset += wsize; 9021da177e4SLinus Torvalds nbytes -= wsize; 9031da177e4SLinus Torvalds } while (nbytes != 0); 9041da177e4SLinus Torvalds 9051da177e4SLinus Torvalds return 0; 9061da177e4SLinus Torvalds 9071da177e4SLinus Torvalds out_bad: 9081da177e4SLinus Torvalds while (!list_empty(&list)) { 9091da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9101da177e4SLinus Torvalds list_del(&data->pages); 9118aca67f0STrond Myklebust nfs_writedata_release(data); 9121da177e4SLinus Torvalds } 91361822ab5STrond Myklebust nfs_redirty_request(req); 9141da177e4SLinus Torvalds return -ENOMEM; 9151da177e4SLinus Torvalds } 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds /* 9181da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9191da177e4SLinus Torvalds * The page must have been locked by the caller. 9201da177e4SLinus Torvalds * 9211da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9221da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9231da177e4SLinus Torvalds * that has been written but not committed. 9241da177e4SLinus Torvalds */ 9258d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9261da177e4SLinus Torvalds { 9271da177e4SLinus Torvalds struct nfs_page *req; 9281da177e4SLinus Torvalds struct page **pages; 9291da177e4SLinus Torvalds struct nfs_write_data *data; 9301da177e4SLinus Torvalds 9318d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 9321da177e4SLinus Torvalds if (!data) 9331da177e4SLinus Torvalds goto out_bad; 9341da177e4SLinus Torvalds 9351da177e4SLinus Torvalds pages = data->pagevec; 9361da177e4SLinus Torvalds while (!list_empty(head)) { 9371da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9381da177e4SLinus Torvalds nfs_list_remove_request(req); 9391da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9401da177e4SLinus Torvalds ClearPageError(req->wb_page); 9411da177e4SLinus Torvalds *pages++ = req->wb_page; 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9441da177e4SLinus Torvalds 9451da177e4SLinus Torvalds /* Set up the argument struct */ 946788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9471da177e4SLinus Torvalds 9481da177e4SLinus Torvalds return 0; 9491da177e4SLinus Torvalds out_bad: 9501da177e4SLinus Torvalds while (!list_empty(head)) { 95110afec90STrond Myklebust req = nfs_list_entry(head->next); 9521da177e4SLinus Torvalds nfs_list_remove_request(req); 95361822ab5STrond Myklebust nfs_redirty_request(req); 9541da177e4SLinus Torvalds } 9551da177e4SLinus Torvalds return -ENOMEM; 9561da177e4SLinus Torvalds } 9571da177e4SLinus Torvalds 958c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 959c63c7b05STrond Myklebust struct inode *inode, int ioflags) 9601da177e4SLinus Torvalds { 961bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 9621da177e4SLinus Torvalds 963bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 964c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 965bcb71bbaSTrond Myklebust else 966c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 9671da177e4SLinus Torvalds } 9681da177e4SLinus Torvalds 9691da177e4SLinus Torvalds /* 9701da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 9711da177e4SLinus Torvalds */ 972788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 9731da177e4SLinus Torvalds { 974788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 9751da177e4SLinus Torvalds struct nfs_page *req = data->req; 9761da177e4SLinus Torvalds struct page *page = req->wb_page; 9771da177e4SLinus Torvalds 9781da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 97988be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 98088be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 9811da177e4SLinus Torvalds req->wb_bytes, 9821da177e4SLinus Torvalds (long long)req_offset(req)); 9831da177e4SLinus Torvalds 984788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 985788e7a89STrond Myklebust return; 986788e7a89STrond Myklebust 987788e7a89STrond Myklebust if (task->tk_status < 0) { 988a301b777STrond Myklebust nfs_set_pageerror(page); 9897b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 990788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 9918e821cadSTrond Myklebust goto out; 9928e821cadSTrond Myklebust } 9938e821cadSTrond Myklebust 9948e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 995587142f8STrond Myklebust struct inode *inode = page->mapping->host; 9968e821cadSTrond Myklebust 997587142f8STrond Myklebust spin_lock(&inode->i_lock); 9988e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 9998e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 10008e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 10011da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10021da177e4SLinus Torvalds dprintk(" defer commit\n"); 10031da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10048e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 10058e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 10061da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10071da177e4SLinus Torvalds } 1008587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10091da177e4SLinus Torvalds } else 10101da177e4SLinus Torvalds dprintk(" OK\n"); 10111da177e4SLinus Torvalds 10128e821cadSTrond Myklebust out: 10131da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 10141da177e4SLinus Torvalds nfs_writepage_release(req); 10151da177e4SLinus Torvalds } 10161da177e4SLinus Torvalds 1017788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1018788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1019788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1020788e7a89STrond Myklebust }; 1021788e7a89STrond Myklebust 10221da177e4SLinus Torvalds /* 10231da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10241da177e4SLinus Torvalds * 10251da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10261da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10271da177e4SLinus Torvalds * as the page has a write request pending. 10281da177e4SLinus Torvalds */ 1029788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10301da177e4SLinus Torvalds { 1031788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10321da177e4SLinus Torvalds struct nfs_page *req; 10331da177e4SLinus Torvalds struct page *page; 10341da177e4SLinus Torvalds 1035788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1036788e7a89STrond Myklebust return; 1037788e7a89STrond Myklebust 10381da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 10391da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 10401da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10411da177e4SLinus Torvalds nfs_list_remove_request(req); 10421da177e4SLinus Torvalds page = req->wb_page; 10431da177e4SLinus Torvalds 10441da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 104588be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 104688be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 10471da177e4SLinus Torvalds req->wb_bytes, 10481da177e4SLinus Torvalds (long long)req_offset(req)); 10491da177e4SLinus Torvalds 1050788e7a89STrond Myklebust if (task->tk_status < 0) { 1051a301b777STrond Myklebust nfs_set_pageerror(page); 10527b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 1053788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 10548e821cadSTrond Myklebust goto remove_request; 10551da177e4SLinus Torvalds } 10561da177e4SLinus Torvalds 10578e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 10581da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10591da177e4SLinus Torvalds nfs_mark_request_commit(req); 10608e821cadSTrond Myklebust nfs_end_page_writeback(page); 10611da177e4SLinus Torvalds dprintk(" marked for commit\n"); 10628e821cadSTrond Myklebust goto next; 10638e821cadSTrond Myklebust } 106444dd151dSTrond Myklebust /* Set the PG_uptodate flag? */ 106544dd151dSTrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 10668e821cadSTrond Myklebust dprintk(" OK\n"); 10678e821cadSTrond Myklebust remove_request: 10688e821cadSTrond Myklebust nfs_end_page_writeback(page); 10691da177e4SLinus Torvalds nfs_inode_remove_request(req); 10701da177e4SLinus Torvalds next: 10719fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 10721da177e4SLinus Torvalds } 10731da177e4SLinus Torvalds } 10741da177e4SLinus Torvalds 1075788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1076788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1077788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1078788e7a89STrond Myklebust }; 1079788e7a89STrond Myklebust 1080788e7a89STrond Myklebust 10811da177e4SLinus Torvalds /* 10821da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 10831da177e4SLinus Torvalds */ 1084462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 10851da177e4SLinus Torvalds { 10861da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 10871da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1088788e7a89STrond Myklebust int status; 10891da177e4SLinus Torvalds 1090a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 10911da177e4SLinus Torvalds task->tk_pid, task->tk_status); 10921da177e4SLinus Torvalds 1093f551e44fSChuck Lever /* 1094f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1095f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1096f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1097f551e44fSChuck Lever * another writer had changed the file, but some applications 1098f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1099f551e44fSChuck Lever */ 1100788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1101788e7a89STrond Myklebust if (status != 0) 1102788e7a89STrond Myklebust return status; 110391d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 110491d5b470SChuck Lever 11051da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11061da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 11071da177e4SLinus Torvalds /* We tried a write call, but the server did not 11081da177e4SLinus Torvalds * commit data to stable storage even though we 11091da177e4SLinus Torvalds * requested it. 11101da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11111da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11121da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11131da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11141da177e4SLinus Torvalds */ 11151da177e4SLinus Torvalds static unsigned long complain; 11161da177e4SLinus Torvalds 11171da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11181da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11191da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 112054ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 11211da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11221da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds } 11251da177e4SLinus Torvalds #endif 11261da177e4SLinus Torvalds /* Is this a short write? */ 11271da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11281da177e4SLinus Torvalds static unsigned long complain; 11291da177e4SLinus Torvalds 113091d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 113191d5b470SChuck Lever 11321da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11331da177e4SLinus Torvalds if (resp->count != 0) { 11341da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11351da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11361da177e4SLinus Torvalds /* Resend from where the server left off */ 11371da177e4SLinus Torvalds argp->offset += resp->count; 11381da177e4SLinus Torvalds argp->pgbase += resp->count; 11391da177e4SLinus Torvalds argp->count -= resp->count; 11401da177e4SLinus Torvalds } else { 11411da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11421da177e4SLinus Torvalds * headaches in the case of a server crash. 11431da177e4SLinus Torvalds */ 11441da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11451da177e4SLinus Torvalds } 11461da177e4SLinus Torvalds rpc_restart_call(task); 1147788e7a89STrond Myklebust return -EAGAIN; 11481da177e4SLinus Torvalds } 11491da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11501da177e4SLinus Torvalds printk(KERN_WARNING 11511da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 11521da177e4SLinus Torvalds argp->count); 11531da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11541da177e4SLinus Torvalds } 11551da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 11561da177e4SLinus Torvalds task->tk_status = -EIO; 11571da177e4SLinus Torvalds } 1158788e7a89STrond Myklebust return 0; 11591da177e4SLinus Torvalds } 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds 11621da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1163383ba719STrond Myklebust void nfs_commit_release(void *data) 11641da177e4SLinus Torvalds { 1165383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1166383ba719STrond Myklebust 1167383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 11681da177e4SLinus Torvalds nfs_commit_free(wdata); 11691da177e4SLinus Torvalds } 11701da177e4SLinus Torvalds 11711da177e4SLinus Torvalds /* 11721da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 11731da177e4SLinus Torvalds */ 11741da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1175788e7a89STrond Myklebust struct nfs_write_data *data, 1176788e7a89STrond Myklebust int how) 11771da177e4SLinus Torvalds { 117884115e1cSTrond Myklebust struct nfs_page *first = nfs_list_entry(head->next); 117984115e1cSTrond Myklebust struct inode *inode = first->wb_context->path.dentry->d_inode; 118084115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 11813ff7576dSTrond Myklebust int priority = flush_task_priority(how); 118207737691STrond Myklebust struct rpc_task *task; 1183bdc7f021STrond Myklebust struct rpc_message msg = { 1184bdc7f021STrond Myklebust .rpc_argp = &data->args, 1185bdc7f021STrond Myklebust .rpc_resp = &data->res, 1186bdc7f021STrond Myklebust .rpc_cred = first->wb_context->cred, 1187bdc7f021STrond Myklebust }; 118884115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 118907737691STrond Myklebust .task = &data->task, 119084115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 1191bdc7f021STrond Myklebust .rpc_message = &msg, 119284115e1cSTrond Myklebust .callback_ops = &nfs_commit_ops, 119384115e1cSTrond Myklebust .callback_data = data, 1194101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 119584115e1cSTrond Myklebust .flags = flags, 11963ff7576dSTrond Myklebust .priority = priority, 119784115e1cSTrond Myklebust }; 11981da177e4SLinus Torvalds 11991da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12001da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12011da177e4SLinus Torvalds 12021da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds data->inode = inode; 1205bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 12083da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 12093da28eb1STrond Myklebust data->args.offset = 0; 12103da28eb1STrond Myklebust data->args.count = 0; 1211383ba719STrond Myklebust data->args.context = get_nfs_open_context(first->wb_context); 12123da28eb1STrond Myklebust data->res.count = 0; 12131da177e4SLinus Torvalds data->res.fattr = &data->fattr; 12141da177e4SLinus Torvalds data->res.verf = &data->verf; 12150e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12161da177e4SLinus Torvalds 1217788e7a89STrond Myklebust /* Set up the initial task struct. */ 1218bdc7f021STrond Myklebust NFS_PROTO(inode)->commit_setup(data, &msg); 12191da177e4SLinus Torvalds 1220a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1221bdc7f021STrond Myklebust 122207737691STrond Myklebust task = rpc_run_task(&task_setup_data); 122307737691STrond Myklebust if (!IS_ERR(task)) 122407737691STrond Myklebust rpc_put_task(task); 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds /* 12281da177e4SLinus Torvalds * Commit dirty pages 12291da177e4SLinus Torvalds */ 12301da177e4SLinus Torvalds static int 123140859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 12321da177e4SLinus Torvalds { 12331da177e4SLinus Torvalds struct nfs_write_data *data; 12341da177e4SLinus Torvalds struct nfs_page *req; 12351da177e4SLinus Torvalds 1236e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds if (!data) 12391da177e4SLinus Torvalds goto out_bad; 12401da177e4SLinus Torvalds 12411da177e4SLinus Torvalds /* Set up the argument struct */ 12421da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds return 0; 12451da177e4SLinus Torvalds out_bad: 12461da177e4SLinus Torvalds while (!list_empty(head)) { 12471da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12481da177e4SLinus Torvalds nfs_list_remove_request(req); 12491da177e4SLinus Torvalds nfs_mark_request_commit(req); 125083715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1251c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1252c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12539fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds return -ENOMEM; 12561da177e4SLinus Torvalds } 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds /* 12591da177e4SLinus Torvalds * COMMIT call returned 12601da177e4SLinus Torvalds */ 1261788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 12621da177e4SLinus Torvalds { 1263963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12641da177e4SLinus Torvalds struct nfs_page *req; 12651da177e4SLinus Torvalds 1266a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 12671da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12681da177e4SLinus Torvalds 1269788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1270788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1271788e7a89STrond Myklebust return; 1272788e7a89STrond Myklebust 12731da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12741da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12751da177e4SLinus Torvalds nfs_list_remove_request(req); 1276612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 1277fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1278c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1279c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12801da177e4SLinus Torvalds 12811da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 128288be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 128388be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 12841da177e4SLinus Torvalds req->wb_bytes, 12851da177e4SLinus Torvalds (long long)req_offset(req)); 12861da177e4SLinus Torvalds if (task->tk_status < 0) { 12877b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 12881da177e4SLinus Torvalds nfs_inode_remove_request(req); 12891da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 12901da177e4SLinus Torvalds goto next; 12911da177e4SLinus Torvalds } 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 12941da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 12951da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 12961da177e4SLinus Torvalds /* We have a match */ 129744dd151dSTrond Myklebust /* Set the PG_uptodate flag */ 129844dd151dSTrond Myklebust nfs_mark_uptodate(req->wb_page, req->wb_pgbase, 129944dd151dSTrond Myklebust req->wb_bytes); 13001da177e4SLinus Torvalds nfs_inode_remove_request(req); 13011da177e4SLinus Torvalds dprintk(" OK\n"); 13021da177e4SLinus Torvalds goto next; 13031da177e4SLinus Torvalds } 13041da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13051da177e4SLinus Torvalds dprintk(" mismatch\n"); 1306*6d884e8fSFred nfs_mark_request_dirty(req); 13071da177e4SLinus Torvalds next: 13089fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 13091da177e4SLinus Torvalds } 13101da177e4SLinus Torvalds } 1311788e7a89STrond Myklebust 1312788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1313788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1314788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1315788e7a89STrond Myklebust }; 13161da177e4SLinus Torvalds 13173da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13181da177e4SLinus Torvalds { 13191da177e4SLinus Torvalds LIST_HEAD(head); 13207d46a49fSTrond Myklebust int res; 13211da177e4SLinus Torvalds 1322587142f8STrond Myklebust spin_lock(&inode->i_lock); 13233da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1324587142f8STrond Myklebust spin_unlock(&inode->i_lock); 13251da177e4SLinus Torvalds if (res) { 13267d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 13271da177e4SLinus Torvalds if (error < 0) 13281da177e4SLinus Torvalds return error; 13293da28eb1STrond Myklebust } 13301da177e4SLinus Torvalds return res; 13311da177e4SLinus Torvalds } 1332c63c7b05STrond Myklebust #else 1333c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1334c63c7b05STrond Myklebust { 1335c63c7b05STrond Myklebust return 0; 1336c63c7b05STrond Myklebust } 13371da177e4SLinus Torvalds #endif 13381da177e4SLinus Torvalds 13391c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 13401da177e4SLinus Torvalds { 13411c75950bSTrond Myklebust struct inode *inode = mapping->host; 1342ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 13431c75950bSTrond Myklebust unsigned int npages = 0; 1344c42de9ddSTrond Myklebust LIST_HEAD(head); 134570b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 13463f442547STrond Myklebust long pages, ret; 13471da177e4SLinus Torvalds 13481c75950bSTrond Myklebust /* FIXME */ 13491c75950bSTrond Myklebust if (wbc->range_cyclic) 13501c75950bSTrond Myklebust idx_start = 0; 13511c75950bSTrond Myklebust else { 13521c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 13531c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 13541c75950bSTrond Myklebust if (idx_end > idx_start) { 1355ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 13561c75950bSTrond Myklebust npages = l_npages; 13571c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1358ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 13591c75950bSTrond Myklebust npages = 0; 13601c75950bSTrond Myklebust } 13611c75950bSTrond Myklebust } 1362c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1363587142f8STrond Myklebust spin_lock(&inode->i_lock); 13641da177e4SLinus Torvalds do { 1365c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1366c42de9ddSTrond Myklebust if (ret != 0) 1367c42de9ddSTrond Myklebust continue; 1368c42de9ddSTrond Myklebust if (nocommit) 1369c42de9ddSTrond Myklebust break; 1370d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1371724c439cSTrond Myklebust if (pages == 0) 1372c42de9ddSTrond Myklebust break; 1373d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1374587142f8STrond Myklebust spin_unlock(&inode->i_lock); 137583715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1376e8e058e8STrond Myklebust ret = pages; 1377587142f8STrond Myklebust spin_lock(&inode->i_lock); 1378d2ccddf0STrond Myklebust continue; 1379d2ccddf0STrond Myklebust } 1380d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1381587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1382c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1383587142f8STrond Myklebust spin_lock(&inode->i_lock); 1384587142f8STrond Myklebust 1385c42de9ddSTrond Myklebust } while (ret >= 0); 1386587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1387c42de9ddSTrond Myklebust return ret; 13881da177e4SLinus Torvalds } 13891da177e4SLinus Torvalds 139034901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 13911c75950bSTrond Myklebust { 13921c75950bSTrond Myklebust int ret; 13931c75950bSTrond Myklebust 139434901f70STrond Myklebust ret = nfs_writepages(mapping, wbc); 139561822ab5STrond Myklebust if (ret < 0) 139661822ab5STrond Myklebust goto out; 139734901f70STrond Myklebust ret = nfs_sync_mapping_wait(mapping, wbc, how); 1398ed90ef51STrond Myklebust if (ret < 0) 1399ed90ef51STrond Myklebust goto out; 14001c75950bSTrond Myklebust return 0; 140161822ab5STrond Myklebust out: 1402e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 14031c75950bSTrond Myklebust return ret; 14041c75950bSTrond Myklebust } 14051c75950bSTrond Myklebust 140634901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ 140734901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how) 140834901f70STrond Myklebust { 140934901f70STrond Myklebust struct writeback_control wbc = { 141034901f70STrond Myklebust .bdi = mapping->backing_dev_info, 141134901f70STrond Myklebust .sync_mode = WB_SYNC_NONE, 141234901f70STrond Myklebust .nr_to_write = LONG_MAX, 141334901f70STrond Myklebust .for_writepages = 1, 141434901f70STrond Myklebust .range_cyclic = 1, 141534901f70STrond Myklebust }; 141634901f70STrond Myklebust int ret; 141734901f70STrond Myklebust 141834901f70STrond Myklebust ret = __nfs_write_mapping(mapping, &wbc, how); 141934901f70STrond Myklebust if (ret < 0) 142034901f70STrond Myklebust return ret; 142134901f70STrond Myklebust wbc.sync_mode = WB_SYNC_ALL; 142234901f70STrond Myklebust return __nfs_write_mapping(mapping, &wbc, how); 142334901f70STrond Myklebust } 142434901f70STrond Myklebust 1425ed90ef51STrond Myklebust /* 1426ed90ef51STrond Myklebust * flush the inode to disk. 1427ed90ef51STrond Myklebust */ 1428ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode) 14291c75950bSTrond Myklebust { 1430ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, 0); 1431ed90ef51STrond Myklebust } 14321c75950bSTrond Myklebust 1433ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode) 1434ed90ef51STrond Myklebust { 1435ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); 14361c75950bSTrond Myklebust } 14371c75950bSTrond Myklebust 14381b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 14391b3b4a1aSTrond Myklebust { 14401b3b4a1aSTrond Myklebust struct nfs_page *req; 14411b3b4a1aSTrond Myklebust loff_t range_start = page_offset(page); 14421b3b4a1aSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14431b3b4a1aSTrond Myklebust struct writeback_control wbc = { 14441b3b4a1aSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14451b3b4a1aSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14461b3b4a1aSTrond Myklebust .nr_to_write = LONG_MAX, 14471b3b4a1aSTrond Myklebust .range_start = range_start, 14481b3b4a1aSTrond Myklebust .range_end = range_end, 14491b3b4a1aSTrond Myklebust }; 14501b3b4a1aSTrond Myklebust int ret = 0; 14511b3b4a1aSTrond Myklebust 14521b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 14531b3b4a1aSTrond Myklebust for (;;) { 14541b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 14551b3b4a1aSTrond Myklebust if (req == NULL) 14561b3b4a1aSTrond Myklebust goto out; 14571b3b4a1aSTrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 14581b3b4a1aSTrond Myklebust nfs_release_request(req); 14591b3b4a1aSTrond Myklebust break; 14601b3b4a1aSTrond Myklebust } 14611b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 14621b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 14631b3b4a1aSTrond Myklebust /* 14641b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 14651b3b4a1aSTrond Myklebust * page as being dirty 14661b3b4a1aSTrond Myklebust */ 14671b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 14681b3b4a1aSTrond Myklebust nfs_unlock_request(req); 14691b3b4a1aSTrond Myklebust break; 14701b3b4a1aSTrond Myklebust } 14711b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 14721b3b4a1aSTrond Myklebust if (ret < 0) 14731b3b4a1aSTrond Myklebust goto out; 14741b3b4a1aSTrond Myklebust } 14751b3b4a1aSTrond Myklebust if (!PagePrivate(page)) 14761b3b4a1aSTrond Myklebust return 0; 14771b3b4a1aSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); 14781b3b4a1aSTrond Myklebust out: 14791b3b4a1aSTrond Myklebust return ret; 14801b3b4a1aSTrond Myklebust } 14811b3b4a1aSTrond Myklebust 14825334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page, 14835334eb13SAdrian Bunk int how) 14841c75950bSTrond Myklebust { 14851c75950bSTrond Myklebust loff_t range_start = page_offset(page); 14861c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14874d770ccfSTrond Myklebust struct writeback_control wbc = { 14884d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14894d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14904d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 14914d770ccfSTrond Myklebust .range_start = range_start, 14924d770ccfSTrond Myklebust .range_end = range_end, 14934d770ccfSTrond Myklebust }; 14944d770ccfSTrond Myklebust int ret; 14951c75950bSTrond Myklebust 14964d770ccfSTrond Myklebust BUG_ON(!PageLocked(page)); 1497c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 14984d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 14994d770ccfSTrond Myklebust if (ret < 0) 15004d770ccfSTrond Myklebust goto out; 15014d770ccfSTrond Myklebust } 1502f40313acSTrond Myklebust if (!PagePrivate(page)) 1503f40313acSTrond Myklebust return 0; 15044d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 15054d770ccfSTrond Myklebust if (ret >= 0) 15064d770ccfSTrond Myklebust return 0; 15074d770ccfSTrond Myklebust out: 1508e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 15094d770ccfSTrond Myklebust return ret; 15101c75950bSTrond Myklebust } 15111c75950bSTrond Myklebust 15121c75950bSTrond Myklebust /* 15131c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15141c75950bSTrond Myklebust */ 15151c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 15161c75950bSTrond Myklebust { 15174d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 15181c75950bSTrond Myklebust } 15191c75950bSTrond Myklebust 1520f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15211da177e4SLinus Torvalds { 15221da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15231da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15241da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 152520c2df83SPaul Mundt NULL); 15261da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15271da177e4SLinus Torvalds return -ENOMEM; 15281da177e4SLinus Torvalds 152993d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15301da177e4SLinus Torvalds nfs_wdata_cachep); 15311da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15321da177e4SLinus Torvalds return -ENOMEM; 15331da177e4SLinus Torvalds 153493d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15351da177e4SLinus Torvalds nfs_wdata_cachep); 15361da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15371da177e4SLinus Torvalds return -ENOMEM; 15381da177e4SLinus Torvalds 153989a09141SPeter Zijlstra /* 154089a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 154189a09141SPeter Zijlstra * 154289a09141SPeter Zijlstra * 64MB: 8192k 154389a09141SPeter Zijlstra * 128MB: 11585k 154489a09141SPeter Zijlstra * 256MB: 16384k 154589a09141SPeter Zijlstra * 512MB: 23170k 154689a09141SPeter Zijlstra * 1GB: 32768k 154789a09141SPeter Zijlstra * 2GB: 46340k 154889a09141SPeter Zijlstra * 4GB: 65536k 154989a09141SPeter Zijlstra * 8GB: 92681k 155089a09141SPeter Zijlstra * 16GB: 131072k 155189a09141SPeter Zijlstra * 155289a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 155389a09141SPeter Zijlstra * Limit the default to 256M 155489a09141SPeter Zijlstra */ 155589a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 155689a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 155789a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 155889a09141SPeter Zijlstra 15591da177e4SLinus Torvalds return 0; 15601da177e4SLinus Torvalds } 15611da177e4SLinus Torvalds 1562266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15631da177e4SLinus Torvalds { 15641da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15651da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15661a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15671da177e4SLinus Torvalds } 15681da177e4SLinus Torvalds 1569