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); 42788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 43788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 44788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 451da177e4SLinus Torvalds 46e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 473feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 481da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 491da177e4SLinus Torvalds 50e9f7bee1STrond Myklebust struct nfs_write_data *nfs_commit_alloc(void) 511da177e4SLinus Torvalds { 52e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5340859d7eSChuck Lever 541da177e4SLinus Torvalds if (p) { 551da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 561da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 571da177e4SLinus Torvalds } 581da177e4SLinus Torvalds return p; 591da177e4SLinus Torvalds } 601da177e4SLinus Torvalds 61*5e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 621da177e4SLinus Torvalds { 6340859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6440859d7eSChuck Lever kfree(p->pagevec); 651da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 661da177e4SLinus Torvalds } 671da177e4SLinus Torvalds 688d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 693feb2d49STrond Myklebust { 70e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 713feb2d49STrond Myklebust 723feb2d49STrond Myklebust if (p) { 733feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 743feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 75e9f7bee1STrond Myklebust p->npages = pagecount; 760d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 770d0b5cb3SChuck Lever p->pagevec = p->page_array; 783feb2d49STrond Myklebust else { 790d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 800d0b5cb3SChuck Lever if (!p->pagevec) { 813feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 823feb2d49STrond Myklebust p = NULL; 833feb2d49STrond Myklebust } 843feb2d49STrond Myklebust } 853feb2d49STrond Myklebust } 863feb2d49STrond Myklebust return p; 873feb2d49STrond Myklebust } 883feb2d49STrond Myklebust 89*5e4424afSTrond Myklebust static void nfs_writedata_free(struct nfs_write_data *p) 903feb2d49STrond Myklebust { 913feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 923feb2d49STrond Myklebust kfree(p->pagevec); 933feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 943feb2d49STrond Myklebust } 953feb2d49STrond Myklebust 96383ba719STrond Myklebust void nfs_writedata_release(void *data) 971da177e4SLinus Torvalds { 98383ba719STrond Myklebust struct nfs_write_data *wdata = data; 99383ba719STrond Myklebust 100383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 1011da177e4SLinus Torvalds nfs_writedata_free(wdata); 1021da177e4SLinus Torvalds } 1031da177e4SLinus Torvalds 1047b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1057b159fc1STrond Myklebust { 1067b159fc1STrond Myklebust ctx->error = error; 1077b159fc1STrond Myklebust smp_wmb(); 1087b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1097b159fc1STrond Myklebust } 1107b159fc1STrond Myklebust 111277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 112277459d2STrond Myklebust { 113277459d2STrond Myklebust struct nfs_page *req = NULL; 114277459d2STrond Myklebust 115277459d2STrond Myklebust if (PagePrivate(page)) { 116277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 117277459d2STrond Myklebust if (req != NULL) 118c03b4024STrond Myklebust kref_get(&req->wb_kref); 119277459d2STrond Myklebust } 120277459d2STrond Myklebust return req; 121277459d2STrond Myklebust } 122277459d2STrond Myklebust 123277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 124277459d2STrond Myklebust { 125587142f8STrond Myklebust struct inode *inode = page->mapping->host; 126277459d2STrond Myklebust struct nfs_page *req = NULL; 127277459d2STrond Myklebust 128587142f8STrond Myklebust spin_lock(&inode->i_lock); 129277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 130587142f8STrond Myklebust spin_unlock(&inode->i_lock); 131277459d2STrond Myklebust return req; 132277459d2STrond Myklebust } 133277459d2STrond Myklebust 1341da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1351da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1361da177e4SLinus Torvalds { 1371da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1381da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 139ca52fec1STrond Myklebust pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1401da177e4SLinus Torvalds 1411da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1421da177e4SLinus Torvalds return; 1431da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1441da177e4SLinus Torvalds if (i_size >= end) 1451da177e4SLinus Torvalds return; 14691d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1471da177e4SLinus Torvalds i_size_write(inode, end); 1481da177e4SLinus Torvalds } 1491da177e4SLinus Torvalds 150a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 151a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 152a301b777STrond Myklebust { 153a301b777STrond Myklebust SetPageError(page); 154a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 155a301b777STrond Myklebust } 156a301b777STrond Myklebust 1571da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1581da177e4SLinus Torvalds * covers the full page. 1591da177e4SLinus Torvalds */ 1601da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1611da177e4SLinus Torvalds { 1621da177e4SLinus Torvalds if (PageUptodate(page)) 1631da177e4SLinus Torvalds return; 1641da177e4SLinus Torvalds if (base != 0) 1651da177e4SLinus Torvalds return; 16649a70f27STrond Myklebust if (count != nfs_page_length(page)) 1671da177e4SLinus Torvalds return; 1681da177e4SLinus Torvalds SetPageUptodate(page); 1691da177e4SLinus Torvalds } 1701da177e4SLinus Torvalds 171e21195a7STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1721da177e4SLinus Torvalds unsigned int offset, unsigned int count) 1731da177e4SLinus Torvalds { 1741da177e4SLinus Torvalds struct nfs_page *req; 175e21195a7STrond Myklebust int ret; 1761da177e4SLinus Torvalds 177e21195a7STrond Myklebust for (;;) { 178e21195a7STrond Myklebust req = nfs_update_request(ctx, page, offset, count); 179e21195a7STrond Myklebust if (!IS_ERR(req)) 180e21195a7STrond Myklebust break; 181e21195a7STrond Myklebust ret = PTR_ERR(req); 182e21195a7STrond Myklebust if (ret != -EBUSY) 183e21195a7STrond Myklebust return ret; 184e21195a7STrond Myklebust ret = nfs_wb_page(page->mapping->host, page); 185e21195a7STrond Myklebust if (ret != 0) 186e21195a7STrond Myklebust return ret; 187e21195a7STrond Myklebust } 1881da177e4SLinus Torvalds /* Update file length */ 1891da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 190acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 191abd3e641STrond Myklebust return 0; 1921da177e4SLinus Torvalds } 1931da177e4SLinus Torvalds 1941da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1951da177e4SLinus Torvalds { 1961da177e4SLinus Torvalds if (wbc->for_reclaim) 197c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 1981da177e4SLinus Torvalds if (wbc->for_kupdate) 1991da177e4SLinus Torvalds return FLUSH_LOWPRI; 2001da177e4SLinus Torvalds return 0; 2011da177e4SLinus Torvalds } 2021da177e4SLinus Torvalds 2031da177e4SLinus Torvalds /* 20489a09141SPeter Zijlstra * NFS congestion control 20589a09141SPeter Zijlstra */ 20689a09141SPeter Zijlstra 20789a09141SPeter Zijlstra int nfs_congestion_kb; 20889a09141SPeter Zijlstra 20989a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 21089a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 21189a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 21289a09141SPeter Zijlstra 2135a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 21489a09141SPeter Zijlstra { 2155a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2165a6d41b3STrond Myklebust 2175a6d41b3STrond Myklebust if (!ret) { 21889a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 21989a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 22089a09141SPeter Zijlstra 221277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 22289a09141SPeter Zijlstra NFS_CONGESTION_ON_THRESH) 22389a09141SPeter Zijlstra set_bdi_congested(&nfss->backing_dev_info, WRITE); 22489a09141SPeter Zijlstra } 2255a6d41b3STrond Myklebust return ret; 22689a09141SPeter Zijlstra } 22789a09141SPeter Zijlstra 22889a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 22989a09141SPeter Zijlstra { 23089a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 23189a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 23289a09141SPeter Zijlstra 23389a09141SPeter Zijlstra end_page_writeback(page); 234c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 23589a09141SPeter Zijlstra clear_bdi_congested(&nfss->backing_dev_info, WRITE); 23689a09141SPeter Zijlstra } 23789a09141SPeter Zijlstra 23889a09141SPeter Zijlstra /* 239e261f51fSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 2409cccef95STrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 241e261f51fSTrond Myklebust */ 242c63c7b05STrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 243c63c7b05STrond Myklebust struct page *page) 244e261f51fSTrond Myklebust { 245587142f8STrond Myklebust struct inode *inode = page->mapping->host; 246e261f51fSTrond Myklebust struct nfs_page *req; 247e261f51fSTrond Myklebust int ret; 248e261f51fSTrond Myklebust 249587142f8STrond Myklebust spin_lock(&inode->i_lock); 250e261f51fSTrond Myklebust for(;;) { 251e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 252e261f51fSTrond Myklebust if (req == NULL) { 253587142f8STrond Myklebust spin_unlock(&inode->i_lock); 2549cccef95STrond Myklebust return 0; 255e261f51fSTrond Myklebust } 256acee478aSTrond Myklebust if (nfs_set_page_tag_locked(req)) 257e261f51fSTrond Myklebust break; 258e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 259acee478aSTrond Myklebust * then the call to nfs_set_page_tag_locked() will always 260e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 261e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 262e261f51fSTrond Myklebust */ 263587142f8STrond Myklebust spin_unlock(&inode->i_lock); 264e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 265e261f51fSTrond Myklebust nfs_release_request(req); 266e261f51fSTrond Myklebust if (ret != 0) 267e261f51fSTrond Myklebust return ret; 268587142f8STrond Myklebust spin_lock(&inode->i_lock); 269e261f51fSTrond Myklebust } 270612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 271612c9384STrond Myklebust /* This request is marked for commit */ 272587142f8STrond Myklebust spin_unlock(&inode->i_lock); 273acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 274c63c7b05STrond Myklebust nfs_pageio_complete(pgio); 2759cccef95STrond Myklebust return 0; 276612c9384STrond Myklebust } 277c63c7b05STrond Myklebust if (nfs_set_page_writeback(page) != 0) { 278587142f8STrond Myklebust spin_unlock(&inode->i_lock); 279c63c7b05STrond Myklebust BUG(); 280c63c7b05STrond Myklebust } 281587142f8STrond Myklebust spin_unlock(&inode->i_lock); 282c63c7b05STrond Myklebust nfs_pageio_add_request(pgio, req); 2839cccef95STrond Myklebust return 0; 284e261f51fSTrond Myklebust } 285e261f51fSTrond Myklebust 286f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 287f758c885STrond Myklebust { 288f758c885STrond Myklebust struct inode *inode = page->mapping->host; 289f758c885STrond Myklebust 290f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 291f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 292f758c885STrond Myklebust 293f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 294f758c885STrond Myklebust return nfs_page_async_flush(pgio, page); 295f758c885STrond Myklebust } 296f758c885STrond Myklebust 297e261f51fSTrond Myklebust /* 2981da177e4SLinus Torvalds * Write an mmapped page to the server. 2991da177e4SLinus Torvalds */ 3004d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3011da177e4SLinus Torvalds { 302f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 303e261f51fSTrond Myklebust int err; 3041da177e4SLinus Torvalds 305f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 306f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 307f758c885STrond Myklebust nfs_pageio_complete(&pgio); 308f758c885STrond Myklebust if (err < 0) 3094d770ccfSTrond Myklebust return err; 310f758c885STrond Myklebust if (pgio.pg_error < 0) 311f758c885STrond Myklebust return pgio.pg_error; 312f758c885STrond Myklebust return 0; 3134d770ccfSTrond Myklebust } 3144d770ccfSTrond Myklebust 3154d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3164d770ccfSTrond Myklebust { 317f758c885STrond Myklebust int ret; 3184d770ccfSTrond Myklebust 319f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3201da177e4SLinus Torvalds unlock_page(page); 321f758c885STrond Myklebust return ret; 322f758c885STrond Myklebust } 323f758c885STrond Myklebust 324f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 325f758c885STrond Myklebust { 326f758c885STrond Myklebust int ret; 327f758c885STrond Myklebust 328f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 329f758c885STrond Myklebust unlock_page(page); 330f758c885STrond Myklebust return ret; 3311da177e4SLinus Torvalds } 3321da177e4SLinus Torvalds 3331da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3341da177e4SLinus Torvalds { 3351da177e4SLinus Torvalds struct inode *inode = mapping->host; 336c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3371da177e4SLinus Torvalds int err; 3381da177e4SLinus Torvalds 33991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 34091d5b470SChuck Lever 341c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 342f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 343c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 344f758c885STrond Myklebust if (err < 0) 3451da177e4SLinus Torvalds return err; 346f758c885STrond Myklebust if (pgio.pg_error < 0) 347c63c7b05STrond Myklebust return pgio.pg_error; 348c63c7b05STrond Myklebust return 0; 3491da177e4SLinus Torvalds } 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds /* 3521da177e4SLinus Torvalds * Insert a write request into an inode 3531da177e4SLinus Torvalds */ 3541da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3551da177e4SLinus Torvalds { 3561da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3571da177e4SLinus Torvalds int error; 3581da177e4SLinus Torvalds 3591da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 3601da177e4SLinus Torvalds BUG_ON(error == -EEXIST); 3611da177e4SLinus Torvalds if (error) 3621da177e4SLinus Torvalds return error; 3631da177e4SLinus Torvalds if (!nfsi->npages) { 3641da177e4SLinus Torvalds igrab(inode); 3651da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3661da177e4SLinus Torvalds nfsi->change_attr++; 3671da177e4SLinus Torvalds } 368deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 369277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3701da177e4SLinus Torvalds nfsi->npages++; 371c03b4024STrond Myklebust kref_get(&req->wb_kref); 372acee478aSTrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED); 3731da177e4SLinus Torvalds return 0; 3741da177e4SLinus Torvalds } 3751da177e4SLinus Torvalds 3761da177e4SLinus Torvalds /* 37789a09141SPeter Zijlstra * Remove a write request from an inode 3781da177e4SLinus Torvalds */ 3791da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 3801da177e4SLinus Torvalds { 38188be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 3821da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3831da177e4SLinus Torvalds 3841da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 3851da177e4SLinus Torvalds 386587142f8STrond Myklebust spin_lock(&inode->i_lock); 387277459d2STrond Myklebust set_page_private(req->wb_page, 0); 388deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 3891da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 3901da177e4SLinus Torvalds nfsi->npages--; 3911da177e4SLinus Torvalds if (!nfsi->npages) { 392587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3931da177e4SLinus Torvalds iput(inode); 3941da177e4SLinus Torvalds } else 395587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3961da177e4SLinus Torvalds nfs_clear_request(req); 3971da177e4SLinus Torvalds nfs_release_request(req); 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds 40061822ab5STrond Myklebust static void 40161822ab5STrond Myklebust nfs_redirty_request(struct nfs_page *req) 40261822ab5STrond Myklebust { 40361822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 40461822ab5STrond Myklebust } 40561822ab5STrond Myklebust 4061da177e4SLinus Torvalds /* 4071da177e4SLinus Torvalds * Check if a request is dirty 4081da177e4SLinus Torvalds */ 4091da177e4SLinus Torvalds static inline int 4101da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 4111da177e4SLinus Torvalds { 4125a6d41b3STrond Myklebust struct page *page = req->wb_page; 4135a6d41b3STrond Myklebust 414612c9384STrond Myklebust if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags)) 4155a6d41b3STrond Myklebust return 0; 4165a6d41b3STrond Myklebust return !PageWriteback(req->wb_page); 4171da177e4SLinus Torvalds } 4181da177e4SLinus Torvalds 4191da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4201da177e4SLinus Torvalds /* 4211da177e4SLinus Torvalds * Add a request to the inode's commit list. 4221da177e4SLinus Torvalds */ 4231da177e4SLinus Torvalds static void 4241da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4251da177e4SLinus Torvalds { 42688be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4271da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4281da177e4SLinus Torvalds 429587142f8STrond Myklebust spin_lock(&inode->i_lock); 4301da177e4SLinus Torvalds nfsi->ncommit++; 431612c9384STrond Myklebust set_bit(PG_NEED_COMMIT, &(req)->wb_flags); 4325c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4335c369683STrond Myklebust req->wb_index, 4345c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 435587142f8STrond Myklebust spin_unlock(&inode->i_lock); 436fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 437c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 438a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4391da177e4SLinus Torvalds } 4408e821cadSTrond Myklebust 4418e821cadSTrond Myklebust static inline 4428e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4438e821cadSTrond Myklebust { 4448e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4458e821cadSTrond Myklebust } 4468e821cadSTrond Myklebust 4478e821cadSTrond Myklebust static inline 4488e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4498e821cadSTrond Myklebust { 450612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4518e821cadSTrond Myklebust nfs_mark_request_commit(req); 4528e821cadSTrond Myklebust return 1; 4538e821cadSTrond Myklebust } 4548e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4558e821cadSTrond Myklebust nfs_redirty_request(req); 4568e821cadSTrond Myklebust return 1; 4578e821cadSTrond Myklebust } 4588e821cadSTrond Myklebust return 0; 4598e821cadSTrond Myklebust } 4608e821cadSTrond Myklebust #else 4618e821cadSTrond Myklebust static inline void 4628e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4638e821cadSTrond Myklebust { 4648e821cadSTrond Myklebust } 4658e821cadSTrond Myklebust 4668e821cadSTrond Myklebust static inline 4678e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4688e821cadSTrond Myklebust { 4698e821cadSTrond Myklebust return 0; 4708e821cadSTrond Myklebust } 4718e821cadSTrond Myklebust 4728e821cadSTrond Myklebust static inline 4738e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4748e821cadSTrond Myklebust { 4758e821cadSTrond Myklebust return 0; 4768e821cadSTrond Myklebust } 4771da177e4SLinus Torvalds #endif 4781da177e4SLinus Torvalds 4791da177e4SLinus Torvalds /* 4801da177e4SLinus Torvalds * Wait for a request to complete. 4811da177e4SLinus Torvalds * 482150030b7SMatthew Wilcox * Interruptible by fatal signals only. 4831da177e4SLinus Torvalds */ 484ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) 4851da177e4SLinus Torvalds { 4861da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4871da177e4SLinus Torvalds struct nfs_page *req; 488ca52fec1STrond Myklebust pgoff_t idx_end, next; 4891da177e4SLinus Torvalds unsigned int res = 0; 4901da177e4SLinus Torvalds int error; 4911da177e4SLinus Torvalds 4921da177e4SLinus Torvalds if (npages == 0) 4931da177e4SLinus Torvalds idx_end = ~0; 4941da177e4SLinus Torvalds else 4951da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds next = idx_start; 4989fd367f0STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { 4991da177e4SLinus Torvalds if (req->wb_index > idx_end) 5001da177e4SLinus Torvalds break; 5011da177e4SLinus Torvalds 5021da177e4SLinus Torvalds next = req->wb_index + 1; 503c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5041da177e4SLinus Torvalds 505c03b4024STrond Myklebust kref_get(&req->wb_kref); 506587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5071da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5081da177e4SLinus Torvalds nfs_release_request(req); 509587142f8STrond Myklebust spin_lock(&inode->i_lock); 5101da177e4SLinus Torvalds if (error < 0) 5111da177e4SLinus Torvalds return error; 5121da177e4SLinus Torvalds res++; 5131da177e4SLinus Torvalds } 5141da177e4SLinus Torvalds return res; 5151da177e4SLinus Torvalds } 5161da177e4SLinus Torvalds 51783715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 51883715ad5STrond Myklebust { 51983715ad5STrond Myklebust struct nfs_page *req; 52083715ad5STrond Myklebust 52183715ad5STrond Myklebust while(!list_empty(head)) { 52283715ad5STrond Myklebust req = nfs_list_entry(head->next); 523b6dff26aSTrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 524c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 525c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 52683715ad5STrond Myklebust nfs_list_remove_request(req); 527612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 52883715ad5STrond Myklebust nfs_inode_remove_request(req); 529b6dff26aSTrond Myklebust nfs_unlock_request(req); 53083715ad5STrond Myklebust } 53183715ad5STrond Myklebust } 53283715ad5STrond Myklebust 5331da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5341da177e4SLinus Torvalds /* 5351da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5361da177e4SLinus Torvalds * @inode: NFS inode to scan 5371da177e4SLinus Torvalds * @dst: destination list 5381da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5391da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5401da177e4SLinus Torvalds * 5411da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5421da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5431da177e4SLinus Torvalds */ 5441da177e4SLinus Torvalds static int 545ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5461da177e4SLinus Torvalds { 5471da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5483da28eb1STrond Myklebust int res = 0; 5493da28eb1STrond Myklebust 5503da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 5515c369683STrond Myklebust res = nfs_scan_list(nfsi, dst, idx_start, npages, 5525c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 5531da177e4SLinus Torvalds nfsi->ncommit -= res; 5543da28eb1STrond Myklebust } 5551da177e4SLinus Torvalds return res; 5561da177e4SLinus Torvalds } 557c42de9ddSTrond Myklebust #else 558ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 559c42de9ddSTrond Myklebust { 560c42de9ddSTrond Myklebust return 0; 561c42de9ddSTrond Myklebust } 5621da177e4SLinus Torvalds #endif 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds /* 5651da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 5661da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 5671da177e4SLinus Torvalds * the calling process. 5681da177e4SLinus Torvalds * 5691da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 5701da177e4SLinus Torvalds */ 5711da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 572e21195a7STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 5731da177e4SLinus Torvalds { 57489a09141SPeter Zijlstra struct address_space *mapping = page->mapping; 57589a09141SPeter Zijlstra struct inode *inode = mapping->host; 5761da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 577ca52fec1STrond Myklebust pgoff_t rqend, end; 5781da177e4SLinus Torvalds 5791da177e4SLinus Torvalds end = offset + bytes; 5801da177e4SLinus Torvalds 5811da177e4SLinus Torvalds for (;;) { 5821da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 5831da177e4SLinus Torvalds * A request for the page we wish to update 5841da177e4SLinus Torvalds */ 585587142f8STrond Myklebust spin_lock(&inode->i_lock); 586277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 5871da177e4SLinus Torvalds if (req) { 588acee478aSTrond Myklebust if (!nfs_set_page_tag_locked(req)) { 5891da177e4SLinus Torvalds int error; 590277459d2STrond Myklebust 591587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5921da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5931da177e4SLinus Torvalds nfs_release_request(req); 5941dd594b2SNeil Brown if (error < 0) { 5951dd594b2SNeil Brown if (new) 5961dd594b2SNeil Brown nfs_release_request(new); 5971da177e4SLinus Torvalds return ERR_PTR(error); 5981dd594b2SNeil Brown } 5991da177e4SLinus Torvalds continue; 6001da177e4SLinus Torvalds } 601587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6021da177e4SLinus Torvalds if (new) 6031da177e4SLinus Torvalds nfs_release_request(new); 6041da177e4SLinus Torvalds break; 6051da177e4SLinus Torvalds } 6061da177e4SLinus Torvalds 6071da177e4SLinus Torvalds if (new) { 6081da177e4SLinus Torvalds int error; 6091da177e4SLinus Torvalds nfs_lock_request_dontget(new); 6101da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 6111da177e4SLinus Torvalds if (error) { 612587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6131da177e4SLinus Torvalds nfs_unlock_request(new); 6141da177e4SLinus Torvalds return ERR_PTR(error); 6151da177e4SLinus Torvalds } 616587142f8STrond Myklebust spin_unlock(&inode->i_lock); 61761e930a9STrond Myklebust req = new; 61861e930a9STrond Myklebust goto zero_page; 6191da177e4SLinus Torvalds } 620587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6211da177e4SLinus Torvalds 6221da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 6231da177e4SLinus Torvalds if (IS_ERR(new)) 6241da177e4SLinus Torvalds return new; 6251da177e4SLinus Torvalds } 6261da177e4SLinus Torvalds 6271da177e4SLinus Torvalds /* We have a request for our page. 6281da177e4SLinus Torvalds * If the creds don't match, or the 6291da177e4SLinus Torvalds * page addresses don't match, 6301da177e4SLinus Torvalds * tell the caller to wait on the conflicting 6311da177e4SLinus Torvalds * request. 6321da177e4SLinus Torvalds */ 6331da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 6341da177e4SLinus Torvalds if (req->wb_context != ctx 6351da177e4SLinus Torvalds || req->wb_page != page 6361da177e4SLinus Torvalds || !nfs_dirty_request(req) 6371da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 638acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 6391da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 6401da177e4SLinus Torvalds } 6411da177e4SLinus Torvalds 6421da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6431da177e4SLinus Torvalds if (offset < req->wb_offset) { 6441da177e4SLinus Torvalds req->wb_offset = offset; 6451da177e4SLinus Torvalds req->wb_pgbase = offset; 64661e930a9STrond Myklebust req->wb_bytes = max(end, rqend) - req->wb_offset; 64761e930a9STrond Myklebust goto zero_page; 6481da177e4SLinus Torvalds } 6491da177e4SLinus Torvalds 6501da177e4SLinus Torvalds if (end > rqend) 6511da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 6521da177e4SLinus Torvalds 6531da177e4SLinus Torvalds return req; 65461e930a9STrond Myklebust zero_page: 65561e930a9STrond Myklebust /* If this page might potentially be marked as up to date, 65661e930a9STrond Myklebust * then we need to zero any uninitalised data. */ 65761e930a9STrond Myklebust if (req->wb_pgbase == 0 && req->wb_bytes != PAGE_CACHE_SIZE 65861e930a9STrond Myklebust && !PageUptodate(req->wb_page)) 659eebd2aa3SChristoph Lameter zero_user_segment(req->wb_page, req->wb_bytes, PAGE_CACHE_SIZE); 66061e930a9STrond Myklebust return req; 6611da177e4SLinus Torvalds } 6621da177e4SLinus Torvalds 6631da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6641da177e4SLinus Torvalds { 665cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 6661da177e4SLinus Torvalds struct nfs_page *req; 6671a54533eSTrond Myklebust int do_flush, status; 6681da177e4SLinus Torvalds /* 6691da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6701da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6711da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 6721da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 6731da177e4SLinus Torvalds * Also do the same if we find a request from an existing 6741da177e4SLinus Torvalds * dropped page. 6751da177e4SLinus Torvalds */ 6761a54533eSTrond Myklebust do { 677277459d2STrond Myklebust req = nfs_page_find_request(page); 6781a54533eSTrond Myklebust if (req == NULL) 6791a54533eSTrond Myklebust return 0; 6801a54533eSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx 681e261f51fSTrond Myklebust || !nfs_dirty_request(req); 6821da177e4SLinus Torvalds nfs_release_request(req); 6831a54533eSTrond Myklebust if (!do_flush) 6841a54533eSTrond Myklebust return 0; 685277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 6861a54533eSTrond Myklebust } while (status == 0); 6871a54533eSTrond Myklebust return status; 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds /* 6915d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 6925d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 6935d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 6945d47a356STrond Myklebust */ 6955d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 6965d47a356STrond Myklebust { 6975d47a356STrond Myklebust return PageUptodate(page) && 6985d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 6995d47a356STrond Myklebust } 7005d47a356STrond Myklebust 7015d47a356STrond Myklebust /* 7021da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7031da177e4SLinus Torvalds * 7041da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7051da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7061da177e4SLinus Torvalds */ 7071da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7081da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7091da177e4SLinus Torvalds { 710cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7111da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7121da177e4SLinus Torvalds int status = 0; 7131da177e4SLinus Torvalds 71491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 71591d5b470SChuck Lever 7161da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 71701cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 71801cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7190bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 7201da177e4SLinus Torvalds 7211da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7225d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7235d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7245d47a356STrond Myklebust * inefficiencies. 7251da177e4SLinus Torvalds */ 7265d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7275d47a356STrond Myklebust inode->i_flock == NULL && 7284b5621f6STrond Myklebust !(file->f_flags & O_SYNC)) { 72949a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7301da177e4SLinus Torvalds offset = 0; 7311da177e4SLinus Torvalds } 7321da177e4SLinus Torvalds 733e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 734e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7351da177e4SLinus Torvalds 7361da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 7371da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7381da177e4SLinus Torvalds if (status < 0) 739a301b777STrond Myklebust nfs_set_pageerror(page); 7401da177e4SLinus Torvalds return status; 7411da177e4SLinus Torvalds } 7421da177e4SLinus Torvalds 7431da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7441da177e4SLinus Torvalds { 7458e821cadSTrond Myklebust 74644dd151dSTrond Myklebust if (PageError(req->wb_page)) { 74744dd151dSTrond Myklebust nfs_end_page_writeback(req->wb_page); 74844dd151dSTrond Myklebust nfs_inode_remove_request(req); 74944dd151dSTrond Myklebust } else if (!nfs_reschedule_unstable_write(req)) { 75044dd151dSTrond Myklebust /* Set the PG_uptodate flag */ 75144dd151dSTrond Myklebust nfs_mark_uptodate(req->wb_page, req->wb_pgbase, req->wb_bytes); 75289a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7531da177e4SLinus Torvalds nfs_inode_remove_request(req); 7548e821cadSTrond Myklebust } else 7558e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 7569fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 7571da177e4SLinus Torvalds } 7581da177e4SLinus Torvalds 7593ff7576dSTrond Myklebust static int flush_task_priority(int how) 7601da177e4SLinus Torvalds { 7611da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7621da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7631da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7641da177e4SLinus Torvalds case FLUSH_LOWPRI: 7651da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7661da177e4SLinus Torvalds } 7671da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7681da177e4SLinus Torvalds } 7691da177e4SLinus Torvalds 7701da177e4SLinus Torvalds /* 7711da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 7721da177e4SLinus Torvalds */ 7731da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 7741da177e4SLinus Torvalds struct nfs_write_data *data, 775788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7761da177e4SLinus Torvalds unsigned int count, unsigned int offset, 7771da177e4SLinus Torvalds int how) 7781da177e4SLinus Torvalds { 77984115e1cSTrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 78084115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 7813ff7576dSTrond Myklebust int priority = flush_task_priority(how); 78207737691STrond Myklebust struct rpc_task *task; 783bdc7f021STrond Myklebust struct rpc_message msg = { 784bdc7f021STrond Myklebust .rpc_argp = &data->args, 785bdc7f021STrond Myklebust .rpc_resp = &data->res, 786bdc7f021STrond Myklebust .rpc_cred = req->wb_context->cred, 787bdc7f021STrond Myklebust }; 78884115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 78984115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 79007737691STrond Myklebust .task = &data->task, 791bdc7f021STrond Myklebust .rpc_message = &msg, 79284115e1cSTrond Myklebust .callback_ops = call_ops, 79384115e1cSTrond Myklebust .callback_data = data, 794101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 79584115e1cSTrond Myklebust .flags = flags, 7963ff7576dSTrond Myklebust .priority = priority, 79784115e1cSTrond Myklebust }; 7981da177e4SLinus Torvalds 7991da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8001da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8011da177e4SLinus Torvalds 8021da177e4SLinus Torvalds data->req = req; 80388be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 804bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 8051da177e4SLinus Torvalds 8061da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8071da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8081da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8091da177e4SLinus Torvalds data->args.pages = data->pagevec; 8101da177e4SLinus Torvalds data->args.count = count; 811383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 812bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 813bdc7f021STrond Myklebust if (how & FLUSH_STABLE) { 814bdc7f021STrond Myklebust data->args.stable = NFS_DATA_SYNC; 815bdc7f021STrond Myklebust if (!NFS_I(inode)->ncommit) 816bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 817bdc7f021STrond Myklebust } 8181da177e4SLinus Torvalds 8191da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8201da177e4SLinus Torvalds data->res.count = count; 8211da177e4SLinus Torvalds data->res.verf = &data->verf; 8220e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8231da177e4SLinus Torvalds 824788e7a89STrond Myklebust /* Set up the initial task struct. */ 825bdc7f021STrond Myklebust NFS_PROTO(inode)->write_setup(data, &msg); 8261da177e4SLinus Torvalds 827a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 828a3f565b1SChuck Lever "(req %s/%Ld, %u bytes @ offset %Lu)\n", 8290bbacc40SChuck Lever data->task.tk_pid, 8301da177e4SLinus Torvalds inode->i_sb->s_id, 8311da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8321da177e4SLinus Torvalds count, 8331da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8341da177e4SLinus Torvalds 83507737691STrond Myklebust task = rpc_run_task(&task_setup_data); 83607737691STrond Myklebust if (!IS_ERR(task)) 83707737691STrond Myklebust rpc_put_task(task); 8381da177e4SLinus Torvalds } 8391da177e4SLinus Torvalds 8401da177e4SLinus Torvalds /* 8411da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8421da177e4SLinus Torvalds * contiguous dirty area on one page. 8431da177e4SLinus Torvalds */ 8448d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8451da177e4SLinus Torvalds { 8461da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 8471da177e4SLinus Torvalds struct page *page = req->wb_page; 8481da177e4SLinus Torvalds struct nfs_write_data *data; 849e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 850e9f7bee1STrond Myklebust unsigned int offset; 8511da177e4SLinus Torvalds int requests = 0; 8521da177e4SLinus Torvalds LIST_HEAD(list); 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds nfs_list_remove_request(req); 8551da177e4SLinus Torvalds 856bcb71bbaSTrond Myklebust nbytes = count; 857e9f7bee1STrond Myklebust do { 858e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 859e9f7bee1STrond Myklebust 8608d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 8611da177e4SLinus Torvalds if (!data) 8621da177e4SLinus Torvalds goto out_bad; 8631da177e4SLinus Torvalds list_add(&data->pages, &list); 8641da177e4SLinus Torvalds requests++; 865e9f7bee1STrond Myklebust nbytes -= len; 866e9f7bee1STrond Myklebust } while (nbytes != 0); 8671da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 8681da177e4SLinus Torvalds 8691da177e4SLinus Torvalds ClearPageError(page); 8701da177e4SLinus Torvalds offset = 0; 871bcb71bbaSTrond Myklebust nbytes = count; 8721da177e4SLinus Torvalds do { 8731da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8741da177e4SLinus Torvalds list_del_init(&data->pages); 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds data->pagevec[0] = page; 8771da177e4SLinus Torvalds 878bcb71bbaSTrond Myklebust if (nbytes < wsize) 879bcb71bbaSTrond Myklebust wsize = nbytes; 880788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 881788e7a89STrond Myklebust wsize, offset, how); 8821da177e4SLinus Torvalds offset += wsize; 8831da177e4SLinus Torvalds nbytes -= wsize; 8841da177e4SLinus Torvalds } while (nbytes != 0); 8851da177e4SLinus Torvalds 8861da177e4SLinus Torvalds return 0; 8871da177e4SLinus Torvalds 8881da177e4SLinus Torvalds out_bad: 8891da177e4SLinus Torvalds while (!list_empty(&list)) { 8901da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8911da177e4SLinus Torvalds list_del(&data->pages); 8928aca67f0STrond Myklebust nfs_writedata_release(data); 8931da177e4SLinus Torvalds } 89461822ab5STrond Myklebust nfs_redirty_request(req); 8956d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 8969fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 8971da177e4SLinus Torvalds return -ENOMEM; 8981da177e4SLinus Torvalds } 8991da177e4SLinus Torvalds 9001da177e4SLinus Torvalds /* 9011da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9021da177e4SLinus Torvalds * The page must have been locked by the caller. 9031da177e4SLinus Torvalds * 9041da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9051da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9061da177e4SLinus Torvalds * that has been written but not committed. 9071da177e4SLinus Torvalds */ 9088d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9091da177e4SLinus Torvalds { 9101da177e4SLinus Torvalds struct nfs_page *req; 9111da177e4SLinus Torvalds struct page **pages; 9121da177e4SLinus Torvalds struct nfs_write_data *data; 9131da177e4SLinus Torvalds 9148d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 9151da177e4SLinus Torvalds if (!data) 9161da177e4SLinus Torvalds goto out_bad; 9171da177e4SLinus Torvalds 9181da177e4SLinus Torvalds pages = data->pagevec; 9191da177e4SLinus Torvalds while (!list_empty(head)) { 9201da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9211da177e4SLinus Torvalds nfs_list_remove_request(req); 9221da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9231da177e4SLinus Torvalds ClearPageError(req->wb_page); 9241da177e4SLinus Torvalds *pages++ = req->wb_page; 9251da177e4SLinus Torvalds } 9261da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds /* Set up the argument struct */ 929788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9301da177e4SLinus Torvalds 9311da177e4SLinus Torvalds return 0; 9321da177e4SLinus Torvalds out_bad: 9331da177e4SLinus Torvalds while (!list_empty(head)) { 93410afec90STrond Myklebust req = nfs_list_entry(head->next); 9351da177e4SLinus Torvalds nfs_list_remove_request(req); 93661822ab5STrond Myklebust nfs_redirty_request(req); 9376d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 9389fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 9391da177e4SLinus Torvalds } 9401da177e4SLinus Torvalds return -ENOMEM; 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds 943c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 944c63c7b05STrond Myklebust struct inode *inode, int ioflags) 9451da177e4SLinus Torvalds { 946bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 9471da177e4SLinus Torvalds 948bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 949c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 950bcb71bbaSTrond Myklebust else 951c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 9521da177e4SLinus Torvalds } 9531da177e4SLinus Torvalds 9541da177e4SLinus Torvalds /* 9551da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 9561da177e4SLinus Torvalds */ 957788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 9581da177e4SLinus Torvalds { 959788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 9601da177e4SLinus Torvalds struct nfs_page *req = data->req; 9611da177e4SLinus Torvalds struct page *page = req->wb_page; 9621da177e4SLinus Torvalds 9631da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 96488be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 96588be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 9661da177e4SLinus Torvalds req->wb_bytes, 9671da177e4SLinus Torvalds (long long)req_offset(req)); 9681da177e4SLinus Torvalds 969788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 970788e7a89STrond Myklebust return; 971788e7a89STrond Myklebust 972788e7a89STrond Myklebust if (task->tk_status < 0) { 973a301b777STrond Myklebust nfs_set_pageerror(page); 9747b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 975788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 9768e821cadSTrond Myklebust goto out; 9778e821cadSTrond Myklebust } 9788e821cadSTrond Myklebust 9798e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 980587142f8STrond Myklebust struct inode *inode = page->mapping->host; 9818e821cadSTrond Myklebust 982587142f8STrond Myklebust spin_lock(&inode->i_lock); 9838e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 9848e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 9858e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 9861da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 9871da177e4SLinus Torvalds dprintk(" defer commit\n"); 9881da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 9898e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 9908e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 9911da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 9921da177e4SLinus Torvalds } 993587142f8STrond Myklebust spin_unlock(&inode->i_lock); 9941da177e4SLinus Torvalds } else 9951da177e4SLinus Torvalds dprintk(" OK\n"); 9961da177e4SLinus Torvalds 9978e821cadSTrond Myklebust out: 9981da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 9991da177e4SLinus Torvalds nfs_writepage_release(req); 10001da177e4SLinus Torvalds } 10011da177e4SLinus Torvalds 1002788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1003788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1004788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1005788e7a89STrond Myklebust }; 1006788e7a89STrond Myklebust 10071da177e4SLinus Torvalds /* 10081da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10091da177e4SLinus Torvalds * 10101da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10111da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10121da177e4SLinus Torvalds * as the page has a write request pending. 10131da177e4SLinus Torvalds */ 1014788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10151da177e4SLinus Torvalds { 1016788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10171da177e4SLinus Torvalds struct nfs_page *req; 10181da177e4SLinus Torvalds struct page *page; 10191da177e4SLinus Torvalds 1020788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1021788e7a89STrond Myklebust return; 1022788e7a89STrond Myklebust 10231da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 10241da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 10251da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10261da177e4SLinus Torvalds nfs_list_remove_request(req); 10271da177e4SLinus Torvalds page = req->wb_page; 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 103088be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 103188be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 10321da177e4SLinus Torvalds req->wb_bytes, 10331da177e4SLinus Torvalds (long long)req_offset(req)); 10341da177e4SLinus Torvalds 1035788e7a89STrond Myklebust if (task->tk_status < 0) { 1036a301b777STrond Myklebust nfs_set_pageerror(page); 10377b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 1038788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 10398e821cadSTrond Myklebust goto remove_request; 10401da177e4SLinus Torvalds } 10411da177e4SLinus Torvalds 10428e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 10431da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10441da177e4SLinus Torvalds nfs_mark_request_commit(req); 10458e821cadSTrond Myklebust nfs_end_page_writeback(page); 10461da177e4SLinus Torvalds dprintk(" marked for commit\n"); 10478e821cadSTrond Myklebust goto next; 10488e821cadSTrond Myklebust } 104944dd151dSTrond Myklebust /* Set the PG_uptodate flag? */ 105044dd151dSTrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 10518e821cadSTrond Myklebust dprintk(" OK\n"); 10528e821cadSTrond Myklebust remove_request: 10538e821cadSTrond Myklebust nfs_end_page_writeback(page); 10541da177e4SLinus Torvalds nfs_inode_remove_request(req); 10551da177e4SLinus Torvalds next: 10569fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 10571da177e4SLinus Torvalds } 10581da177e4SLinus Torvalds } 10591da177e4SLinus Torvalds 1060788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1061788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1062788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1063788e7a89STrond Myklebust }; 1064788e7a89STrond Myklebust 1065788e7a89STrond Myklebust 10661da177e4SLinus Torvalds /* 10671da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 10681da177e4SLinus Torvalds */ 1069462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 10701da177e4SLinus Torvalds { 10711da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 10721da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1073788e7a89STrond Myklebust int status; 10741da177e4SLinus Torvalds 1075a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 10761da177e4SLinus Torvalds task->tk_pid, task->tk_status); 10771da177e4SLinus Torvalds 1078f551e44fSChuck Lever /* 1079f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1080f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1081f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1082f551e44fSChuck Lever * another writer had changed the file, but some applications 1083f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1084f551e44fSChuck Lever */ 1085788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1086788e7a89STrond Myklebust if (status != 0) 1087788e7a89STrond Myklebust return status; 108891d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 108991d5b470SChuck Lever 10901da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 10911da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 10921da177e4SLinus Torvalds /* We tried a write call, but the server did not 10931da177e4SLinus Torvalds * commit data to stable storage even though we 10941da177e4SLinus Torvalds * requested it. 10951da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 10961da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 10971da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 10981da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 10991da177e4SLinus Torvalds */ 11001da177e4SLinus Torvalds static unsigned long complain; 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11031da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11041da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 110554ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 11061da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11071da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11081da177e4SLinus Torvalds } 11091da177e4SLinus Torvalds } 11101da177e4SLinus Torvalds #endif 11111da177e4SLinus Torvalds /* Is this a short write? */ 11121da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11131da177e4SLinus Torvalds static unsigned long complain; 11141da177e4SLinus Torvalds 111591d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 111691d5b470SChuck Lever 11171da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11181da177e4SLinus Torvalds if (resp->count != 0) { 11191da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11201da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11211da177e4SLinus Torvalds /* Resend from where the server left off */ 11221da177e4SLinus Torvalds argp->offset += resp->count; 11231da177e4SLinus Torvalds argp->pgbase += resp->count; 11241da177e4SLinus Torvalds argp->count -= resp->count; 11251da177e4SLinus Torvalds } else { 11261da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11271da177e4SLinus Torvalds * headaches in the case of a server crash. 11281da177e4SLinus Torvalds */ 11291da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11301da177e4SLinus Torvalds } 11311da177e4SLinus Torvalds rpc_restart_call(task); 1132788e7a89STrond Myklebust return -EAGAIN; 11331da177e4SLinus Torvalds } 11341da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11351da177e4SLinus Torvalds printk(KERN_WARNING 11361da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 11371da177e4SLinus Torvalds argp->count); 11381da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11391da177e4SLinus Torvalds } 11401da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 11411da177e4SLinus Torvalds task->tk_status = -EIO; 11421da177e4SLinus Torvalds } 1143788e7a89STrond Myklebust return 0; 11441da177e4SLinus Torvalds } 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds 11471da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1148383ba719STrond Myklebust void nfs_commit_release(void *data) 11491da177e4SLinus Torvalds { 1150383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1151383ba719STrond Myklebust 1152383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 11531da177e4SLinus Torvalds nfs_commit_free(wdata); 11541da177e4SLinus Torvalds } 11551da177e4SLinus Torvalds 11561da177e4SLinus Torvalds /* 11571da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 11581da177e4SLinus Torvalds */ 11591da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1160788e7a89STrond Myklebust struct nfs_write_data *data, 1161788e7a89STrond Myklebust int how) 11621da177e4SLinus Torvalds { 116384115e1cSTrond Myklebust struct nfs_page *first = nfs_list_entry(head->next); 116484115e1cSTrond Myklebust struct inode *inode = first->wb_context->path.dentry->d_inode; 116584115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 11663ff7576dSTrond Myklebust int priority = flush_task_priority(how); 116707737691STrond Myklebust struct rpc_task *task; 1168bdc7f021STrond Myklebust struct rpc_message msg = { 1169bdc7f021STrond Myklebust .rpc_argp = &data->args, 1170bdc7f021STrond Myklebust .rpc_resp = &data->res, 1171bdc7f021STrond Myklebust .rpc_cred = first->wb_context->cred, 1172bdc7f021STrond Myklebust }; 117384115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 117407737691STrond Myklebust .task = &data->task, 117584115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 1176bdc7f021STrond Myklebust .rpc_message = &msg, 117784115e1cSTrond Myklebust .callback_ops = &nfs_commit_ops, 117884115e1cSTrond Myklebust .callback_data = data, 1179101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 118084115e1cSTrond Myklebust .flags = flags, 11813ff7576dSTrond Myklebust .priority = priority, 118284115e1cSTrond Myklebust }; 11831da177e4SLinus Torvalds 11841da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 11851da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 11861da177e4SLinus Torvalds 11871da177e4SLinus Torvalds list_splice_init(head, &data->pages); 11881da177e4SLinus Torvalds 11891da177e4SLinus Torvalds data->inode = inode; 1190bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 11933da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 11943da28eb1STrond Myklebust data->args.offset = 0; 11953da28eb1STrond Myklebust data->args.count = 0; 1196383ba719STrond Myklebust data->args.context = get_nfs_open_context(first->wb_context); 11973da28eb1STrond Myklebust data->res.count = 0; 11981da177e4SLinus Torvalds data->res.fattr = &data->fattr; 11991da177e4SLinus Torvalds data->res.verf = &data->verf; 12000e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12011da177e4SLinus Torvalds 1202788e7a89STrond Myklebust /* Set up the initial task struct. */ 1203bdc7f021STrond Myklebust NFS_PROTO(inode)->commit_setup(data, &msg); 12041da177e4SLinus Torvalds 1205a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1206bdc7f021STrond Myklebust 120707737691STrond Myklebust task = rpc_run_task(&task_setup_data); 120807737691STrond Myklebust if (!IS_ERR(task)) 120907737691STrond Myklebust rpc_put_task(task); 12101da177e4SLinus Torvalds } 12111da177e4SLinus Torvalds 12121da177e4SLinus Torvalds /* 12131da177e4SLinus Torvalds * Commit dirty pages 12141da177e4SLinus Torvalds */ 12151da177e4SLinus Torvalds static int 121640859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 12171da177e4SLinus Torvalds { 12181da177e4SLinus Torvalds struct nfs_write_data *data; 12191da177e4SLinus Torvalds struct nfs_page *req; 12201da177e4SLinus Torvalds 1221e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds if (!data) 12241da177e4SLinus Torvalds goto out_bad; 12251da177e4SLinus Torvalds 12261da177e4SLinus Torvalds /* Set up the argument struct */ 12271da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds return 0; 12301da177e4SLinus Torvalds out_bad: 12311da177e4SLinus Torvalds while (!list_empty(head)) { 12321da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12331da177e4SLinus Torvalds nfs_list_remove_request(req); 12341da177e4SLinus Torvalds nfs_mark_request_commit(req); 123583715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1236c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1237c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12389fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12391da177e4SLinus Torvalds } 12401da177e4SLinus Torvalds return -ENOMEM; 12411da177e4SLinus Torvalds } 12421da177e4SLinus Torvalds 12431da177e4SLinus Torvalds /* 12441da177e4SLinus Torvalds * COMMIT call returned 12451da177e4SLinus Torvalds */ 1246788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 12471da177e4SLinus Torvalds { 1248963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12491da177e4SLinus Torvalds struct nfs_page *req; 12501da177e4SLinus Torvalds 1251a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 12521da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12531da177e4SLinus Torvalds 1254788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1255788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1256788e7a89STrond Myklebust return; 1257788e7a89STrond Myklebust 12581da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12591da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12601da177e4SLinus Torvalds nfs_list_remove_request(req); 1261612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 1262fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1263c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1264c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12651da177e4SLinus Torvalds 12661da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 126788be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 126888be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 12691da177e4SLinus Torvalds req->wb_bytes, 12701da177e4SLinus Torvalds (long long)req_offset(req)); 12711da177e4SLinus Torvalds if (task->tk_status < 0) { 12727b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 12731da177e4SLinus Torvalds nfs_inode_remove_request(req); 12741da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 12751da177e4SLinus Torvalds goto next; 12761da177e4SLinus Torvalds } 12771da177e4SLinus Torvalds 12781da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 12791da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 12801da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 12811da177e4SLinus Torvalds /* We have a match */ 128244dd151dSTrond Myklebust /* Set the PG_uptodate flag */ 128344dd151dSTrond Myklebust nfs_mark_uptodate(req->wb_page, req->wb_pgbase, 128444dd151dSTrond Myklebust req->wb_bytes); 12851da177e4SLinus Torvalds nfs_inode_remove_request(req); 12861da177e4SLinus Torvalds dprintk(" OK\n"); 12871da177e4SLinus Torvalds goto next; 12881da177e4SLinus Torvalds } 12891da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 12901da177e4SLinus Torvalds dprintk(" mismatch\n"); 129161822ab5STrond Myklebust nfs_redirty_request(req); 12921da177e4SLinus Torvalds next: 12939fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12941da177e4SLinus Torvalds } 12951da177e4SLinus Torvalds } 1296788e7a89STrond Myklebust 1297788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1298788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1299788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1300788e7a89STrond Myklebust }; 13011da177e4SLinus Torvalds 13023da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13031da177e4SLinus Torvalds { 13041da177e4SLinus Torvalds LIST_HEAD(head); 13057d46a49fSTrond Myklebust int res; 13061da177e4SLinus Torvalds 1307587142f8STrond Myklebust spin_lock(&inode->i_lock); 13083da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1309587142f8STrond Myklebust spin_unlock(&inode->i_lock); 13101da177e4SLinus Torvalds if (res) { 13117d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 13121da177e4SLinus Torvalds if (error < 0) 13131da177e4SLinus Torvalds return error; 13143da28eb1STrond Myklebust } 13151da177e4SLinus Torvalds return res; 13161da177e4SLinus Torvalds } 1317c63c7b05STrond Myklebust #else 1318c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1319c63c7b05STrond Myklebust { 1320c63c7b05STrond Myklebust return 0; 1321c63c7b05STrond Myklebust } 13221da177e4SLinus Torvalds #endif 13231da177e4SLinus Torvalds 13241c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 13251da177e4SLinus Torvalds { 13261c75950bSTrond Myklebust struct inode *inode = mapping->host; 1327ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 13281c75950bSTrond Myklebust unsigned int npages = 0; 1329c42de9ddSTrond Myklebust LIST_HEAD(head); 133070b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 13313f442547STrond Myklebust long pages, ret; 13321da177e4SLinus Torvalds 13331c75950bSTrond Myklebust /* FIXME */ 13341c75950bSTrond Myklebust if (wbc->range_cyclic) 13351c75950bSTrond Myklebust idx_start = 0; 13361c75950bSTrond Myklebust else { 13371c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 13381c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 13391c75950bSTrond Myklebust if (idx_end > idx_start) { 1340ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 13411c75950bSTrond Myklebust npages = l_npages; 13421c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1343ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 13441c75950bSTrond Myklebust npages = 0; 13451c75950bSTrond Myklebust } 13461c75950bSTrond Myklebust } 1347c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1348587142f8STrond Myklebust spin_lock(&inode->i_lock); 13491da177e4SLinus Torvalds do { 1350c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1351c42de9ddSTrond Myklebust if (ret != 0) 1352c42de9ddSTrond Myklebust continue; 1353c42de9ddSTrond Myklebust if (nocommit) 1354c42de9ddSTrond Myklebust break; 1355d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1356724c439cSTrond Myklebust if (pages == 0) 1357c42de9ddSTrond Myklebust break; 1358d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1359587142f8STrond Myklebust spin_unlock(&inode->i_lock); 136083715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1361e8e058e8STrond Myklebust ret = pages; 1362587142f8STrond Myklebust spin_lock(&inode->i_lock); 1363d2ccddf0STrond Myklebust continue; 1364d2ccddf0STrond Myklebust } 1365d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1366587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1367c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1368587142f8STrond Myklebust spin_lock(&inode->i_lock); 1369587142f8STrond Myklebust 1370c42de9ddSTrond Myklebust } while (ret >= 0); 1371587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1372c42de9ddSTrond Myklebust return ret; 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds 137534901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 13761c75950bSTrond Myklebust { 13771c75950bSTrond Myklebust int ret; 13781c75950bSTrond Myklebust 137934901f70STrond Myklebust ret = nfs_writepages(mapping, wbc); 138061822ab5STrond Myklebust if (ret < 0) 138161822ab5STrond Myklebust goto out; 138234901f70STrond Myklebust ret = nfs_sync_mapping_wait(mapping, wbc, how); 1383ed90ef51STrond Myklebust if (ret < 0) 1384ed90ef51STrond Myklebust goto out; 13851c75950bSTrond Myklebust return 0; 138661822ab5STrond Myklebust out: 1387e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 13881c75950bSTrond Myklebust return ret; 13891c75950bSTrond Myklebust } 13901c75950bSTrond Myklebust 139134901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ 139234901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how) 139334901f70STrond Myklebust { 139434901f70STrond Myklebust struct writeback_control wbc = { 139534901f70STrond Myklebust .bdi = mapping->backing_dev_info, 139634901f70STrond Myklebust .sync_mode = WB_SYNC_NONE, 139734901f70STrond Myklebust .nr_to_write = LONG_MAX, 139834901f70STrond Myklebust .for_writepages = 1, 139934901f70STrond Myklebust .range_cyclic = 1, 140034901f70STrond Myklebust }; 140134901f70STrond Myklebust int ret; 140234901f70STrond Myklebust 140334901f70STrond Myklebust ret = __nfs_write_mapping(mapping, &wbc, how); 140434901f70STrond Myklebust if (ret < 0) 140534901f70STrond Myklebust return ret; 140634901f70STrond Myklebust wbc.sync_mode = WB_SYNC_ALL; 140734901f70STrond Myklebust return __nfs_write_mapping(mapping, &wbc, how); 140834901f70STrond Myklebust } 140934901f70STrond Myklebust 1410ed90ef51STrond Myklebust /* 1411ed90ef51STrond Myklebust * flush the inode to disk. 1412ed90ef51STrond Myklebust */ 1413ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode) 14141c75950bSTrond Myklebust { 1415ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, 0); 1416ed90ef51STrond Myklebust } 14171c75950bSTrond Myklebust 1418ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode) 1419ed90ef51STrond Myklebust { 1420ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); 14211c75950bSTrond Myklebust } 14221c75950bSTrond Myklebust 14231b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 14241b3b4a1aSTrond Myklebust { 14251b3b4a1aSTrond Myklebust struct nfs_page *req; 14261b3b4a1aSTrond Myklebust loff_t range_start = page_offset(page); 14271b3b4a1aSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14281b3b4a1aSTrond Myklebust struct writeback_control wbc = { 14291b3b4a1aSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14301b3b4a1aSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14311b3b4a1aSTrond Myklebust .nr_to_write = LONG_MAX, 14321b3b4a1aSTrond Myklebust .range_start = range_start, 14331b3b4a1aSTrond Myklebust .range_end = range_end, 14341b3b4a1aSTrond Myklebust }; 14351b3b4a1aSTrond Myklebust int ret = 0; 14361b3b4a1aSTrond Myklebust 14371b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 14381b3b4a1aSTrond Myklebust for (;;) { 14391b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 14401b3b4a1aSTrond Myklebust if (req == NULL) 14411b3b4a1aSTrond Myklebust goto out; 14421b3b4a1aSTrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 14431b3b4a1aSTrond Myklebust nfs_release_request(req); 14441b3b4a1aSTrond Myklebust break; 14451b3b4a1aSTrond Myklebust } 14461b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 14471b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 14481b3b4a1aSTrond Myklebust /* 14491b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 14501b3b4a1aSTrond Myklebust * page as being dirty 14511b3b4a1aSTrond Myklebust */ 14521b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 14531b3b4a1aSTrond Myklebust nfs_unlock_request(req); 14541b3b4a1aSTrond Myklebust break; 14551b3b4a1aSTrond Myklebust } 14561b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 14571b3b4a1aSTrond Myklebust if (ret < 0) 14581b3b4a1aSTrond Myklebust goto out; 14591b3b4a1aSTrond Myklebust } 14601b3b4a1aSTrond Myklebust if (!PagePrivate(page)) 14611b3b4a1aSTrond Myklebust return 0; 14621b3b4a1aSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); 14631b3b4a1aSTrond Myklebust out: 14641b3b4a1aSTrond Myklebust return ret; 14651b3b4a1aSTrond Myklebust } 14661b3b4a1aSTrond Myklebust 14675334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page, 14685334eb13SAdrian Bunk int how) 14691c75950bSTrond Myklebust { 14701c75950bSTrond Myklebust loff_t range_start = page_offset(page); 14711c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14724d770ccfSTrond Myklebust struct writeback_control wbc = { 14734d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14744d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14754d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 14764d770ccfSTrond Myklebust .range_start = range_start, 14774d770ccfSTrond Myklebust .range_end = range_end, 14784d770ccfSTrond Myklebust }; 14794d770ccfSTrond Myklebust int ret; 14801c75950bSTrond Myklebust 14814d770ccfSTrond Myklebust BUG_ON(!PageLocked(page)); 1482c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 14834d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 14844d770ccfSTrond Myklebust if (ret < 0) 14854d770ccfSTrond Myklebust goto out; 14864d770ccfSTrond Myklebust } 1487f40313acSTrond Myklebust if (!PagePrivate(page)) 1488f40313acSTrond Myklebust return 0; 14894d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 14904d770ccfSTrond Myklebust if (ret >= 0) 14914d770ccfSTrond Myklebust return 0; 14924d770ccfSTrond Myklebust out: 1493e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 14944d770ccfSTrond Myklebust return ret; 14951c75950bSTrond Myklebust } 14961c75950bSTrond Myklebust 14971c75950bSTrond Myklebust /* 14981c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 14991c75950bSTrond Myklebust */ 15001c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 15011c75950bSTrond Myklebust { 15024d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 15031c75950bSTrond Myklebust } 15041c75950bSTrond Myklebust 1505f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15061da177e4SLinus Torvalds { 15071da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15081da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15091da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 151020c2df83SPaul Mundt NULL); 15111da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15121da177e4SLinus Torvalds return -ENOMEM; 15131da177e4SLinus Torvalds 151493d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15151da177e4SLinus Torvalds nfs_wdata_cachep); 15161da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15171da177e4SLinus Torvalds return -ENOMEM; 15181da177e4SLinus Torvalds 151993d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15201da177e4SLinus Torvalds nfs_wdata_cachep); 15211da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15221da177e4SLinus Torvalds return -ENOMEM; 15231da177e4SLinus Torvalds 152489a09141SPeter Zijlstra /* 152589a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 152689a09141SPeter Zijlstra * 152789a09141SPeter Zijlstra * 64MB: 8192k 152889a09141SPeter Zijlstra * 128MB: 11585k 152989a09141SPeter Zijlstra * 256MB: 16384k 153089a09141SPeter Zijlstra * 512MB: 23170k 153189a09141SPeter Zijlstra * 1GB: 32768k 153289a09141SPeter Zijlstra * 2GB: 46340k 153389a09141SPeter Zijlstra * 4GB: 65536k 153489a09141SPeter Zijlstra * 8GB: 92681k 153589a09141SPeter Zijlstra * 16GB: 131072k 153689a09141SPeter Zijlstra * 153789a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 153889a09141SPeter Zijlstra * Limit the default to 256M 153989a09141SPeter Zijlstra */ 154089a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 154189a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 154289a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 154389a09141SPeter Zijlstra 15441da177e4SLinus Torvalds return 0; 15451da177e4SLinus Torvalds } 15461da177e4SLinus Torvalds 1547266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15481da177e4SLinus Torvalds { 15491da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15501da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15511a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15521da177e4SLinus Torvalds } 15531da177e4SLinus Torvalds 1554