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 6110afec90STrond Myklebust static void nfs_commit_rcu_free(struct rcu_head *head) 621da177e4SLinus Torvalds { 638aca67f0STrond Myklebust struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); 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 698aca67f0STrond Myklebust void nfs_commit_free(struct nfs_write_data *wdata) 708aca67f0STrond Myklebust { 718aca67f0STrond Myklebust call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free); 728aca67f0STrond Myklebust } 738aca67f0STrond Myklebust 748d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 753feb2d49STrond Myklebust { 76e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 773feb2d49STrond Myklebust 783feb2d49STrond Myklebust if (p) { 793feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 803feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 81e9f7bee1STrond Myklebust p->npages = pagecount; 820d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 830d0b5cb3SChuck Lever p->pagevec = p->page_array; 843feb2d49STrond Myklebust else { 850d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 860d0b5cb3SChuck Lever if (!p->pagevec) { 873feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 883feb2d49STrond Myklebust p = NULL; 893feb2d49STrond Myklebust } 903feb2d49STrond Myklebust } 913feb2d49STrond Myklebust } 923feb2d49STrond Myklebust return p; 933feb2d49STrond Myklebust } 943feb2d49STrond Myklebust 958aca67f0STrond Myklebust static void nfs_writedata_rcu_free(struct rcu_head *head) 963feb2d49STrond Myklebust { 978aca67f0STrond Myklebust struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); 983feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 993feb2d49STrond Myklebust kfree(p->pagevec); 1003feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 1013feb2d49STrond Myklebust } 1023feb2d49STrond Myklebust 1038aca67f0STrond Myklebust static void nfs_writedata_free(struct nfs_write_data *wdata) 1048aca67f0STrond Myklebust { 1058aca67f0STrond Myklebust call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free); 1068aca67f0STrond Myklebust } 1078aca67f0STrond Myklebust 108963d8fe5STrond Myklebust void nfs_writedata_release(void *wdata) 1091da177e4SLinus Torvalds { 1101da177e4SLinus Torvalds nfs_writedata_free(wdata); 1111da177e4SLinus Torvalds } 1121da177e4SLinus Torvalds 1137b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1147b159fc1STrond Myklebust { 1157b159fc1STrond Myklebust ctx->error = error; 1167b159fc1STrond Myklebust smp_wmb(); 1177b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1187b159fc1STrond Myklebust } 1197b159fc1STrond Myklebust 120277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 121277459d2STrond Myklebust { 122277459d2STrond Myklebust struct nfs_page *req = NULL; 123277459d2STrond Myklebust 124277459d2STrond Myklebust if (PagePrivate(page)) { 125277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 126277459d2STrond Myklebust if (req != NULL) 127c03b4024STrond Myklebust kref_get(&req->wb_kref); 128277459d2STrond Myklebust } 129277459d2STrond Myklebust return req; 130277459d2STrond Myklebust } 131277459d2STrond Myklebust 132277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 133277459d2STrond Myklebust { 134587142f8STrond Myklebust struct inode *inode = page->mapping->host; 135277459d2STrond Myklebust struct nfs_page *req = NULL; 136277459d2STrond Myklebust 137587142f8STrond Myklebust spin_lock(&inode->i_lock); 138277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 139587142f8STrond Myklebust spin_unlock(&inode->i_lock); 140277459d2STrond Myklebust return req; 141277459d2STrond Myklebust } 142277459d2STrond Myklebust 1431da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1441da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1451da177e4SLinus Torvalds { 1461da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1471da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 148ca52fec1STrond Myklebust pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1491da177e4SLinus Torvalds 1501da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1511da177e4SLinus Torvalds return; 1521da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1531da177e4SLinus Torvalds if (i_size >= end) 1541da177e4SLinus Torvalds return; 15591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1561da177e4SLinus Torvalds i_size_write(inode, end); 1571da177e4SLinus Torvalds } 1581da177e4SLinus Torvalds 159a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 160a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 161a301b777STrond Myklebust { 162a301b777STrond Myklebust SetPageError(page); 163a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 164a301b777STrond Myklebust } 165a301b777STrond Myklebust 1661da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1671da177e4SLinus Torvalds * covers the full page. 1681da177e4SLinus Torvalds */ 1691da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1701da177e4SLinus Torvalds { 1711da177e4SLinus Torvalds if (PageUptodate(page)) 1721da177e4SLinus Torvalds return; 1731da177e4SLinus Torvalds if (base != 0) 1741da177e4SLinus Torvalds return; 17549a70f27STrond Myklebust if (count != nfs_page_length(page)) 1761da177e4SLinus Torvalds return; 1771da177e4SLinus Torvalds SetPageUptodate(page); 1781da177e4SLinus Torvalds } 1791da177e4SLinus Torvalds 180e21195a7STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1811da177e4SLinus Torvalds unsigned int offset, unsigned int count) 1821da177e4SLinus Torvalds { 1831da177e4SLinus Torvalds struct nfs_page *req; 184e21195a7STrond Myklebust int ret; 1851da177e4SLinus Torvalds 186e21195a7STrond Myklebust for (;;) { 187e21195a7STrond Myklebust req = nfs_update_request(ctx, page, offset, count); 188e21195a7STrond Myklebust if (!IS_ERR(req)) 189e21195a7STrond Myklebust break; 190e21195a7STrond Myklebust ret = PTR_ERR(req); 191e21195a7STrond Myklebust if (ret != -EBUSY) 192e21195a7STrond Myklebust return ret; 193e21195a7STrond Myklebust ret = nfs_wb_page(page->mapping->host, page); 194e21195a7STrond Myklebust if (ret != 0) 195e21195a7STrond Myklebust return ret; 196e21195a7STrond Myklebust } 1971da177e4SLinus Torvalds /* Update file length */ 1981da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 199acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 200abd3e641STrond Myklebust return 0; 2011da177e4SLinus Torvalds } 2021da177e4SLinus Torvalds 2031da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2041da177e4SLinus Torvalds { 2051da177e4SLinus Torvalds if (wbc->for_reclaim) 206c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 2071da177e4SLinus Torvalds if (wbc->for_kupdate) 2081da177e4SLinus Torvalds return FLUSH_LOWPRI; 2091da177e4SLinus Torvalds return 0; 2101da177e4SLinus Torvalds } 2111da177e4SLinus Torvalds 2121da177e4SLinus Torvalds /* 21389a09141SPeter Zijlstra * NFS congestion control 21489a09141SPeter Zijlstra */ 21589a09141SPeter Zijlstra 21689a09141SPeter Zijlstra int nfs_congestion_kb; 21789a09141SPeter Zijlstra 21889a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 21989a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 22089a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 22189a09141SPeter Zijlstra 2225a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 22389a09141SPeter Zijlstra { 2245a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2255a6d41b3STrond Myklebust 2265a6d41b3STrond Myklebust if (!ret) { 22789a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 22889a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 22989a09141SPeter Zijlstra 230277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 23189a09141SPeter Zijlstra NFS_CONGESTION_ON_THRESH) 23289a09141SPeter Zijlstra set_bdi_congested(&nfss->backing_dev_info, WRITE); 23389a09141SPeter Zijlstra } 2345a6d41b3STrond Myklebust return ret; 23589a09141SPeter Zijlstra } 23689a09141SPeter Zijlstra 23789a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 23889a09141SPeter Zijlstra { 23989a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 24089a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 24189a09141SPeter Zijlstra 24289a09141SPeter Zijlstra end_page_writeback(page); 243c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 24489a09141SPeter Zijlstra clear_bdi_congested(&nfss->backing_dev_info, WRITE); 24589a09141SPeter Zijlstra } 24689a09141SPeter Zijlstra 24789a09141SPeter Zijlstra /* 248e261f51fSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 2499cccef95STrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 250e261f51fSTrond Myklebust */ 251c63c7b05STrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 252c63c7b05STrond Myklebust struct page *page) 253e261f51fSTrond Myklebust { 254587142f8STrond Myklebust struct inode *inode = page->mapping->host; 255e261f51fSTrond Myklebust struct nfs_page *req; 256e261f51fSTrond Myklebust int ret; 257e261f51fSTrond Myklebust 258587142f8STrond Myklebust spin_lock(&inode->i_lock); 259e261f51fSTrond Myklebust for(;;) { 260e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 261e261f51fSTrond Myklebust if (req == NULL) { 262587142f8STrond Myklebust spin_unlock(&inode->i_lock); 2639cccef95STrond Myklebust return 0; 264e261f51fSTrond Myklebust } 265acee478aSTrond Myklebust if (nfs_set_page_tag_locked(req)) 266e261f51fSTrond Myklebust break; 267e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 268acee478aSTrond Myklebust * then the call to nfs_set_page_tag_locked() will always 269e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 270e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 271e261f51fSTrond Myklebust */ 272587142f8STrond Myklebust spin_unlock(&inode->i_lock); 273e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 274e261f51fSTrond Myklebust nfs_release_request(req); 275e261f51fSTrond Myklebust if (ret != 0) 276e261f51fSTrond Myklebust return ret; 277587142f8STrond Myklebust spin_lock(&inode->i_lock); 278e261f51fSTrond Myklebust } 279612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 280612c9384STrond Myklebust /* This request is marked for commit */ 281587142f8STrond Myklebust spin_unlock(&inode->i_lock); 282acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 283c63c7b05STrond Myklebust nfs_pageio_complete(pgio); 2849cccef95STrond Myklebust return 0; 285612c9384STrond Myklebust } 286c63c7b05STrond Myklebust if (nfs_set_page_writeback(page) != 0) { 287587142f8STrond Myklebust spin_unlock(&inode->i_lock); 288c63c7b05STrond Myklebust BUG(); 289c63c7b05STrond Myklebust } 290587142f8STrond Myklebust spin_unlock(&inode->i_lock); 291c63c7b05STrond Myklebust nfs_pageio_add_request(pgio, req); 2929cccef95STrond Myklebust return 0; 293e261f51fSTrond Myklebust } 294e261f51fSTrond Myklebust 295f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 296f758c885STrond Myklebust { 297f758c885STrond Myklebust struct inode *inode = page->mapping->host; 298f758c885STrond Myklebust 299f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 300f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 301f758c885STrond Myklebust 302f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 303f758c885STrond Myklebust return nfs_page_async_flush(pgio, page); 304f758c885STrond Myklebust } 305f758c885STrond Myklebust 306e261f51fSTrond Myklebust /* 3071da177e4SLinus Torvalds * Write an mmapped page to the server. 3081da177e4SLinus Torvalds */ 3094d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3101da177e4SLinus Torvalds { 311f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 312e261f51fSTrond Myklebust int err; 3131da177e4SLinus Torvalds 314f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 315f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 316f758c885STrond Myklebust nfs_pageio_complete(&pgio); 317f758c885STrond Myklebust if (err < 0) 3184d770ccfSTrond Myklebust return err; 319f758c885STrond Myklebust if (pgio.pg_error < 0) 320f758c885STrond Myklebust return pgio.pg_error; 321f758c885STrond Myklebust return 0; 3224d770ccfSTrond Myklebust } 3234d770ccfSTrond Myklebust 3244d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3254d770ccfSTrond Myklebust { 326f758c885STrond Myklebust int ret; 3274d770ccfSTrond Myklebust 328f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3291da177e4SLinus Torvalds unlock_page(page); 330f758c885STrond Myklebust return ret; 331f758c885STrond Myklebust } 332f758c885STrond Myklebust 333f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 334f758c885STrond Myklebust { 335f758c885STrond Myklebust int ret; 336f758c885STrond Myklebust 337f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 338f758c885STrond Myklebust unlock_page(page); 339f758c885STrond Myklebust return ret; 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3431da177e4SLinus Torvalds { 3441da177e4SLinus Torvalds struct inode *inode = mapping->host; 345c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3461da177e4SLinus Torvalds int err; 3471da177e4SLinus Torvalds 34891d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 34991d5b470SChuck Lever 350c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 351f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 352c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 353f758c885STrond Myklebust if (err < 0) 3541da177e4SLinus Torvalds return err; 355f758c885STrond Myklebust if (pgio.pg_error < 0) 356c63c7b05STrond Myklebust return pgio.pg_error; 357c63c7b05STrond Myklebust return 0; 3581da177e4SLinus Torvalds } 3591da177e4SLinus Torvalds 3601da177e4SLinus Torvalds /* 3611da177e4SLinus Torvalds * Insert a write request into an inode 3621da177e4SLinus Torvalds */ 3631da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3641da177e4SLinus Torvalds { 3651da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3661da177e4SLinus Torvalds int error; 3671da177e4SLinus Torvalds 3681da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 3691da177e4SLinus Torvalds BUG_ON(error == -EEXIST); 3701da177e4SLinus Torvalds if (error) 3711da177e4SLinus Torvalds return error; 3721da177e4SLinus Torvalds if (!nfsi->npages) { 3731da177e4SLinus Torvalds igrab(inode); 3741da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3751da177e4SLinus Torvalds nfsi->change_attr++; 3761da177e4SLinus Torvalds } 377deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 378277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3791da177e4SLinus Torvalds nfsi->npages++; 380c03b4024STrond Myklebust kref_get(&req->wb_kref); 381acee478aSTrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED); 3821da177e4SLinus Torvalds return 0; 3831da177e4SLinus Torvalds } 3841da177e4SLinus Torvalds 3851da177e4SLinus Torvalds /* 38689a09141SPeter Zijlstra * Remove a write request from an inode 3871da177e4SLinus Torvalds */ 3881da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 3891da177e4SLinus Torvalds { 39088be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 3911da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3921da177e4SLinus Torvalds 3931da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 3941da177e4SLinus Torvalds 395587142f8STrond Myklebust spin_lock(&inode->i_lock); 396277459d2STrond Myklebust set_page_private(req->wb_page, 0); 397deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 3981da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 3991da177e4SLinus Torvalds nfsi->npages--; 4001da177e4SLinus Torvalds if (!nfsi->npages) { 401587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4021da177e4SLinus Torvalds iput(inode); 4031da177e4SLinus Torvalds } else 404587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4051da177e4SLinus Torvalds nfs_clear_request(req); 4061da177e4SLinus Torvalds nfs_release_request(req); 4071da177e4SLinus Torvalds } 4081da177e4SLinus Torvalds 40961822ab5STrond Myklebust static void 41061822ab5STrond Myklebust nfs_redirty_request(struct nfs_page *req) 41161822ab5STrond Myklebust { 41261822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 41361822ab5STrond Myklebust } 41461822ab5STrond Myklebust 4151da177e4SLinus Torvalds /* 4161da177e4SLinus Torvalds * Check if a request is dirty 4171da177e4SLinus Torvalds */ 4181da177e4SLinus Torvalds static inline int 4191da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 4201da177e4SLinus Torvalds { 4215a6d41b3STrond Myklebust struct page *page = req->wb_page; 4225a6d41b3STrond Myklebust 423612c9384STrond Myklebust if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags)) 4245a6d41b3STrond Myklebust return 0; 4255a6d41b3STrond Myklebust return !PageWriteback(req->wb_page); 4261da177e4SLinus Torvalds } 4271da177e4SLinus Torvalds 4281da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4291da177e4SLinus Torvalds /* 4301da177e4SLinus Torvalds * Add a request to the inode's commit list. 4311da177e4SLinus Torvalds */ 4321da177e4SLinus Torvalds static void 4331da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4341da177e4SLinus Torvalds { 43588be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4361da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4371da177e4SLinus Torvalds 438587142f8STrond Myklebust spin_lock(&inode->i_lock); 4391da177e4SLinus Torvalds nfsi->ncommit++; 440612c9384STrond Myklebust set_bit(PG_NEED_COMMIT, &(req)->wb_flags); 4415c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4425c369683STrond Myklebust req->wb_index, 4435c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 444587142f8STrond Myklebust spin_unlock(&inode->i_lock); 445fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 446c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 447a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4481da177e4SLinus Torvalds } 4498e821cadSTrond Myklebust 4508e821cadSTrond Myklebust static inline 4518e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4528e821cadSTrond Myklebust { 4538e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4548e821cadSTrond Myklebust } 4558e821cadSTrond Myklebust 4568e821cadSTrond Myklebust static inline 4578e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4588e821cadSTrond Myklebust { 459612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4608e821cadSTrond Myklebust nfs_mark_request_commit(req); 4618e821cadSTrond Myklebust return 1; 4628e821cadSTrond Myklebust } 4638e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4648e821cadSTrond Myklebust nfs_redirty_request(req); 4658e821cadSTrond Myklebust return 1; 4668e821cadSTrond Myklebust } 4678e821cadSTrond Myklebust return 0; 4688e821cadSTrond Myklebust } 4698e821cadSTrond Myklebust #else 4708e821cadSTrond Myklebust static inline void 4718e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4728e821cadSTrond Myklebust { 4738e821cadSTrond Myklebust } 4748e821cadSTrond Myklebust 4758e821cadSTrond Myklebust static inline 4768e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4778e821cadSTrond Myklebust { 4788e821cadSTrond Myklebust return 0; 4798e821cadSTrond Myklebust } 4808e821cadSTrond Myklebust 4818e821cadSTrond Myklebust static inline 4828e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4838e821cadSTrond Myklebust { 4848e821cadSTrond Myklebust return 0; 4858e821cadSTrond Myklebust } 4861da177e4SLinus Torvalds #endif 4871da177e4SLinus Torvalds 4881da177e4SLinus Torvalds /* 4891da177e4SLinus Torvalds * Wait for a request to complete. 4901da177e4SLinus Torvalds * 4911da177e4SLinus Torvalds * Interruptible by signals only if mounted with intr flag. 4921da177e4SLinus Torvalds */ 493ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) 4941da177e4SLinus Torvalds { 4951da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4961da177e4SLinus Torvalds struct nfs_page *req; 497ca52fec1STrond Myklebust pgoff_t idx_end, next; 4981da177e4SLinus Torvalds unsigned int res = 0; 4991da177e4SLinus Torvalds int error; 5001da177e4SLinus Torvalds 5011da177e4SLinus Torvalds if (npages == 0) 5021da177e4SLinus Torvalds idx_end = ~0; 5031da177e4SLinus Torvalds else 5041da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5051da177e4SLinus Torvalds 5061da177e4SLinus Torvalds next = idx_start; 5079fd367f0STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { 5081da177e4SLinus Torvalds if (req->wb_index > idx_end) 5091da177e4SLinus Torvalds break; 5101da177e4SLinus Torvalds 5111da177e4SLinus Torvalds next = req->wb_index + 1; 512c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5131da177e4SLinus Torvalds 514c03b4024STrond Myklebust kref_get(&req->wb_kref); 515587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5161da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5171da177e4SLinus Torvalds nfs_release_request(req); 518587142f8STrond Myklebust spin_lock(&inode->i_lock); 5191da177e4SLinus Torvalds if (error < 0) 5201da177e4SLinus Torvalds return error; 5211da177e4SLinus Torvalds res++; 5221da177e4SLinus Torvalds } 5231da177e4SLinus Torvalds return res; 5241da177e4SLinus Torvalds } 5251da177e4SLinus Torvalds 52683715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 52783715ad5STrond Myklebust { 52883715ad5STrond Myklebust struct nfs_page *req; 52983715ad5STrond Myklebust 53083715ad5STrond Myklebust while(!list_empty(head)) { 53183715ad5STrond Myklebust req = nfs_list_entry(head->next); 532b6dff26aSTrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 533c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 534c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 53583715ad5STrond Myklebust nfs_list_remove_request(req); 536612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 53783715ad5STrond Myklebust nfs_inode_remove_request(req); 538b6dff26aSTrond Myklebust nfs_unlock_request(req); 53983715ad5STrond Myklebust } 54083715ad5STrond Myklebust } 54183715ad5STrond Myklebust 5421da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5431da177e4SLinus Torvalds /* 5441da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5451da177e4SLinus Torvalds * @inode: NFS inode to scan 5461da177e4SLinus Torvalds * @dst: destination list 5471da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5481da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5491da177e4SLinus Torvalds * 5501da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5511da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5521da177e4SLinus Torvalds */ 5531da177e4SLinus Torvalds static int 554ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5551da177e4SLinus Torvalds { 5561da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5573da28eb1STrond Myklebust int res = 0; 5583da28eb1STrond Myklebust 5593da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 5605c369683STrond Myklebust res = nfs_scan_list(nfsi, dst, idx_start, npages, 5615c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 5621da177e4SLinus Torvalds nfsi->ncommit -= res; 5633da28eb1STrond Myklebust } 5641da177e4SLinus Torvalds return res; 5651da177e4SLinus Torvalds } 566c42de9ddSTrond Myklebust #else 567ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 568c42de9ddSTrond Myklebust { 569c42de9ddSTrond Myklebust return 0; 570c42de9ddSTrond Myklebust } 5711da177e4SLinus Torvalds #endif 5721da177e4SLinus Torvalds 5731da177e4SLinus Torvalds /* 5741da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 5751da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 5761da177e4SLinus Torvalds * the calling process. 5771da177e4SLinus Torvalds * 5781da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 5791da177e4SLinus Torvalds */ 5801da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 581e21195a7STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 5821da177e4SLinus Torvalds { 58389a09141SPeter Zijlstra struct address_space *mapping = page->mapping; 58489a09141SPeter Zijlstra struct inode *inode = mapping->host; 5851da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 586ca52fec1STrond Myklebust pgoff_t rqend, end; 5871da177e4SLinus Torvalds 5881da177e4SLinus Torvalds end = offset + bytes; 5891da177e4SLinus Torvalds 5901da177e4SLinus Torvalds for (;;) { 5911da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 5921da177e4SLinus Torvalds * A request for the page we wish to update 5931da177e4SLinus Torvalds */ 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) { 6041dd594b2SNeil Brown if (new) 6051dd594b2SNeil Brown nfs_release_request(new); 6061da177e4SLinus Torvalds return ERR_PTR(error); 6071dd594b2SNeil Brown } 6081da177e4SLinus Torvalds continue; 6091da177e4SLinus Torvalds } 610587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6111da177e4SLinus Torvalds if (new) 6121da177e4SLinus Torvalds nfs_release_request(new); 6131da177e4SLinus Torvalds break; 6141da177e4SLinus Torvalds } 6151da177e4SLinus Torvalds 6161da177e4SLinus Torvalds if (new) { 6171da177e4SLinus Torvalds int error; 6181da177e4SLinus Torvalds nfs_lock_request_dontget(new); 6191da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 6201da177e4SLinus Torvalds if (error) { 621587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6221da177e4SLinus Torvalds nfs_unlock_request(new); 6231da177e4SLinus Torvalds return ERR_PTR(error); 6241da177e4SLinus Torvalds } 625587142f8STrond Myklebust spin_unlock(&inode->i_lock); 62661e930a9STrond Myklebust req = new; 62761e930a9STrond Myklebust goto zero_page; 6281da177e4SLinus Torvalds } 629587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6301da177e4SLinus Torvalds 6311da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 6321da177e4SLinus Torvalds if (IS_ERR(new)) 6331da177e4SLinus Torvalds return new; 6341da177e4SLinus Torvalds } 6351da177e4SLinus Torvalds 6361da177e4SLinus Torvalds /* We have a request for our page. 6371da177e4SLinus Torvalds * If the creds don't match, or the 6381da177e4SLinus Torvalds * page addresses don't match, 6391da177e4SLinus Torvalds * tell the caller to wait on the conflicting 6401da177e4SLinus Torvalds * request. 6411da177e4SLinus Torvalds */ 6421da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 6431da177e4SLinus Torvalds if (req->wb_context != ctx 6441da177e4SLinus Torvalds || req->wb_page != page 6451da177e4SLinus Torvalds || !nfs_dirty_request(req) 6461da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 647acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 6481da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 6491da177e4SLinus Torvalds } 6501da177e4SLinus Torvalds 6511da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6521da177e4SLinus Torvalds if (offset < req->wb_offset) { 6531da177e4SLinus Torvalds req->wb_offset = offset; 6541da177e4SLinus Torvalds req->wb_pgbase = offset; 65561e930a9STrond Myklebust req->wb_bytes = max(end, rqend) - req->wb_offset; 65661e930a9STrond Myklebust goto zero_page; 6571da177e4SLinus Torvalds } 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds if (end > rqend) 6601da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds return req; 66361e930a9STrond Myklebust zero_page: 66461e930a9STrond Myklebust /* If this page might potentially be marked as up to date, 66561e930a9STrond Myklebust * then we need to zero any uninitalised data. */ 66661e930a9STrond Myklebust if (req->wb_pgbase == 0 && req->wb_bytes != PAGE_CACHE_SIZE 66761e930a9STrond Myklebust && !PageUptodate(req->wb_page)) 66861e930a9STrond Myklebust zero_user_page(req->wb_page, req->wb_bytes, 66961e930a9STrond Myklebust PAGE_CACHE_SIZE - req->wb_bytes, 67061e930a9STrond Myklebust KM_USER0); 67161e930a9STrond Myklebust return req; 6721da177e4SLinus Torvalds } 6731da177e4SLinus Torvalds 6741da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6751da177e4SLinus Torvalds { 676cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 6771da177e4SLinus Torvalds struct nfs_page *req; 6781a54533eSTrond Myklebust int do_flush, status; 6791da177e4SLinus Torvalds /* 6801da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6811da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6821da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 6831da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 6841da177e4SLinus Torvalds * Also do the same if we find a request from an existing 6851da177e4SLinus Torvalds * dropped page. 6861da177e4SLinus Torvalds */ 6871a54533eSTrond Myklebust do { 688277459d2STrond Myklebust req = nfs_page_find_request(page); 6891a54533eSTrond Myklebust if (req == NULL) 6901a54533eSTrond Myklebust return 0; 6911a54533eSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx 692e261f51fSTrond Myklebust || !nfs_dirty_request(req); 6931da177e4SLinus Torvalds nfs_release_request(req); 6941a54533eSTrond Myklebust if (!do_flush) 6951a54533eSTrond Myklebust return 0; 696277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 6971a54533eSTrond Myklebust } while (status == 0); 6981a54533eSTrond Myklebust return status; 6991da177e4SLinus Torvalds } 7001da177e4SLinus Torvalds 7011da177e4SLinus Torvalds /* 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 7221da177e4SLinus Torvalds * is entirely in cache, it may be more efficient to avoid 7231da177e4SLinus Torvalds * fragmenting write requests. 7241da177e4SLinus Torvalds */ 725ab0a3dbeSTrond Myklebust if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { 72649a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7271da177e4SLinus Torvalds offset = 0; 7281da177e4SLinus Torvalds } 7291da177e4SLinus Torvalds 730e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 731e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7321da177e4SLinus Torvalds 7331da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 7341da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7351da177e4SLinus Torvalds if (status < 0) 736a301b777STrond Myklebust nfs_set_pageerror(page); 7371da177e4SLinus Torvalds return status; 7381da177e4SLinus Torvalds } 7391da177e4SLinus Torvalds 7401da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7411da177e4SLinus Torvalds { 7428e821cadSTrond Myklebust 74344dd151dSTrond Myklebust if (PageError(req->wb_page)) { 74444dd151dSTrond Myklebust nfs_end_page_writeback(req->wb_page); 74544dd151dSTrond Myklebust nfs_inode_remove_request(req); 74644dd151dSTrond Myklebust } else if (!nfs_reschedule_unstable_write(req)) { 74744dd151dSTrond Myklebust /* Set the PG_uptodate flag */ 74844dd151dSTrond Myklebust nfs_mark_uptodate(req->wb_page, req->wb_pgbase, req->wb_bytes); 74989a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7501da177e4SLinus Torvalds nfs_inode_remove_request(req); 7518e821cadSTrond Myklebust } else 7528e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 7539fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 7541da177e4SLinus Torvalds } 7551da177e4SLinus Torvalds 7563ff7576dSTrond Myklebust static int flush_task_priority(int how) 7571da177e4SLinus Torvalds { 7581da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7591da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7601da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7611da177e4SLinus Torvalds case FLUSH_LOWPRI: 7621da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7631da177e4SLinus Torvalds } 7641da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7651da177e4SLinus Torvalds } 7661da177e4SLinus Torvalds 7671da177e4SLinus Torvalds /* 7681da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 7691da177e4SLinus Torvalds */ 7701da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 7711da177e4SLinus Torvalds struct nfs_write_data *data, 772788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7731da177e4SLinus Torvalds unsigned int count, unsigned int offset, 7741da177e4SLinus Torvalds int how) 7751da177e4SLinus Torvalds { 77684115e1cSTrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 77784115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 7783ff7576dSTrond Myklebust int priority = flush_task_priority(how); 77907737691STrond Myklebust struct rpc_task *task; 780bdc7f021STrond Myklebust struct rpc_message msg = { 781bdc7f021STrond Myklebust .rpc_argp = &data->args, 782bdc7f021STrond Myklebust .rpc_resp = &data->res, 783bdc7f021STrond Myklebust .rpc_cred = req->wb_context->cred, 784bdc7f021STrond Myklebust }; 78584115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 78684115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 78707737691STrond Myklebust .task = &data->task, 788bdc7f021STrond Myklebust .rpc_message = &msg, 78984115e1cSTrond Myklebust .callback_ops = call_ops, 79084115e1cSTrond Myklebust .callback_data = data, 79184115e1cSTrond Myklebust .flags = flags, 7923ff7576dSTrond Myklebust .priority = priority, 79384115e1cSTrond Myklebust }; 7941da177e4SLinus Torvalds 7951da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 7961da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 7971da177e4SLinus Torvalds 7981da177e4SLinus Torvalds data->req = req; 79988be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 800bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 8011da177e4SLinus Torvalds 8021da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8031da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8041da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8051da177e4SLinus Torvalds data->args.pages = data->pagevec; 8061da177e4SLinus Torvalds data->args.count = count; 8071da177e4SLinus Torvalds data->args.context = req->wb_context; 808bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 809bdc7f021STrond Myklebust if (how & FLUSH_STABLE) { 810bdc7f021STrond Myklebust data->args.stable = NFS_DATA_SYNC; 811bdc7f021STrond Myklebust if (!NFS_I(inode)->ncommit) 812bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 813bdc7f021STrond Myklebust } 8141da177e4SLinus Torvalds 8151da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8161da177e4SLinus Torvalds data->res.count = count; 8171da177e4SLinus Torvalds data->res.verf = &data->verf; 8180e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8191da177e4SLinus Torvalds 820788e7a89STrond Myklebust /* Set up the initial task struct. */ 821bdc7f021STrond Myklebust NFS_PROTO(inode)->write_setup(data, &msg); 8221da177e4SLinus Torvalds 823a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 824a3f565b1SChuck Lever "(req %s/%Ld, %u bytes @ offset %Lu)\n", 8250bbacc40SChuck Lever data->task.tk_pid, 8261da177e4SLinus Torvalds inode->i_sb->s_id, 8271da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8281da177e4SLinus Torvalds count, 8291da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8301da177e4SLinus Torvalds 83107737691STrond Myklebust task = rpc_run_task(&task_setup_data); 83207737691STrond Myklebust if (!IS_ERR(task)) 83307737691STrond Myklebust rpc_put_task(task); 8341da177e4SLinus Torvalds } 8351da177e4SLinus Torvalds 8361da177e4SLinus Torvalds /* 8371da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8381da177e4SLinus Torvalds * contiguous dirty area on one page. 8391da177e4SLinus Torvalds */ 8408d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8411da177e4SLinus Torvalds { 8421da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 8431da177e4SLinus Torvalds struct page *page = req->wb_page; 8441da177e4SLinus Torvalds struct nfs_write_data *data; 845e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 846e9f7bee1STrond Myklebust unsigned int offset; 8471da177e4SLinus Torvalds int requests = 0; 8481da177e4SLinus Torvalds LIST_HEAD(list); 8491da177e4SLinus Torvalds 8501da177e4SLinus Torvalds nfs_list_remove_request(req); 8511da177e4SLinus Torvalds 852bcb71bbaSTrond Myklebust nbytes = count; 853e9f7bee1STrond Myklebust do { 854e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 855e9f7bee1STrond Myklebust 8568d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 8571da177e4SLinus Torvalds if (!data) 8581da177e4SLinus Torvalds goto out_bad; 8591da177e4SLinus Torvalds list_add(&data->pages, &list); 8601da177e4SLinus Torvalds requests++; 861e9f7bee1STrond Myklebust nbytes -= len; 862e9f7bee1STrond Myklebust } while (nbytes != 0); 8631da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 8641da177e4SLinus Torvalds 8651da177e4SLinus Torvalds ClearPageError(page); 8661da177e4SLinus Torvalds offset = 0; 867bcb71bbaSTrond Myklebust nbytes = count; 8681da177e4SLinus Torvalds do { 8691da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8701da177e4SLinus Torvalds list_del_init(&data->pages); 8711da177e4SLinus Torvalds 8721da177e4SLinus Torvalds data->pagevec[0] = page; 8731da177e4SLinus Torvalds 874bcb71bbaSTrond Myklebust if (nbytes < wsize) 875bcb71bbaSTrond Myklebust wsize = nbytes; 876788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 877788e7a89STrond Myklebust wsize, offset, how); 8781da177e4SLinus Torvalds offset += wsize; 8791da177e4SLinus Torvalds nbytes -= wsize; 8801da177e4SLinus Torvalds } while (nbytes != 0); 8811da177e4SLinus Torvalds 8821da177e4SLinus Torvalds return 0; 8831da177e4SLinus Torvalds 8841da177e4SLinus Torvalds out_bad: 8851da177e4SLinus Torvalds while (!list_empty(&list)) { 8861da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8871da177e4SLinus Torvalds list_del(&data->pages); 8888aca67f0STrond Myklebust nfs_writedata_release(data); 8891da177e4SLinus Torvalds } 89061822ab5STrond Myklebust nfs_redirty_request(req); 8916d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 8929fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 8931da177e4SLinus Torvalds return -ENOMEM; 8941da177e4SLinus Torvalds } 8951da177e4SLinus Torvalds 8961da177e4SLinus Torvalds /* 8971da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 8981da177e4SLinus Torvalds * The page must have been locked by the caller. 8991da177e4SLinus Torvalds * 9001da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9011da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9021da177e4SLinus Torvalds * that has been written but not committed. 9031da177e4SLinus Torvalds */ 9048d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9051da177e4SLinus Torvalds { 9061da177e4SLinus Torvalds struct nfs_page *req; 9071da177e4SLinus Torvalds struct page **pages; 9081da177e4SLinus Torvalds struct nfs_write_data *data; 9091da177e4SLinus Torvalds 9108d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 9111da177e4SLinus Torvalds if (!data) 9121da177e4SLinus Torvalds goto out_bad; 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds pages = data->pagevec; 9151da177e4SLinus Torvalds while (!list_empty(head)) { 9161da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9171da177e4SLinus Torvalds nfs_list_remove_request(req); 9181da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9191da177e4SLinus Torvalds ClearPageError(req->wb_page); 9201da177e4SLinus Torvalds *pages++ = req->wb_page; 9211da177e4SLinus Torvalds } 9221da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9231da177e4SLinus Torvalds 9241da177e4SLinus Torvalds /* Set up the argument struct */ 925788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds return 0; 9281da177e4SLinus Torvalds out_bad: 9291da177e4SLinus Torvalds while (!list_empty(head)) { 93010afec90STrond Myklebust req = nfs_list_entry(head->next); 9311da177e4SLinus Torvalds nfs_list_remove_request(req); 93261822ab5STrond Myklebust nfs_redirty_request(req); 9336d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 9349fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 9351da177e4SLinus Torvalds } 9361da177e4SLinus Torvalds return -ENOMEM; 9371da177e4SLinus Torvalds } 9381da177e4SLinus Torvalds 939c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 940c63c7b05STrond Myklebust struct inode *inode, int ioflags) 9411da177e4SLinus Torvalds { 942*bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 9431da177e4SLinus Torvalds 944bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 945c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 946bcb71bbaSTrond Myklebust else 947c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 9481da177e4SLinus Torvalds } 9491da177e4SLinus Torvalds 9501da177e4SLinus Torvalds /* 9511da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 9521da177e4SLinus Torvalds */ 953788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 9541da177e4SLinus Torvalds { 955788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 9561da177e4SLinus Torvalds struct nfs_page *req = data->req; 9571da177e4SLinus Torvalds struct page *page = req->wb_page; 9581da177e4SLinus Torvalds 9591da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 96088be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 96188be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 9621da177e4SLinus Torvalds req->wb_bytes, 9631da177e4SLinus Torvalds (long long)req_offset(req)); 9641da177e4SLinus Torvalds 965788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 966788e7a89STrond Myklebust return; 967788e7a89STrond Myklebust 968788e7a89STrond Myklebust if (task->tk_status < 0) { 969a301b777STrond Myklebust nfs_set_pageerror(page); 9707b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 971788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 9728e821cadSTrond Myklebust goto out; 9738e821cadSTrond Myklebust } 9748e821cadSTrond Myklebust 9758e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 976587142f8STrond Myklebust struct inode *inode = page->mapping->host; 9778e821cadSTrond Myklebust 978587142f8STrond Myklebust spin_lock(&inode->i_lock); 9798e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 9808e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 9818e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 9821da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 9831da177e4SLinus Torvalds dprintk(" defer commit\n"); 9841da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 9858e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 9868e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 9871da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 9881da177e4SLinus Torvalds } 989587142f8STrond Myklebust spin_unlock(&inode->i_lock); 9901da177e4SLinus Torvalds } else 9911da177e4SLinus Torvalds dprintk(" OK\n"); 9921da177e4SLinus Torvalds 9938e821cadSTrond Myklebust out: 9941da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 9951da177e4SLinus Torvalds nfs_writepage_release(req); 9961da177e4SLinus Torvalds } 9971da177e4SLinus Torvalds 998788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 999788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1000788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1001788e7a89STrond Myklebust }; 1002788e7a89STrond Myklebust 10031da177e4SLinus Torvalds /* 10041da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10051da177e4SLinus Torvalds * 10061da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10071da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10081da177e4SLinus Torvalds * as the page has a write request pending. 10091da177e4SLinus Torvalds */ 1010788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10111da177e4SLinus Torvalds { 1012788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10131da177e4SLinus Torvalds struct nfs_page *req; 10141da177e4SLinus Torvalds struct page *page; 10151da177e4SLinus Torvalds 1016788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1017788e7a89STrond Myklebust return; 1018788e7a89STrond Myklebust 10191da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 10201da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 10211da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10221da177e4SLinus Torvalds nfs_list_remove_request(req); 10231da177e4SLinus Torvalds page = req->wb_page; 10241da177e4SLinus Torvalds 10251da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 102688be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 102788be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 10281da177e4SLinus Torvalds req->wb_bytes, 10291da177e4SLinus Torvalds (long long)req_offset(req)); 10301da177e4SLinus Torvalds 1031788e7a89STrond Myklebust if (task->tk_status < 0) { 1032a301b777STrond Myklebust nfs_set_pageerror(page); 10337b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 1034788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 10358e821cadSTrond Myklebust goto remove_request; 10361da177e4SLinus Torvalds } 10371da177e4SLinus Torvalds 10388e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 10391da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10401da177e4SLinus Torvalds nfs_mark_request_commit(req); 10418e821cadSTrond Myklebust nfs_end_page_writeback(page); 10421da177e4SLinus Torvalds dprintk(" marked for commit\n"); 10438e821cadSTrond Myklebust goto next; 10448e821cadSTrond Myklebust } 104544dd151dSTrond Myklebust /* Set the PG_uptodate flag? */ 104644dd151dSTrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 10478e821cadSTrond Myklebust dprintk(" OK\n"); 10488e821cadSTrond Myklebust remove_request: 10498e821cadSTrond Myklebust nfs_end_page_writeback(page); 10501da177e4SLinus Torvalds nfs_inode_remove_request(req); 10511da177e4SLinus Torvalds next: 10529fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 10531da177e4SLinus Torvalds } 10541da177e4SLinus Torvalds } 10551da177e4SLinus Torvalds 1056788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1057788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1058788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1059788e7a89STrond Myklebust }; 1060788e7a89STrond Myklebust 1061788e7a89STrond Myklebust 10621da177e4SLinus Torvalds /* 10631da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 10641da177e4SLinus Torvalds */ 1065462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 10661da177e4SLinus Torvalds { 10671da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 10681da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1069788e7a89STrond Myklebust int status; 10701da177e4SLinus Torvalds 1071a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 10721da177e4SLinus Torvalds task->tk_pid, task->tk_status); 10731da177e4SLinus Torvalds 1074f551e44fSChuck Lever /* 1075f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1076f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1077f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1078f551e44fSChuck Lever * another writer had changed the file, but some applications 1079f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1080f551e44fSChuck Lever */ 1081788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1082788e7a89STrond Myklebust if (status != 0) 1083788e7a89STrond Myklebust return status; 108491d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 108591d5b470SChuck Lever 10861da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 10871da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 10881da177e4SLinus Torvalds /* We tried a write call, but the server did not 10891da177e4SLinus Torvalds * commit data to stable storage even though we 10901da177e4SLinus Torvalds * requested it. 10911da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 10921da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 10931da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 10941da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 10951da177e4SLinus Torvalds */ 10961da177e4SLinus Torvalds static unsigned long complain; 10971da177e4SLinus Torvalds 10981da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 10991da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11001da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 110154ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 11021da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11031da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11041da177e4SLinus Torvalds } 11051da177e4SLinus Torvalds } 11061da177e4SLinus Torvalds #endif 11071da177e4SLinus Torvalds /* Is this a short write? */ 11081da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11091da177e4SLinus Torvalds static unsigned long complain; 11101da177e4SLinus Torvalds 111191d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 111291d5b470SChuck Lever 11131da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11141da177e4SLinus Torvalds if (resp->count != 0) { 11151da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11161da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11171da177e4SLinus Torvalds /* Resend from where the server left off */ 11181da177e4SLinus Torvalds argp->offset += resp->count; 11191da177e4SLinus Torvalds argp->pgbase += resp->count; 11201da177e4SLinus Torvalds argp->count -= resp->count; 11211da177e4SLinus Torvalds } else { 11221da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11231da177e4SLinus Torvalds * headaches in the case of a server crash. 11241da177e4SLinus Torvalds */ 11251da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11261da177e4SLinus Torvalds } 11271da177e4SLinus Torvalds rpc_restart_call(task); 1128788e7a89STrond Myklebust return -EAGAIN; 11291da177e4SLinus Torvalds } 11301da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11311da177e4SLinus Torvalds printk(KERN_WARNING 11321da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 11331da177e4SLinus Torvalds argp->count); 11341da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11351da177e4SLinus Torvalds } 11361da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 11371da177e4SLinus Torvalds task->tk_status = -EIO; 11381da177e4SLinus Torvalds } 1139788e7a89STrond Myklebust return 0; 11401da177e4SLinus Torvalds } 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds 11431da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1144963d8fe5STrond Myklebust void nfs_commit_release(void *wdata) 11451da177e4SLinus Torvalds { 11461da177e4SLinus Torvalds nfs_commit_free(wdata); 11471da177e4SLinus Torvalds } 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds /* 11501da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 11511da177e4SLinus Torvalds */ 11521da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1153788e7a89STrond Myklebust struct nfs_write_data *data, 1154788e7a89STrond Myklebust int how) 11551da177e4SLinus Torvalds { 115684115e1cSTrond Myklebust struct nfs_page *first = nfs_list_entry(head->next); 115784115e1cSTrond Myklebust struct inode *inode = first->wb_context->path.dentry->d_inode; 115884115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 11593ff7576dSTrond Myklebust int priority = flush_task_priority(how); 116007737691STrond Myklebust struct rpc_task *task; 1161bdc7f021STrond Myklebust struct rpc_message msg = { 1162bdc7f021STrond Myklebust .rpc_argp = &data->args, 1163bdc7f021STrond Myklebust .rpc_resp = &data->res, 1164bdc7f021STrond Myklebust .rpc_cred = first->wb_context->cred, 1165bdc7f021STrond Myklebust }; 116684115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 116707737691STrond Myklebust .task = &data->task, 116884115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 1169bdc7f021STrond Myklebust .rpc_message = &msg, 117084115e1cSTrond Myklebust .callback_ops = &nfs_commit_ops, 117184115e1cSTrond Myklebust .callback_data = data, 117284115e1cSTrond Myklebust .flags = flags, 11733ff7576dSTrond Myklebust .priority = priority, 117484115e1cSTrond Myklebust }; 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 11771da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 11781da177e4SLinus Torvalds 11791da177e4SLinus Torvalds list_splice_init(head, &data->pages); 11801da177e4SLinus Torvalds 11811da177e4SLinus Torvalds data->inode = inode; 1182bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 11831da177e4SLinus Torvalds 11841da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 11853da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 11863da28eb1STrond Myklebust data->args.offset = 0; 11873da28eb1STrond Myklebust data->args.count = 0; 11883da28eb1STrond Myklebust data->res.count = 0; 11891da177e4SLinus Torvalds data->res.fattr = &data->fattr; 11901da177e4SLinus Torvalds data->res.verf = &data->verf; 11910e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 11921da177e4SLinus Torvalds 1193788e7a89STrond Myklebust /* Set up the initial task struct. */ 1194bdc7f021STrond Myklebust NFS_PROTO(inode)->commit_setup(data, &msg); 11951da177e4SLinus Torvalds 1196a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1197bdc7f021STrond Myklebust 119807737691STrond Myklebust task = rpc_run_task(&task_setup_data); 119907737691STrond Myklebust if (!IS_ERR(task)) 120007737691STrond Myklebust rpc_put_task(task); 12011da177e4SLinus Torvalds } 12021da177e4SLinus Torvalds 12031da177e4SLinus Torvalds /* 12041da177e4SLinus Torvalds * Commit dirty pages 12051da177e4SLinus Torvalds */ 12061da177e4SLinus Torvalds static int 120740859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 12081da177e4SLinus Torvalds { 12091da177e4SLinus Torvalds struct nfs_write_data *data; 12101da177e4SLinus Torvalds struct nfs_page *req; 12111da177e4SLinus Torvalds 1212e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 12131da177e4SLinus Torvalds 12141da177e4SLinus Torvalds if (!data) 12151da177e4SLinus Torvalds goto out_bad; 12161da177e4SLinus Torvalds 12171da177e4SLinus Torvalds /* Set up the argument struct */ 12181da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds return 0; 12211da177e4SLinus Torvalds out_bad: 12221da177e4SLinus Torvalds while (!list_empty(head)) { 12231da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12241da177e4SLinus Torvalds nfs_list_remove_request(req); 12251da177e4SLinus Torvalds nfs_mark_request_commit(req); 122683715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1227c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1228c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12299fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12301da177e4SLinus Torvalds } 12311da177e4SLinus Torvalds return -ENOMEM; 12321da177e4SLinus Torvalds } 12331da177e4SLinus Torvalds 12341da177e4SLinus Torvalds /* 12351da177e4SLinus Torvalds * COMMIT call returned 12361da177e4SLinus Torvalds */ 1237788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 12381da177e4SLinus Torvalds { 1239963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12401da177e4SLinus Torvalds struct nfs_page *req; 12411da177e4SLinus Torvalds 1242a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 12431da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12441da177e4SLinus Torvalds 1245788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1246788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1247788e7a89STrond Myklebust return; 1248788e7a89STrond Myklebust 12491da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12501da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12511da177e4SLinus Torvalds nfs_list_remove_request(req); 1252612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 1253fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1254c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1255c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 125888be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 125988be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 12601da177e4SLinus Torvalds req->wb_bytes, 12611da177e4SLinus Torvalds (long long)req_offset(req)); 12621da177e4SLinus Torvalds if (task->tk_status < 0) { 12637b159fc1STrond Myklebust nfs_context_set_write_error(req->wb_context, task->tk_status); 12641da177e4SLinus Torvalds nfs_inode_remove_request(req); 12651da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 12661da177e4SLinus Torvalds goto next; 12671da177e4SLinus Torvalds } 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 12701da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 12711da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 12721da177e4SLinus Torvalds /* We have a match */ 127344dd151dSTrond Myklebust /* Set the PG_uptodate flag */ 127444dd151dSTrond Myklebust nfs_mark_uptodate(req->wb_page, req->wb_pgbase, 127544dd151dSTrond Myklebust req->wb_bytes); 12761da177e4SLinus Torvalds nfs_inode_remove_request(req); 12771da177e4SLinus Torvalds dprintk(" OK\n"); 12781da177e4SLinus Torvalds goto next; 12791da177e4SLinus Torvalds } 12801da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 12811da177e4SLinus Torvalds dprintk(" mismatch\n"); 128261822ab5STrond Myklebust nfs_redirty_request(req); 12831da177e4SLinus Torvalds next: 12849fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12851da177e4SLinus Torvalds } 12861da177e4SLinus Torvalds } 1287788e7a89STrond Myklebust 1288788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1289788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1290788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1291788e7a89STrond Myklebust }; 12921da177e4SLinus Torvalds 12933da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 12941da177e4SLinus Torvalds { 12951da177e4SLinus Torvalds LIST_HEAD(head); 12967d46a49fSTrond Myklebust int res; 12971da177e4SLinus Torvalds 1298587142f8STrond Myklebust spin_lock(&inode->i_lock); 12993da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1300587142f8STrond Myklebust spin_unlock(&inode->i_lock); 13011da177e4SLinus Torvalds if (res) { 13027d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 13031da177e4SLinus Torvalds if (error < 0) 13041da177e4SLinus Torvalds return error; 13053da28eb1STrond Myklebust } 13061da177e4SLinus Torvalds return res; 13071da177e4SLinus Torvalds } 1308c63c7b05STrond Myklebust #else 1309c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1310c63c7b05STrond Myklebust { 1311c63c7b05STrond Myklebust return 0; 1312c63c7b05STrond Myklebust } 13131da177e4SLinus Torvalds #endif 13141da177e4SLinus Torvalds 13151c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 13161da177e4SLinus Torvalds { 13171c75950bSTrond Myklebust struct inode *inode = mapping->host; 1318ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 13191c75950bSTrond Myklebust unsigned int npages = 0; 1320c42de9ddSTrond Myklebust LIST_HEAD(head); 132170b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 13223f442547STrond Myklebust long pages, ret; 13231da177e4SLinus Torvalds 13241c75950bSTrond Myklebust /* FIXME */ 13251c75950bSTrond Myklebust if (wbc->range_cyclic) 13261c75950bSTrond Myklebust idx_start = 0; 13271c75950bSTrond Myklebust else { 13281c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 13291c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 13301c75950bSTrond Myklebust if (idx_end > idx_start) { 1331ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 13321c75950bSTrond Myklebust npages = l_npages; 13331c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1334ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 13351c75950bSTrond Myklebust npages = 0; 13361c75950bSTrond Myklebust } 13371c75950bSTrond Myklebust } 1338c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1339587142f8STrond Myklebust spin_lock(&inode->i_lock); 13401da177e4SLinus Torvalds do { 1341c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1342c42de9ddSTrond Myklebust if (ret != 0) 1343c42de9ddSTrond Myklebust continue; 1344c42de9ddSTrond Myklebust if (nocommit) 1345c42de9ddSTrond Myklebust break; 1346d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1347724c439cSTrond Myklebust if (pages == 0) 1348c42de9ddSTrond Myklebust break; 1349d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1350587142f8STrond Myklebust spin_unlock(&inode->i_lock); 135183715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1352e8e058e8STrond Myklebust ret = pages; 1353587142f8STrond Myklebust spin_lock(&inode->i_lock); 1354d2ccddf0STrond Myklebust continue; 1355d2ccddf0STrond Myklebust } 1356d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1357587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1358c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1359587142f8STrond Myklebust spin_lock(&inode->i_lock); 1360587142f8STrond Myklebust 1361c42de9ddSTrond Myklebust } while (ret >= 0); 1362587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1363c42de9ddSTrond Myklebust return ret; 13641da177e4SLinus Torvalds } 13651da177e4SLinus Torvalds 136634901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 13671c75950bSTrond Myklebust { 13681c75950bSTrond Myklebust int ret; 13691c75950bSTrond Myklebust 137034901f70STrond Myklebust ret = nfs_writepages(mapping, wbc); 137161822ab5STrond Myklebust if (ret < 0) 137261822ab5STrond Myklebust goto out; 137334901f70STrond Myklebust ret = nfs_sync_mapping_wait(mapping, wbc, how); 1374ed90ef51STrond Myklebust if (ret < 0) 1375ed90ef51STrond Myklebust goto out; 13761c75950bSTrond Myklebust return 0; 137761822ab5STrond Myklebust out: 1378e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 13791c75950bSTrond Myklebust return ret; 13801c75950bSTrond Myklebust } 13811c75950bSTrond Myklebust 138234901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ 138334901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how) 138434901f70STrond Myklebust { 138534901f70STrond Myklebust struct writeback_control wbc = { 138634901f70STrond Myklebust .bdi = mapping->backing_dev_info, 138734901f70STrond Myklebust .sync_mode = WB_SYNC_NONE, 138834901f70STrond Myklebust .nr_to_write = LONG_MAX, 138934901f70STrond Myklebust .for_writepages = 1, 139034901f70STrond Myklebust .range_cyclic = 1, 139134901f70STrond Myklebust }; 139234901f70STrond Myklebust int ret; 139334901f70STrond Myklebust 139434901f70STrond Myklebust ret = __nfs_write_mapping(mapping, &wbc, how); 139534901f70STrond Myklebust if (ret < 0) 139634901f70STrond Myklebust return ret; 139734901f70STrond Myklebust wbc.sync_mode = WB_SYNC_ALL; 139834901f70STrond Myklebust return __nfs_write_mapping(mapping, &wbc, how); 139934901f70STrond Myklebust } 140034901f70STrond Myklebust 1401ed90ef51STrond Myklebust /* 1402ed90ef51STrond Myklebust * flush the inode to disk. 1403ed90ef51STrond Myklebust */ 1404ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode) 14051c75950bSTrond Myklebust { 1406ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, 0); 1407ed90ef51STrond Myklebust } 14081c75950bSTrond Myklebust 1409ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode) 1410ed90ef51STrond Myklebust { 1411ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); 14121c75950bSTrond Myklebust } 14131c75950bSTrond Myklebust 14141b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 14151b3b4a1aSTrond Myklebust { 14161b3b4a1aSTrond Myklebust struct nfs_page *req; 14171b3b4a1aSTrond Myklebust loff_t range_start = page_offset(page); 14181b3b4a1aSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14191b3b4a1aSTrond Myklebust struct writeback_control wbc = { 14201b3b4a1aSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14211b3b4a1aSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14221b3b4a1aSTrond Myklebust .nr_to_write = LONG_MAX, 14231b3b4a1aSTrond Myklebust .range_start = range_start, 14241b3b4a1aSTrond Myklebust .range_end = range_end, 14251b3b4a1aSTrond Myklebust }; 14261b3b4a1aSTrond Myklebust int ret = 0; 14271b3b4a1aSTrond Myklebust 14281b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 14291b3b4a1aSTrond Myklebust for (;;) { 14301b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 14311b3b4a1aSTrond Myklebust if (req == NULL) 14321b3b4a1aSTrond Myklebust goto out; 14331b3b4a1aSTrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 14341b3b4a1aSTrond Myklebust nfs_release_request(req); 14351b3b4a1aSTrond Myklebust break; 14361b3b4a1aSTrond Myklebust } 14371b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 14381b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 14391b3b4a1aSTrond Myklebust /* 14401b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 14411b3b4a1aSTrond Myklebust * page as being dirty 14421b3b4a1aSTrond Myklebust */ 14431b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 14441b3b4a1aSTrond Myklebust nfs_unlock_request(req); 14451b3b4a1aSTrond Myklebust break; 14461b3b4a1aSTrond Myklebust } 14471b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 14481b3b4a1aSTrond Myklebust if (ret < 0) 14491b3b4a1aSTrond Myklebust goto out; 14501b3b4a1aSTrond Myklebust } 14511b3b4a1aSTrond Myklebust if (!PagePrivate(page)) 14521b3b4a1aSTrond Myklebust return 0; 14531b3b4a1aSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); 14541b3b4a1aSTrond Myklebust out: 14551b3b4a1aSTrond Myklebust return ret; 14561b3b4a1aSTrond Myklebust } 14571b3b4a1aSTrond Myklebust 14585334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page, 14595334eb13SAdrian Bunk int how) 14601c75950bSTrond Myklebust { 14611c75950bSTrond Myklebust loff_t range_start = page_offset(page); 14621c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14634d770ccfSTrond Myklebust struct writeback_control wbc = { 14644d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14654d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14664d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 14674d770ccfSTrond Myklebust .range_start = range_start, 14684d770ccfSTrond Myklebust .range_end = range_end, 14694d770ccfSTrond Myklebust }; 14704d770ccfSTrond Myklebust int ret; 14711c75950bSTrond Myklebust 14724d770ccfSTrond Myklebust BUG_ON(!PageLocked(page)); 1473c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 14744d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 14754d770ccfSTrond Myklebust if (ret < 0) 14764d770ccfSTrond Myklebust goto out; 14774d770ccfSTrond Myklebust } 1478f40313acSTrond Myklebust if (!PagePrivate(page)) 1479f40313acSTrond Myklebust return 0; 14804d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 14814d770ccfSTrond Myklebust if (ret >= 0) 14824d770ccfSTrond Myklebust return 0; 14834d770ccfSTrond Myklebust out: 1484e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 14854d770ccfSTrond Myklebust return ret; 14861c75950bSTrond Myklebust } 14871c75950bSTrond Myklebust 14881c75950bSTrond Myklebust /* 14891c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 14901c75950bSTrond Myklebust */ 14911c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 14921c75950bSTrond Myklebust { 14934d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 14941c75950bSTrond Myklebust } 14951c75950bSTrond Myklebust 1496f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 14971da177e4SLinus Torvalds { 14981da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 14991da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15001da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 150120c2df83SPaul Mundt NULL); 15021da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15031da177e4SLinus Torvalds return -ENOMEM; 15041da177e4SLinus Torvalds 150593d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15061da177e4SLinus Torvalds nfs_wdata_cachep); 15071da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15081da177e4SLinus Torvalds return -ENOMEM; 15091da177e4SLinus Torvalds 151093d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15111da177e4SLinus Torvalds nfs_wdata_cachep); 15121da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15131da177e4SLinus Torvalds return -ENOMEM; 15141da177e4SLinus Torvalds 151589a09141SPeter Zijlstra /* 151689a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 151789a09141SPeter Zijlstra * 151889a09141SPeter Zijlstra * 64MB: 8192k 151989a09141SPeter Zijlstra * 128MB: 11585k 152089a09141SPeter Zijlstra * 256MB: 16384k 152189a09141SPeter Zijlstra * 512MB: 23170k 152289a09141SPeter Zijlstra * 1GB: 32768k 152389a09141SPeter Zijlstra * 2GB: 46340k 152489a09141SPeter Zijlstra * 4GB: 65536k 152589a09141SPeter Zijlstra * 8GB: 92681k 152689a09141SPeter Zijlstra * 16GB: 131072k 152789a09141SPeter Zijlstra * 152889a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 152989a09141SPeter Zijlstra * Limit the default to 256M 153089a09141SPeter Zijlstra */ 153189a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 153289a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 153389a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 153489a09141SPeter Zijlstra 15351da177e4SLinus Torvalds return 0; 15361da177e4SLinus Torvalds } 15371da177e4SLinus Torvalds 1538266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15391da177e4SLinus Torvalds { 15401da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15411da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15421a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15431da177e4SLinus Torvalds } 15441da177e4SLinus Torvalds 1545