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 113277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 114277459d2STrond Myklebust { 115277459d2STrond Myklebust struct nfs_page *req = NULL; 116277459d2STrond Myklebust 117277459d2STrond Myklebust if (PagePrivate(page)) { 118277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 119277459d2STrond Myklebust if (req != NULL) 120c03b4024STrond Myklebust kref_get(&req->wb_kref); 121277459d2STrond Myklebust } 122277459d2STrond Myklebust return req; 123277459d2STrond Myklebust } 124277459d2STrond Myklebust 125277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 126277459d2STrond Myklebust { 127587142f8STrond Myklebust struct inode *inode = page->mapping->host; 128277459d2STrond Myklebust struct nfs_page *req = NULL; 129277459d2STrond Myklebust 130587142f8STrond Myklebust spin_lock(&inode->i_lock); 131277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 132587142f8STrond Myklebust spin_unlock(&inode->i_lock); 133277459d2STrond Myklebust return req; 134277459d2STrond Myklebust } 135277459d2STrond Myklebust 1361da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1371da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1381da177e4SLinus Torvalds { 1391da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1401da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 141ca52fec1STrond Myklebust pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1421da177e4SLinus Torvalds 1431da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1441da177e4SLinus Torvalds return; 1451da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1461da177e4SLinus Torvalds if (i_size >= end) 1471da177e4SLinus Torvalds return; 14891d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1491da177e4SLinus Torvalds i_size_write(inode, end); 1501da177e4SLinus Torvalds } 1511da177e4SLinus Torvalds 152a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 153a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 154a301b777STrond Myklebust { 155a301b777STrond Myklebust SetPageError(page); 156a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 157a301b777STrond Myklebust } 158a301b777STrond Myklebust 1591da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1601da177e4SLinus Torvalds * covers the full page. 1611da177e4SLinus Torvalds */ 1621da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1631da177e4SLinus Torvalds { 1641da177e4SLinus Torvalds if (PageUptodate(page)) 1651da177e4SLinus Torvalds return; 1661da177e4SLinus Torvalds if (base != 0) 1671da177e4SLinus Torvalds return; 16849a70f27STrond Myklebust if (count != nfs_page_length(page)) 1691da177e4SLinus Torvalds return; 17049a70f27STrond Myklebust if (count != PAGE_CACHE_SIZE) 17160945cb7SNate Diller zero_user_page(page, count, PAGE_CACHE_SIZE - count, KM_USER0); 1721da177e4SLinus Torvalds SetPageUptodate(page); 1731da177e4SLinus Torvalds } 1741da177e4SLinus Torvalds 175e21195a7STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1761da177e4SLinus Torvalds unsigned int offset, unsigned int count) 1771da177e4SLinus Torvalds { 1781da177e4SLinus Torvalds struct nfs_page *req; 179e21195a7STrond Myklebust int ret; 1801da177e4SLinus Torvalds 181e21195a7STrond Myklebust for (;;) { 182e21195a7STrond Myklebust req = nfs_update_request(ctx, page, offset, count); 183e21195a7STrond Myklebust if (!IS_ERR(req)) 184e21195a7STrond Myklebust break; 185e21195a7STrond Myklebust ret = PTR_ERR(req); 186e21195a7STrond Myklebust if (ret != -EBUSY) 187e21195a7STrond Myklebust return ret; 188e21195a7STrond Myklebust ret = nfs_wb_page(page->mapping->host, page); 189e21195a7STrond Myklebust if (ret != 0) 190e21195a7STrond Myklebust return ret; 191e21195a7STrond Myklebust } 1921da177e4SLinus Torvalds /* Update file length */ 1931da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 1941da177e4SLinus Torvalds nfs_unlock_request(req); 195abd3e641STrond Myklebust return 0; 1961da177e4SLinus Torvalds } 1971da177e4SLinus Torvalds 1981da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1991da177e4SLinus Torvalds { 2001da177e4SLinus Torvalds if (wbc->for_reclaim) 201c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 2021da177e4SLinus Torvalds if (wbc->for_kupdate) 2031da177e4SLinus Torvalds return FLUSH_LOWPRI; 2041da177e4SLinus Torvalds return 0; 2051da177e4SLinus Torvalds } 2061da177e4SLinus Torvalds 2071da177e4SLinus Torvalds /* 20889a09141SPeter Zijlstra * NFS congestion control 20989a09141SPeter Zijlstra */ 21089a09141SPeter Zijlstra 21189a09141SPeter Zijlstra int nfs_congestion_kb; 21289a09141SPeter Zijlstra 21389a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 21489a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 21589a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 21689a09141SPeter Zijlstra 2175a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 21889a09141SPeter Zijlstra { 2195a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2205a6d41b3STrond Myklebust 2215a6d41b3STrond Myklebust if (!ret) { 22289a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 22389a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 22489a09141SPeter Zijlstra 225277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 22689a09141SPeter Zijlstra NFS_CONGESTION_ON_THRESH) 22789a09141SPeter Zijlstra set_bdi_congested(&nfss->backing_dev_info, WRITE); 22889a09141SPeter Zijlstra } 2295a6d41b3STrond Myklebust return ret; 23089a09141SPeter Zijlstra } 23189a09141SPeter Zijlstra 23289a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 23389a09141SPeter Zijlstra { 23489a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 23589a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 23689a09141SPeter Zijlstra 23789a09141SPeter Zijlstra end_page_writeback(page); 238277866a0SPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) { 23989a09141SPeter Zijlstra clear_bdi_congested(&nfss->backing_dev_info, WRITE); 24089a09141SPeter Zijlstra congestion_end(WRITE); 24189a09141SPeter Zijlstra } 24289a09141SPeter Zijlstra } 24389a09141SPeter Zijlstra 24489a09141SPeter Zijlstra /* 245e261f51fSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 2469cccef95STrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 247e261f51fSTrond Myklebust */ 248c63c7b05STrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 249c63c7b05STrond Myklebust struct page *page) 250e261f51fSTrond Myklebust { 251587142f8STrond Myklebust struct inode *inode = page->mapping->host; 252587142f8STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 253e261f51fSTrond Myklebust struct nfs_page *req; 254e261f51fSTrond Myklebust int ret; 255e261f51fSTrond Myklebust 256587142f8STrond Myklebust spin_lock(&inode->i_lock); 257e261f51fSTrond Myklebust for(;;) { 258e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 259e261f51fSTrond Myklebust if (req == NULL) { 260587142f8STrond Myklebust spin_unlock(&inode->i_lock); 2619cccef95STrond Myklebust return 0; 262e261f51fSTrond Myklebust } 263e261f51fSTrond Myklebust if (nfs_lock_request_dontget(req)) 264e261f51fSTrond Myklebust break; 265e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 266e261f51fSTrond Myklebust * then the call to nfs_lock_request_dontget() will always 267e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 268e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 269e261f51fSTrond Myklebust */ 270587142f8STrond Myklebust spin_unlock(&inode->i_lock); 271e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 272e261f51fSTrond Myklebust nfs_release_request(req); 273e261f51fSTrond Myklebust if (ret != 0) 274e261f51fSTrond Myklebust return ret; 275587142f8STrond Myklebust spin_lock(&inode->i_lock); 276e261f51fSTrond Myklebust } 277612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 278612c9384STrond Myklebust /* This request is marked for commit */ 279587142f8STrond Myklebust spin_unlock(&inode->i_lock); 280612c9384STrond Myklebust nfs_unlock_request(req); 281c63c7b05STrond Myklebust nfs_pageio_complete(pgio); 2829cccef95STrond Myklebust return 0; 283612c9384STrond Myklebust } 284c63c7b05STrond Myklebust if (nfs_set_page_writeback(page) != 0) { 285587142f8STrond Myklebust spin_unlock(&inode->i_lock); 286c63c7b05STrond Myklebust BUG(); 287c63c7b05STrond Myklebust } 288c63c7b05STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, 2899fd367f0STrond Myklebust NFS_PAGE_TAG_LOCKED); 290587142f8STrond Myklebust spin_unlock(&inode->i_lock); 291c63c7b05STrond Myklebust nfs_pageio_add_request(pgio, req); 2929cccef95STrond Myklebust return 0; 293e261f51fSTrond Myklebust } 294e261f51fSTrond Myklebust 295*f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 296*f758c885STrond Myklebust { 297*f758c885STrond Myklebust struct inode *inode = page->mapping->host; 298*f758c885STrond Myklebust 299*f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 300*f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 301*f758c885STrond Myklebust 302*f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 303*f758c885STrond Myklebust return nfs_page_async_flush(pgio, page); 304*f758c885STrond Myklebust } 305*f758c885STrond 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 { 311*f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 312e261f51fSTrond Myklebust int err; 3131da177e4SLinus Torvalds 314*f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 315*f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 316*f758c885STrond Myklebust nfs_pageio_complete(&pgio); 317*f758c885STrond Myklebust if (err < 0) 3184d770ccfSTrond Myklebust return err; 319*f758c885STrond Myklebust if (pgio.pg_error < 0) 320*f758c885STrond Myklebust return pgio.pg_error; 321*f758c885STrond Myklebust return 0; 3224d770ccfSTrond Myklebust } 3234d770ccfSTrond Myklebust 3244d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3254d770ccfSTrond Myklebust { 326*f758c885STrond Myklebust int ret; 3274d770ccfSTrond Myklebust 328*f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3291da177e4SLinus Torvalds unlock_page(page); 330*f758c885STrond Myklebust return ret; 331*f758c885STrond Myklebust } 332*f758c885STrond Myklebust 333*f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 334*f758c885STrond Myklebust { 335*f758c885STrond Myklebust int ret; 336*f758c885STrond Myklebust 337*f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 338*f758c885STrond Myklebust unlock_page(page); 339*f758c885STrond 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)); 351*f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 352c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 353*f758c885STrond Myklebust if (err < 0) 3541da177e4SLinus Torvalds return err; 355*f758c885STrond 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 nfs_begin_data_update(inode); 3751da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3761da177e4SLinus Torvalds nfsi->change_attr++; 3771da177e4SLinus Torvalds } 378deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 379277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3801da177e4SLinus Torvalds nfsi->npages++; 381c03b4024STrond Myklebust kref_get(&req->wb_kref); 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); 402951a143bSTrond Myklebust nfs_end_data_update(inode); 4031da177e4SLinus Torvalds iput(inode); 4041da177e4SLinus Torvalds } else 405587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4061da177e4SLinus Torvalds nfs_clear_request(req); 4071da177e4SLinus Torvalds nfs_release_request(req); 4081da177e4SLinus Torvalds } 4091da177e4SLinus Torvalds 41061822ab5STrond Myklebust static void 41161822ab5STrond Myklebust nfs_redirty_request(struct nfs_page *req) 41261822ab5STrond Myklebust { 41361822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 41461822ab5STrond Myklebust } 41561822ab5STrond Myklebust 4161da177e4SLinus Torvalds /* 4171da177e4SLinus Torvalds * Check if a request is dirty 4181da177e4SLinus Torvalds */ 4191da177e4SLinus Torvalds static inline int 4201da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 4211da177e4SLinus Torvalds { 4225a6d41b3STrond Myklebust struct page *page = req->wb_page; 4235a6d41b3STrond Myklebust 424612c9384STrond Myklebust if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags)) 4255a6d41b3STrond Myklebust return 0; 4265a6d41b3STrond Myklebust return !PageWriteback(req->wb_page); 4271da177e4SLinus Torvalds } 4281da177e4SLinus Torvalds 4291da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4301da177e4SLinus Torvalds /* 4311da177e4SLinus Torvalds * Add a request to the inode's commit list. 4321da177e4SLinus Torvalds */ 4331da177e4SLinus Torvalds static void 4341da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4351da177e4SLinus Torvalds { 43688be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4371da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4381da177e4SLinus Torvalds 439587142f8STrond Myklebust spin_lock(&inode->i_lock); 4401da177e4SLinus Torvalds nfsi->ncommit++; 441612c9384STrond Myklebust set_bit(PG_NEED_COMMIT, &(req)->wb_flags); 4425c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4435c369683STrond Myklebust req->wb_index, 4445c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 445587142f8STrond Myklebust spin_unlock(&inode->i_lock); 446fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 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); 53383715ad5STrond Myklebust nfs_list_remove_request(req); 534612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 53583715ad5STrond Myklebust nfs_inode_remove_request(req); 536b6dff26aSTrond Myklebust nfs_unlock_request(req); 53783715ad5STrond Myklebust } 53883715ad5STrond Myklebust } 53983715ad5STrond Myklebust 5401da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5411da177e4SLinus Torvalds /* 5421da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5431da177e4SLinus Torvalds * @inode: NFS inode to scan 5441da177e4SLinus Torvalds * @dst: destination list 5451da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5461da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5471da177e4SLinus Torvalds * 5481da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5491da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5501da177e4SLinus Torvalds */ 5511da177e4SLinus Torvalds static int 552ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5531da177e4SLinus Torvalds { 5541da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5553da28eb1STrond Myklebust int res = 0; 5563da28eb1STrond Myklebust 5573da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 5585c369683STrond Myklebust res = nfs_scan_list(nfsi, dst, idx_start, npages, 5595c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 5601da177e4SLinus Torvalds nfsi->ncommit -= res; 5613da28eb1STrond Myklebust } 5621da177e4SLinus Torvalds return res; 5631da177e4SLinus Torvalds } 564c42de9ddSTrond Myklebust #else 565ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 566c42de9ddSTrond Myklebust { 567c42de9ddSTrond Myklebust return 0; 568c42de9ddSTrond Myklebust } 5691da177e4SLinus Torvalds #endif 5701da177e4SLinus Torvalds 5711da177e4SLinus Torvalds /* 5721da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 5731da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 5741da177e4SLinus Torvalds * the calling process. 5751da177e4SLinus Torvalds * 5761da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 5771da177e4SLinus Torvalds */ 5781da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 579e21195a7STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 5801da177e4SLinus Torvalds { 58189a09141SPeter Zijlstra struct address_space *mapping = page->mapping; 58289a09141SPeter Zijlstra struct inode *inode = mapping->host; 5831da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 584ca52fec1STrond Myklebust pgoff_t rqend, end; 5851da177e4SLinus Torvalds 5861da177e4SLinus Torvalds end = offset + bytes; 5871da177e4SLinus Torvalds 5881da177e4SLinus Torvalds for (;;) { 5891da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 5901da177e4SLinus Torvalds * A request for the page we wish to update 5911da177e4SLinus Torvalds */ 592587142f8STrond Myklebust spin_lock(&inode->i_lock); 593277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 5941da177e4SLinus Torvalds if (req) { 5951da177e4SLinus Torvalds if (!nfs_lock_request_dontget(req)) { 5961da177e4SLinus Torvalds int error; 597277459d2STrond Myklebust 598587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5991da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6001da177e4SLinus Torvalds nfs_release_request(req); 6011dd594b2SNeil Brown if (error < 0) { 6021dd594b2SNeil Brown if (new) 6031dd594b2SNeil Brown nfs_release_request(new); 6041da177e4SLinus Torvalds return ERR_PTR(error); 6051dd594b2SNeil Brown } 6061da177e4SLinus Torvalds continue; 6071da177e4SLinus Torvalds } 608587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6091da177e4SLinus Torvalds if (new) 6101da177e4SLinus Torvalds nfs_release_request(new); 6111da177e4SLinus Torvalds break; 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds 6141da177e4SLinus Torvalds if (new) { 6151da177e4SLinus Torvalds int error; 6161da177e4SLinus Torvalds nfs_lock_request_dontget(new); 6171da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 6181da177e4SLinus Torvalds if (error) { 619587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6201da177e4SLinus Torvalds nfs_unlock_request(new); 6211da177e4SLinus Torvalds return ERR_PTR(error); 6221da177e4SLinus Torvalds } 623587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6241da177e4SLinus Torvalds return new; 6251da177e4SLinus Torvalds } 626587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6271da177e4SLinus Torvalds 6281da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 6291da177e4SLinus Torvalds if (IS_ERR(new)) 6301da177e4SLinus Torvalds return new; 6311da177e4SLinus Torvalds } 6321da177e4SLinus Torvalds 6331da177e4SLinus Torvalds /* We have a request for our page. 6341da177e4SLinus Torvalds * If the creds don't match, or the 6351da177e4SLinus Torvalds * page addresses don't match, 6361da177e4SLinus Torvalds * tell the caller to wait on the conflicting 6371da177e4SLinus Torvalds * request. 6381da177e4SLinus Torvalds */ 6391da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 6401da177e4SLinus Torvalds if (req->wb_context != ctx 6411da177e4SLinus Torvalds || req->wb_page != page 6421da177e4SLinus Torvalds || !nfs_dirty_request(req) 6431da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 6441da177e4SLinus Torvalds nfs_unlock_request(req); 6451da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 6461da177e4SLinus Torvalds } 6471da177e4SLinus Torvalds 6481da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6491da177e4SLinus Torvalds if (offset < req->wb_offset) { 6501da177e4SLinus Torvalds req->wb_offset = offset; 6511da177e4SLinus Torvalds req->wb_pgbase = offset; 6521da177e4SLinus Torvalds req->wb_bytes = rqend - req->wb_offset; 6531da177e4SLinus Torvalds } 6541da177e4SLinus Torvalds 6551da177e4SLinus Torvalds if (end > rqend) 6561da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 6571da177e4SLinus Torvalds 6581da177e4SLinus Torvalds return req; 6591da177e4SLinus Torvalds } 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6621da177e4SLinus Torvalds { 6631da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 6641da177e4SLinus Torvalds struct nfs_page *req; 6651a54533eSTrond Myklebust int do_flush, status; 6661da177e4SLinus Torvalds /* 6671da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6681da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6691da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 6701da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 6711da177e4SLinus Torvalds * Also do the same if we find a request from an existing 6721da177e4SLinus Torvalds * dropped page. 6731da177e4SLinus Torvalds */ 6741a54533eSTrond Myklebust do { 675277459d2STrond Myklebust req = nfs_page_find_request(page); 6761a54533eSTrond Myklebust if (req == NULL) 6771a54533eSTrond Myklebust return 0; 6781a54533eSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx 679e261f51fSTrond Myklebust || !nfs_dirty_request(req); 6801da177e4SLinus Torvalds nfs_release_request(req); 6811a54533eSTrond Myklebust if (!do_flush) 6821a54533eSTrond Myklebust return 0; 683277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 6841a54533eSTrond Myklebust } while (status == 0); 6851a54533eSTrond Myklebust return status; 6861da177e4SLinus Torvalds } 6871da177e4SLinus Torvalds 6881da177e4SLinus Torvalds /* 6891da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 6901da177e4SLinus Torvalds * 6911da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 6921da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 6931da177e4SLinus Torvalds */ 6941da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 6951da177e4SLinus Torvalds unsigned int offset, unsigned int count) 6961da177e4SLinus Torvalds { 6971da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 6981da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 6991da177e4SLinus Torvalds int status = 0; 7001da177e4SLinus Torvalds 70191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 70291d5b470SChuck Lever 7031da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 70401cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 70501cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7060bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 7071da177e4SLinus Torvalds 7081da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7091da177e4SLinus Torvalds * is entirely in cache, it may be more efficient to avoid 7101da177e4SLinus Torvalds * fragmenting write requests. 7111da177e4SLinus Torvalds */ 712ab0a3dbeSTrond Myklebust if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { 71349a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7141da177e4SLinus Torvalds offset = 0; 7151da177e4SLinus Torvalds } 7161da177e4SLinus Torvalds 717e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 718e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7191da177e4SLinus Torvalds 7201da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 7211da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7221da177e4SLinus Torvalds if (status < 0) 723a301b777STrond Myklebust nfs_set_pageerror(page); 7241da177e4SLinus Torvalds return status; 7251da177e4SLinus Torvalds } 7261da177e4SLinus Torvalds 7271da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7281da177e4SLinus Torvalds { 7298e821cadSTrond Myklebust 73044dd151dSTrond Myklebust if (PageError(req->wb_page)) { 73144dd151dSTrond Myklebust nfs_end_page_writeback(req->wb_page); 73244dd151dSTrond Myklebust nfs_inode_remove_request(req); 73344dd151dSTrond Myklebust } else if (!nfs_reschedule_unstable_write(req)) { 73444dd151dSTrond Myklebust /* Set the PG_uptodate flag */ 73544dd151dSTrond Myklebust nfs_mark_uptodate(req->wb_page, req->wb_pgbase, req->wb_bytes); 73689a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7371da177e4SLinus Torvalds nfs_inode_remove_request(req); 7388e821cadSTrond Myklebust } else 7398e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 7409fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 7411da177e4SLinus Torvalds } 7421da177e4SLinus Torvalds 7431da177e4SLinus Torvalds static inline int flush_task_priority(int how) 7441da177e4SLinus Torvalds { 7451da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7461da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7471da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7481da177e4SLinus Torvalds case FLUSH_LOWPRI: 7491da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7501da177e4SLinus Torvalds } 7511da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7521da177e4SLinus Torvalds } 7531da177e4SLinus Torvalds 7541da177e4SLinus Torvalds /* 7551da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 7561da177e4SLinus Torvalds */ 7571da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 7581da177e4SLinus Torvalds struct nfs_write_data *data, 759788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7601da177e4SLinus Torvalds unsigned int count, unsigned int offset, 7611da177e4SLinus Torvalds int how) 7621da177e4SLinus Torvalds { 7631da177e4SLinus Torvalds struct inode *inode; 764788e7a89STrond Myklebust int flags; 7651da177e4SLinus Torvalds 7661da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 7671da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 7681da177e4SLinus Torvalds 7691da177e4SLinus Torvalds data->req = req; 77088be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 7711da177e4SLinus Torvalds data->cred = req->wb_context->cred; 7721da177e4SLinus Torvalds 7731da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 7741da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 7751da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 7761da177e4SLinus Torvalds data->args.pages = data->pagevec; 7771da177e4SLinus Torvalds data->args.count = count; 7781da177e4SLinus Torvalds data->args.context = req->wb_context; 7791da177e4SLinus Torvalds 7801da177e4SLinus Torvalds data->res.fattr = &data->fattr; 7811da177e4SLinus Torvalds data->res.count = count; 7821da177e4SLinus Torvalds data->res.verf = &data->verf; 7830e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 7841da177e4SLinus Torvalds 785788e7a89STrond Myklebust /* Set up the initial task struct. */ 786788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 787788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); 7881da177e4SLinus Torvalds NFS_PROTO(inode)->write_setup(data, how); 7891da177e4SLinus Torvalds 7901da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 7911da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 7921da177e4SLinus Torvalds 793a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 794a3f565b1SChuck Lever "(req %s/%Ld, %u bytes @ offset %Lu)\n", 7950bbacc40SChuck Lever data->task.tk_pid, 7961da177e4SLinus Torvalds inode->i_sb->s_id, 7971da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 7981da177e4SLinus Torvalds count, 7991da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds 8021da177e4SLinus Torvalds static void nfs_execute_write(struct nfs_write_data *data) 8031da177e4SLinus Torvalds { 8041da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(data->inode); 8051da177e4SLinus Torvalds sigset_t oldset; 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 8081da177e4SLinus Torvalds rpc_execute(&data->task); 8091da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 8101da177e4SLinus Torvalds } 8111da177e4SLinus Torvalds 8121da177e4SLinus Torvalds /* 8131da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8141da177e4SLinus Torvalds * contiguous dirty area on one page. 8151da177e4SLinus Torvalds */ 8168d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8171da177e4SLinus Torvalds { 8181da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 8191da177e4SLinus Torvalds struct page *page = req->wb_page; 8201da177e4SLinus Torvalds struct nfs_write_data *data; 821e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 822e9f7bee1STrond Myklebust unsigned int offset; 8231da177e4SLinus Torvalds int requests = 0; 8241da177e4SLinus Torvalds LIST_HEAD(list); 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds nfs_list_remove_request(req); 8271da177e4SLinus Torvalds 828bcb71bbaSTrond Myklebust nbytes = count; 829e9f7bee1STrond Myklebust do { 830e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 831e9f7bee1STrond Myklebust 8328d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 8331da177e4SLinus Torvalds if (!data) 8341da177e4SLinus Torvalds goto out_bad; 8351da177e4SLinus Torvalds list_add(&data->pages, &list); 8361da177e4SLinus Torvalds requests++; 837e9f7bee1STrond Myklebust nbytes -= len; 838e9f7bee1STrond Myklebust } while (nbytes != 0); 8391da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 8401da177e4SLinus Torvalds 8411da177e4SLinus Torvalds ClearPageError(page); 8421da177e4SLinus Torvalds offset = 0; 843bcb71bbaSTrond Myklebust nbytes = count; 8441da177e4SLinus Torvalds do { 8451da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8461da177e4SLinus Torvalds list_del_init(&data->pages); 8471da177e4SLinus Torvalds 8481da177e4SLinus Torvalds data->pagevec[0] = page; 8491da177e4SLinus Torvalds 850bcb71bbaSTrond Myklebust if (nbytes < wsize) 851bcb71bbaSTrond Myklebust wsize = nbytes; 852788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 853788e7a89STrond Myklebust wsize, offset, how); 8541da177e4SLinus Torvalds offset += wsize; 8551da177e4SLinus Torvalds nbytes -= wsize; 8561da177e4SLinus Torvalds nfs_execute_write(data); 8571da177e4SLinus Torvalds } while (nbytes != 0); 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds return 0; 8601da177e4SLinus Torvalds 8611da177e4SLinus Torvalds out_bad: 8621da177e4SLinus Torvalds while (!list_empty(&list)) { 8631da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8641da177e4SLinus Torvalds list_del(&data->pages); 8658aca67f0STrond Myklebust nfs_writedata_release(data); 8661da177e4SLinus Torvalds } 86761822ab5STrond Myklebust nfs_redirty_request(req); 8686d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 8699fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 8701da177e4SLinus Torvalds return -ENOMEM; 8711da177e4SLinus Torvalds } 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds /* 8741da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 8751da177e4SLinus Torvalds * The page must have been locked by the caller. 8761da177e4SLinus Torvalds * 8771da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 8781da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 8791da177e4SLinus Torvalds * that has been written but not committed. 8801da177e4SLinus Torvalds */ 8818d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8821da177e4SLinus Torvalds { 8831da177e4SLinus Torvalds struct nfs_page *req; 8841da177e4SLinus Torvalds struct page **pages; 8851da177e4SLinus Torvalds struct nfs_write_data *data; 8861da177e4SLinus Torvalds 8878d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 8881da177e4SLinus Torvalds if (!data) 8891da177e4SLinus Torvalds goto out_bad; 8901da177e4SLinus Torvalds 8911da177e4SLinus Torvalds pages = data->pagevec; 8921da177e4SLinus Torvalds while (!list_empty(head)) { 8931da177e4SLinus Torvalds req = nfs_list_entry(head->next); 8941da177e4SLinus Torvalds nfs_list_remove_request(req); 8951da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 8961da177e4SLinus Torvalds ClearPageError(req->wb_page); 8971da177e4SLinus Torvalds *pages++ = req->wb_page; 8981da177e4SLinus Torvalds } 8991da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9001da177e4SLinus Torvalds 9011da177e4SLinus Torvalds /* Set up the argument struct */ 902788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9031da177e4SLinus Torvalds 9041da177e4SLinus Torvalds nfs_execute_write(data); 9051da177e4SLinus Torvalds return 0; 9061da177e4SLinus Torvalds out_bad: 9071da177e4SLinus Torvalds while (!list_empty(head)) { 90810afec90STrond Myklebust req = nfs_list_entry(head->next); 9091da177e4SLinus Torvalds nfs_list_remove_request(req); 91061822ab5STrond Myklebust nfs_redirty_request(req); 9116d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 9129fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 9131da177e4SLinus Torvalds } 9141da177e4SLinus Torvalds return -ENOMEM; 9151da177e4SLinus Torvalds } 9161da177e4SLinus Torvalds 917c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 918c63c7b05STrond Myklebust struct inode *inode, int ioflags) 9191da177e4SLinus Torvalds { 9207d46a49fSTrond Myklebust int wsize = NFS_SERVER(inode)->wsize; 9211da177e4SLinus Torvalds 922bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 923c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 924bcb71bbaSTrond Myklebust else 925c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 9261da177e4SLinus Torvalds } 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds /* 9291da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 9301da177e4SLinus Torvalds */ 931788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 9321da177e4SLinus Torvalds { 933788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 9341da177e4SLinus Torvalds struct nfs_page *req = data->req; 9351da177e4SLinus Torvalds struct page *page = req->wb_page; 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 93888be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 93988be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 9401da177e4SLinus Torvalds req->wb_bytes, 9411da177e4SLinus Torvalds (long long)req_offset(req)); 9421da177e4SLinus Torvalds 943788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 944788e7a89STrond Myklebust return; 945788e7a89STrond Myklebust 946788e7a89STrond Myklebust if (task->tk_status < 0) { 947a301b777STrond Myklebust nfs_set_pageerror(page); 948788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 949788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 9508e821cadSTrond Myklebust goto out; 9518e821cadSTrond Myklebust } 9528e821cadSTrond Myklebust 9538e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 954587142f8STrond Myklebust struct inode *inode = page->mapping->host; 9558e821cadSTrond Myklebust 956587142f8STrond Myklebust spin_lock(&inode->i_lock); 9578e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 9588e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 9598e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 9601da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 9611da177e4SLinus Torvalds dprintk(" defer commit\n"); 9621da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 9638e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 9648e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 9651da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 9661da177e4SLinus Torvalds } 967587142f8STrond Myklebust spin_unlock(&inode->i_lock); 9681da177e4SLinus Torvalds } else 9691da177e4SLinus Torvalds dprintk(" OK\n"); 9701da177e4SLinus Torvalds 9718e821cadSTrond Myklebust out: 9721da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 9731da177e4SLinus Torvalds nfs_writepage_release(req); 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 976788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 977788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 978788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 979788e7a89STrond Myklebust }; 980788e7a89STrond Myklebust 9811da177e4SLinus Torvalds /* 9821da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 9831da177e4SLinus Torvalds * 9841da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 9851da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 9861da177e4SLinus Torvalds * as the page has a write request pending. 9871da177e4SLinus Torvalds */ 988788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 9891da177e4SLinus Torvalds { 990788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 9911da177e4SLinus Torvalds struct nfs_page *req; 9921da177e4SLinus Torvalds struct page *page; 9931da177e4SLinus Torvalds 994788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 995788e7a89STrond Myklebust return; 996788e7a89STrond Myklebust 9971da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 9981da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 9991da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10001da177e4SLinus Torvalds nfs_list_remove_request(req); 10011da177e4SLinus Torvalds page = req->wb_page; 10021da177e4SLinus Torvalds 10031da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 100488be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 100588be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 10061da177e4SLinus Torvalds req->wb_bytes, 10071da177e4SLinus Torvalds (long long)req_offset(req)); 10081da177e4SLinus Torvalds 1009788e7a89STrond Myklebust if (task->tk_status < 0) { 1010a301b777STrond Myklebust nfs_set_pageerror(page); 1011788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 1012788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 10138e821cadSTrond Myklebust goto remove_request; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds 10168e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 10171da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10181da177e4SLinus Torvalds nfs_mark_request_commit(req); 10198e821cadSTrond Myklebust nfs_end_page_writeback(page); 10201da177e4SLinus Torvalds dprintk(" marked for commit\n"); 10218e821cadSTrond Myklebust goto next; 10228e821cadSTrond Myklebust } 102344dd151dSTrond Myklebust /* Set the PG_uptodate flag? */ 102444dd151dSTrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 10258e821cadSTrond Myklebust dprintk(" OK\n"); 10268e821cadSTrond Myklebust remove_request: 10278e821cadSTrond Myklebust nfs_end_page_writeback(page); 10281da177e4SLinus Torvalds nfs_inode_remove_request(req); 10291da177e4SLinus Torvalds next: 10309fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 10311da177e4SLinus Torvalds } 10321da177e4SLinus Torvalds } 10331da177e4SLinus Torvalds 1034788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1035788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1036788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1037788e7a89STrond Myklebust }; 1038788e7a89STrond Myklebust 1039788e7a89STrond Myklebust 10401da177e4SLinus Torvalds /* 10411da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 10421da177e4SLinus Torvalds */ 1043462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 10441da177e4SLinus Torvalds { 10451da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 10461da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1047788e7a89STrond Myklebust int status; 10481da177e4SLinus Torvalds 1049a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 10501da177e4SLinus Torvalds task->tk_pid, task->tk_status); 10511da177e4SLinus Torvalds 1052f551e44fSChuck Lever /* 1053f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1054f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1055f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1056f551e44fSChuck Lever * another writer had changed the file, but some applications 1057f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1058f551e44fSChuck Lever */ 1059788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1060788e7a89STrond Myklebust if (status != 0) 1061788e7a89STrond Myklebust return status; 106291d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 106391d5b470SChuck Lever 10641da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 10651da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 10661da177e4SLinus Torvalds /* We tried a write call, but the server did not 10671da177e4SLinus Torvalds * commit data to stable storage even though we 10681da177e4SLinus Torvalds * requested it. 10691da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 10701da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 10711da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 10721da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 10731da177e4SLinus Torvalds */ 10741da177e4SLinus Torvalds static unsigned long complain; 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 10771da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 10781da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 107954ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 10801da177e4SLinus Torvalds resp->verf->committed, argp->stable); 10811da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 10821da177e4SLinus Torvalds } 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds #endif 10851da177e4SLinus Torvalds /* Is this a short write? */ 10861da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 10871da177e4SLinus Torvalds static unsigned long complain; 10881da177e4SLinus Torvalds 108991d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 109091d5b470SChuck Lever 10911da177e4SLinus Torvalds /* Has the server at least made some progress? */ 10921da177e4SLinus Torvalds if (resp->count != 0) { 10931da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 10941da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 10951da177e4SLinus Torvalds /* Resend from where the server left off */ 10961da177e4SLinus Torvalds argp->offset += resp->count; 10971da177e4SLinus Torvalds argp->pgbase += resp->count; 10981da177e4SLinus Torvalds argp->count -= resp->count; 10991da177e4SLinus Torvalds } else { 11001da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11011da177e4SLinus Torvalds * headaches in the case of a server crash. 11021da177e4SLinus Torvalds */ 11031da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11041da177e4SLinus Torvalds } 11051da177e4SLinus Torvalds rpc_restart_call(task); 1106788e7a89STrond Myklebust return -EAGAIN; 11071da177e4SLinus Torvalds } 11081da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11091da177e4SLinus Torvalds printk(KERN_WARNING 11101da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 11111da177e4SLinus Torvalds argp->count); 11121da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11131da177e4SLinus Torvalds } 11141da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 11151da177e4SLinus Torvalds task->tk_status = -EIO; 11161da177e4SLinus Torvalds } 1117788e7a89STrond Myklebust return 0; 11181da177e4SLinus Torvalds } 11191da177e4SLinus Torvalds 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1122963d8fe5STrond Myklebust void nfs_commit_release(void *wdata) 11231da177e4SLinus Torvalds { 11241da177e4SLinus Torvalds nfs_commit_free(wdata); 11251da177e4SLinus Torvalds } 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds /* 11281da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 11291da177e4SLinus Torvalds */ 11301da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1131788e7a89STrond Myklebust struct nfs_write_data *data, 1132788e7a89STrond Myklebust int how) 11331da177e4SLinus Torvalds { 11343da28eb1STrond Myklebust struct nfs_page *first; 11351da177e4SLinus Torvalds struct inode *inode; 1136788e7a89STrond Myklebust int flags; 11371da177e4SLinus Torvalds 11381da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 11391da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 11401da177e4SLinus Torvalds 11411da177e4SLinus Torvalds list_splice_init(head, &data->pages); 11421da177e4SLinus Torvalds first = nfs_list_entry(data->pages.next); 114388be9f99STrond Myklebust inode = first->wb_context->path.dentry->d_inode; 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds data->inode = inode; 11461da177e4SLinus Torvalds data->cred = first->wb_context->cred; 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 11493da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 11503da28eb1STrond Myklebust data->args.offset = 0; 11513da28eb1STrond Myklebust data->args.count = 0; 11523da28eb1STrond Myklebust data->res.count = 0; 11531da177e4SLinus Torvalds data->res.fattr = &data->fattr; 11541da177e4SLinus Torvalds data->res.verf = &data->verf; 11550e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 11561da177e4SLinus Torvalds 1157788e7a89STrond Myklebust /* Set up the initial task struct. */ 1158788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 1159788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data); 11601da177e4SLinus Torvalds NFS_PROTO(inode)->commit_setup(data, how); 11611da177e4SLinus Torvalds 11621da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 11631da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 11641da177e4SLinus Torvalds 1165a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds /* 11691da177e4SLinus Torvalds * Commit dirty pages 11701da177e4SLinus Torvalds */ 11711da177e4SLinus Torvalds static int 117240859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 11731da177e4SLinus Torvalds { 11741da177e4SLinus Torvalds struct nfs_write_data *data; 11751da177e4SLinus Torvalds struct nfs_page *req; 11761da177e4SLinus Torvalds 1177e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 11781da177e4SLinus Torvalds 11791da177e4SLinus Torvalds if (!data) 11801da177e4SLinus Torvalds goto out_bad; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds /* Set up the argument struct */ 11831da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds nfs_execute_write(data); 11861da177e4SLinus Torvalds return 0; 11871da177e4SLinus Torvalds out_bad: 11881da177e4SLinus Torvalds while (!list_empty(head)) { 11891da177e4SLinus Torvalds req = nfs_list_entry(head->next); 11901da177e4SLinus Torvalds nfs_list_remove_request(req); 11911da177e4SLinus Torvalds nfs_mark_request_commit(req); 119283715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 11939fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 11941da177e4SLinus Torvalds } 11951da177e4SLinus Torvalds return -ENOMEM; 11961da177e4SLinus Torvalds } 11971da177e4SLinus Torvalds 11981da177e4SLinus Torvalds /* 11991da177e4SLinus Torvalds * COMMIT call returned 12001da177e4SLinus Torvalds */ 1201788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 12021da177e4SLinus Torvalds { 1203963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12041da177e4SLinus Torvalds struct nfs_page *req; 12051da177e4SLinus Torvalds 1206a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 12071da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12081da177e4SLinus Torvalds 1209788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1210788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1211788e7a89STrond Myklebust return; 1212788e7a89STrond Myklebust 12131da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12141da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12151da177e4SLinus Torvalds nfs_list_remove_request(req); 1216612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 1217fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 122088be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 122188be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 12221da177e4SLinus Torvalds req->wb_bytes, 12231da177e4SLinus Torvalds (long long)req_offset(req)); 12241da177e4SLinus Torvalds if (task->tk_status < 0) { 12251da177e4SLinus Torvalds req->wb_context->error = task->tk_status; 12261da177e4SLinus Torvalds nfs_inode_remove_request(req); 12271da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 12281da177e4SLinus Torvalds goto next; 12291da177e4SLinus Torvalds } 12301da177e4SLinus Torvalds 12311da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 12321da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 12331da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 12341da177e4SLinus Torvalds /* We have a match */ 123544dd151dSTrond Myklebust /* Set the PG_uptodate flag */ 123644dd151dSTrond Myklebust nfs_mark_uptodate(req->wb_page, req->wb_pgbase, 123744dd151dSTrond Myklebust req->wb_bytes); 12381da177e4SLinus Torvalds nfs_inode_remove_request(req); 12391da177e4SLinus Torvalds dprintk(" OK\n"); 12401da177e4SLinus Torvalds goto next; 12411da177e4SLinus Torvalds } 12421da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 12431da177e4SLinus Torvalds dprintk(" mismatch\n"); 124461822ab5STrond Myklebust nfs_redirty_request(req); 12451da177e4SLinus Torvalds next: 12469fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12471da177e4SLinus Torvalds } 12481da177e4SLinus Torvalds } 1249788e7a89STrond Myklebust 1250788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1251788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1252788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1253788e7a89STrond Myklebust }; 12541da177e4SLinus Torvalds 12553da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 12561da177e4SLinus Torvalds { 12571da177e4SLinus Torvalds LIST_HEAD(head); 12587d46a49fSTrond Myklebust int res; 12591da177e4SLinus Torvalds 1260587142f8STrond Myklebust spin_lock(&inode->i_lock); 12613da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1262587142f8STrond Myklebust spin_unlock(&inode->i_lock); 12631da177e4SLinus Torvalds if (res) { 12647d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 12651da177e4SLinus Torvalds if (error < 0) 12661da177e4SLinus Torvalds return error; 12673da28eb1STrond Myklebust } 12681da177e4SLinus Torvalds return res; 12691da177e4SLinus Torvalds } 1270c63c7b05STrond Myklebust #else 1271c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1272c63c7b05STrond Myklebust { 1273c63c7b05STrond Myklebust return 0; 1274c63c7b05STrond Myklebust } 12751da177e4SLinus Torvalds #endif 12761da177e4SLinus Torvalds 12771c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 12781da177e4SLinus Torvalds { 12791c75950bSTrond Myklebust struct inode *inode = mapping->host; 1280ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 12811c75950bSTrond Myklebust unsigned int npages = 0; 1282c42de9ddSTrond Myklebust LIST_HEAD(head); 128370b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 12843f442547STrond Myklebust long pages, ret; 12851da177e4SLinus Torvalds 12861c75950bSTrond Myklebust /* FIXME */ 12871c75950bSTrond Myklebust if (wbc->range_cyclic) 12881c75950bSTrond Myklebust idx_start = 0; 12891c75950bSTrond Myklebust else { 12901c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 12911c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 12921c75950bSTrond Myklebust if (idx_end > idx_start) { 1293ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 12941c75950bSTrond Myklebust npages = l_npages; 12951c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1296ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 12971c75950bSTrond Myklebust npages = 0; 12981c75950bSTrond Myklebust } 12991c75950bSTrond Myklebust } 1300c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1301587142f8STrond Myklebust spin_lock(&inode->i_lock); 13021da177e4SLinus Torvalds do { 1303c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1304c42de9ddSTrond Myklebust if (ret != 0) 1305c42de9ddSTrond Myklebust continue; 1306c42de9ddSTrond Myklebust if (nocommit) 1307c42de9ddSTrond Myklebust break; 1308d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1309724c439cSTrond Myklebust if (pages == 0) 1310c42de9ddSTrond Myklebust break; 1311d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1312587142f8STrond Myklebust spin_unlock(&inode->i_lock); 131383715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1314e8e058e8STrond Myklebust ret = pages; 1315587142f8STrond Myklebust spin_lock(&inode->i_lock); 1316d2ccddf0STrond Myklebust continue; 1317d2ccddf0STrond Myklebust } 1318d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1319587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1320c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1321587142f8STrond Myklebust spin_lock(&inode->i_lock); 1322587142f8STrond Myklebust 1323c42de9ddSTrond Myklebust } while (ret >= 0); 1324587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1325c42de9ddSTrond Myklebust return ret; 13261da177e4SLinus Torvalds } 13271da177e4SLinus Torvalds 13281c75950bSTrond Myklebust /* 13291c75950bSTrond Myklebust * flush the inode to disk. 13301c75950bSTrond Myklebust */ 13311c75950bSTrond Myklebust int nfs_wb_all(struct inode *inode) 13321c75950bSTrond Myklebust { 13331c75950bSTrond Myklebust struct address_space *mapping = inode->i_mapping; 13341c75950bSTrond Myklebust struct writeback_control wbc = { 13351c75950bSTrond Myklebust .bdi = mapping->backing_dev_info, 13361c75950bSTrond Myklebust .sync_mode = WB_SYNC_ALL, 13371c75950bSTrond Myklebust .nr_to_write = LONG_MAX, 133861822ab5STrond Myklebust .for_writepages = 1, 13391c75950bSTrond Myklebust .range_cyclic = 1, 13401c75950bSTrond Myklebust }; 13411c75950bSTrond Myklebust int ret; 13421c75950bSTrond Myklebust 1343c63c7b05STrond Myklebust ret = nfs_writepages(mapping, &wbc); 134461822ab5STrond Myklebust if (ret < 0) 134561822ab5STrond Myklebust goto out; 13461c75950bSTrond Myklebust ret = nfs_sync_mapping_wait(mapping, &wbc, 0); 13471c75950bSTrond Myklebust if (ret >= 0) 13481c75950bSTrond Myklebust return 0; 134961822ab5STrond Myklebust out: 1350e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 13511c75950bSTrond Myklebust return ret; 13521c75950bSTrond Myklebust } 13531c75950bSTrond Myklebust 13541c75950bSTrond Myklebust int nfs_sync_mapping_range(struct address_space *mapping, loff_t range_start, loff_t range_end, int how) 13551c75950bSTrond Myklebust { 13561c75950bSTrond Myklebust struct writeback_control wbc = { 13571c75950bSTrond Myklebust .bdi = mapping->backing_dev_info, 13581c75950bSTrond Myklebust .sync_mode = WB_SYNC_ALL, 13591c75950bSTrond Myklebust .nr_to_write = LONG_MAX, 13601c75950bSTrond Myklebust .range_start = range_start, 13611c75950bSTrond Myklebust .range_end = range_end, 136261822ab5STrond Myklebust .for_writepages = 1, 13631c75950bSTrond Myklebust }; 13641c75950bSTrond Myklebust int ret; 13651c75950bSTrond Myklebust 1366c63c7b05STrond Myklebust ret = nfs_writepages(mapping, &wbc); 136761822ab5STrond Myklebust if (ret < 0) 136861822ab5STrond Myklebust goto out; 13691c75950bSTrond Myklebust ret = nfs_sync_mapping_wait(mapping, &wbc, how); 13701c75950bSTrond Myklebust if (ret >= 0) 13711c75950bSTrond Myklebust return 0; 137261822ab5STrond Myklebust out: 1373e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 13741c75950bSTrond Myklebust return ret; 13751c75950bSTrond Myklebust } 13761c75950bSTrond Myklebust 13771b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 13781b3b4a1aSTrond Myklebust { 13791b3b4a1aSTrond Myklebust struct nfs_page *req; 13801b3b4a1aSTrond Myklebust loff_t range_start = page_offset(page); 13811b3b4a1aSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 13821b3b4a1aSTrond Myklebust struct writeback_control wbc = { 13831b3b4a1aSTrond Myklebust .bdi = page->mapping->backing_dev_info, 13841b3b4a1aSTrond Myklebust .sync_mode = WB_SYNC_ALL, 13851b3b4a1aSTrond Myklebust .nr_to_write = LONG_MAX, 13861b3b4a1aSTrond Myklebust .range_start = range_start, 13871b3b4a1aSTrond Myklebust .range_end = range_end, 13881b3b4a1aSTrond Myklebust }; 13891b3b4a1aSTrond Myklebust int ret = 0; 13901b3b4a1aSTrond Myklebust 13911b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 13921b3b4a1aSTrond Myklebust for (;;) { 13931b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 13941b3b4a1aSTrond Myklebust if (req == NULL) 13951b3b4a1aSTrond Myklebust goto out; 13961b3b4a1aSTrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 13971b3b4a1aSTrond Myklebust nfs_release_request(req); 13981b3b4a1aSTrond Myklebust break; 13991b3b4a1aSTrond Myklebust } 14001b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 14011b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 14021b3b4a1aSTrond Myklebust /* 14031b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 14041b3b4a1aSTrond Myklebust * page as being dirty 14051b3b4a1aSTrond Myklebust */ 14061b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 14071b3b4a1aSTrond Myklebust nfs_unlock_request(req); 14081b3b4a1aSTrond Myklebust break; 14091b3b4a1aSTrond Myklebust } 14101b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 14111b3b4a1aSTrond Myklebust if (ret < 0) 14121b3b4a1aSTrond Myklebust goto out; 14131b3b4a1aSTrond Myklebust } 14141b3b4a1aSTrond Myklebust if (!PagePrivate(page)) 14151b3b4a1aSTrond Myklebust return 0; 14161b3b4a1aSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); 14171b3b4a1aSTrond Myklebust out: 14181b3b4a1aSTrond Myklebust return ret; 14191b3b4a1aSTrond Myklebust } 14201b3b4a1aSTrond Myklebust 142161822ab5STrond Myklebust int nfs_wb_page_priority(struct inode *inode, struct page *page, int how) 14221c75950bSTrond Myklebust { 14231c75950bSTrond Myklebust loff_t range_start = page_offset(page); 14241c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14254d770ccfSTrond Myklebust struct writeback_control wbc = { 14264d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14274d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14284d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 14294d770ccfSTrond Myklebust .range_start = range_start, 14304d770ccfSTrond Myklebust .range_end = range_end, 14314d770ccfSTrond Myklebust }; 14324d770ccfSTrond Myklebust int ret; 14331c75950bSTrond Myklebust 14344d770ccfSTrond Myklebust BUG_ON(!PageLocked(page)); 1435c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 14364d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 14374d770ccfSTrond Myklebust if (ret < 0) 14384d770ccfSTrond Myklebust goto out; 14394d770ccfSTrond Myklebust } 1440f40313acSTrond Myklebust if (!PagePrivate(page)) 1441f40313acSTrond Myklebust return 0; 14424d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 14434d770ccfSTrond Myklebust if (ret >= 0) 14444d770ccfSTrond Myklebust return 0; 14454d770ccfSTrond Myklebust out: 1446e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 14474d770ccfSTrond Myklebust return ret; 14481c75950bSTrond Myklebust } 14491c75950bSTrond Myklebust 14501c75950bSTrond Myklebust /* 14511c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 14521c75950bSTrond Myklebust */ 14531c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 14541c75950bSTrond Myklebust { 14554d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 14561c75950bSTrond Myklebust } 14571c75950bSTrond Myklebust 1458f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 14591da177e4SLinus Torvalds { 14601da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 14611da177e4SLinus Torvalds sizeof(struct nfs_write_data), 14621da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 146320c2df83SPaul Mundt NULL); 14641da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 14651da177e4SLinus Torvalds return -ENOMEM; 14661da177e4SLinus Torvalds 146793d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 14681da177e4SLinus Torvalds nfs_wdata_cachep); 14691da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 14701da177e4SLinus Torvalds return -ENOMEM; 14711da177e4SLinus Torvalds 147293d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 14731da177e4SLinus Torvalds nfs_wdata_cachep); 14741da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 14751da177e4SLinus Torvalds return -ENOMEM; 14761da177e4SLinus Torvalds 147789a09141SPeter Zijlstra /* 147889a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 147989a09141SPeter Zijlstra * 148089a09141SPeter Zijlstra * 64MB: 8192k 148189a09141SPeter Zijlstra * 128MB: 11585k 148289a09141SPeter Zijlstra * 256MB: 16384k 148389a09141SPeter Zijlstra * 512MB: 23170k 148489a09141SPeter Zijlstra * 1GB: 32768k 148589a09141SPeter Zijlstra * 2GB: 46340k 148689a09141SPeter Zijlstra * 4GB: 65536k 148789a09141SPeter Zijlstra * 8GB: 92681k 148889a09141SPeter Zijlstra * 16GB: 131072k 148989a09141SPeter Zijlstra * 149089a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 149189a09141SPeter Zijlstra * Limit the default to 256M 149289a09141SPeter Zijlstra */ 149389a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 149489a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 149589a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 149689a09141SPeter Zijlstra 14971da177e4SLinus Torvalds return 0; 14981da177e4SLinus Torvalds } 14991da177e4SLinus Torvalds 1500266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15011da177e4SLinus Torvalds { 15021da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15031da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15041a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15051da177e4SLinus Torvalds } 15061da177e4SLinus Torvalds 1507