11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 181da177e4SLinus Torvalds #include <linux/nfs_fs.h> 191da177e4SLinus Torvalds #include <linux/nfs_mount.h> 201da177e4SLinus Torvalds #include <linux/nfs_page.h> 213fcfab16SAndrew Morton #include <linux/backing-dev.h> 223fcfab16SAndrew Morton 231da177e4SLinus Torvalds #include <asm/uaccess.h> 241da177e4SLinus Torvalds 251da177e4SLinus Torvalds #include "delegation.h" 2649a70f27STrond Myklebust #include "internal.h" 2791d5b470SChuck Lever #include "iostat.h" 281da177e4SLinus Torvalds 291da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 321da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds /* 351da177e4SLinus Torvalds * Local function declarations 361da177e4SLinus Torvalds */ 371da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context*, 381da177e4SLinus Torvalds struct page *, 391da177e4SLinus Torvalds unsigned int, unsigned int); 40c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, 41c63c7b05STrond Myklebust struct inode *inode, int ioflags); 42f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 43788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 44788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 461da177e4SLinus Torvalds 47e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 483feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 491da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 501da177e4SLinus Torvalds 51c9d8f89dSTrond Myklebust struct nfs_write_data *nfs_commitdata_alloc(void) 521da177e4SLinus Torvalds { 53e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5440859d7eSChuck Lever 551da177e4SLinus Torvalds if (p) { 561da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 571da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 581da177e4SLinus Torvalds } 591da177e4SLinus Torvalds return p; 601da177e4SLinus Torvalds } 611da177e4SLinus Torvalds 625e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 631da177e4SLinus Torvalds { 6440859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6540859d7eSChuck Lever kfree(p->pagevec); 661da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 671da177e4SLinus Torvalds } 681da177e4SLinus Torvalds 698d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 703feb2d49STrond Myklebust { 71e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 723feb2d49STrond Myklebust 733feb2d49STrond Myklebust if (p) { 743feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 753feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 76e9f7bee1STrond Myklebust p->npages = pagecount; 770d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 780d0b5cb3SChuck Lever p->pagevec = p->page_array; 793feb2d49STrond Myklebust else { 800d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 810d0b5cb3SChuck Lever if (!p->pagevec) { 823feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 833feb2d49STrond Myklebust p = NULL; 843feb2d49STrond Myklebust } 853feb2d49STrond Myklebust } 863feb2d49STrond Myklebust } 873feb2d49STrond Myklebust return p; 883feb2d49STrond Myklebust } 893feb2d49STrond Myklebust 905e4424afSTrond Myklebust static void nfs_writedata_free(struct nfs_write_data *p) 913feb2d49STrond Myklebust { 923feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 933feb2d49STrond Myklebust kfree(p->pagevec); 943feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 953feb2d49STrond Myklebust } 963feb2d49STrond Myklebust 97383ba719STrond Myklebust void nfs_writedata_release(void *data) 981da177e4SLinus Torvalds { 99383ba719STrond Myklebust struct nfs_write_data *wdata = data; 100383ba719STrond Myklebust 101383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 1021da177e4SLinus Torvalds nfs_writedata_free(wdata); 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds 1057b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1067b159fc1STrond Myklebust { 1077b159fc1STrond Myklebust ctx->error = error; 1087b159fc1STrond Myklebust smp_wmb(); 1097b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1107b159fc1STrond Myklebust } 1117b159fc1STrond Myklebust 112277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 113277459d2STrond Myklebust { 114277459d2STrond Myklebust struct nfs_page *req = NULL; 115277459d2STrond Myklebust 116277459d2STrond Myklebust if (PagePrivate(page)) { 117277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 118277459d2STrond Myklebust if (req != NULL) 119c03b4024STrond Myklebust kref_get(&req->wb_kref); 120277459d2STrond Myklebust } 121277459d2STrond Myklebust return req; 122277459d2STrond Myklebust } 123277459d2STrond Myklebust 124277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 125277459d2STrond Myklebust { 126587142f8STrond Myklebust struct inode *inode = page->mapping->host; 127277459d2STrond Myklebust struct nfs_page *req = NULL; 128277459d2STrond Myklebust 129587142f8STrond Myklebust spin_lock(&inode->i_lock); 130277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 131587142f8STrond Myklebust spin_unlock(&inode->i_lock); 132277459d2STrond Myklebust return req; 133277459d2STrond Myklebust } 134277459d2STrond Myklebust 1351da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1361da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1371da177e4SLinus Torvalds { 1381da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1391da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 140ca52fec1STrond Myklebust pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1411da177e4SLinus Torvalds 1421da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1431da177e4SLinus Torvalds return; 1441da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1451da177e4SLinus Torvalds if (i_size >= end) 1461da177e4SLinus Torvalds return; 14791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1481da177e4SLinus Torvalds i_size_write(inode, end); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 152a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 153a301b777STrond Myklebust { 154a301b777STrond Myklebust SetPageError(page); 155a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 156a301b777STrond Myklebust } 157a301b777STrond Myklebust 1581da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1591da177e4SLinus Torvalds * covers the full page. 1601da177e4SLinus Torvalds */ 1611da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1621da177e4SLinus Torvalds { 1631da177e4SLinus Torvalds if (PageUptodate(page)) 1641da177e4SLinus Torvalds return; 1651da177e4SLinus Torvalds if (base != 0) 1661da177e4SLinus Torvalds return; 16749a70f27STrond Myklebust if (count != nfs_page_length(page)) 1681da177e4SLinus Torvalds return; 1691da177e4SLinus Torvalds SetPageUptodate(page); 1701da177e4SLinus Torvalds } 1711da177e4SLinus Torvalds 172e21195a7STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1731da177e4SLinus Torvalds unsigned int offset, unsigned int count) 1741da177e4SLinus Torvalds { 1751da177e4SLinus Torvalds struct nfs_page *req; 176e21195a7STrond Myklebust int ret; 1771da177e4SLinus Torvalds 178e21195a7STrond Myklebust for (;;) { 179e21195a7STrond Myklebust req = nfs_update_request(ctx, page, offset, count); 180e21195a7STrond Myklebust if (!IS_ERR(req)) 181e21195a7STrond Myklebust break; 182e21195a7STrond Myklebust ret = PTR_ERR(req); 183e21195a7STrond Myklebust if (ret != -EBUSY) 184e21195a7STrond Myklebust return ret; 185e21195a7STrond Myklebust ret = nfs_wb_page(page->mapping->host, page); 186e21195a7STrond Myklebust if (ret != 0) 187e21195a7STrond Myklebust return ret; 188e21195a7STrond Myklebust } 1891da177e4SLinus Torvalds /* Update file length */ 1901da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 1917e5f6146STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 192acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 193abd3e641STrond Myklebust return 0; 1941da177e4SLinus Torvalds } 1951da177e4SLinus Torvalds 1961da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1971da177e4SLinus Torvalds { 1981da177e4SLinus Torvalds if (wbc->for_reclaim) 199c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 2001da177e4SLinus Torvalds if (wbc->for_kupdate) 2011da177e4SLinus Torvalds return FLUSH_LOWPRI; 2021da177e4SLinus Torvalds return 0; 2031da177e4SLinus Torvalds } 2041da177e4SLinus Torvalds 2051da177e4SLinus Torvalds /* 20689a09141SPeter Zijlstra * NFS congestion control 20789a09141SPeter Zijlstra */ 20889a09141SPeter Zijlstra 20989a09141SPeter Zijlstra int nfs_congestion_kb; 21089a09141SPeter Zijlstra 21189a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 21289a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 21389a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 21489a09141SPeter Zijlstra 2155a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 21689a09141SPeter Zijlstra { 2175a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2185a6d41b3STrond Myklebust 2195a6d41b3STrond Myklebust if (!ret) { 22089a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 22189a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 22289a09141SPeter Zijlstra 223277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 22489a09141SPeter Zijlstra NFS_CONGESTION_ON_THRESH) 22589a09141SPeter Zijlstra set_bdi_congested(&nfss->backing_dev_info, WRITE); 22689a09141SPeter Zijlstra } 2275a6d41b3STrond Myklebust return ret; 22889a09141SPeter Zijlstra } 22989a09141SPeter Zijlstra 23089a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 23189a09141SPeter Zijlstra { 23289a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 23389a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 23489a09141SPeter Zijlstra 23589a09141SPeter Zijlstra end_page_writeback(page); 236c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 23789a09141SPeter Zijlstra clear_bdi_congested(&nfss->backing_dev_info, WRITE); 23889a09141SPeter Zijlstra } 23989a09141SPeter Zijlstra 24089a09141SPeter Zijlstra /* 241e261f51fSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 2429cccef95STrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 243e261f51fSTrond Myklebust */ 244c63c7b05STrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 245c63c7b05STrond Myklebust struct page *page) 246e261f51fSTrond Myklebust { 247587142f8STrond Myklebust struct inode *inode = page->mapping->host; 248e261f51fSTrond Myklebust struct nfs_page *req; 249e261f51fSTrond Myklebust int ret; 250e261f51fSTrond Myklebust 251587142f8STrond Myklebust spin_lock(&inode->i_lock); 252e261f51fSTrond Myklebust for(;;) { 253e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 254e261f51fSTrond Myklebust if (req == NULL) { 255587142f8STrond Myklebust spin_unlock(&inode->i_lock); 2569cccef95STrond Myklebust return 0; 257e261f51fSTrond Myklebust } 258acee478aSTrond Myklebust if (nfs_set_page_tag_locked(req)) 259e261f51fSTrond Myklebust break; 260e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 261acee478aSTrond Myklebust * then the call to nfs_set_page_tag_locked() will always 262e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 263e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 264e261f51fSTrond Myklebust */ 265587142f8STrond Myklebust spin_unlock(&inode->i_lock); 266e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 267e261f51fSTrond Myklebust nfs_release_request(req); 268e261f51fSTrond Myklebust if (ret != 0) 269e261f51fSTrond Myklebust return ret; 270587142f8STrond Myklebust spin_lock(&inode->i_lock); 271e261f51fSTrond Myklebust } 272612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 273612c9384STrond Myklebust /* This request is marked for commit */ 274587142f8STrond Myklebust spin_unlock(&inode->i_lock); 275acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 276c63c7b05STrond Myklebust nfs_pageio_complete(pgio); 2779cccef95STrond Myklebust return 0; 278612c9384STrond Myklebust } 279c63c7b05STrond Myklebust if (nfs_set_page_writeback(page) != 0) { 280587142f8STrond Myklebust spin_unlock(&inode->i_lock); 281c63c7b05STrond Myklebust BUG(); 282c63c7b05STrond Myklebust } 283587142f8STrond Myklebust spin_unlock(&inode->i_lock); 284f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 285f8512ad0SFred Isaman nfs_redirty_request(req); 286f8512ad0SFred Isaman return pgio->pg_error; 287f8512ad0SFred Isaman } 2889cccef95STrond Myklebust return 0; 289e261f51fSTrond Myklebust } 290e261f51fSTrond Myklebust 291f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 292f758c885STrond Myklebust { 293f758c885STrond Myklebust struct inode *inode = page->mapping->host; 294f758c885STrond Myklebust 295f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 296f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 297f758c885STrond Myklebust 298f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 299f758c885STrond Myklebust return nfs_page_async_flush(pgio, page); 300f758c885STrond Myklebust } 301f758c885STrond Myklebust 302e261f51fSTrond Myklebust /* 3031da177e4SLinus Torvalds * Write an mmapped page to the server. 3041da177e4SLinus Torvalds */ 3054d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3061da177e4SLinus Torvalds { 307f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 308e261f51fSTrond Myklebust int err; 3091da177e4SLinus Torvalds 310f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 311f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 312f758c885STrond Myklebust nfs_pageio_complete(&pgio); 313f758c885STrond Myklebust if (err < 0) 3144d770ccfSTrond Myklebust return err; 315f758c885STrond Myklebust if (pgio.pg_error < 0) 316f758c885STrond Myklebust return pgio.pg_error; 317f758c885STrond Myklebust return 0; 3184d770ccfSTrond Myklebust } 3194d770ccfSTrond Myklebust 3204d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3214d770ccfSTrond Myklebust { 322f758c885STrond Myklebust int ret; 3234d770ccfSTrond Myklebust 324f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3251da177e4SLinus Torvalds unlock_page(page); 326f758c885STrond Myklebust return ret; 327f758c885STrond Myklebust } 328f758c885STrond Myklebust 329f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 330f758c885STrond Myklebust { 331f758c885STrond Myklebust int ret; 332f758c885STrond Myklebust 333f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 334f758c885STrond Myklebust unlock_page(page); 335f758c885STrond Myklebust return ret; 3361da177e4SLinus Torvalds } 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3391da177e4SLinus Torvalds { 3401da177e4SLinus Torvalds struct inode *inode = mapping->host; 341c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3421da177e4SLinus Torvalds int err; 3431da177e4SLinus Torvalds 34491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 34591d5b470SChuck Lever 346c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 347f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 348c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 349f758c885STrond Myklebust if (err < 0) 3501da177e4SLinus Torvalds return err; 351f758c885STrond Myklebust if (pgio.pg_error < 0) 352c63c7b05STrond Myklebust return pgio.pg_error; 353c63c7b05STrond Myklebust return 0; 3541da177e4SLinus Torvalds } 3551da177e4SLinus Torvalds 3561da177e4SLinus Torvalds /* 3571da177e4SLinus Torvalds * Insert a write request into an inode 3581da177e4SLinus Torvalds */ 35927852596SNick Piggin static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3601da177e4SLinus Torvalds { 3611da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3621da177e4SLinus Torvalds int error; 3631da177e4SLinus Torvalds 3641da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 36527852596SNick Piggin BUG_ON(error); 3661da177e4SLinus Torvalds if (!nfsi->npages) { 3671da177e4SLinus Torvalds igrab(inode); 3681da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3691da177e4SLinus Torvalds nfsi->change_attr++; 3701da177e4SLinus Torvalds } 371deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 372277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3731da177e4SLinus Torvalds nfsi->npages++; 374c03b4024STrond Myklebust kref_get(&req->wb_kref); 37527852596SNick Piggin radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, 37627852596SNick Piggin NFS_PAGE_TAG_LOCKED); 3771da177e4SLinus Torvalds } 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds /* 38089a09141SPeter Zijlstra * Remove a write request from an inode 3811da177e4SLinus Torvalds */ 3821da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 3831da177e4SLinus Torvalds { 38488be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 3851da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3861da177e4SLinus Torvalds 3871da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 3881da177e4SLinus Torvalds 389587142f8STrond Myklebust spin_lock(&inode->i_lock); 390277459d2STrond Myklebust set_page_private(req->wb_page, 0); 391deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 3921da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 3931da177e4SLinus Torvalds nfsi->npages--; 3941da177e4SLinus Torvalds if (!nfsi->npages) { 395587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3961da177e4SLinus Torvalds iput(inode); 3971da177e4SLinus Torvalds } else 398587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3991da177e4SLinus Torvalds nfs_clear_request(req); 4001da177e4SLinus Torvalds nfs_release_request(req); 4011da177e4SLinus Torvalds } 4021da177e4SLinus Torvalds 40361822ab5STrond Myklebust static void 4046d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 40561822ab5STrond Myklebust { 40661822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 40761822ab5STrond Myklebust } 40861822ab5STrond Myklebust 4091da177e4SLinus Torvalds /* 4101da177e4SLinus Torvalds * Check if a request is dirty 4111da177e4SLinus Torvalds */ 4121da177e4SLinus Torvalds static inline int 4131da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 4141da177e4SLinus Torvalds { 4155a6d41b3STrond Myklebust struct page *page = req->wb_page; 4165a6d41b3STrond Myklebust 417612c9384STrond Myklebust if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags)) 4185a6d41b3STrond Myklebust return 0; 41938def50fSFred Isaman return !PageWriteback(page); 4201da177e4SLinus Torvalds } 4211da177e4SLinus Torvalds 4221da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4231da177e4SLinus Torvalds /* 4241da177e4SLinus Torvalds * Add a request to the inode's commit list. 4251da177e4SLinus Torvalds */ 4261da177e4SLinus Torvalds static void 4271da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4281da177e4SLinus Torvalds { 42988be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4301da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4311da177e4SLinus Torvalds 432587142f8STrond Myklebust spin_lock(&inode->i_lock); 4331da177e4SLinus Torvalds nfsi->ncommit++; 434612c9384STrond Myklebust set_bit(PG_NEED_COMMIT, &(req)->wb_flags); 4355c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4365c369683STrond Myklebust req->wb_index, 4375c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 438587142f8STrond Myklebust spin_unlock(&inode->i_lock); 439fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 440c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 441a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4421da177e4SLinus Torvalds } 4438e821cadSTrond Myklebust 4448e821cadSTrond Myklebust static inline 4458e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4468e821cadSTrond Myklebust { 4478e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4488e821cadSTrond Myklebust } 4498e821cadSTrond Myklebust 4508e821cadSTrond Myklebust static inline 4518e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4528e821cadSTrond Myklebust { 453612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4548e821cadSTrond Myklebust nfs_mark_request_commit(req); 4558e821cadSTrond Myklebust return 1; 4568e821cadSTrond Myklebust } 4578e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4586d884e8fSFred nfs_mark_request_dirty(req); 4598e821cadSTrond Myklebust return 1; 4608e821cadSTrond Myklebust } 4618e821cadSTrond Myklebust return 0; 4628e821cadSTrond Myklebust } 4638e821cadSTrond Myklebust #else 4648e821cadSTrond Myklebust static inline void 4658e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4668e821cadSTrond Myklebust { 4678e821cadSTrond Myklebust } 4688e821cadSTrond Myklebust 4698e821cadSTrond Myklebust static inline 4708e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4718e821cadSTrond Myklebust { 4728e821cadSTrond Myklebust return 0; 4738e821cadSTrond Myklebust } 4748e821cadSTrond Myklebust 4758e821cadSTrond Myklebust static inline 4768e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4778e821cadSTrond Myklebust { 4788e821cadSTrond Myklebust return 0; 4798e821cadSTrond Myklebust } 4801da177e4SLinus Torvalds #endif 4811da177e4SLinus Torvalds 4821da177e4SLinus Torvalds /* 4831da177e4SLinus Torvalds * Wait for a request to complete. 4841da177e4SLinus Torvalds * 485150030b7SMatthew Wilcox * Interruptible by fatal signals only. 4861da177e4SLinus Torvalds */ 487ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) 4881da177e4SLinus Torvalds { 4891da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4901da177e4SLinus Torvalds struct nfs_page *req; 491ca52fec1STrond Myklebust pgoff_t idx_end, next; 4921da177e4SLinus Torvalds unsigned int res = 0; 4931da177e4SLinus Torvalds int error; 4941da177e4SLinus Torvalds 4951da177e4SLinus Torvalds if (npages == 0) 4961da177e4SLinus Torvalds idx_end = ~0; 4971da177e4SLinus Torvalds else 4981da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 4991da177e4SLinus Torvalds 5001da177e4SLinus Torvalds next = idx_start; 5019fd367f0STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { 5021da177e4SLinus Torvalds if (req->wb_index > idx_end) 5031da177e4SLinus Torvalds break; 5041da177e4SLinus Torvalds 5051da177e4SLinus Torvalds next = req->wb_index + 1; 506c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5071da177e4SLinus Torvalds 508c03b4024STrond Myklebust kref_get(&req->wb_kref); 509587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5101da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5111da177e4SLinus Torvalds nfs_release_request(req); 512587142f8STrond Myklebust spin_lock(&inode->i_lock); 5131da177e4SLinus Torvalds if (error < 0) 5141da177e4SLinus Torvalds return error; 5151da177e4SLinus Torvalds res++; 5161da177e4SLinus Torvalds } 5171da177e4SLinus Torvalds return res; 5181da177e4SLinus Torvalds } 5191da177e4SLinus Torvalds 52083715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 52183715ad5STrond Myklebust { 52283715ad5STrond Myklebust struct nfs_page *req; 52383715ad5STrond Myklebust 52483715ad5STrond Myklebust while(!list_empty(head)) { 52583715ad5STrond Myklebust req = nfs_list_entry(head->next); 526b6dff26aSTrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 527c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 528c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 52983715ad5STrond Myklebust nfs_list_remove_request(req); 530612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 53183715ad5STrond Myklebust nfs_inode_remove_request(req); 532b6dff26aSTrond Myklebust nfs_unlock_request(req); 53383715ad5STrond Myklebust } 53483715ad5STrond Myklebust } 53583715ad5STrond Myklebust 5361da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5371da177e4SLinus Torvalds /* 5381da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5391da177e4SLinus Torvalds * @inode: NFS inode to scan 5401da177e4SLinus Torvalds * @dst: destination list 5411da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5421da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5431da177e4SLinus Torvalds * 5441da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5451da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5461da177e4SLinus Torvalds */ 5471da177e4SLinus Torvalds static int 548ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5491da177e4SLinus Torvalds { 5501da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5513da28eb1STrond Myklebust int res = 0; 5523da28eb1STrond Myklebust 5533da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 5545c369683STrond Myklebust res = nfs_scan_list(nfsi, dst, idx_start, npages, 5555c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 5561da177e4SLinus Torvalds nfsi->ncommit -= res; 5573da28eb1STrond Myklebust } 5581da177e4SLinus Torvalds return res; 5591da177e4SLinus Torvalds } 560c42de9ddSTrond Myklebust #else 561ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 562c42de9ddSTrond Myklebust { 563c42de9ddSTrond Myklebust return 0; 564c42de9ddSTrond Myklebust } 5651da177e4SLinus Torvalds #endif 5661da177e4SLinus Torvalds 5671da177e4SLinus Torvalds /* 5681da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 5691da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 5701da177e4SLinus Torvalds * the calling process. 5711da177e4SLinus Torvalds * 5721da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 5731da177e4SLinus Torvalds */ 5741da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 575e21195a7STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 5761da177e4SLinus Torvalds { 57789a09141SPeter Zijlstra struct address_space *mapping = page->mapping; 57889a09141SPeter Zijlstra struct inode *inode = mapping->host; 5791da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 580ca52fec1STrond Myklebust pgoff_t rqend, end; 5811da177e4SLinus Torvalds 5821da177e4SLinus Torvalds end = offset + bytes; 5831da177e4SLinus Torvalds 5841da177e4SLinus Torvalds for (;;) { 5851da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 5861da177e4SLinus Torvalds * A request for the page we wish to update 5871da177e4SLinus Torvalds */ 588587142f8STrond Myklebust spin_lock(&inode->i_lock); 589277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 5901da177e4SLinus Torvalds if (req) { 591acee478aSTrond Myklebust if (!nfs_set_page_tag_locked(req)) { 5921da177e4SLinus Torvalds int error; 593277459d2STrond Myklebust 594587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5951da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5961da177e4SLinus Torvalds nfs_release_request(req); 5971dd594b2SNeil Brown if (error < 0) { 59827852596SNick Piggin if (new) { 59927852596SNick Piggin radix_tree_preload_end(); 6001dd594b2SNeil Brown nfs_release_request(new); 60127852596SNick Piggin } 6021da177e4SLinus Torvalds return ERR_PTR(error); 6031dd594b2SNeil Brown } 6041da177e4SLinus Torvalds continue; 6051da177e4SLinus Torvalds } 606587142f8STrond Myklebust spin_unlock(&inode->i_lock); 60727852596SNick Piggin if (new) { 60827852596SNick Piggin radix_tree_preload_end(); 6091da177e4SLinus Torvalds nfs_release_request(new); 61027852596SNick Piggin } 6111da177e4SLinus Torvalds break; 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds 6141da177e4SLinus Torvalds if (new) { 6151da177e4SLinus Torvalds nfs_lock_request_dontget(new); 61627852596SNick Piggin nfs_inode_add_request(inode, new); 617587142f8STrond Myklebust spin_unlock(&inode->i_lock); 61827852596SNick Piggin radix_tree_preload_end(); 61961e930a9STrond Myklebust req = new; 620efc91ed0STrond Myklebust goto out; 6211da177e4SLinus Torvalds } 622587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 6251da177e4SLinus Torvalds if (IS_ERR(new)) 6261da177e4SLinus Torvalds return new; 627f3d47a3aSTrond Myklebust if (radix_tree_preload(GFP_NOFS)) { 628f3d47a3aSTrond Myklebust nfs_release_request(new); 629f3d47a3aSTrond Myklebust return ERR_PTR(-ENOMEM); 630f3d47a3aSTrond Myklebust } 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) { 644acee478aSTrond Myklebust nfs_clear_page_tag_locked(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; 65261e930a9STrond Myklebust req->wb_bytes = max(end, rqend) - req->wb_offset; 653efc91ed0STrond Myklebust goto out; 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds 6561da177e4SLinus Torvalds if (end > rqend) 6571da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 6581da177e4SLinus Torvalds 659efc91ed0STrond Myklebust out: 66061e930a9STrond Myklebust return req; 6611da177e4SLinus Torvalds } 6621da177e4SLinus Torvalds 6631da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6641da177e4SLinus Torvalds { 665cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 6661da177e4SLinus Torvalds struct nfs_page *req; 6671a54533eSTrond Myklebust int do_flush, status; 6681da177e4SLinus Torvalds /* 6691da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6701da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6711da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 6721da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 6731da177e4SLinus Torvalds * Also do the same if we find a request from an existing 6741da177e4SLinus Torvalds * dropped page. 6751da177e4SLinus Torvalds */ 6761a54533eSTrond Myklebust do { 677277459d2STrond Myklebust req = nfs_page_find_request(page); 6781a54533eSTrond Myklebust if (req == NULL) 6791a54533eSTrond Myklebust return 0; 6801a54533eSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx 681e261f51fSTrond Myklebust || !nfs_dirty_request(req); 6821da177e4SLinus Torvalds nfs_release_request(req); 6831a54533eSTrond Myklebust if (!do_flush) 6841a54533eSTrond Myklebust return 0; 685277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 6861a54533eSTrond Myklebust } while (status == 0); 6871a54533eSTrond Myklebust return status; 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds /* 6915d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 6925d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 6935d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 6945d47a356STrond Myklebust */ 6955d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 6965d47a356STrond Myklebust { 6975d47a356STrond Myklebust return PageUptodate(page) && 6985d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 6995d47a356STrond Myklebust } 7005d47a356STrond Myklebust 7015d47a356STrond Myklebust /* 7021da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7031da177e4SLinus Torvalds * 7041da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7051da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7061da177e4SLinus Torvalds */ 7071da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7081da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7091da177e4SLinus Torvalds { 710cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7111da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7121da177e4SLinus Torvalds int status = 0; 7131da177e4SLinus Torvalds 71491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 71591d5b470SChuck Lever 716*48186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 71701cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 71801cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7190bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 7201da177e4SLinus Torvalds 7211da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7225d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7235d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7245d47a356STrond Myklebust * inefficiencies. 7251da177e4SLinus Torvalds */ 7265d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7275d47a356STrond Myklebust inode->i_flock == NULL && 7284b5621f6STrond Myklebust !(file->f_flags & O_SYNC)) { 72949a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7301da177e4SLinus Torvalds offset = 0; 7311da177e4SLinus Torvalds } 7321da177e4SLinus Torvalds 733e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 73403fa9e84STrond Myklebust if (status < 0) 73503fa9e84STrond Myklebust nfs_set_pageerror(page); 73603fa9e84STrond Myklebust else 737e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7381da177e4SLinus Torvalds 739*48186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 7401da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7411da177e4SLinus Torvalds return status; 7421da177e4SLinus Torvalds } 7431da177e4SLinus Torvalds 7441da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7451da177e4SLinus Torvalds { 7468e821cadSTrond Myklebust 7477e5f6146STrond Myklebust if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) { 74889a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7491da177e4SLinus Torvalds nfs_inode_remove_request(req); 7508e821cadSTrond Myklebust } else 7518e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 7529fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 7531da177e4SLinus Torvalds } 7541da177e4SLinus Torvalds 7553ff7576dSTrond Myklebust static int flush_task_priority(int how) 7561da177e4SLinus Torvalds { 7571da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7581da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7591da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7601da177e4SLinus Torvalds case FLUSH_LOWPRI: 7611da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7621da177e4SLinus Torvalds } 7631da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7641da177e4SLinus Torvalds } 7651da177e4SLinus Torvalds 7661da177e4SLinus Torvalds /* 7671da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 7681da177e4SLinus Torvalds */ 769dbae4c73STrond Myklebust static int nfs_write_rpcsetup(struct nfs_page *req, 7701da177e4SLinus Torvalds struct nfs_write_data *data, 771788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7721da177e4SLinus Torvalds unsigned int count, unsigned int offset, 7731da177e4SLinus Torvalds int how) 7741da177e4SLinus Torvalds { 77584115e1cSTrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 77684115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 7773ff7576dSTrond Myklebust int priority = flush_task_priority(how); 77807737691STrond Myklebust struct rpc_task *task; 779bdc7f021STrond Myklebust struct rpc_message msg = { 780bdc7f021STrond Myklebust .rpc_argp = &data->args, 781bdc7f021STrond Myklebust .rpc_resp = &data->res, 782bdc7f021STrond Myklebust .rpc_cred = req->wb_context->cred, 783bdc7f021STrond Myklebust }; 78484115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 78584115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 78607737691STrond Myklebust .task = &data->task, 787bdc7f021STrond Myklebust .rpc_message = &msg, 78884115e1cSTrond Myklebust .callback_ops = call_ops, 78984115e1cSTrond Myklebust .callback_data = data, 790101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 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; 807383ba719STrond Myklebust data->args.context = get_nfs_open_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 " 824*48186c7dSChuck Lever "(req %s/%lld, %u bytes @ offset %llu)\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); 832dbae4c73STrond Myklebust if (IS_ERR(task)) 833dbae4c73STrond Myklebust return PTR_ERR(task); 83407737691STrond Myklebust rpc_put_task(task); 835dbae4c73STrond Myklebust return 0; 8361da177e4SLinus Torvalds } 8371da177e4SLinus Torvalds 8386d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 8396d884e8fSFred * call this on each, which will prepare them to be retried on next 8406d884e8fSFred * writeback using standard nfs. 8416d884e8fSFred */ 8426d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 8436d884e8fSFred { 8446d884e8fSFred nfs_mark_request_dirty(req); 8456d884e8fSFred nfs_end_page_writeback(req->wb_page); 8466d884e8fSFred nfs_clear_page_tag_locked(req); 8476d884e8fSFred } 8486d884e8fSFred 8491da177e4SLinus Torvalds /* 8501da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8511da177e4SLinus Torvalds * contiguous dirty area on one page. 8521da177e4SLinus Torvalds */ 8538d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8541da177e4SLinus Torvalds { 8551da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 8561da177e4SLinus Torvalds struct page *page = req->wb_page; 8571da177e4SLinus Torvalds struct nfs_write_data *data; 858e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 859e9f7bee1STrond Myklebust unsigned int offset; 8601da177e4SLinus Torvalds int requests = 0; 861dbae4c73STrond Myklebust int ret = 0; 8621da177e4SLinus Torvalds LIST_HEAD(list); 8631da177e4SLinus Torvalds 8641da177e4SLinus Torvalds nfs_list_remove_request(req); 8651da177e4SLinus Torvalds 866bcb71bbaSTrond Myklebust nbytes = count; 867e9f7bee1STrond Myklebust do { 868e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 869e9f7bee1STrond Myklebust 8708d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 8711da177e4SLinus Torvalds if (!data) 8721da177e4SLinus Torvalds goto out_bad; 8731da177e4SLinus Torvalds list_add(&data->pages, &list); 8741da177e4SLinus Torvalds requests++; 875e9f7bee1STrond Myklebust nbytes -= len; 876e9f7bee1STrond Myklebust } while (nbytes != 0); 8771da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 8781da177e4SLinus Torvalds 8791da177e4SLinus Torvalds ClearPageError(page); 8801da177e4SLinus Torvalds offset = 0; 881bcb71bbaSTrond Myklebust nbytes = count; 8821da177e4SLinus Torvalds do { 883dbae4c73STrond Myklebust int ret2; 884dbae4c73STrond Myklebust 8851da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8861da177e4SLinus Torvalds list_del_init(&data->pages); 8871da177e4SLinus Torvalds 8881da177e4SLinus Torvalds data->pagevec[0] = page; 8891da177e4SLinus Torvalds 890bcb71bbaSTrond Myklebust if (nbytes < wsize) 891bcb71bbaSTrond Myklebust wsize = nbytes; 892dbae4c73STrond Myklebust ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 893788e7a89STrond Myklebust wsize, offset, how); 894dbae4c73STrond Myklebust if (ret == 0) 895dbae4c73STrond Myklebust ret = ret2; 8961da177e4SLinus Torvalds offset += wsize; 8971da177e4SLinus Torvalds nbytes -= wsize; 8981da177e4SLinus Torvalds } while (nbytes != 0); 8991da177e4SLinus Torvalds 900dbae4c73STrond Myklebust return ret; 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds out_bad: 9031da177e4SLinus Torvalds while (!list_empty(&list)) { 9041da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9051da177e4SLinus Torvalds list_del(&data->pages); 9068aca67f0STrond Myklebust nfs_writedata_release(data); 9071da177e4SLinus Torvalds } 90861822ab5STrond Myklebust nfs_redirty_request(req); 9091da177e4SLinus Torvalds return -ENOMEM; 9101da177e4SLinus Torvalds } 9111da177e4SLinus Torvalds 9121da177e4SLinus Torvalds /* 9131da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9141da177e4SLinus Torvalds * The page must have been locked by the caller. 9151da177e4SLinus Torvalds * 9161da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9171da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9181da177e4SLinus Torvalds * that has been written but not committed. 9191da177e4SLinus Torvalds */ 9208d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9211da177e4SLinus Torvalds { 9221da177e4SLinus Torvalds struct nfs_page *req; 9231da177e4SLinus Torvalds struct page **pages; 9241da177e4SLinus Torvalds struct nfs_write_data *data; 9251da177e4SLinus Torvalds 9268d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 9271da177e4SLinus Torvalds if (!data) 9281da177e4SLinus Torvalds goto out_bad; 9291da177e4SLinus Torvalds 9301da177e4SLinus Torvalds pages = data->pagevec; 9311da177e4SLinus Torvalds while (!list_empty(head)) { 9321da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9331da177e4SLinus Torvalds nfs_list_remove_request(req); 9341da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9351da177e4SLinus Torvalds ClearPageError(req->wb_page); 9361da177e4SLinus Torvalds *pages++ = req->wb_page; 9371da177e4SLinus Torvalds } 9381da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9391da177e4SLinus Torvalds 9401da177e4SLinus Torvalds /* Set up the argument struct */ 941dbae4c73STrond Myklebust return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9421da177e4SLinus Torvalds out_bad: 9431da177e4SLinus Torvalds while (!list_empty(head)) { 94410afec90STrond Myklebust req = nfs_list_entry(head->next); 9451da177e4SLinus Torvalds nfs_list_remove_request(req); 94661822ab5STrond Myklebust nfs_redirty_request(req); 9471da177e4SLinus Torvalds } 9481da177e4SLinus Torvalds return -ENOMEM; 9491da177e4SLinus Torvalds } 9501da177e4SLinus Torvalds 951c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 952c63c7b05STrond Myklebust struct inode *inode, int ioflags) 9531da177e4SLinus Torvalds { 954bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 9551da177e4SLinus Torvalds 956bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 957c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 958bcb71bbaSTrond Myklebust else 959c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 9601da177e4SLinus Torvalds } 9611da177e4SLinus Torvalds 9621da177e4SLinus Torvalds /* 9631da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 9641da177e4SLinus Torvalds */ 965788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 9661da177e4SLinus Torvalds { 967788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 9681da177e4SLinus Torvalds 969*48186c7dSChuck Lever dprintk("NFS: %5u write(%s/%lld %d@%lld)", 970*48186c7dSChuck Lever task->tk_pid, 971*48186c7dSChuck Lever data->req->wb_context->path.dentry->d_inode->i_sb->s_id, 972*48186c7dSChuck Lever (long long) 973*48186c7dSChuck Lever NFS_FILEID(data->req->wb_context->path.dentry->d_inode), 974*48186c7dSChuck Lever data->req->wb_bytes, (long long)req_offset(data->req)); 9751da177e4SLinus Torvalds 976c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 977c9d8f89dSTrond Myklebust } 978788e7a89STrond Myklebust 979c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata) 980c9d8f89dSTrond Myklebust { 981c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 982c9d8f89dSTrond Myklebust struct nfs_page *req = data->req; 983c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 984c9d8f89dSTrond Myklebust int status = data->task.tk_status; 985c9d8f89dSTrond Myklebust 986c9d8f89dSTrond Myklebust if (status < 0) { 987a301b777STrond Myklebust nfs_set_pageerror(page); 988c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 989c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 9908e821cadSTrond Myklebust goto out; 9918e821cadSTrond Myklebust } 9928e821cadSTrond Myklebust 9938e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 994587142f8STrond Myklebust struct inode *inode = page->mapping->host; 9958e821cadSTrond Myklebust 996587142f8STrond Myklebust spin_lock(&inode->i_lock); 9978e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 9988e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 9998e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 10001da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10011da177e4SLinus Torvalds dprintk(" defer commit\n"); 10021da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10038e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 10048e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 10051da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10061da177e4SLinus Torvalds } 1007587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10081da177e4SLinus Torvalds } else 10091da177e4SLinus Torvalds dprintk(" OK\n"); 10101da177e4SLinus Torvalds 10118e821cadSTrond Myklebust out: 10121da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 10131da177e4SLinus Torvalds nfs_writepage_release(req); 1014c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 10151da177e4SLinus Torvalds } 10161da177e4SLinus Torvalds 1017788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1018788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1019c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_partial, 1020788e7a89STrond Myklebust }; 1021788e7a89STrond Myklebust 10221da177e4SLinus Torvalds /* 10231da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10241da177e4SLinus Torvalds * 10251da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10261da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10271da177e4SLinus Torvalds * as the page has a write request pending. 10281da177e4SLinus Torvalds */ 1029788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10301da177e4SLinus Torvalds { 1031788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10321da177e4SLinus Torvalds 1033c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1034c9d8f89dSTrond Myklebust } 1035c9d8f89dSTrond Myklebust 1036c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata) 1037c9d8f89dSTrond Myklebust { 1038c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1039c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1040788e7a89STrond Myklebust 10411da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 10421da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 1043c9d8f89dSTrond Myklebust struct nfs_page *req = nfs_list_entry(data->pages.next); 1044c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1045c9d8f89dSTrond Myklebust 10461da177e4SLinus Torvalds nfs_list_remove_request(req); 10471da177e4SLinus Torvalds 1048*48186c7dSChuck Lever dprintk("NFS: %5u write (%s/%lld %d@%lld)", 1049*48186c7dSChuck Lever data->task.tk_pid, 105088be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 105188be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 10521da177e4SLinus Torvalds req->wb_bytes, 10531da177e4SLinus Torvalds (long long)req_offset(req)); 10541da177e4SLinus Torvalds 1055c9d8f89dSTrond Myklebust if (status < 0) { 1056a301b777STrond Myklebust nfs_set_pageerror(page); 1057c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1058c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 10598e821cadSTrond Myklebust goto remove_request; 10601da177e4SLinus Torvalds } 10611da177e4SLinus Torvalds 10628e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 10631da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10641da177e4SLinus Torvalds nfs_mark_request_commit(req); 10658e821cadSTrond Myklebust nfs_end_page_writeback(page); 10661da177e4SLinus Torvalds dprintk(" marked for commit\n"); 10678e821cadSTrond Myklebust goto next; 10688e821cadSTrond Myklebust } 10698e821cadSTrond Myklebust dprintk(" OK\n"); 10708e821cadSTrond Myklebust remove_request: 10718e821cadSTrond Myklebust nfs_end_page_writeback(page); 10721da177e4SLinus Torvalds nfs_inode_remove_request(req); 10731da177e4SLinus Torvalds next: 10749fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 10751da177e4SLinus Torvalds } 1076c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 1079788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1080788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1081c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_full, 1082788e7a89STrond Myklebust }; 1083788e7a89STrond Myklebust 1084788e7a89STrond Myklebust 10851da177e4SLinus Torvalds /* 10861da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 10871da177e4SLinus Torvalds */ 1088462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 10891da177e4SLinus Torvalds { 10901da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 10911da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1092788e7a89STrond Myklebust int status; 10931da177e4SLinus Torvalds 1094a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 10951da177e4SLinus Torvalds task->tk_pid, task->tk_status); 10961da177e4SLinus Torvalds 1097f551e44fSChuck Lever /* 1098f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1099f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1100f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1101f551e44fSChuck Lever * another writer had changed the file, but some applications 1102f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1103f551e44fSChuck Lever */ 1104788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1105788e7a89STrond Myklebust if (status != 0) 1106788e7a89STrond Myklebust return status; 110791d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 110891d5b470SChuck Lever 11091da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11101da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 11111da177e4SLinus Torvalds /* We tried a write call, but the server did not 11121da177e4SLinus Torvalds * commit data to stable storage even though we 11131da177e4SLinus Torvalds * requested it. 11141da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11151da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11161da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11171da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11181da177e4SLinus Torvalds */ 11191da177e4SLinus Torvalds static unsigned long complain; 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11221da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11231da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 112454ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 11251da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11261da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11271da177e4SLinus Torvalds } 11281da177e4SLinus Torvalds } 11291da177e4SLinus Torvalds #endif 11301da177e4SLinus Torvalds /* Is this a short write? */ 11311da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11321da177e4SLinus Torvalds static unsigned long complain; 11331da177e4SLinus Torvalds 113491d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 113591d5b470SChuck Lever 11361da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11371da177e4SLinus Torvalds if (resp->count != 0) { 11381da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11391da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11401da177e4SLinus Torvalds /* Resend from where the server left off */ 11411da177e4SLinus Torvalds argp->offset += resp->count; 11421da177e4SLinus Torvalds argp->pgbase += resp->count; 11431da177e4SLinus Torvalds argp->count -= resp->count; 11441da177e4SLinus Torvalds } else { 11451da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11461da177e4SLinus Torvalds * headaches in the case of a server crash. 11471da177e4SLinus Torvalds */ 11481da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11491da177e4SLinus Torvalds } 11501da177e4SLinus Torvalds rpc_restart_call(task); 1151788e7a89STrond Myklebust return -EAGAIN; 11521da177e4SLinus Torvalds } 11531da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11541da177e4SLinus Torvalds printk(KERN_WARNING 11551da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 11561da177e4SLinus Torvalds argp->count); 11571da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11581da177e4SLinus Torvalds } 11591da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 11601da177e4SLinus Torvalds task->tk_status = -EIO; 11611da177e4SLinus Torvalds } 1162788e7a89STrond Myklebust return 0; 11631da177e4SLinus Torvalds } 11641da177e4SLinus Torvalds 11651da177e4SLinus Torvalds 11661da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1167c9d8f89dSTrond Myklebust void nfs_commitdata_release(void *data) 11681da177e4SLinus Torvalds { 1169383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1170383ba719STrond Myklebust 1171383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 11721da177e4SLinus Torvalds nfs_commit_free(wdata); 11731da177e4SLinus Torvalds } 11741da177e4SLinus Torvalds 11751da177e4SLinus Torvalds /* 11761da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 11771da177e4SLinus Torvalds */ 1178dbae4c73STrond Myklebust static int nfs_commit_rpcsetup(struct list_head *head, 1179788e7a89STrond Myklebust struct nfs_write_data *data, 1180788e7a89STrond Myklebust int how) 11811da177e4SLinus Torvalds { 118284115e1cSTrond Myklebust struct nfs_page *first = nfs_list_entry(head->next); 118384115e1cSTrond Myklebust struct inode *inode = first->wb_context->path.dentry->d_inode; 118484115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 11853ff7576dSTrond Myklebust int priority = flush_task_priority(how); 118607737691STrond Myklebust struct rpc_task *task; 1187bdc7f021STrond Myklebust struct rpc_message msg = { 1188bdc7f021STrond Myklebust .rpc_argp = &data->args, 1189bdc7f021STrond Myklebust .rpc_resp = &data->res, 1190bdc7f021STrond Myklebust .rpc_cred = first->wb_context->cred, 1191bdc7f021STrond Myklebust }; 119284115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 119307737691STrond Myklebust .task = &data->task, 119484115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 1195bdc7f021STrond Myklebust .rpc_message = &msg, 119684115e1cSTrond Myklebust .callback_ops = &nfs_commit_ops, 119784115e1cSTrond Myklebust .callback_data = data, 1198101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 119984115e1cSTrond Myklebust .flags = flags, 12003ff7576dSTrond Myklebust .priority = priority, 120184115e1cSTrond Myklebust }; 12021da177e4SLinus Torvalds 12031da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12041da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12051da177e4SLinus Torvalds 12061da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds data->inode = inode; 1209bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 12123da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 12133da28eb1STrond Myklebust data->args.offset = 0; 12143da28eb1STrond Myklebust data->args.count = 0; 1215383ba719STrond Myklebust data->args.context = get_nfs_open_context(first->wb_context); 12163da28eb1STrond Myklebust data->res.count = 0; 12171da177e4SLinus Torvalds data->res.fattr = &data->fattr; 12181da177e4SLinus Torvalds data->res.verf = &data->verf; 12190e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12201da177e4SLinus Torvalds 1221788e7a89STrond Myklebust /* Set up the initial task struct. */ 1222bdc7f021STrond Myklebust NFS_PROTO(inode)->commit_setup(data, &msg); 12231da177e4SLinus Torvalds 1224a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1225bdc7f021STrond Myklebust 122607737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1227dbae4c73STrond Myklebust if (IS_ERR(task)) 1228dbae4c73STrond Myklebust return PTR_ERR(task); 122907737691STrond Myklebust rpc_put_task(task); 1230dbae4c73STrond Myklebust return 0; 12311da177e4SLinus Torvalds } 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds /* 12341da177e4SLinus Torvalds * Commit dirty pages 12351da177e4SLinus Torvalds */ 12361da177e4SLinus Torvalds static int 123740859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 12381da177e4SLinus Torvalds { 12391da177e4SLinus Torvalds struct nfs_write_data *data; 12401da177e4SLinus Torvalds struct nfs_page *req; 12411da177e4SLinus Torvalds 1242c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds if (!data) 12451da177e4SLinus Torvalds goto out_bad; 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds /* Set up the argument struct */ 1248dbae4c73STrond Myklebust return nfs_commit_rpcsetup(head, data, how); 12491da177e4SLinus Torvalds out_bad: 12501da177e4SLinus Torvalds while (!list_empty(head)) { 12511da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12521da177e4SLinus Torvalds nfs_list_remove_request(req); 12531da177e4SLinus Torvalds nfs_mark_request_commit(req); 125483715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1255c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1256c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12579fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds return -ENOMEM; 12601da177e4SLinus Torvalds } 12611da177e4SLinus Torvalds 12621da177e4SLinus Torvalds /* 12631da177e4SLinus Torvalds * COMMIT call returned 12641da177e4SLinus Torvalds */ 1265788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 12661da177e4SLinus Torvalds { 1267963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12681da177e4SLinus Torvalds 1269a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 12701da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12711da177e4SLinus Torvalds 1272788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1273788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1274788e7a89STrond Myklebust return; 1275c9d8f89dSTrond Myklebust } 1276c9d8f89dSTrond Myklebust 1277c9d8f89dSTrond Myklebust static void nfs_commit_release(void *calldata) 1278c9d8f89dSTrond Myklebust { 1279c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1280c9d8f89dSTrond Myklebust struct nfs_page *req; 1281c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1282788e7a89STrond Myklebust 12831da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12841da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12851da177e4SLinus Torvalds nfs_list_remove_request(req); 1286612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 1287fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1288c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1289c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12901da177e4SLinus Torvalds 1291*48186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 129288be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 129388be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 12941da177e4SLinus Torvalds req->wb_bytes, 12951da177e4SLinus Torvalds (long long)req_offset(req)); 1296c9d8f89dSTrond Myklebust if (status < 0) { 1297c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 12981da177e4SLinus Torvalds nfs_inode_remove_request(req); 1299c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 13001da177e4SLinus Torvalds goto next; 13011da177e4SLinus Torvalds } 13021da177e4SLinus Torvalds 13031da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13041da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13051da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 13061da177e4SLinus Torvalds /* We have a match */ 13071da177e4SLinus Torvalds nfs_inode_remove_request(req); 13081da177e4SLinus Torvalds dprintk(" OK\n"); 13091da177e4SLinus Torvalds goto next; 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13121da177e4SLinus Torvalds dprintk(" mismatch\n"); 13136d884e8fSFred nfs_mark_request_dirty(req); 13141da177e4SLinus Torvalds next: 13159fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 13161da177e4SLinus Torvalds } 1317c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 13181da177e4SLinus Torvalds } 1319788e7a89STrond Myklebust 1320788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1321788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1322788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1323788e7a89STrond Myklebust }; 13241da177e4SLinus Torvalds 13253da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13261da177e4SLinus Torvalds { 13271da177e4SLinus Torvalds LIST_HEAD(head); 13287d46a49fSTrond Myklebust int res; 13291da177e4SLinus Torvalds 1330587142f8STrond Myklebust spin_lock(&inode->i_lock); 13313da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1332587142f8STrond Myklebust spin_unlock(&inode->i_lock); 13331da177e4SLinus Torvalds if (res) { 13347d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 13351da177e4SLinus Torvalds if (error < 0) 13361da177e4SLinus Torvalds return error; 13373da28eb1STrond Myklebust } 13381da177e4SLinus Torvalds return res; 13391da177e4SLinus Torvalds } 1340c63c7b05STrond Myklebust #else 1341c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1342c63c7b05STrond Myklebust { 1343c63c7b05STrond Myklebust return 0; 1344c63c7b05STrond Myklebust } 13451da177e4SLinus Torvalds #endif 13461da177e4SLinus Torvalds 13471c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 13481da177e4SLinus Torvalds { 13491c75950bSTrond Myklebust struct inode *inode = mapping->host; 1350ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 13511c75950bSTrond Myklebust unsigned int npages = 0; 1352c42de9ddSTrond Myklebust LIST_HEAD(head); 135370b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 13543f442547STrond Myklebust long pages, ret; 13551da177e4SLinus Torvalds 13561c75950bSTrond Myklebust /* FIXME */ 13571c75950bSTrond Myklebust if (wbc->range_cyclic) 13581c75950bSTrond Myklebust idx_start = 0; 13591c75950bSTrond Myklebust else { 13601c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 13611c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 13621c75950bSTrond Myklebust if (idx_end > idx_start) { 1363ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 13641c75950bSTrond Myklebust npages = l_npages; 13651c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1366ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 13671c75950bSTrond Myklebust npages = 0; 13681c75950bSTrond Myklebust } 13691c75950bSTrond Myklebust } 1370c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1371587142f8STrond Myklebust spin_lock(&inode->i_lock); 13721da177e4SLinus Torvalds do { 1373c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1374c42de9ddSTrond Myklebust if (ret != 0) 1375c42de9ddSTrond Myklebust continue; 1376c42de9ddSTrond Myklebust if (nocommit) 1377c42de9ddSTrond Myklebust break; 1378d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1379724c439cSTrond Myklebust if (pages == 0) 1380c42de9ddSTrond Myklebust break; 1381d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1382587142f8STrond Myklebust spin_unlock(&inode->i_lock); 138383715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1384e8e058e8STrond Myklebust ret = pages; 1385587142f8STrond Myklebust spin_lock(&inode->i_lock); 1386d2ccddf0STrond Myklebust continue; 1387d2ccddf0STrond Myklebust } 1388d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1389587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1390c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1391587142f8STrond Myklebust spin_lock(&inode->i_lock); 1392587142f8STrond Myklebust 1393c42de9ddSTrond Myklebust } while (ret >= 0); 1394587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1395c42de9ddSTrond Myklebust return ret; 13961da177e4SLinus Torvalds } 13971da177e4SLinus Torvalds 139834901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 13991c75950bSTrond Myklebust { 14001c75950bSTrond Myklebust int ret; 14011c75950bSTrond Myklebust 140234901f70STrond Myklebust ret = nfs_writepages(mapping, wbc); 140361822ab5STrond Myklebust if (ret < 0) 140461822ab5STrond Myklebust goto out; 140534901f70STrond Myklebust ret = nfs_sync_mapping_wait(mapping, wbc, how); 1406ed90ef51STrond Myklebust if (ret < 0) 1407ed90ef51STrond Myklebust goto out; 14081c75950bSTrond Myklebust return 0; 140961822ab5STrond Myklebust out: 1410e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 14111c75950bSTrond Myklebust return ret; 14121c75950bSTrond Myklebust } 14131c75950bSTrond Myklebust 141434901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ 141534901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how) 141634901f70STrond Myklebust { 141734901f70STrond Myklebust struct writeback_control wbc = { 141834901f70STrond Myklebust .bdi = mapping->backing_dev_info, 141934901f70STrond Myklebust .sync_mode = WB_SYNC_NONE, 142034901f70STrond Myklebust .nr_to_write = LONG_MAX, 142134901f70STrond Myklebust .for_writepages = 1, 142234901f70STrond Myklebust .range_cyclic = 1, 142334901f70STrond Myklebust }; 142434901f70STrond Myklebust int ret; 142534901f70STrond Myklebust 142634901f70STrond Myklebust ret = __nfs_write_mapping(mapping, &wbc, how); 142734901f70STrond Myklebust if (ret < 0) 142834901f70STrond Myklebust return ret; 142934901f70STrond Myklebust wbc.sync_mode = WB_SYNC_ALL; 143034901f70STrond Myklebust return __nfs_write_mapping(mapping, &wbc, how); 143134901f70STrond Myklebust } 143234901f70STrond Myklebust 1433ed90ef51STrond Myklebust /* 1434ed90ef51STrond Myklebust * flush the inode to disk. 1435ed90ef51STrond Myklebust */ 1436ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode) 14371c75950bSTrond Myklebust { 1438ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, 0); 1439ed90ef51STrond Myklebust } 14401c75950bSTrond Myklebust 1441ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode) 1442ed90ef51STrond Myklebust { 1443ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); 14441c75950bSTrond Myklebust } 14451c75950bSTrond Myklebust 14461b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 14471b3b4a1aSTrond Myklebust { 14481b3b4a1aSTrond Myklebust struct nfs_page *req; 14491b3b4a1aSTrond Myklebust loff_t range_start = page_offset(page); 14501b3b4a1aSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14511b3b4a1aSTrond Myklebust struct writeback_control wbc = { 14521b3b4a1aSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14531b3b4a1aSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14541b3b4a1aSTrond Myklebust .nr_to_write = LONG_MAX, 14551b3b4a1aSTrond Myklebust .range_start = range_start, 14561b3b4a1aSTrond Myklebust .range_end = range_end, 14571b3b4a1aSTrond Myklebust }; 14581b3b4a1aSTrond Myklebust int ret = 0; 14591b3b4a1aSTrond Myklebust 14601b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 14611b3b4a1aSTrond Myklebust for (;;) { 14621b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 14631b3b4a1aSTrond Myklebust if (req == NULL) 14641b3b4a1aSTrond Myklebust goto out; 14651b3b4a1aSTrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 14661b3b4a1aSTrond Myklebust nfs_release_request(req); 14671b3b4a1aSTrond Myklebust break; 14681b3b4a1aSTrond Myklebust } 14691b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 14701b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 14711b3b4a1aSTrond Myklebust /* 14721b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 14731b3b4a1aSTrond Myklebust * page as being dirty 14741b3b4a1aSTrond Myklebust */ 14751b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 14761b3b4a1aSTrond Myklebust nfs_unlock_request(req); 14771b3b4a1aSTrond Myklebust break; 14781b3b4a1aSTrond Myklebust } 14791b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 14801b3b4a1aSTrond Myklebust if (ret < 0) 14811b3b4a1aSTrond Myklebust goto out; 14821b3b4a1aSTrond Myklebust } 14831b3b4a1aSTrond Myklebust if (!PagePrivate(page)) 14841b3b4a1aSTrond Myklebust return 0; 14851b3b4a1aSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); 14861b3b4a1aSTrond Myklebust out: 14871b3b4a1aSTrond Myklebust return ret; 14881b3b4a1aSTrond Myklebust } 14891b3b4a1aSTrond Myklebust 14905334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page, 14915334eb13SAdrian Bunk int how) 14921c75950bSTrond Myklebust { 14931c75950bSTrond Myklebust loff_t range_start = page_offset(page); 14941c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14954d770ccfSTrond Myklebust struct writeback_control wbc = { 14964d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14974d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14984d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 14994d770ccfSTrond Myklebust .range_start = range_start, 15004d770ccfSTrond Myklebust .range_end = range_end, 15014d770ccfSTrond Myklebust }; 15024d770ccfSTrond Myklebust int ret; 15031c75950bSTrond Myklebust 150473e3302fSTrond Myklebust do { 1505c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 15064d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 15074d770ccfSTrond Myklebust if (ret < 0) 150873e3302fSTrond Myklebust goto out_error; 150973e3302fSTrond Myklebust } else if (!PagePrivate(page)) 151073e3302fSTrond Myklebust break; 15114d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 151273e3302fSTrond Myklebust if (ret < 0) 151373e3302fSTrond Myklebust goto out_error; 151473e3302fSTrond Myklebust } while (PagePrivate(page)); 15154d770ccfSTrond Myklebust return 0; 151673e3302fSTrond Myklebust out_error: 1517e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 15184d770ccfSTrond Myklebust return ret; 15191c75950bSTrond Myklebust } 15201c75950bSTrond Myklebust 15211c75950bSTrond Myklebust /* 15221c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15231c75950bSTrond Myklebust */ 15241c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 15251c75950bSTrond Myklebust { 15264d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 15271c75950bSTrond Myklebust } 15281c75950bSTrond Myklebust 1529f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15301da177e4SLinus Torvalds { 15311da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15321da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15331da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 153420c2df83SPaul Mundt NULL); 15351da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15361da177e4SLinus Torvalds return -ENOMEM; 15371da177e4SLinus Torvalds 153893d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15391da177e4SLinus Torvalds nfs_wdata_cachep); 15401da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15411da177e4SLinus Torvalds return -ENOMEM; 15421da177e4SLinus Torvalds 154393d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15441da177e4SLinus Torvalds nfs_wdata_cachep); 15451da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15461da177e4SLinus Torvalds return -ENOMEM; 15471da177e4SLinus Torvalds 154889a09141SPeter Zijlstra /* 154989a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 155089a09141SPeter Zijlstra * 155189a09141SPeter Zijlstra * 64MB: 8192k 155289a09141SPeter Zijlstra * 128MB: 11585k 155389a09141SPeter Zijlstra * 256MB: 16384k 155489a09141SPeter Zijlstra * 512MB: 23170k 155589a09141SPeter Zijlstra * 1GB: 32768k 155689a09141SPeter Zijlstra * 2GB: 46340k 155789a09141SPeter Zijlstra * 4GB: 65536k 155889a09141SPeter Zijlstra * 8GB: 92681k 155989a09141SPeter Zijlstra * 16GB: 131072k 156089a09141SPeter Zijlstra * 156189a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 156289a09141SPeter Zijlstra * Limit the default to 256M 156389a09141SPeter Zijlstra */ 156489a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 156589a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 156689a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 156789a09141SPeter Zijlstra 15681da177e4SLinus Torvalds return 0; 15691da177e4SLinus Torvalds } 15701da177e4SLinus Torvalds 1571266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15721da177e4SLinus Torvalds { 15731da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15741da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15751a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15761da177e4SLinus Torvalds } 15771da177e4SLinus Torvalds 1578