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 */ 37c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, 38c63c7b05STrond Myklebust struct inode *inode, int ioflags); 39f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 40788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 41788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 42788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 431da177e4SLinus Torvalds 44e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 453feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 461da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 471da177e4SLinus Torvalds 48c9d8f89dSTrond Myklebust struct nfs_write_data *nfs_commitdata_alloc(void) 491da177e4SLinus Torvalds { 50e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5140859d7eSChuck Lever 521da177e4SLinus Torvalds if (p) { 531da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 541da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 551da177e4SLinus Torvalds } 561da177e4SLinus Torvalds return p; 571da177e4SLinus Torvalds } 581da177e4SLinus Torvalds 595e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 601da177e4SLinus Torvalds { 6140859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6240859d7eSChuck Lever kfree(p->pagevec); 631da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 641da177e4SLinus Torvalds } 651da177e4SLinus Torvalds 668d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 673feb2d49STrond Myklebust { 68e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 693feb2d49STrond Myklebust 703feb2d49STrond Myklebust if (p) { 713feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 723feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 73e9f7bee1STrond Myklebust p->npages = pagecount; 740d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 750d0b5cb3SChuck Lever p->pagevec = p->page_array; 763feb2d49STrond Myklebust else { 770d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 780d0b5cb3SChuck Lever if (!p->pagevec) { 793feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 803feb2d49STrond Myklebust p = NULL; 813feb2d49STrond Myklebust } 823feb2d49STrond Myklebust } 833feb2d49STrond Myklebust } 843feb2d49STrond Myklebust return p; 853feb2d49STrond Myklebust } 863feb2d49STrond Myklebust 875e4424afSTrond Myklebust static void nfs_writedata_free(struct nfs_write_data *p) 883feb2d49STrond Myklebust { 893feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 903feb2d49STrond Myklebust kfree(p->pagevec); 913feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 923feb2d49STrond Myklebust } 933feb2d49STrond Myklebust 94383ba719STrond Myklebust void nfs_writedata_release(void *data) 951da177e4SLinus Torvalds { 96383ba719STrond Myklebust struct nfs_write_data *wdata = data; 97383ba719STrond Myklebust 98383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 991da177e4SLinus Torvalds nfs_writedata_free(wdata); 1001da177e4SLinus Torvalds } 1011da177e4SLinus Torvalds 1027b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1037b159fc1STrond Myklebust { 1047b159fc1STrond Myklebust ctx->error = error; 1057b159fc1STrond Myklebust smp_wmb(); 1067b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1077b159fc1STrond Myklebust } 1087b159fc1STrond Myklebust 109277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 110277459d2STrond Myklebust { 111277459d2STrond Myklebust struct nfs_page *req = NULL; 112277459d2STrond Myklebust 113277459d2STrond Myklebust if (PagePrivate(page)) { 114277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 115277459d2STrond Myklebust if (req != NULL) 116c03b4024STrond Myklebust kref_get(&req->wb_kref); 117277459d2STrond Myklebust } 118277459d2STrond Myklebust return req; 119277459d2STrond Myklebust } 120277459d2STrond Myklebust 121277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 122277459d2STrond Myklebust { 123587142f8STrond Myklebust struct inode *inode = page->mapping->host; 124277459d2STrond Myklebust struct nfs_page *req = NULL; 125277459d2STrond Myklebust 126587142f8STrond Myklebust spin_lock(&inode->i_lock); 127277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 128587142f8STrond Myklebust spin_unlock(&inode->i_lock); 129277459d2STrond Myklebust return req; 130277459d2STrond Myklebust } 131277459d2STrond Myklebust 1321da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1331da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1341da177e4SLinus Torvalds { 1351da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 136a3d01454STrond Myklebust loff_t end, i_size; 137a3d01454STrond Myklebust pgoff_t end_index; 1381da177e4SLinus Torvalds 139a3d01454STrond Myklebust spin_lock(&inode->i_lock); 140a3d01454STrond Myklebust i_size = i_size_read(inode); 141a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1421da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 143a3d01454STrond Myklebust goto out; 1441da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1451da177e4SLinus Torvalds if (i_size >= end) 146a3d01454STrond Myklebust goto out; 1471da177e4SLinus Torvalds i_size_write(inode, end); 148a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 149a3d01454STrond Myklebust out: 150a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1511da177e4SLinus Torvalds } 1521da177e4SLinus Torvalds 153a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 154a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 155a301b777STrond Myklebust { 156a301b777STrond Myklebust SetPageError(page); 157a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 158a301b777STrond Myklebust } 159a301b777STrond Myklebust 1601da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1611da177e4SLinus Torvalds * covers the full page. 1621da177e4SLinus Torvalds */ 1631da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1641da177e4SLinus Torvalds { 1651da177e4SLinus Torvalds if (PageUptodate(page)) 1661da177e4SLinus Torvalds return; 1671da177e4SLinus Torvalds if (base != 0) 1681da177e4SLinus Torvalds return; 16949a70f27STrond Myklebust if (count != nfs_page_length(page)) 1701da177e4SLinus Torvalds return; 1711da177e4SLinus Torvalds SetPageUptodate(page); 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1751da177e4SLinus Torvalds { 1761da177e4SLinus Torvalds if (wbc->for_reclaim) 177c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 1781da177e4SLinus Torvalds if (wbc->for_kupdate) 1791da177e4SLinus Torvalds return FLUSH_LOWPRI; 1801da177e4SLinus Torvalds return 0; 1811da177e4SLinus Torvalds } 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds /* 18489a09141SPeter Zijlstra * NFS congestion control 18589a09141SPeter Zijlstra */ 18689a09141SPeter Zijlstra 18789a09141SPeter Zijlstra int nfs_congestion_kb; 18889a09141SPeter Zijlstra 18989a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 19089a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 19189a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 19289a09141SPeter Zijlstra 1935a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 19489a09141SPeter Zijlstra { 1955a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 1965a6d41b3STrond Myklebust 1975a6d41b3STrond Myklebust if (!ret) { 19889a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 19989a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 20089a09141SPeter Zijlstra 201277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 20289a09141SPeter Zijlstra NFS_CONGESTION_ON_THRESH) 20389a09141SPeter Zijlstra set_bdi_congested(&nfss->backing_dev_info, WRITE); 20489a09141SPeter Zijlstra } 2055a6d41b3STrond Myklebust return ret; 20689a09141SPeter Zijlstra } 20789a09141SPeter Zijlstra 20889a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 20989a09141SPeter Zijlstra { 21089a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 21189a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 21289a09141SPeter Zijlstra 21389a09141SPeter Zijlstra end_page_writeback(page); 214c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 21589a09141SPeter Zijlstra clear_bdi_congested(&nfss->backing_dev_info, WRITE); 21689a09141SPeter Zijlstra } 21789a09141SPeter Zijlstra 21889a09141SPeter Zijlstra /* 219e261f51fSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 2209cccef95STrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 221e261f51fSTrond Myklebust */ 222c63c7b05STrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 223c63c7b05STrond Myklebust struct page *page) 224e261f51fSTrond Myklebust { 225587142f8STrond Myklebust struct inode *inode = page->mapping->host; 226e261f51fSTrond Myklebust struct nfs_page *req; 227e261f51fSTrond Myklebust int ret; 228e261f51fSTrond Myklebust 229587142f8STrond Myklebust spin_lock(&inode->i_lock); 230e261f51fSTrond Myklebust for(;;) { 231e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 232e261f51fSTrond Myklebust if (req == NULL) { 233587142f8STrond Myklebust spin_unlock(&inode->i_lock); 2349cccef95STrond Myklebust return 0; 235e261f51fSTrond Myklebust } 236acee478aSTrond Myklebust if (nfs_set_page_tag_locked(req)) 237e261f51fSTrond Myklebust break; 238e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 239acee478aSTrond Myklebust * then the call to nfs_set_page_tag_locked() will always 240e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 241e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 242e261f51fSTrond Myklebust */ 243587142f8STrond Myklebust spin_unlock(&inode->i_lock); 244e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 245e261f51fSTrond Myklebust nfs_release_request(req); 246e261f51fSTrond Myklebust if (ret != 0) 247e261f51fSTrond Myklebust return ret; 248587142f8STrond Myklebust spin_lock(&inode->i_lock); 249e261f51fSTrond Myklebust } 250e468bae9STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 251587142f8STrond Myklebust spin_unlock(&inode->i_lock); 252e468bae9STrond Myklebust BUG(); 253612c9384STrond Myklebust } 254c63c7b05STrond Myklebust if (nfs_set_page_writeback(page) != 0) { 255587142f8STrond Myklebust spin_unlock(&inode->i_lock); 256c63c7b05STrond Myklebust BUG(); 257c63c7b05STrond Myklebust } 258587142f8STrond Myklebust spin_unlock(&inode->i_lock); 259f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 260f8512ad0SFred Isaman nfs_redirty_request(req); 261f8512ad0SFred Isaman return pgio->pg_error; 262f8512ad0SFred Isaman } 2639cccef95STrond Myklebust return 0; 264e261f51fSTrond Myklebust } 265e261f51fSTrond Myklebust 266f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 267f758c885STrond Myklebust { 268f758c885STrond Myklebust struct inode *inode = page->mapping->host; 269f758c885STrond Myklebust 270f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 271f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 272f758c885STrond Myklebust 273f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 274f758c885STrond Myklebust return nfs_page_async_flush(pgio, page); 275f758c885STrond Myklebust } 276f758c885STrond Myklebust 277e261f51fSTrond Myklebust /* 2781da177e4SLinus Torvalds * Write an mmapped page to the server. 2791da177e4SLinus Torvalds */ 2804d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 2811da177e4SLinus Torvalds { 282f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 283e261f51fSTrond Myklebust int err; 2841da177e4SLinus Torvalds 285f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 286f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 287f758c885STrond Myklebust nfs_pageio_complete(&pgio); 288f758c885STrond Myklebust if (err < 0) 2894d770ccfSTrond Myklebust return err; 290f758c885STrond Myklebust if (pgio.pg_error < 0) 291f758c885STrond Myklebust return pgio.pg_error; 292f758c885STrond Myklebust return 0; 2934d770ccfSTrond Myklebust } 2944d770ccfSTrond Myklebust 2954d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 2964d770ccfSTrond Myklebust { 297f758c885STrond Myklebust int ret; 2984d770ccfSTrond Myklebust 299f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3001da177e4SLinus Torvalds unlock_page(page); 301f758c885STrond Myklebust return ret; 302f758c885STrond Myklebust } 303f758c885STrond Myklebust 304f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 305f758c885STrond Myklebust { 306f758c885STrond Myklebust int ret; 307f758c885STrond Myklebust 308f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 309f758c885STrond Myklebust unlock_page(page); 310f758c885STrond Myklebust return ret; 3111da177e4SLinus Torvalds } 3121da177e4SLinus Torvalds 3131da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3141da177e4SLinus Torvalds { 3151da177e4SLinus Torvalds struct inode *inode = mapping->host; 316c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3171da177e4SLinus Torvalds int err; 3181da177e4SLinus Torvalds 31991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 32091d5b470SChuck Lever 321c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 322f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 323c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 324f758c885STrond Myklebust if (err < 0) 3251da177e4SLinus Torvalds return err; 326f758c885STrond Myklebust if (pgio.pg_error < 0) 327c63c7b05STrond Myklebust return pgio.pg_error; 328c63c7b05STrond Myklebust return 0; 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds 3311da177e4SLinus Torvalds /* 3321da177e4SLinus Torvalds * Insert a write request into an inode 3331da177e4SLinus Torvalds */ 334e7d39069STrond Myklebust static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3351da177e4SLinus Torvalds { 3361da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3371da177e4SLinus Torvalds int error; 3381da177e4SLinus Torvalds 339e7d39069STrond Myklebust error = radix_tree_preload(GFP_NOFS); 340e7d39069STrond Myklebust if (error != 0) 341e7d39069STrond Myklebust goto out; 342e7d39069STrond Myklebust 343e7d39069STrond Myklebust /* Lock the request! */ 344e7d39069STrond Myklebust nfs_lock_request_dontget(req); 345e7d39069STrond Myklebust 346e7d39069STrond Myklebust spin_lock(&inode->i_lock); 3471da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 34827852596SNick Piggin BUG_ON(error); 3491da177e4SLinus Torvalds if (!nfsi->npages) { 3501da177e4SLinus Torvalds igrab(inode); 3511da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3521da177e4SLinus Torvalds nfsi->change_attr++; 3531da177e4SLinus Torvalds } 354deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 355277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3561da177e4SLinus Torvalds nfsi->npages++; 357c03b4024STrond Myklebust kref_get(&req->wb_kref); 35827852596SNick Piggin radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, 35927852596SNick Piggin NFS_PAGE_TAG_LOCKED); 360e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 361e7d39069STrond Myklebust radix_tree_preload_end(); 362e7d39069STrond Myklebust out: 363e7d39069STrond Myklebust return error; 3641da177e4SLinus Torvalds } 3651da177e4SLinus Torvalds 3661da177e4SLinus Torvalds /* 36789a09141SPeter Zijlstra * Remove a write request from an inode 3681da177e4SLinus Torvalds */ 3691da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 3701da177e4SLinus Torvalds { 37188be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 3721da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3731da177e4SLinus Torvalds 3741da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 3751da177e4SLinus Torvalds 376587142f8STrond Myklebust spin_lock(&inode->i_lock); 377277459d2STrond Myklebust set_page_private(req->wb_page, 0); 378deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 3791da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 3801da177e4SLinus Torvalds nfsi->npages--; 3811da177e4SLinus Torvalds if (!nfsi->npages) { 382587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3831da177e4SLinus Torvalds iput(inode); 3841da177e4SLinus Torvalds } else 385587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3861da177e4SLinus Torvalds nfs_clear_request(req); 3871da177e4SLinus Torvalds nfs_release_request(req); 3881da177e4SLinus Torvalds } 3891da177e4SLinus Torvalds 39061822ab5STrond Myklebust static void 3916d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 39261822ab5STrond Myklebust { 39361822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 39461822ab5STrond Myklebust } 39561822ab5STrond Myklebust 3961da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 3971da177e4SLinus Torvalds /* 3981da177e4SLinus Torvalds * Add a request to the inode's commit list. 3991da177e4SLinus Torvalds */ 4001da177e4SLinus Torvalds static void 4011da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4021da177e4SLinus Torvalds { 40388be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4041da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4051da177e4SLinus Torvalds 406587142f8STrond Myklebust spin_lock(&inode->i_lock); 4071da177e4SLinus Torvalds nfsi->ncommit++; 408e468bae9STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 4095c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4105c369683STrond Myklebust req->wb_index, 4115c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 412587142f8STrond Myklebust spin_unlock(&inode->i_lock); 413fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 414c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 415a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4161da177e4SLinus Torvalds } 4178e821cadSTrond Myklebust 418e468bae9STrond Myklebust static int 419e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 420e468bae9STrond Myklebust { 421e468bae9STrond Myklebust struct page *page = req->wb_page; 422e468bae9STrond Myklebust 423e468bae9STrond Myklebust if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { 424e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 425e468bae9STrond Myklebust dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); 426e468bae9STrond Myklebust return 1; 427e468bae9STrond Myklebust } 428e468bae9STrond Myklebust return 0; 429e468bae9STrond Myklebust } 430e468bae9STrond Myklebust 4318e821cadSTrond Myklebust static inline 4328e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4338e821cadSTrond Myklebust { 4348e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4358e821cadSTrond Myklebust } 4368e821cadSTrond Myklebust 4378e821cadSTrond Myklebust static inline 4388e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4398e821cadSTrond Myklebust { 440e468bae9STrond Myklebust if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4418e821cadSTrond Myklebust nfs_mark_request_commit(req); 4428e821cadSTrond Myklebust return 1; 4438e821cadSTrond Myklebust } 4448e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4456d884e8fSFred nfs_mark_request_dirty(req); 4468e821cadSTrond Myklebust return 1; 4478e821cadSTrond Myklebust } 4488e821cadSTrond Myklebust return 0; 4498e821cadSTrond Myklebust } 4508e821cadSTrond Myklebust #else 4518e821cadSTrond Myklebust static inline void 4528e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4538e821cadSTrond Myklebust { 4548e821cadSTrond Myklebust } 4558e821cadSTrond Myklebust 456e468bae9STrond Myklebust static inline int 457e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 458e468bae9STrond Myklebust { 459e468bae9STrond Myklebust return 0; 460e468bae9STrond Myklebust } 461e468bae9STrond Myklebust 4628e821cadSTrond Myklebust static inline 4638e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4648e821cadSTrond Myklebust { 4658e821cadSTrond Myklebust return 0; 4668e821cadSTrond Myklebust } 4678e821cadSTrond Myklebust 4688e821cadSTrond Myklebust static inline 4698e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4708e821cadSTrond Myklebust { 4718e821cadSTrond Myklebust return 0; 4728e821cadSTrond Myklebust } 4731da177e4SLinus Torvalds #endif 4741da177e4SLinus Torvalds 4751da177e4SLinus Torvalds /* 4761da177e4SLinus Torvalds * Wait for a request to complete. 4771da177e4SLinus Torvalds * 478150030b7SMatthew Wilcox * Interruptible by fatal signals only. 4791da177e4SLinus Torvalds */ 480ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) 4811da177e4SLinus Torvalds { 4821da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4831da177e4SLinus Torvalds struct nfs_page *req; 484ca52fec1STrond Myklebust pgoff_t idx_end, next; 4851da177e4SLinus Torvalds unsigned int res = 0; 4861da177e4SLinus Torvalds int error; 4871da177e4SLinus Torvalds 4881da177e4SLinus Torvalds if (npages == 0) 4891da177e4SLinus Torvalds idx_end = ~0; 4901da177e4SLinus Torvalds else 4911da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 4921da177e4SLinus Torvalds 4931da177e4SLinus Torvalds next = idx_start; 4949fd367f0STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { 4951da177e4SLinus Torvalds if (req->wb_index > idx_end) 4961da177e4SLinus Torvalds break; 4971da177e4SLinus Torvalds 4981da177e4SLinus Torvalds next = req->wb_index + 1; 499c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5001da177e4SLinus Torvalds 501c03b4024STrond Myklebust kref_get(&req->wb_kref); 502587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5031da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5041da177e4SLinus Torvalds nfs_release_request(req); 505587142f8STrond Myklebust spin_lock(&inode->i_lock); 5061da177e4SLinus Torvalds if (error < 0) 5071da177e4SLinus Torvalds return error; 5081da177e4SLinus Torvalds res++; 5091da177e4SLinus Torvalds } 5101da177e4SLinus Torvalds return res; 5111da177e4SLinus Torvalds } 5121da177e4SLinus Torvalds 51383715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 51483715ad5STrond Myklebust { 51583715ad5STrond Myklebust struct nfs_page *req; 51683715ad5STrond Myklebust 51783715ad5STrond Myklebust while(!list_empty(head)) { 51883715ad5STrond Myklebust req = nfs_list_entry(head->next); 51983715ad5STrond Myklebust nfs_list_remove_request(req); 520e468bae9STrond Myklebust nfs_clear_request_commit(req); 52183715ad5STrond Myklebust nfs_inode_remove_request(req); 522b6dff26aSTrond Myklebust nfs_unlock_request(req); 52383715ad5STrond Myklebust } 52483715ad5STrond Myklebust } 52583715ad5STrond Myklebust 5261da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5271da177e4SLinus Torvalds /* 5281da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5291da177e4SLinus Torvalds * @inode: NFS inode to scan 5301da177e4SLinus Torvalds * @dst: destination list 5311da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5321da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5331da177e4SLinus Torvalds * 5341da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5351da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5361da177e4SLinus Torvalds */ 5371da177e4SLinus Torvalds static int 538ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5391da177e4SLinus Torvalds { 5401da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5413da28eb1STrond Myklebust int res = 0; 5423da28eb1STrond Myklebust 5433da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 5445c369683STrond Myklebust res = nfs_scan_list(nfsi, dst, idx_start, npages, 5455c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 5461da177e4SLinus Torvalds nfsi->ncommit -= res; 5473da28eb1STrond Myklebust } 5481da177e4SLinus Torvalds return res; 5491da177e4SLinus Torvalds } 550c42de9ddSTrond Myklebust #else 551ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 552c42de9ddSTrond Myklebust { 553c42de9ddSTrond Myklebust return 0; 554c42de9ddSTrond Myklebust } 5551da177e4SLinus Torvalds #endif 5561da177e4SLinus Torvalds 5571da177e4SLinus Torvalds /* 558e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 559e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 5601da177e4SLinus Torvalds * 561e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 562e7d39069STrond Myklebust * to disk. 5631da177e4SLinus Torvalds */ 564e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 565e7d39069STrond Myklebust struct page *page, 566e7d39069STrond Myklebust unsigned int offset, 567e7d39069STrond Myklebust unsigned int bytes) 5681da177e4SLinus Torvalds { 569e7d39069STrond Myklebust struct nfs_page *req; 570e7d39069STrond Myklebust unsigned int rqend; 571e7d39069STrond Myklebust unsigned int end; 5721da177e4SLinus Torvalds int error; 573277459d2STrond Myklebust 574e7d39069STrond Myklebust if (!PagePrivate(page)) 575e7d39069STrond Myklebust return NULL; 576e7d39069STrond Myklebust 577e7d39069STrond Myklebust end = offset + bytes; 578e7d39069STrond Myklebust spin_lock(&inode->i_lock); 579e7d39069STrond Myklebust 580e7d39069STrond Myklebust for (;;) { 581e7d39069STrond Myklebust req = nfs_page_find_request_locked(page); 582e7d39069STrond Myklebust if (req == NULL) 583e7d39069STrond Myklebust goto out_unlock; 584e7d39069STrond Myklebust 585e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 586e7d39069STrond Myklebust /* 587e7d39069STrond Myklebust * Tell the caller to flush out the request if 588e7d39069STrond Myklebust * the offsets are non-contiguous. 589e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 590e7d39069STrond Myklebust * have flushed out requests having wrong owners. 591e7d39069STrond Myklebust */ 592e468bae9STrond Myklebust if (offset > rqend 593e7d39069STrond Myklebust || end < req->wb_offset) 594e7d39069STrond Myklebust goto out_flushme; 595e7d39069STrond Myklebust 596e7d39069STrond Myklebust if (nfs_set_page_tag_locked(req)) 597e7d39069STrond Myklebust break; 598e7d39069STrond Myklebust 599e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 600587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6011da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6021da177e4SLinus Torvalds nfs_release_request(req); 603e7d39069STrond Myklebust if (error != 0) 604e7d39069STrond Myklebust goto out_err; 605e7d39069STrond Myklebust spin_lock(&inode->i_lock); 6061da177e4SLinus Torvalds } 6071da177e4SLinus Torvalds 608e468bae9STrond Myklebust if (nfs_clear_request_commit(req)) 609e468bae9STrond Myklebust radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, 610e468bae9STrond Myklebust req->wb_index, NFS_PAGE_TAG_COMMIT); 611e468bae9STrond Myklebust 6121da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6131da177e4SLinus Torvalds if (offset < req->wb_offset) { 6141da177e4SLinus Torvalds req->wb_offset = offset; 6151da177e4SLinus Torvalds req->wb_pgbase = offset; 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds if (end > rqend) 6181da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 619e7d39069STrond Myklebust else 620e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 621e7d39069STrond Myklebust out_unlock: 622e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 623e7d39069STrond Myklebust return req; 624e7d39069STrond Myklebust out_flushme: 625e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 626e7d39069STrond Myklebust nfs_release_request(req); 627e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 628e7d39069STrond Myklebust out_err: 629e7d39069STrond Myklebust return ERR_PTR(error); 630e7d39069STrond Myklebust } 6311da177e4SLinus Torvalds 632e7d39069STrond Myklebust /* 633e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 634e7d39069STrond Myklebust * 635e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 636e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 637e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 638e7d39069STrond Myklebust */ 639e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 640e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 641e7d39069STrond Myklebust { 642e7d39069STrond Myklebust struct inode *inode = page->mapping->host; 643e7d39069STrond Myklebust struct nfs_page *req; 644e7d39069STrond Myklebust int error; 645e7d39069STrond Myklebust 646e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 647e7d39069STrond Myklebust if (req != NULL) 648e7d39069STrond Myklebust goto out; 649e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 650e7d39069STrond Myklebust if (IS_ERR(req)) 651e7d39069STrond Myklebust goto out; 652e7d39069STrond Myklebust error = nfs_inode_add_request(inode, req); 653e7d39069STrond Myklebust if (error != 0) { 654e7d39069STrond Myklebust nfs_release_request(req); 655e7d39069STrond Myklebust req = ERR_PTR(error); 656e7d39069STrond Myklebust } 657efc91ed0STrond Myklebust out: 65861e930a9STrond Myklebust return req; 6591da177e4SLinus Torvalds } 6601da177e4SLinus Torvalds 661e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 662e7d39069STrond Myklebust unsigned int offset, unsigned int count) 663e7d39069STrond Myklebust { 664e7d39069STrond Myklebust struct nfs_page *req; 665e7d39069STrond Myklebust 666e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 667e7d39069STrond Myklebust if (IS_ERR(req)) 668e7d39069STrond Myklebust return PTR_ERR(req); 669e7d39069STrond Myklebust /* Update file length */ 670e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 671e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 672e7d39069STrond Myklebust nfs_clear_page_tag_locked(req); 673e7d39069STrond Myklebust return 0; 674e7d39069STrond Myklebust } 675e7d39069STrond Myklebust 6761da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6771da177e4SLinus Torvalds { 678cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 6791da177e4SLinus Torvalds struct nfs_page *req; 6801a54533eSTrond Myklebust int do_flush, status; 6811da177e4SLinus Torvalds /* 6821da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6831da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6841da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 6851da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 6861da177e4SLinus Torvalds * Also do the same if we find a request from an existing 6871da177e4SLinus Torvalds * dropped page. 6881da177e4SLinus Torvalds */ 6891a54533eSTrond Myklebust do { 690277459d2STrond Myklebust req = nfs_page_find_request(page); 6911a54533eSTrond Myklebust if (req == NULL) 6921a54533eSTrond Myklebust return 0; 693e468bae9STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 6941da177e4SLinus Torvalds nfs_release_request(req); 6951a54533eSTrond Myklebust if (!do_flush) 6961a54533eSTrond Myklebust return 0; 697277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 6981a54533eSTrond Myklebust } while (status == 0); 6991a54533eSTrond Myklebust return status; 7001da177e4SLinus Torvalds } 7011da177e4SLinus Torvalds 7021da177e4SLinus Torvalds /* 7035d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 7045d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 7055d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 7065d47a356STrond Myklebust */ 7075d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 7085d47a356STrond Myklebust { 7095d47a356STrond Myklebust return PageUptodate(page) && 7105d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 7115d47a356STrond Myklebust } 7125d47a356STrond Myklebust 7135d47a356STrond Myklebust /* 7141da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7151da177e4SLinus Torvalds * 7161da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7171da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7181da177e4SLinus Torvalds */ 7191da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7201da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7211da177e4SLinus Torvalds { 722cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7231da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7241da177e4SLinus Torvalds int status = 0; 7251da177e4SLinus Torvalds 72691d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 72791d5b470SChuck Lever 72848186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 72901cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 73001cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7310bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 7321da177e4SLinus Torvalds 7331da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7345d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7355d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7365d47a356STrond Myklebust * inefficiencies. 7371da177e4SLinus Torvalds */ 7385d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7395d47a356STrond Myklebust inode->i_flock == NULL && 7404b5621f6STrond Myklebust !(file->f_flags & O_SYNC)) { 74149a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7421da177e4SLinus Torvalds offset = 0; 7431da177e4SLinus Torvalds } 7441da177e4SLinus Torvalds 745e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 74603fa9e84STrond Myklebust if (status < 0) 74703fa9e84STrond Myklebust nfs_set_pageerror(page); 74803fa9e84STrond Myklebust else 749e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7501da177e4SLinus Torvalds 75148186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 7521da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7531da177e4SLinus Torvalds return status; 7541da177e4SLinus Torvalds } 7551da177e4SLinus Torvalds 7561da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7571da177e4SLinus Torvalds { 7588e821cadSTrond Myklebust 7597e5f6146STrond Myklebust if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) { 76089a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7611da177e4SLinus Torvalds nfs_inode_remove_request(req); 7628e821cadSTrond Myklebust } else 7638e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 7649fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 7651da177e4SLinus Torvalds } 7661da177e4SLinus Torvalds 7673ff7576dSTrond Myklebust static int flush_task_priority(int how) 7681da177e4SLinus Torvalds { 7691da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7701da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7711da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7721da177e4SLinus Torvalds case FLUSH_LOWPRI: 7731da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7761da177e4SLinus Torvalds } 7771da177e4SLinus Torvalds 7781da177e4SLinus Torvalds /* 7791da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 7801da177e4SLinus Torvalds */ 781dbae4c73STrond Myklebust static int nfs_write_rpcsetup(struct nfs_page *req, 7821da177e4SLinus Torvalds struct nfs_write_data *data, 783788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7841da177e4SLinus Torvalds unsigned int count, unsigned int offset, 7851da177e4SLinus Torvalds int how) 7861da177e4SLinus Torvalds { 78784115e1cSTrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 78884115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 7893ff7576dSTrond Myklebust int priority = flush_task_priority(how); 79007737691STrond Myklebust struct rpc_task *task; 791bdc7f021STrond Myklebust struct rpc_message msg = { 792bdc7f021STrond Myklebust .rpc_argp = &data->args, 793bdc7f021STrond Myklebust .rpc_resp = &data->res, 794bdc7f021STrond Myklebust .rpc_cred = req->wb_context->cred, 795bdc7f021STrond Myklebust }; 79684115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 79784115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 79807737691STrond Myklebust .task = &data->task, 799bdc7f021STrond Myklebust .rpc_message = &msg, 80084115e1cSTrond Myklebust .callback_ops = call_ops, 80184115e1cSTrond Myklebust .callback_data = data, 802101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 80384115e1cSTrond Myklebust .flags = flags, 8043ff7576dSTrond Myklebust .priority = priority, 80584115e1cSTrond Myklebust }; 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8081da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds data->req = req; 81188be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 812bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 8131da177e4SLinus Torvalds 8141da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8151da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8161da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8171da177e4SLinus Torvalds data->args.pages = data->pagevec; 8181da177e4SLinus Torvalds data->args.count = count; 819383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 820bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 821bdc7f021STrond Myklebust if (how & FLUSH_STABLE) { 822bdc7f021STrond Myklebust data->args.stable = NFS_DATA_SYNC; 823bdc7f021STrond Myklebust if (!NFS_I(inode)->ncommit) 824bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 825bdc7f021STrond Myklebust } 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8281da177e4SLinus Torvalds data->res.count = count; 8291da177e4SLinus Torvalds data->res.verf = &data->verf; 8300e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8311da177e4SLinus Torvalds 832788e7a89STrond Myklebust /* Set up the initial task struct. */ 833bdc7f021STrond Myklebust NFS_PROTO(inode)->write_setup(data, &msg); 8341da177e4SLinus Torvalds 835a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 83648186c7dSChuck Lever "(req %s/%lld, %u bytes @ offset %llu)\n", 8370bbacc40SChuck Lever data->task.tk_pid, 8381da177e4SLinus Torvalds inode->i_sb->s_id, 8391da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8401da177e4SLinus Torvalds count, 8411da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8421da177e4SLinus Torvalds 84307737691STrond Myklebust task = rpc_run_task(&task_setup_data); 844dbae4c73STrond Myklebust if (IS_ERR(task)) 845dbae4c73STrond Myklebust return PTR_ERR(task); 84607737691STrond Myklebust rpc_put_task(task); 847dbae4c73STrond Myklebust return 0; 8481da177e4SLinus Torvalds } 8491da177e4SLinus Torvalds 8506d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 8516d884e8fSFred * call this on each, which will prepare them to be retried on next 8526d884e8fSFred * writeback using standard nfs. 8536d884e8fSFred */ 8546d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 8556d884e8fSFred { 8566d884e8fSFred nfs_mark_request_dirty(req); 8576d884e8fSFred nfs_end_page_writeback(req->wb_page); 8586d884e8fSFred nfs_clear_page_tag_locked(req); 8596d884e8fSFred } 8606d884e8fSFred 8611da177e4SLinus Torvalds /* 8621da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8631da177e4SLinus Torvalds * contiguous dirty area on one page. 8641da177e4SLinus Torvalds */ 8658d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8661da177e4SLinus Torvalds { 8671da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 8681da177e4SLinus Torvalds struct page *page = req->wb_page; 8691da177e4SLinus Torvalds struct nfs_write_data *data; 870e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 871e9f7bee1STrond Myklebust unsigned int offset; 8721da177e4SLinus Torvalds int requests = 0; 873dbae4c73STrond Myklebust int ret = 0; 8741da177e4SLinus Torvalds LIST_HEAD(list); 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds nfs_list_remove_request(req); 8771da177e4SLinus Torvalds 878bcb71bbaSTrond Myklebust nbytes = count; 879e9f7bee1STrond Myklebust do { 880e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 881e9f7bee1STrond Myklebust 8828d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 8831da177e4SLinus Torvalds if (!data) 8841da177e4SLinus Torvalds goto out_bad; 8851da177e4SLinus Torvalds list_add(&data->pages, &list); 8861da177e4SLinus Torvalds requests++; 887e9f7bee1STrond Myklebust nbytes -= len; 888e9f7bee1STrond Myklebust } while (nbytes != 0); 8891da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 8901da177e4SLinus Torvalds 8911da177e4SLinus Torvalds ClearPageError(page); 8921da177e4SLinus Torvalds offset = 0; 893bcb71bbaSTrond Myklebust nbytes = count; 8941da177e4SLinus Torvalds do { 895dbae4c73STrond Myklebust int ret2; 896dbae4c73STrond Myklebust 8971da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8981da177e4SLinus Torvalds list_del_init(&data->pages); 8991da177e4SLinus Torvalds 9001da177e4SLinus Torvalds data->pagevec[0] = page; 9011da177e4SLinus Torvalds 902bcb71bbaSTrond Myklebust if (nbytes < wsize) 903bcb71bbaSTrond Myklebust wsize = nbytes; 904dbae4c73STrond Myklebust ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 905788e7a89STrond Myklebust wsize, offset, how); 906dbae4c73STrond Myklebust if (ret == 0) 907dbae4c73STrond Myklebust ret = ret2; 9081da177e4SLinus Torvalds offset += wsize; 9091da177e4SLinus Torvalds nbytes -= wsize; 9101da177e4SLinus Torvalds } while (nbytes != 0); 9111da177e4SLinus Torvalds 912dbae4c73STrond Myklebust return ret; 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds out_bad: 9151da177e4SLinus Torvalds while (!list_empty(&list)) { 9161da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9171da177e4SLinus Torvalds list_del(&data->pages); 9188aca67f0STrond Myklebust nfs_writedata_release(data); 9191da177e4SLinus Torvalds } 92061822ab5STrond Myklebust nfs_redirty_request(req); 9211da177e4SLinus Torvalds return -ENOMEM; 9221da177e4SLinus Torvalds } 9231da177e4SLinus Torvalds 9241da177e4SLinus Torvalds /* 9251da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9261da177e4SLinus Torvalds * The page must have been locked by the caller. 9271da177e4SLinus Torvalds * 9281da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9291da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9301da177e4SLinus Torvalds * that has been written but not committed. 9311da177e4SLinus Torvalds */ 9328d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9331da177e4SLinus Torvalds { 9341da177e4SLinus Torvalds struct nfs_page *req; 9351da177e4SLinus Torvalds struct page **pages; 9361da177e4SLinus Torvalds struct nfs_write_data *data; 9371da177e4SLinus Torvalds 9388d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 9391da177e4SLinus Torvalds if (!data) 9401da177e4SLinus Torvalds goto out_bad; 9411da177e4SLinus Torvalds 9421da177e4SLinus Torvalds pages = data->pagevec; 9431da177e4SLinus Torvalds while (!list_empty(head)) { 9441da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9451da177e4SLinus Torvalds nfs_list_remove_request(req); 9461da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9471da177e4SLinus Torvalds ClearPageError(req->wb_page); 9481da177e4SLinus Torvalds *pages++ = req->wb_page; 9491da177e4SLinus Torvalds } 9501da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9511da177e4SLinus Torvalds 9521da177e4SLinus Torvalds /* Set up the argument struct */ 953dbae4c73STrond Myklebust return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9541da177e4SLinus Torvalds out_bad: 9551da177e4SLinus Torvalds while (!list_empty(head)) { 95610afec90STrond Myklebust req = nfs_list_entry(head->next); 9571da177e4SLinus Torvalds nfs_list_remove_request(req); 95861822ab5STrond Myklebust nfs_redirty_request(req); 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds return -ENOMEM; 9611da177e4SLinus Torvalds } 9621da177e4SLinus Torvalds 963c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 964c63c7b05STrond Myklebust struct inode *inode, int ioflags) 9651da177e4SLinus Torvalds { 966bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 9671da177e4SLinus Torvalds 968bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 969c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 970bcb71bbaSTrond Myklebust else 971c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 9721da177e4SLinus Torvalds } 9731da177e4SLinus Torvalds 9741da177e4SLinus Torvalds /* 9751da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 9761da177e4SLinus Torvalds */ 977788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 9781da177e4SLinus Torvalds { 979788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 9801da177e4SLinus Torvalds 98148186c7dSChuck Lever dprintk("NFS: %5u write(%s/%lld %d@%lld)", 98248186c7dSChuck Lever task->tk_pid, 98348186c7dSChuck Lever data->req->wb_context->path.dentry->d_inode->i_sb->s_id, 98448186c7dSChuck Lever (long long) 98548186c7dSChuck Lever NFS_FILEID(data->req->wb_context->path.dentry->d_inode), 98648186c7dSChuck Lever data->req->wb_bytes, (long long)req_offset(data->req)); 9871da177e4SLinus Torvalds 988c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 989c9d8f89dSTrond Myklebust } 990788e7a89STrond Myklebust 991c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata) 992c9d8f89dSTrond Myklebust { 993c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 994c9d8f89dSTrond Myklebust struct nfs_page *req = data->req; 995c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 996c9d8f89dSTrond Myklebust int status = data->task.tk_status; 997c9d8f89dSTrond Myklebust 998c9d8f89dSTrond Myklebust if (status < 0) { 999a301b777STrond Myklebust nfs_set_pageerror(page); 1000c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1001c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 10028e821cadSTrond Myklebust goto out; 10038e821cadSTrond Myklebust } 10048e821cadSTrond Myklebust 10058e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 1006587142f8STrond Myklebust struct inode *inode = page->mapping->host; 10078e821cadSTrond Myklebust 1008587142f8STrond Myklebust spin_lock(&inode->i_lock); 10098e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 10108e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 10118e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 10121da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10131da177e4SLinus Torvalds dprintk(" defer commit\n"); 10141da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10158e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 10168e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 10171da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10181da177e4SLinus Torvalds } 1019587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10201da177e4SLinus Torvalds } else 10211da177e4SLinus Torvalds dprintk(" OK\n"); 10221da177e4SLinus Torvalds 10238e821cadSTrond Myklebust out: 10241da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 10251da177e4SLinus Torvalds nfs_writepage_release(req); 1026c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds 1029788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1030788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1031c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_partial, 1032788e7a89STrond Myklebust }; 1033788e7a89STrond Myklebust 10341da177e4SLinus Torvalds /* 10351da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10361da177e4SLinus Torvalds * 10371da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10381da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10391da177e4SLinus Torvalds * as the page has a write request pending. 10401da177e4SLinus Torvalds */ 1041788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10421da177e4SLinus Torvalds { 1043788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10441da177e4SLinus Torvalds 1045c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1046c9d8f89dSTrond Myklebust } 1047c9d8f89dSTrond Myklebust 1048c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata) 1049c9d8f89dSTrond Myklebust { 1050c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1051c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1052788e7a89STrond Myklebust 10531da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 10541da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 1055c9d8f89dSTrond Myklebust struct nfs_page *req = nfs_list_entry(data->pages.next); 1056c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1057c9d8f89dSTrond Myklebust 10581da177e4SLinus Torvalds nfs_list_remove_request(req); 10591da177e4SLinus Torvalds 106048186c7dSChuck Lever dprintk("NFS: %5u write (%s/%lld %d@%lld)", 106148186c7dSChuck Lever data->task.tk_pid, 106288be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 106388be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 10641da177e4SLinus Torvalds req->wb_bytes, 10651da177e4SLinus Torvalds (long long)req_offset(req)); 10661da177e4SLinus Torvalds 1067c9d8f89dSTrond Myklebust if (status < 0) { 1068a301b777STrond Myklebust nfs_set_pageerror(page); 1069c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1070c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 10718e821cadSTrond Myklebust goto remove_request; 10721da177e4SLinus Torvalds } 10731da177e4SLinus Torvalds 10748e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 10751da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10761da177e4SLinus Torvalds nfs_mark_request_commit(req); 10778e821cadSTrond Myklebust nfs_end_page_writeback(page); 10781da177e4SLinus Torvalds dprintk(" marked for commit\n"); 10798e821cadSTrond Myklebust goto next; 10808e821cadSTrond Myklebust } 10818e821cadSTrond Myklebust dprintk(" OK\n"); 10828e821cadSTrond Myklebust remove_request: 10838e821cadSTrond Myklebust nfs_end_page_writeback(page); 10841da177e4SLinus Torvalds nfs_inode_remove_request(req); 10851da177e4SLinus Torvalds next: 10869fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 10871da177e4SLinus Torvalds } 1088c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 10891da177e4SLinus Torvalds } 10901da177e4SLinus Torvalds 1091788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1092788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1093c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_full, 1094788e7a89STrond Myklebust }; 1095788e7a89STrond Myklebust 1096788e7a89STrond Myklebust 10971da177e4SLinus Torvalds /* 10981da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 10991da177e4SLinus Torvalds */ 1100462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 11011da177e4SLinus Torvalds { 11021da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 11031da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1104788e7a89STrond Myklebust int status; 11051da177e4SLinus Torvalds 1106a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 11071da177e4SLinus Torvalds task->tk_pid, task->tk_status); 11081da177e4SLinus Torvalds 1109f551e44fSChuck Lever /* 1110f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1111f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1112f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1113f551e44fSChuck Lever * another writer had changed the file, but some applications 1114f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1115f551e44fSChuck Lever */ 1116788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1117788e7a89STrond Myklebust if (status != 0) 1118788e7a89STrond Myklebust return status; 111991d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 112091d5b470SChuck Lever 11211da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11221da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 11231da177e4SLinus Torvalds /* We tried a write call, but the server did not 11241da177e4SLinus Torvalds * commit data to stable storage even though we 11251da177e4SLinus Torvalds * requested it. 11261da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11271da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11281da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11291da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11301da177e4SLinus Torvalds */ 11311da177e4SLinus Torvalds static unsigned long complain; 11321da177e4SLinus Torvalds 11331da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11341da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11351da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 113654ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 11371da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11381da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11391da177e4SLinus Torvalds } 11401da177e4SLinus Torvalds } 11411da177e4SLinus Torvalds #endif 11421da177e4SLinus Torvalds /* Is this a short write? */ 11431da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11441da177e4SLinus Torvalds static unsigned long complain; 11451da177e4SLinus Torvalds 114691d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 114791d5b470SChuck Lever 11481da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11491da177e4SLinus Torvalds if (resp->count != 0) { 11501da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11511da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11521da177e4SLinus Torvalds /* Resend from where the server left off */ 11531da177e4SLinus Torvalds argp->offset += resp->count; 11541da177e4SLinus Torvalds argp->pgbase += resp->count; 11551da177e4SLinus Torvalds argp->count -= resp->count; 11561da177e4SLinus Torvalds } else { 11571da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11581da177e4SLinus Torvalds * headaches in the case of a server crash. 11591da177e4SLinus Torvalds */ 11601da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11611da177e4SLinus Torvalds } 11621da177e4SLinus Torvalds rpc_restart_call(task); 1163788e7a89STrond Myklebust return -EAGAIN; 11641da177e4SLinus Torvalds } 11651da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11661da177e4SLinus Torvalds printk(KERN_WARNING 11671da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 11681da177e4SLinus Torvalds argp->count); 11691da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11701da177e4SLinus Torvalds } 11711da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 11721da177e4SLinus Torvalds task->tk_status = -EIO; 11731da177e4SLinus Torvalds } 1174788e7a89STrond Myklebust return 0; 11751da177e4SLinus Torvalds } 11761da177e4SLinus Torvalds 11771da177e4SLinus Torvalds 11781da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1179c9d8f89dSTrond Myklebust void nfs_commitdata_release(void *data) 11801da177e4SLinus Torvalds { 1181383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1182383ba719STrond Myklebust 1183383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 11841da177e4SLinus Torvalds nfs_commit_free(wdata); 11851da177e4SLinus Torvalds } 11861da177e4SLinus Torvalds 11871da177e4SLinus Torvalds /* 11881da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 11891da177e4SLinus Torvalds */ 1190dbae4c73STrond Myklebust static int nfs_commit_rpcsetup(struct list_head *head, 1191788e7a89STrond Myklebust struct nfs_write_data *data, 1192788e7a89STrond Myklebust int how) 11931da177e4SLinus Torvalds { 119484115e1cSTrond Myklebust struct nfs_page *first = nfs_list_entry(head->next); 119584115e1cSTrond Myklebust struct inode *inode = first->wb_context->path.dentry->d_inode; 119684115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 11973ff7576dSTrond Myklebust int priority = flush_task_priority(how); 119807737691STrond Myklebust struct rpc_task *task; 1199bdc7f021STrond Myklebust struct rpc_message msg = { 1200bdc7f021STrond Myklebust .rpc_argp = &data->args, 1201bdc7f021STrond Myklebust .rpc_resp = &data->res, 1202bdc7f021STrond Myklebust .rpc_cred = first->wb_context->cred, 1203bdc7f021STrond Myklebust }; 120484115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 120507737691STrond Myklebust .task = &data->task, 120684115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 1207bdc7f021STrond Myklebust .rpc_message = &msg, 120884115e1cSTrond Myklebust .callback_ops = &nfs_commit_ops, 120984115e1cSTrond Myklebust .callback_data = data, 1210101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 121184115e1cSTrond Myklebust .flags = flags, 12123ff7576dSTrond Myklebust .priority = priority, 121384115e1cSTrond Myklebust }; 12141da177e4SLinus Torvalds 12151da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12161da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12171da177e4SLinus Torvalds 12181da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds data->inode = inode; 1221bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 12243da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 12253da28eb1STrond Myklebust data->args.offset = 0; 12263da28eb1STrond Myklebust data->args.count = 0; 1227383ba719STrond Myklebust data->args.context = get_nfs_open_context(first->wb_context); 12283da28eb1STrond Myklebust data->res.count = 0; 12291da177e4SLinus Torvalds data->res.fattr = &data->fattr; 12301da177e4SLinus Torvalds data->res.verf = &data->verf; 12310e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12321da177e4SLinus Torvalds 1233788e7a89STrond Myklebust /* Set up the initial task struct. */ 1234bdc7f021STrond Myklebust NFS_PROTO(inode)->commit_setup(data, &msg); 12351da177e4SLinus Torvalds 1236a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1237bdc7f021STrond Myklebust 123807737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1239dbae4c73STrond Myklebust if (IS_ERR(task)) 1240dbae4c73STrond Myklebust return PTR_ERR(task); 124107737691STrond Myklebust rpc_put_task(task); 1242dbae4c73STrond Myklebust return 0; 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds /* 12461da177e4SLinus Torvalds * Commit dirty pages 12471da177e4SLinus Torvalds */ 12481da177e4SLinus Torvalds static int 124940859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 12501da177e4SLinus Torvalds { 12511da177e4SLinus Torvalds struct nfs_write_data *data; 12521da177e4SLinus Torvalds struct nfs_page *req; 12531da177e4SLinus Torvalds 1254c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 12551da177e4SLinus Torvalds 12561da177e4SLinus Torvalds if (!data) 12571da177e4SLinus Torvalds goto out_bad; 12581da177e4SLinus Torvalds 12591da177e4SLinus Torvalds /* Set up the argument struct */ 1260dbae4c73STrond Myklebust return nfs_commit_rpcsetup(head, data, how); 12611da177e4SLinus Torvalds out_bad: 12621da177e4SLinus Torvalds while (!list_empty(head)) { 12631da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12641da177e4SLinus Torvalds nfs_list_remove_request(req); 12651da177e4SLinus Torvalds nfs_mark_request_commit(req); 126683715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1267c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1268c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12699fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12701da177e4SLinus Torvalds } 12711da177e4SLinus Torvalds return -ENOMEM; 12721da177e4SLinus Torvalds } 12731da177e4SLinus Torvalds 12741da177e4SLinus Torvalds /* 12751da177e4SLinus Torvalds * COMMIT call returned 12761da177e4SLinus Torvalds */ 1277788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 12781da177e4SLinus Torvalds { 1279963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12801da177e4SLinus Torvalds 1281a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 12821da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12831da177e4SLinus Torvalds 1284788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1285788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1286788e7a89STrond Myklebust return; 1287c9d8f89dSTrond Myklebust } 1288c9d8f89dSTrond Myklebust 1289c9d8f89dSTrond Myklebust static void nfs_commit_release(void *calldata) 1290c9d8f89dSTrond Myklebust { 1291c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1292c9d8f89dSTrond Myklebust struct nfs_page *req; 1293c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1294788e7a89STrond Myklebust 12951da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12961da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12971da177e4SLinus Torvalds nfs_list_remove_request(req); 1298e468bae9STrond Myklebust nfs_clear_request_commit(req); 12991da177e4SLinus Torvalds 130048186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 130188be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 130288be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 13031da177e4SLinus Torvalds req->wb_bytes, 13041da177e4SLinus Torvalds (long long)req_offset(req)); 1305c9d8f89dSTrond Myklebust if (status < 0) { 1306c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 13071da177e4SLinus Torvalds nfs_inode_remove_request(req); 1308c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 13091da177e4SLinus Torvalds goto next; 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds 13121da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13131da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13141da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 13151da177e4SLinus Torvalds /* We have a match */ 13161da177e4SLinus Torvalds nfs_inode_remove_request(req); 13171da177e4SLinus Torvalds dprintk(" OK\n"); 13181da177e4SLinus Torvalds goto next; 13191da177e4SLinus Torvalds } 13201da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13211da177e4SLinus Torvalds dprintk(" mismatch\n"); 13226d884e8fSFred nfs_mark_request_dirty(req); 13231da177e4SLinus Torvalds next: 13249fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 13251da177e4SLinus Torvalds } 1326c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 13271da177e4SLinus Torvalds } 1328788e7a89STrond Myklebust 1329788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1330788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1331788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1332788e7a89STrond Myklebust }; 13331da177e4SLinus Torvalds 13343da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13351da177e4SLinus Torvalds { 13361da177e4SLinus Torvalds LIST_HEAD(head); 13377d46a49fSTrond Myklebust int res; 13381da177e4SLinus Torvalds 1339587142f8STrond Myklebust spin_lock(&inode->i_lock); 13403da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1341587142f8STrond Myklebust spin_unlock(&inode->i_lock); 13421da177e4SLinus Torvalds if (res) { 13437d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 13441da177e4SLinus Torvalds if (error < 0) 13451da177e4SLinus Torvalds return error; 13463da28eb1STrond Myklebust } 13471da177e4SLinus Torvalds return res; 13481da177e4SLinus Torvalds } 1349c63c7b05STrond Myklebust #else 1350c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1351c63c7b05STrond Myklebust { 1352c63c7b05STrond Myklebust return 0; 1353c63c7b05STrond Myklebust } 13541da177e4SLinus Torvalds #endif 13551da177e4SLinus Torvalds 13561c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 13571da177e4SLinus Torvalds { 13581c75950bSTrond Myklebust struct inode *inode = mapping->host; 1359ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 13601c75950bSTrond Myklebust unsigned int npages = 0; 1361c42de9ddSTrond Myklebust LIST_HEAD(head); 136270b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 13633f442547STrond Myklebust long pages, ret; 13641da177e4SLinus Torvalds 13651c75950bSTrond Myklebust /* FIXME */ 13661c75950bSTrond Myklebust if (wbc->range_cyclic) 13671c75950bSTrond Myklebust idx_start = 0; 13681c75950bSTrond Myklebust else { 13691c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 13701c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 13711c75950bSTrond Myklebust if (idx_end > idx_start) { 1372ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 13731c75950bSTrond Myklebust npages = l_npages; 13741c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1375ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 13761c75950bSTrond Myklebust npages = 0; 13771c75950bSTrond Myklebust } 13781c75950bSTrond Myklebust } 1379c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1380587142f8STrond Myklebust spin_lock(&inode->i_lock); 13811da177e4SLinus Torvalds do { 1382c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1383c42de9ddSTrond Myklebust if (ret != 0) 1384c42de9ddSTrond Myklebust continue; 1385c42de9ddSTrond Myklebust if (nocommit) 1386c42de9ddSTrond Myklebust break; 1387d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1388724c439cSTrond Myklebust if (pages == 0) 1389c42de9ddSTrond Myklebust break; 1390d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1391587142f8STrond Myklebust spin_unlock(&inode->i_lock); 139283715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1393e8e058e8STrond Myklebust ret = pages; 1394587142f8STrond Myklebust spin_lock(&inode->i_lock); 1395d2ccddf0STrond Myklebust continue; 1396d2ccddf0STrond Myklebust } 1397d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1398587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1399c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1400587142f8STrond Myklebust spin_lock(&inode->i_lock); 1401587142f8STrond Myklebust 1402c42de9ddSTrond Myklebust } while (ret >= 0); 1403587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1404c42de9ddSTrond Myklebust return ret; 14051da177e4SLinus Torvalds } 14061da177e4SLinus Torvalds 140734901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 14081c75950bSTrond Myklebust { 14091c75950bSTrond Myklebust int ret; 14101c75950bSTrond Myklebust 141134901f70STrond Myklebust ret = nfs_writepages(mapping, wbc); 141261822ab5STrond Myklebust if (ret < 0) 141361822ab5STrond Myklebust goto out; 141434901f70STrond Myklebust ret = nfs_sync_mapping_wait(mapping, wbc, how); 1415ed90ef51STrond Myklebust if (ret < 0) 1416ed90ef51STrond Myklebust goto out; 14171c75950bSTrond Myklebust return 0; 141861822ab5STrond Myklebust out: 1419e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 14201c75950bSTrond Myklebust return ret; 14211c75950bSTrond Myklebust } 14221c75950bSTrond Myklebust 142334901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ 142434901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how) 142534901f70STrond Myklebust { 142634901f70STrond Myklebust struct writeback_control wbc = { 142734901f70STrond Myklebust .bdi = mapping->backing_dev_info, 142834901f70STrond Myklebust .sync_mode = WB_SYNC_NONE, 142934901f70STrond Myklebust .nr_to_write = LONG_MAX, 1430*d7fb1207STrond Myklebust .range_start = 0, 1431*d7fb1207STrond Myklebust .range_end = LLONG_MAX, 143234901f70STrond Myklebust .for_writepages = 1, 143334901f70STrond Myklebust }; 143434901f70STrond Myklebust int ret; 143534901f70STrond Myklebust 143634901f70STrond Myklebust ret = __nfs_write_mapping(mapping, &wbc, how); 143734901f70STrond Myklebust if (ret < 0) 143834901f70STrond Myklebust return ret; 143934901f70STrond Myklebust wbc.sync_mode = WB_SYNC_ALL; 144034901f70STrond Myklebust return __nfs_write_mapping(mapping, &wbc, how); 144134901f70STrond Myklebust } 144234901f70STrond Myklebust 1443ed90ef51STrond Myklebust /* 1444ed90ef51STrond Myklebust * flush the inode to disk. 1445ed90ef51STrond Myklebust */ 1446ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode) 14471c75950bSTrond Myklebust { 1448ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, 0); 1449ed90ef51STrond Myklebust } 14501c75950bSTrond Myklebust 1451ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode) 1452ed90ef51STrond Myklebust { 1453ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); 14541c75950bSTrond Myklebust } 14551c75950bSTrond Myklebust 14561b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 14571b3b4a1aSTrond Myklebust { 14581b3b4a1aSTrond Myklebust struct nfs_page *req; 14591b3b4a1aSTrond Myklebust loff_t range_start = page_offset(page); 14601b3b4a1aSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14611b3b4a1aSTrond Myklebust struct writeback_control wbc = { 14621b3b4a1aSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14631b3b4a1aSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14641b3b4a1aSTrond Myklebust .nr_to_write = LONG_MAX, 14651b3b4a1aSTrond Myklebust .range_start = range_start, 14661b3b4a1aSTrond Myklebust .range_end = range_end, 14671b3b4a1aSTrond Myklebust }; 14681b3b4a1aSTrond Myklebust int ret = 0; 14691b3b4a1aSTrond Myklebust 14701b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 14711b3b4a1aSTrond Myklebust for (;;) { 14721b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 14731b3b4a1aSTrond Myklebust if (req == NULL) 14741b3b4a1aSTrond Myklebust goto out; 1475e468bae9STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 14761b3b4a1aSTrond Myklebust nfs_release_request(req); 14771b3b4a1aSTrond Myklebust break; 14781b3b4a1aSTrond Myklebust } 14791b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 14801b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 14811b3b4a1aSTrond Myklebust /* 14821b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 14831b3b4a1aSTrond Myklebust * page as being dirty 14841b3b4a1aSTrond Myklebust */ 14851b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 14861b3b4a1aSTrond Myklebust nfs_unlock_request(req); 14871b3b4a1aSTrond Myklebust break; 14881b3b4a1aSTrond Myklebust } 14891b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 14901b3b4a1aSTrond Myklebust if (ret < 0) 14911b3b4a1aSTrond Myklebust goto out; 14921b3b4a1aSTrond Myklebust } 14931b3b4a1aSTrond Myklebust if (!PagePrivate(page)) 14941b3b4a1aSTrond Myklebust return 0; 14951b3b4a1aSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); 14961b3b4a1aSTrond Myklebust out: 14971b3b4a1aSTrond Myklebust return ret; 14981b3b4a1aSTrond Myklebust } 14991b3b4a1aSTrond Myklebust 15005334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page, 15015334eb13SAdrian Bunk int how) 15021c75950bSTrond Myklebust { 15031c75950bSTrond Myklebust loff_t range_start = page_offset(page); 15041c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 15054d770ccfSTrond Myklebust struct writeback_control wbc = { 15064d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 15074d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 15084d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 15094d770ccfSTrond Myklebust .range_start = range_start, 15104d770ccfSTrond Myklebust .range_end = range_end, 15114d770ccfSTrond Myklebust }; 15124d770ccfSTrond Myklebust int ret; 15131c75950bSTrond Myklebust 151473e3302fSTrond Myklebust do { 1515c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 15164d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 15174d770ccfSTrond Myklebust if (ret < 0) 151873e3302fSTrond Myklebust goto out_error; 151973e3302fSTrond Myklebust } else if (!PagePrivate(page)) 152073e3302fSTrond Myklebust break; 15214d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 152273e3302fSTrond Myklebust if (ret < 0) 152373e3302fSTrond Myklebust goto out_error; 152473e3302fSTrond Myklebust } while (PagePrivate(page)); 15254d770ccfSTrond Myklebust return 0; 152673e3302fSTrond Myklebust out_error: 1527e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 15284d770ccfSTrond Myklebust return ret; 15291c75950bSTrond Myklebust } 15301c75950bSTrond Myklebust 15311c75950bSTrond Myklebust /* 15321c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15331c75950bSTrond Myklebust */ 15341c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 15351c75950bSTrond Myklebust { 15364d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 15371c75950bSTrond Myklebust } 15381c75950bSTrond Myklebust 1539f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15401da177e4SLinus Torvalds { 15411da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15421da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15431da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 154420c2df83SPaul Mundt NULL); 15451da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15461da177e4SLinus Torvalds return -ENOMEM; 15471da177e4SLinus Torvalds 154893d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15491da177e4SLinus Torvalds nfs_wdata_cachep); 15501da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15511da177e4SLinus Torvalds return -ENOMEM; 15521da177e4SLinus Torvalds 155393d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15541da177e4SLinus Torvalds nfs_wdata_cachep); 15551da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15561da177e4SLinus Torvalds return -ENOMEM; 15571da177e4SLinus Torvalds 155889a09141SPeter Zijlstra /* 155989a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 156089a09141SPeter Zijlstra * 156189a09141SPeter Zijlstra * 64MB: 8192k 156289a09141SPeter Zijlstra * 128MB: 11585k 156389a09141SPeter Zijlstra * 256MB: 16384k 156489a09141SPeter Zijlstra * 512MB: 23170k 156589a09141SPeter Zijlstra * 1GB: 32768k 156689a09141SPeter Zijlstra * 2GB: 46340k 156789a09141SPeter Zijlstra * 4GB: 65536k 156889a09141SPeter Zijlstra * 8GB: 92681k 156989a09141SPeter Zijlstra * 16GB: 131072k 157089a09141SPeter Zijlstra * 157189a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 157289a09141SPeter Zijlstra * Limit the default to 256M 157389a09141SPeter Zijlstra */ 157489a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 157589a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 157689a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 157789a09141SPeter Zijlstra 15781da177e4SLinus Torvalds return 0; 15791da177e4SLinus Torvalds } 15801da177e4SLinus Torvalds 1581266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15821da177e4SLinus Torvalds { 15831da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15841da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15851a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15861da177e4SLinus Torvalds } 15871da177e4SLinus Torvalds 1588