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> 16074cc1deSTrond Myklebust #include <linux/migrate.h> 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 191da177e4SLinus Torvalds #include <linux/nfs_fs.h> 201da177e4SLinus Torvalds #include <linux/nfs_mount.h> 211da177e4SLinus Torvalds #include <linux/nfs_page.h> 223fcfab16SAndrew Morton #include <linux/backing-dev.h> 23afeacc8cSPaul Gortmaker #include <linux/export.h> 243fcfab16SAndrew Morton 251da177e4SLinus Torvalds #include <asm/uaccess.h> 261da177e4SLinus Torvalds 271da177e4SLinus Torvalds #include "delegation.h" 2849a70f27STrond Myklebust #include "internal.h" 2991d5b470SChuck Lever #include "iostat.h" 30def6ed7eSAndy Adamson #include "nfs4_fs.h" 31074cc1deSTrond Myklebust #include "fscache.h" 3294ad1c80SFred Isaman #include "pnfs.h" 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 351da177e4SLinus Torvalds 361da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 371da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 381da177e4SLinus Torvalds 391da177e4SLinus Torvalds /* 401da177e4SLinus Torvalds * Local function declarations 411da177e4SLinus Torvalds */ 42f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 436c75dc0dSFred Isaman static const struct rpc_call_ops nfs_write_common_ops; 44788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 45061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 46f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 471da177e4SLinus Torvalds 48e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 493feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 500b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 511da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 521da177e4SLinus Torvalds 530b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 541da177e4SLinus Torvalds { 550b7c0153SFred Isaman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5640859d7eSChuck Lever 571da177e4SLinus Torvalds if (p) { 581da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 591da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 601da177e4SLinus Torvalds } 611da177e4SLinus Torvalds return p; 621da177e4SLinus Torvalds } 63e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 641da177e4SLinus Torvalds 650b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 661da177e4SLinus Torvalds { 671da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 681da177e4SLinus Torvalds } 69e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 701da177e4SLinus Torvalds 716c75dc0dSFred Isaman struct nfs_write_header *nfs_writehdr_alloc(void) 723feb2d49STrond Myklebust { 73cd841605SFred Isaman struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 743feb2d49STrond Myklebust 753feb2d49STrond Myklebust if (p) { 76cd841605SFred Isaman struct nfs_pgio_header *hdr = &p->header; 77cd841605SFred Isaman 783feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 79cd841605SFred Isaman INIT_LIST_HEAD(&hdr->pages); 806c75dc0dSFred Isaman INIT_LIST_HEAD(&hdr->rpc_list); 816c75dc0dSFred Isaman spin_lock_init(&hdr->lock); 826c75dc0dSFred Isaman atomic_set(&hdr->refcnt, 0); 833feb2d49STrond Myklebust } 843feb2d49STrond Myklebust return p; 853feb2d49STrond Myklebust } 863feb2d49STrond Myklebust 87*1763da12SFred Isaman static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr, 886c75dc0dSFred Isaman unsigned int pagecount) 896c75dc0dSFred Isaman { 906c75dc0dSFred Isaman struct nfs_write_data *data, *prealloc; 916c75dc0dSFred Isaman 926c75dc0dSFred Isaman prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data; 936c75dc0dSFred Isaman if (prealloc->header == NULL) 946c75dc0dSFred Isaman data = prealloc; 956c75dc0dSFred Isaman else 966c75dc0dSFred Isaman data = kzalloc(sizeof(*data), GFP_KERNEL); 976c75dc0dSFred Isaman if (!data) 986c75dc0dSFred Isaman goto out; 996c75dc0dSFred Isaman 1006c75dc0dSFred Isaman if (nfs_pgarray_set(&data->pages, pagecount)) { 1016c75dc0dSFred Isaman data->header = hdr; 1026c75dc0dSFred Isaman atomic_inc(&hdr->refcnt); 1036c75dc0dSFred Isaman } else { 1046c75dc0dSFred Isaman if (data != prealloc) 1056c75dc0dSFred Isaman kfree(data); 1066c75dc0dSFred Isaman data = NULL; 1076c75dc0dSFred Isaman } 1086c75dc0dSFred Isaman out: 1096c75dc0dSFred Isaman return data; 1106c75dc0dSFred Isaman } 1116c75dc0dSFred Isaman 112cd841605SFred Isaman void nfs_writehdr_free(struct nfs_pgio_header *hdr) 1133feb2d49STrond Myklebust { 114cd841605SFred Isaman struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header); 115cd841605SFred Isaman mempool_free(whdr, nfs_wdata_mempool); 1163feb2d49STrond Myklebust } 1173feb2d49STrond Myklebust 118dce81290STrond Myklebust void nfs_writedata_release(struct nfs_write_data *wdata) 1191da177e4SLinus Torvalds { 1206c75dc0dSFred Isaman struct nfs_pgio_header *hdr = wdata->header; 1216c75dc0dSFred Isaman struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header); 1226c75dc0dSFred Isaman 123383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 12430dd374fSFred Isaman if (wdata->pages.pagevec != wdata->pages.page_array) 12530dd374fSFred Isaman kfree(wdata->pages.pagevec); 1266c75dc0dSFred Isaman if (wdata != &write_header->rpc_data) 1276c75dc0dSFred Isaman kfree(wdata); 1286c75dc0dSFred Isaman else 1296c75dc0dSFred Isaman wdata->header = NULL; 1306c75dc0dSFred Isaman if (atomic_dec_and_test(&hdr->refcnt)) 131061ae2edSFred Isaman hdr->completion_ops->completion(hdr); 1321da177e4SLinus Torvalds } 1331da177e4SLinus Torvalds 1347b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1357b159fc1STrond Myklebust { 1367b159fc1STrond Myklebust ctx->error = error; 1377b159fc1STrond Myklebust smp_wmb(); 1387b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1397b159fc1STrond Myklebust } 1407b159fc1STrond Myklebust 141277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 142277459d2STrond Myklebust { 143277459d2STrond Myklebust struct nfs_page *req = NULL; 144277459d2STrond Myklebust 145277459d2STrond Myklebust if (PagePrivate(page)) { 146277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 147277459d2STrond Myklebust if (req != NULL) 148c03b4024STrond Myklebust kref_get(&req->wb_kref); 149277459d2STrond Myklebust } 150277459d2STrond Myklebust return req; 151277459d2STrond Myklebust } 152277459d2STrond Myklebust 153277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 154277459d2STrond Myklebust { 155587142f8STrond Myklebust struct inode *inode = page->mapping->host; 156277459d2STrond Myklebust struct nfs_page *req = NULL; 157277459d2STrond Myklebust 158587142f8STrond Myklebust spin_lock(&inode->i_lock); 159277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 160587142f8STrond Myklebust spin_unlock(&inode->i_lock); 161277459d2STrond Myklebust return req; 162277459d2STrond Myklebust } 163277459d2STrond Myklebust 1641da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1651da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1661da177e4SLinus Torvalds { 1671da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 168a3d01454STrond Myklebust loff_t end, i_size; 169a3d01454STrond Myklebust pgoff_t end_index; 1701da177e4SLinus Torvalds 171a3d01454STrond Myklebust spin_lock(&inode->i_lock); 172a3d01454STrond Myklebust i_size = i_size_read(inode); 173a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1741da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 175a3d01454STrond Myklebust goto out; 1761da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1771da177e4SLinus Torvalds if (i_size >= end) 178a3d01454STrond Myklebust goto out; 1791da177e4SLinus Torvalds i_size_write(inode, end); 180a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 181a3d01454STrond Myklebust out: 182a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1831da177e4SLinus Torvalds } 1841da177e4SLinus Torvalds 185a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 186a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 187a301b777STrond Myklebust { 188a301b777STrond Myklebust SetPageError(page); 189a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 190a301b777STrond Myklebust } 191a301b777STrond Myklebust 1921da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1931da177e4SLinus Torvalds * covers the full page. 1941da177e4SLinus Torvalds */ 1951da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1961da177e4SLinus Torvalds { 1971da177e4SLinus Torvalds if (PageUptodate(page)) 1981da177e4SLinus Torvalds return; 1991da177e4SLinus Torvalds if (base != 0) 2001da177e4SLinus Torvalds return; 20149a70f27STrond Myklebust if (count != nfs_page_length(page)) 2021da177e4SLinus Torvalds return; 2031da177e4SLinus Torvalds SetPageUptodate(page); 2041da177e4SLinus Torvalds } 2051da177e4SLinus Torvalds 2061da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2071da177e4SLinus Torvalds { 2081da177e4SLinus Torvalds if (wbc->for_reclaim) 209c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 210b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 211b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 212b31268acSTrond Myklebust return FLUSH_COND_STABLE; 2131da177e4SLinus Torvalds } 2141da177e4SLinus Torvalds 2151da177e4SLinus Torvalds /* 21689a09141SPeter Zijlstra * NFS congestion control 21789a09141SPeter Zijlstra */ 21889a09141SPeter Zijlstra 21989a09141SPeter Zijlstra int nfs_congestion_kb; 22089a09141SPeter Zijlstra 22189a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 22289a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 22389a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 22489a09141SPeter Zijlstra 2255a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 22689a09141SPeter Zijlstra { 2275a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2285a6d41b3STrond Myklebust 2295a6d41b3STrond Myklebust if (!ret) { 23089a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 23189a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 23289a09141SPeter Zijlstra 233a6305ddbSTrond Myklebust page_cache_get(page); 234277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2358aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2368aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2378aa7e847SJens Axboe BLK_RW_ASYNC); 2388aa7e847SJens Axboe } 23989a09141SPeter Zijlstra } 2405a6d41b3STrond Myklebust return ret; 24189a09141SPeter Zijlstra } 24289a09141SPeter Zijlstra 24389a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 24489a09141SPeter Zijlstra { 24589a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 24689a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 24789a09141SPeter Zijlstra 24889a09141SPeter Zijlstra end_page_writeback(page); 249a6305ddbSTrond Myklebust page_cache_release(page); 250c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2518aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 25289a09141SPeter Zijlstra } 25389a09141SPeter Zijlstra 254cfb506e1STrond Myklebust static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) 255e261f51fSTrond Myklebust { 256587142f8STrond Myklebust struct inode *inode = page->mapping->host; 257e261f51fSTrond Myklebust struct nfs_page *req; 258e261f51fSTrond Myklebust int ret; 259e261f51fSTrond Myklebust 260587142f8STrond Myklebust spin_lock(&inode->i_lock); 261e261f51fSTrond Myklebust for (;;) { 262e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 263074cc1deSTrond Myklebust if (req == NULL) 264074cc1deSTrond Myklebust break; 2659994b62bSFred Isaman if (nfs_lock_request_dontget(req)) 266e261f51fSTrond Myklebust break; 267e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 2689994b62bSFred Isaman * then the call to nfs_lock_request_dontget() will always 269e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 270e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 271e261f51fSTrond Myklebust */ 272587142f8STrond Myklebust spin_unlock(&inode->i_lock); 273cfb506e1STrond Myklebust if (!nonblock) 274e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 275cfb506e1STrond Myklebust else 276cfb506e1STrond Myklebust ret = -EAGAIN; 277e261f51fSTrond Myklebust nfs_release_request(req); 278e261f51fSTrond Myklebust if (ret != 0) 279074cc1deSTrond Myklebust return ERR_PTR(ret); 280587142f8STrond Myklebust spin_lock(&inode->i_lock); 281e261f51fSTrond Myklebust } 282587142f8STrond Myklebust spin_unlock(&inode->i_lock); 283074cc1deSTrond Myklebust return req; 284612c9384STrond Myklebust } 285074cc1deSTrond Myklebust 286074cc1deSTrond Myklebust /* 287074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 288074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 289074cc1deSTrond Myklebust */ 290074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 291cfb506e1STrond Myklebust struct page *page, bool nonblock) 292074cc1deSTrond Myklebust { 293074cc1deSTrond Myklebust struct nfs_page *req; 294074cc1deSTrond Myklebust int ret = 0; 295074cc1deSTrond Myklebust 296cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, nonblock); 297074cc1deSTrond Myklebust if (!req) 298074cc1deSTrond Myklebust goto out; 299074cc1deSTrond Myklebust ret = PTR_ERR(req); 300074cc1deSTrond Myklebust if (IS_ERR(req)) 301074cc1deSTrond Myklebust goto out; 302074cc1deSTrond Myklebust 303074cc1deSTrond Myklebust ret = nfs_set_page_writeback(page); 304074cc1deSTrond Myklebust BUG_ON(ret != 0); 305074cc1deSTrond Myklebust BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); 306074cc1deSTrond Myklebust 307f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 308f8512ad0SFred Isaman nfs_redirty_request(req); 309074cc1deSTrond Myklebust ret = pgio->pg_error; 310f8512ad0SFred Isaman } 311074cc1deSTrond Myklebust out: 312074cc1deSTrond Myklebust return ret; 313e261f51fSTrond Myklebust } 314e261f51fSTrond Myklebust 315f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 316f758c885STrond Myklebust { 317f758c885STrond Myklebust struct inode *inode = page->mapping->host; 318cfb506e1STrond Myklebust int ret; 319f758c885STrond Myklebust 320f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 321f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 322f758c885STrond Myklebust 323f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 3241b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 325cfb506e1STrond Myklebust if (ret == -EAGAIN) { 326cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 327cfb506e1STrond Myklebust ret = 0; 328cfb506e1STrond Myklebust } 329cfb506e1STrond Myklebust return ret; 330f758c885STrond Myklebust } 331f758c885STrond Myklebust 332e261f51fSTrond Myklebust /* 3331da177e4SLinus Torvalds * Write an mmapped page to the server. 3341da177e4SLinus Torvalds */ 3354d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3361da177e4SLinus Torvalds { 337f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 338e261f51fSTrond Myklebust int err; 3391da177e4SLinus Torvalds 340061ae2edSFred Isaman nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 341061ae2edSFred Isaman &nfs_async_write_completion_ops); 342f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 343f758c885STrond Myklebust nfs_pageio_complete(&pgio); 344f758c885STrond Myklebust if (err < 0) 3454d770ccfSTrond Myklebust return err; 346f758c885STrond Myklebust if (pgio.pg_error < 0) 347f758c885STrond Myklebust return pgio.pg_error; 348f758c885STrond Myklebust return 0; 3494d770ccfSTrond Myklebust } 3504d770ccfSTrond Myklebust 3514d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3524d770ccfSTrond Myklebust { 353f758c885STrond Myklebust int ret; 3544d770ccfSTrond Myklebust 355f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3561da177e4SLinus Torvalds unlock_page(page); 357f758c885STrond Myklebust return ret; 358f758c885STrond Myklebust } 359f758c885STrond Myklebust 360f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 361f758c885STrond Myklebust { 362f758c885STrond Myklebust int ret; 363f758c885STrond Myklebust 364f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 365f758c885STrond Myklebust unlock_page(page); 366f758c885STrond Myklebust return ret; 3671da177e4SLinus Torvalds } 3681da177e4SLinus Torvalds 3691da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3701da177e4SLinus Torvalds { 3711da177e4SLinus Torvalds struct inode *inode = mapping->host; 37272cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 373c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3741da177e4SLinus Torvalds int err; 3751da177e4SLinus Torvalds 37672cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 37772cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 37872cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 37972cb77f4STrond Myklebust if (err) 38072cb77f4STrond Myklebust goto out_err; 38172cb77f4STrond Myklebust 38291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 38391d5b470SChuck Lever 384061ae2edSFred Isaman nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), 385061ae2edSFred Isaman &nfs_async_write_completion_ops); 386f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 387c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 38872cb77f4STrond Myklebust 38972cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 39072cb77f4STrond Myklebust smp_mb__after_clear_bit(); 39172cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 39272cb77f4STrond Myklebust 393f758c885STrond Myklebust if (err < 0) 39472cb77f4STrond Myklebust goto out_err; 39572cb77f4STrond Myklebust err = pgio.pg_error; 39672cb77f4STrond Myklebust if (err < 0) 39772cb77f4STrond Myklebust goto out_err; 398c63c7b05STrond Myklebust return 0; 39972cb77f4STrond Myklebust out_err: 40072cb77f4STrond Myklebust return err; 4011da177e4SLinus Torvalds } 4021da177e4SLinus Torvalds 4031da177e4SLinus Torvalds /* 4041da177e4SLinus Torvalds * Insert a write request into an inode 4051da177e4SLinus Torvalds */ 406d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 4071da177e4SLinus Torvalds { 4081da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 409e7d39069STrond Myklebust 410e7d39069STrond Myklebust /* Lock the request! */ 411e7d39069STrond Myklebust nfs_lock_request_dontget(req); 412e7d39069STrond Myklebust 413e7d39069STrond Myklebust spin_lock(&inode->i_lock); 4144d65c520STrond Myklebust if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE)) 415a9a4a87aSTrond Myklebust inode->i_version++; 4162df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 417deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 418277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 4191da177e4SLinus Torvalds nfsi->npages++; 420c03b4024STrond Myklebust kref_get(&req->wb_kref); 421e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 4221da177e4SLinus Torvalds } 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds /* 42589a09141SPeter Zijlstra * Remove a write request from an inode 4261da177e4SLinus Torvalds */ 4271da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4281da177e4SLinus Torvalds { 4293d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 4301da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4311da177e4SLinus Torvalds 4321da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4331da177e4SLinus Torvalds 434587142f8STrond Myklebust spin_lock(&inode->i_lock); 435277459d2STrond Myklebust set_page_private(req->wb_page, 0); 436deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4372df485a7STrond Myklebust clear_bit(PG_MAPPED, &req->wb_flags); 4381da177e4SLinus Torvalds nfsi->npages--; 439587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4401da177e4SLinus Torvalds nfs_release_request(req); 4411da177e4SLinus Torvalds } 4421da177e4SLinus Torvalds 44361822ab5STrond Myklebust static void 4446d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 44561822ab5STrond Myklebust { 44661822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 44761822ab5STrond Myklebust } 44861822ab5STrond Myklebust 4491da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4508dd37758STrond Myklebust /** 4518dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 4528dd37758STrond Myklebust * @req: pointer to a struct nfs_page 453ea2cf228SFred Isaman * @dst: commit list head 454ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 4558dd37758STrond Myklebust * 456ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 4578dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 4588dd37758STrond Myklebust * the MM page stats. 4598dd37758STrond Myklebust * 460ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 4618dd37758STrond Myklebust * holding the nfs_page lock. 4628dd37758STrond Myklebust */ 4638dd37758STrond Myklebust void 464ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 465ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 4668dd37758STrond Myklebust { 4678dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 468ea2cf228SFred Isaman spin_lock(cinfo->lock); 469ea2cf228SFred Isaman nfs_list_add_request(req, dst); 470ea2cf228SFred Isaman cinfo->mds->ncommit++; 471ea2cf228SFred Isaman spin_unlock(cinfo->lock); 47256f9cd68SFred Isaman if (!cinfo->dreq) { 4738dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 47456f9cd68SFred Isaman inc_bdi_stat(req->wb_page->mapping->backing_dev_info, 47556f9cd68SFred Isaman BDI_RECLAIMABLE); 47656f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 47756f9cd68SFred Isaman I_DIRTY_DATASYNC); 47856f9cd68SFred Isaman } 4798dd37758STrond Myklebust } 4808dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 4818dd37758STrond Myklebust 4828dd37758STrond Myklebust /** 4838dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 4848dd37758STrond Myklebust * @req: pointer to a nfs_page 485ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 4868dd37758STrond Myklebust * 487ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 4888dd37758STrond Myklebust * number of outstanding requests requiring a commit 4898dd37758STrond Myklebust * It does not update the MM page stats. 4908dd37758STrond Myklebust * 491ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 4928dd37758STrond Myklebust */ 4938dd37758STrond Myklebust void 494ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 495ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 4968dd37758STrond Myklebust { 4978dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 4988dd37758STrond Myklebust return; 4998dd37758STrond Myklebust nfs_list_remove_request(req); 500ea2cf228SFred Isaman cinfo->mds->ncommit--; 5018dd37758STrond Myklebust } 5028dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 5038dd37758STrond Myklebust 504ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 505ea2cf228SFred Isaman struct inode *inode) 506ea2cf228SFred Isaman { 507ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 508ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 509ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 510b359f9d0SFred Isaman cinfo->dreq = NULL; 511f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 512ea2cf228SFred Isaman } 513ea2cf228SFred Isaman 514ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 515ea2cf228SFred Isaman struct inode *inode, 516ea2cf228SFred Isaman struct nfs_direct_req *dreq) 517ea2cf228SFred Isaman { 518*1763da12SFred Isaman if (dreq) 519*1763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 520*1763da12SFred Isaman else 521ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 522ea2cf228SFred Isaman } 523ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 5248dd37758STrond Myklebust 5251da177e4SLinus Torvalds /* 5261da177e4SLinus Torvalds * Add a request to the inode's commit list. 5271da177e4SLinus Torvalds */ 528*1763da12SFred Isaman void 529ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 530ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5311da177e4SLinus Torvalds { 532ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 5338dd37758STrond Myklebust return; 534ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 5351da177e4SLinus Torvalds } 5368e821cadSTrond Myklebust 537d6d6dc7cSFred Isaman static void 538d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 539e468bae9STrond Myklebust { 540e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 541e468bae9STrond Myklebust dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); 542e468bae9STrond Myklebust } 543d6d6dc7cSFred Isaman 5448dd37758STrond Myklebust static void 545d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 546d6d6dc7cSFred Isaman { 5478dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 5488dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 549ea2cf228SFred Isaman struct nfs_commit_info cinfo; 550d6d6dc7cSFred Isaman 551ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 552ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 553ea2cf228SFred Isaman spin_lock(cinfo.lock); 554ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 555ea2cf228SFred Isaman spin_unlock(cinfo.lock); 556d6d6dc7cSFred Isaman } 5578dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 5588dd37758STrond Myklebust } 559e468bae9STrond Myklebust } 560e468bae9STrond Myklebust 5618e821cadSTrond Myklebust static inline 5628e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 5638e821cadSTrond Myklebust { 564465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 565cd841605SFred Isaman return data->header->lseg == NULL; 5668e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 5678e821cadSTrond Myklebust } 5688e821cadSTrond Myklebust 5698e821cadSTrond Myklebust #else 570*1763da12SFred Isaman void 571ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 572ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5738e821cadSTrond Myklebust { 5748e821cadSTrond Myklebust } 5758e821cadSTrond Myklebust 5768dd37758STrond Myklebust static void 577e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 578e468bae9STrond Myklebust { 579e468bae9STrond Myklebust } 580e468bae9STrond Myklebust 5818e821cadSTrond Myklebust static inline 5828e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 5838e821cadSTrond Myklebust { 5848e821cadSTrond Myklebust return 0; 5858e821cadSTrond Myklebust } 5868e821cadSTrond Myklebust 5871da177e4SLinus Torvalds #endif 5881da177e4SLinus Torvalds 589061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 5906c75dc0dSFred Isaman { 591ea2cf228SFred Isaman struct nfs_commit_info cinfo; 5926c75dc0dSFred Isaman unsigned long bytes = 0; 5936c75dc0dSFred Isaman 5946c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 5956c75dc0dSFred Isaman goto out; 596ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 5976c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 5986c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 5996c75dc0dSFred Isaman struct page *page = req->wb_page; 6006c75dc0dSFred Isaman 6016c75dc0dSFred Isaman bytes += req->wb_bytes; 6026c75dc0dSFred Isaman nfs_list_remove_request(req); 6036c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 6046c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 6056c75dc0dSFred Isaman nfs_set_pageerror(page); 6066c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 6076c75dc0dSFred Isaman goto remove_req; 6086c75dc0dSFred Isaman } 6096c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { 6106c75dc0dSFred Isaman nfs_mark_request_dirty(req); 6116c75dc0dSFred Isaman goto next; 6126c75dc0dSFred Isaman } 6136c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { 614ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 6156c75dc0dSFred Isaman goto next; 6166c75dc0dSFred Isaman } 6176c75dc0dSFred Isaman remove_req: 6186c75dc0dSFred Isaman nfs_inode_remove_request(req); 6196c75dc0dSFred Isaman next: 6206c75dc0dSFred Isaman nfs_unlock_request(req); 6216c75dc0dSFred Isaman nfs_end_page_writeback(page); 6226c75dc0dSFred Isaman } 6236c75dc0dSFred Isaman out: 6246c75dc0dSFred Isaman hdr->release(hdr); 6256c75dc0dSFred Isaman } 6266c75dc0dSFred Isaman 62747c62564STrond Myklebust #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 628ea2cf228SFred Isaman static unsigned long 629ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 630fb8a1f11STrond Myklebust { 631ea2cf228SFred Isaman return cinfo->mds->ncommit; 632fb8a1f11STrond Myklebust } 633fb8a1f11STrond Myklebust 634ea2cf228SFred Isaman /* cinfo->lock held by caller */ 635*1763da12SFred Isaman int 636ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 637ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 638d6d6dc7cSFred Isaman { 639d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 640d6d6dc7cSFred Isaman int ret = 0; 641d6d6dc7cSFred Isaman 642d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 6438dd37758STrond Myklebust if (!nfs_lock_request(req)) 6448dd37758STrond Myklebust continue; 645ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 6463b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 647ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 6488dd37758STrond Myklebust nfs_list_add_request(req, dst); 649d6d6dc7cSFred Isaman ret++; 650*1763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 651d6d6dc7cSFred Isaman break; 652d6d6dc7cSFred Isaman } 653d6d6dc7cSFred Isaman return ret; 654d6d6dc7cSFred Isaman } 655d6d6dc7cSFred Isaman 6561da177e4SLinus Torvalds /* 6571da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 6581da177e4SLinus Torvalds * @inode: NFS inode to scan 659ea2cf228SFred Isaman * @dst: mds destination list 660ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 6611da177e4SLinus Torvalds * 6621da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 6631da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6641da177e4SLinus Torvalds */ 665*1763da12SFred Isaman int 666ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 667ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 6681da177e4SLinus Torvalds { 669d6d6dc7cSFred Isaman int ret = 0; 670fb8a1f11STrond Myklebust 671ea2cf228SFred Isaman spin_lock(cinfo->lock); 672ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 6738dd37758STrond Myklebust const int max = INT_MAX; 674d6d6dc7cSFred Isaman 675ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 676ea2cf228SFred Isaman cinfo, max); 677ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 678d6d6dc7cSFred Isaman } 679ea2cf228SFred Isaman spin_unlock(cinfo->lock); 680ff778d02STrond Myklebust return ret; 6811da177e4SLinus Torvalds } 682d6d6dc7cSFred Isaman 683c42de9ddSTrond Myklebust #else 684ea2cf228SFred Isaman static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 685fb8a1f11STrond Myklebust { 686fb8a1f11STrond Myklebust return 0; 687fb8a1f11STrond Myklebust } 688fb8a1f11STrond Myklebust 689*1763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 690ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 691c42de9ddSTrond Myklebust { 692c42de9ddSTrond Myklebust return 0; 693c42de9ddSTrond Myklebust } 6941da177e4SLinus Torvalds #endif 6951da177e4SLinus Torvalds 6961da177e4SLinus Torvalds /* 697e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 698e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 6991da177e4SLinus Torvalds * 700e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 701e7d39069STrond Myklebust * to disk. 7021da177e4SLinus Torvalds */ 703e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 704e7d39069STrond Myklebust struct page *page, 705e7d39069STrond Myklebust unsigned int offset, 706e7d39069STrond Myklebust unsigned int bytes) 7071da177e4SLinus Torvalds { 708e7d39069STrond Myklebust struct nfs_page *req; 709e7d39069STrond Myklebust unsigned int rqend; 710e7d39069STrond Myklebust unsigned int end; 7111da177e4SLinus Torvalds int error; 712277459d2STrond Myklebust 713e7d39069STrond Myklebust if (!PagePrivate(page)) 714e7d39069STrond Myklebust return NULL; 715e7d39069STrond Myklebust 716e7d39069STrond Myklebust end = offset + bytes; 717e7d39069STrond Myklebust spin_lock(&inode->i_lock); 718e7d39069STrond Myklebust 719e7d39069STrond Myklebust for (;;) { 720e7d39069STrond Myklebust req = nfs_page_find_request_locked(page); 721e7d39069STrond Myklebust if (req == NULL) 722e7d39069STrond Myklebust goto out_unlock; 723e7d39069STrond Myklebust 724e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 725e7d39069STrond Myklebust /* 726e7d39069STrond Myklebust * Tell the caller to flush out the request if 727e7d39069STrond Myklebust * the offsets are non-contiguous. 728e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 729e7d39069STrond Myklebust * have flushed out requests having wrong owners. 730e7d39069STrond Myklebust */ 731e468bae9STrond Myklebust if (offset > rqend 732e7d39069STrond Myklebust || end < req->wb_offset) 733e7d39069STrond Myklebust goto out_flushme; 734e7d39069STrond Myklebust 7359994b62bSFred Isaman if (nfs_lock_request_dontget(req)) 736e7d39069STrond Myklebust break; 737e7d39069STrond Myklebust 738e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 739587142f8STrond Myklebust spin_unlock(&inode->i_lock); 7401da177e4SLinus Torvalds error = nfs_wait_on_request(req); 7411da177e4SLinus Torvalds nfs_release_request(req); 742e7d39069STrond Myklebust if (error != 0) 743e7d39069STrond Myklebust goto out_err; 744e7d39069STrond Myklebust spin_lock(&inode->i_lock); 7451da177e4SLinus Torvalds } 7461da177e4SLinus Torvalds 7471da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7481da177e4SLinus Torvalds if (offset < req->wb_offset) { 7491da177e4SLinus Torvalds req->wb_offset = offset; 7501da177e4SLinus Torvalds req->wb_pgbase = offset; 7511da177e4SLinus Torvalds } 7521da177e4SLinus Torvalds if (end > rqend) 7531da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 754e7d39069STrond Myklebust else 755e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 756e7d39069STrond Myklebust out_unlock: 757e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 758ca138f36SFred Isaman if (req) 7598dd37758STrond Myklebust nfs_clear_request_commit(req); 760e7d39069STrond Myklebust return req; 761e7d39069STrond Myklebust out_flushme: 762e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 763e7d39069STrond Myklebust nfs_release_request(req); 764e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 765e7d39069STrond Myklebust out_err: 766e7d39069STrond Myklebust return ERR_PTR(error); 767e7d39069STrond Myklebust } 7681da177e4SLinus Torvalds 769e7d39069STrond Myklebust /* 770e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 771e7d39069STrond Myklebust * 772e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 773e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 774e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 775e7d39069STrond Myklebust */ 776e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 777e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 778e7d39069STrond Myklebust { 779e7d39069STrond Myklebust struct inode *inode = page->mapping->host; 780e7d39069STrond Myklebust struct nfs_page *req; 781e7d39069STrond Myklebust 782e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 783e7d39069STrond Myklebust if (req != NULL) 784e7d39069STrond Myklebust goto out; 785e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 786e7d39069STrond Myklebust if (IS_ERR(req)) 787e7d39069STrond Myklebust goto out; 788d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 789efc91ed0STrond Myklebust out: 79061e930a9STrond Myklebust return req; 7911da177e4SLinus Torvalds } 7921da177e4SLinus Torvalds 793e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 794e7d39069STrond Myklebust unsigned int offset, unsigned int count) 795e7d39069STrond Myklebust { 796e7d39069STrond Myklebust struct nfs_page *req; 797e7d39069STrond Myklebust 798e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 799e7d39069STrond Myklebust if (IS_ERR(req)) 800e7d39069STrond Myklebust return PTR_ERR(req); 801e7d39069STrond Myklebust /* Update file length */ 802e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 803e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 804a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 8059994b62bSFred Isaman nfs_unlock_request(req); 806e7d39069STrond Myklebust return 0; 807e7d39069STrond Myklebust } 808e7d39069STrond Myklebust 8091da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 8101da177e4SLinus Torvalds { 811cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 8121da177e4SLinus Torvalds struct nfs_page *req; 8131a54533eSTrond Myklebust int do_flush, status; 8141da177e4SLinus Torvalds /* 8151da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 8161da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 8171da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 8181da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 8191da177e4SLinus Torvalds * Also do the same if we find a request from an existing 8201da177e4SLinus Torvalds * dropped page. 8211da177e4SLinus Torvalds */ 8221a54533eSTrond Myklebust do { 823277459d2STrond Myklebust req = nfs_page_find_request(page); 8241a54533eSTrond Myklebust if (req == NULL) 8251a54533eSTrond Myklebust return 0; 826f11ac8dbSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx || 827f11ac8dbSTrond Myklebust req->wb_lock_context->lockowner != current->files || 828f11ac8dbSTrond Myklebust req->wb_lock_context->pid != current->tgid; 8291da177e4SLinus Torvalds nfs_release_request(req); 8301a54533eSTrond Myklebust if (!do_flush) 8311a54533eSTrond Myklebust return 0; 832277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 8331a54533eSTrond Myklebust } while (status == 0); 8341a54533eSTrond Myklebust return status; 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 8371da177e4SLinus Torvalds /* 8385d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 8395d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 8405d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 8415d47a356STrond Myklebust */ 8425d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 8435d47a356STrond Myklebust { 8445d47a356STrond Myklebust return PageUptodate(page) && 8455d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 8465d47a356STrond Myklebust } 8475d47a356STrond Myklebust 8485d47a356STrond Myklebust /* 8491da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 8501da177e4SLinus Torvalds * 8511da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 8521da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 8531da177e4SLinus Torvalds */ 8541da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 8551da177e4SLinus Torvalds unsigned int offset, unsigned int count) 8561da177e4SLinus Torvalds { 857cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 8581da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 8591da177e4SLinus Torvalds int status = 0; 8601da177e4SLinus Torvalds 86191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 86291d5b470SChuck Lever 86348186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 86401cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 86501cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 8660bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 8671da177e4SLinus Torvalds 8681da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 8695d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 8705d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 8715d47a356STrond Myklebust * inefficiencies. 8721da177e4SLinus Torvalds */ 8735d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 8745d47a356STrond Myklebust inode->i_flock == NULL && 8756b2f3d1fSChristoph Hellwig !(file->f_flags & O_DSYNC)) { 87649a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 8771da177e4SLinus Torvalds offset = 0; 8781da177e4SLinus Torvalds } 8791da177e4SLinus Torvalds 880e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 88103fa9e84STrond Myklebust if (status < 0) 88203fa9e84STrond Myklebust nfs_set_pageerror(page); 88359b7c05fSTrond Myklebust else 88459b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 8851da177e4SLinus Torvalds 88648186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 8871da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 8881da177e4SLinus Torvalds return status; 8891da177e4SLinus Torvalds } 8901da177e4SLinus Torvalds 8913ff7576dSTrond Myklebust static int flush_task_priority(int how) 8921da177e4SLinus Torvalds { 8931da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 8941da177e4SLinus Torvalds case FLUSH_HIGHPRI: 8951da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 8961da177e4SLinus Torvalds case FLUSH_LOWPRI: 8971da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 8981da177e4SLinus Torvalds } 8991da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 9001da177e4SLinus Torvalds } 9011da177e4SLinus Torvalds 902c5996c4eSFred Isaman int nfs_initiate_write(struct rpc_clnt *clnt, 903c5996c4eSFred Isaman struct nfs_write_data *data, 904788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 9051da177e4SLinus Torvalds int how) 9061da177e4SLinus Torvalds { 907cd841605SFred Isaman struct inode *inode = data->header->inode; 9083ff7576dSTrond Myklebust int priority = flush_task_priority(how); 90907737691STrond Myklebust struct rpc_task *task; 910bdc7f021STrond Myklebust struct rpc_message msg = { 911bdc7f021STrond Myklebust .rpc_argp = &data->args, 912bdc7f021STrond Myklebust .rpc_resp = &data->res, 913cd841605SFred Isaman .rpc_cred = data->header->cred, 914bdc7f021STrond Myklebust }; 91584115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 916d138d5d1SAndy Adamson .rpc_client = clnt, 91707737691STrond Myklebust .task = &data->task, 918bdc7f021STrond Myklebust .rpc_message = &msg, 91984115e1cSTrond Myklebust .callback_ops = call_ops, 92084115e1cSTrond Myklebust .callback_data = data, 921101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 9222c61be0aSTrond Myklebust .flags = RPC_TASK_ASYNC, 9233ff7576dSTrond Myklebust .priority = priority, 92484115e1cSTrond Myklebust }; 9252c61be0aSTrond Myklebust int ret = 0; 9261da177e4SLinus Torvalds 927d138d5d1SAndy Adamson /* Set up the initial task struct. */ 928d138d5d1SAndy Adamson NFS_PROTO(inode)->write_setup(data, &msg); 929d138d5d1SAndy Adamson 930d138d5d1SAndy Adamson dprintk("NFS: %5u initiated write call " 931d138d5d1SAndy Adamson "(req %s/%lld, %u bytes @ offset %llu)\n", 932d138d5d1SAndy Adamson data->task.tk_pid, 933d138d5d1SAndy Adamson inode->i_sb->s_id, 934d138d5d1SAndy Adamson (long long)NFS_FILEID(inode), 935d138d5d1SAndy Adamson data->args.count, 936d138d5d1SAndy Adamson (unsigned long long)data->args.offset); 937d138d5d1SAndy Adamson 938d138d5d1SAndy Adamson task = rpc_run_task(&task_setup_data); 939d138d5d1SAndy Adamson if (IS_ERR(task)) { 940d138d5d1SAndy Adamson ret = PTR_ERR(task); 941d138d5d1SAndy Adamson goto out; 942d138d5d1SAndy Adamson } 943d138d5d1SAndy Adamson if (how & FLUSH_SYNC) { 944d138d5d1SAndy Adamson ret = rpc_wait_for_completion_task(task); 945d138d5d1SAndy Adamson if (ret == 0) 946d138d5d1SAndy Adamson ret = task->tk_status; 947d138d5d1SAndy Adamson } 948d138d5d1SAndy Adamson rpc_put_task(task); 949d138d5d1SAndy Adamson out: 950d138d5d1SAndy Adamson return ret; 951d138d5d1SAndy Adamson } 952a69aef14SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_write); 953d138d5d1SAndy Adamson 954d138d5d1SAndy Adamson /* 955d138d5d1SAndy Adamson * Set up the argument/result storage required for the RPC call. 956d138d5d1SAndy Adamson */ 9576c75dc0dSFred Isaman static void nfs_write_rpcsetup(struct nfs_write_data *data, 958d138d5d1SAndy Adamson unsigned int count, unsigned int offset, 959ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 960d138d5d1SAndy Adamson { 9616c75dc0dSFred Isaman struct nfs_page *req = data->header->req; 962d138d5d1SAndy Adamson 9631da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 9641da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 9651da177e4SLinus Torvalds 9666c75dc0dSFred Isaman data->args.fh = NFS_FH(data->header->inode); 9671da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 9682bea038cSBoaz Harrosh /* pnfs_set_layoutcommit needs this */ 9692bea038cSBoaz Harrosh data->mds_offset = data->args.offset; 9701da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 97130dd374fSFred Isaman data->args.pages = data->pages.pagevec; 9721da177e4SLinus Torvalds data->args.count = count; 973383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 974f11ac8dbSTrond Myklebust data->args.lock_context = req->wb_lock_context; 975bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 97687ed5eb4STrond Myklebust switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { 97787ed5eb4STrond Myklebust case 0: 97887ed5eb4STrond Myklebust break; 97987ed5eb4STrond Myklebust case FLUSH_COND_STABLE: 980ea2cf228SFred Isaman if (nfs_reqs_to_commit(cinfo)) 98187ed5eb4STrond Myklebust break; 98287ed5eb4STrond Myklebust default: 983bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 984bdc7f021STrond Myklebust } 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds data->res.fattr = &data->fattr; 9871da177e4SLinus Torvalds data->res.count = count; 9881da177e4SLinus Torvalds data->res.verf = &data->verf; 9890e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 9906e4efd56STrond Myklebust } 9911da177e4SLinus Torvalds 9926e4efd56STrond Myklebust static int nfs_do_write(struct nfs_write_data *data, 9936e4efd56STrond Myklebust const struct rpc_call_ops *call_ops, 9946e4efd56STrond Myklebust int how) 9956e4efd56STrond Myklebust { 996cd841605SFred Isaman struct inode *inode = data->header->inode; 9970382b744SAndy Adamson 998c5996c4eSFred Isaman return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how); 9991da177e4SLinus Torvalds } 10001da177e4SLinus Torvalds 1001275acaafSTrond Myklebust static int nfs_do_multiple_writes(struct list_head *head, 1002275acaafSTrond Myklebust const struct rpc_call_ops *call_ops, 1003275acaafSTrond Myklebust int how) 1004275acaafSTrond Myklebust { 1005275acaafSTrond Myklebust struct nfs_write_data *data; 1006275acaafSTrond Myklebust int ret = 0; 1007275acaafSTrond Myklebust 1008275acaafSTrond Myklebust while (!list_empty(head)) { 1009275acaafSTrond Myklebust int ret2; 1010275acaafSTrond Myklebust 10116c75dc0dSFred Isaman data = list_first_entry(head, struct nfs_write_data, list); 1012275acaafSTrond Myklebust list_del_init(&data->list); 1013275acaafSTrond Myklebust 1014dce81290STrond Myklebust ret2 = nfs_do_write(data, call_ops, how); 1015275acaafSTrond Myklebust if (ret == 0) 1016275acaafSTrond Myklebust ret = ret2; 1017275acaafSTrond Myklebust } 1018275acaafSTrond Myklebust return ret; 1019275acaafSTrond Myklebust } 1020275acaafSTrond Myklebust 10216d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 10226d884e8fSFred * call this on each, which will prepare them to be retried on next 10236d884e8fSFred * writeback using standard nfs. 10246d884e8fSFred */ 10256d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 10266d884e8fSFred { 1027a6305ddbSTrond Myklebust struct page *page = req->wb_page; 1028a6305ddbSTrond Myklebust 10296d884e8fSFred nfs_mark_request_dirty(req); 10309994b62bSFred Isaman nfs_unlock_request(req); 1031a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 10326d884e8fSFred } 10336d884e8fSFred 1034061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 10356c75dc0dSFred Isaman { 10366c75dc0dSFred Isaman struct nfs_page *req; 10376c75dc0dSFred Isaman 10386c75dc0dSFred Isaman while (!list_empty(head)) { 10396c75dc0dSFred Isaman req = nfs_list_entry(head->next); 10406c75dc0dSFred Isaman nfs_list_remove_request(req); 10416c75dc0dSFred Isaman nfs_redirty_request(req); 10426c75dc0dSFred Isaman } 10436c75dc0dSFred Isaman } 10446c75dc0dSFred Isaman 1045061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1046061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1047061ae2edSFred Isaman .completion = nfs_write_completion, 1048061ae2edSFred Isaman }; 1049061ae2edSFred Isaman 10501da177e4SLinus Torvalds /* 10511da177e4SLinus Torvalds * Generate multiple small requests to write out a single 10521da177e4SLinus Torvalds * contiguous dirty area on one page. 10531da177e4SLinus Torvalds */ 10546c75dc0dSFred Isaman static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, 10556c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 10561da177e4SLinus Torvalds { 10576c75dc0dSFred Isaman struct nfs_page *req = hdr->req; 10581da177e4SLinus Torvalds struct page *page = req->wb_page; 10591da177e4SLinus Torvalds struct nfs_write_data *data; 1060d097971dSTrond Myklebust size_t wsize = desc->pg_bsize, nbytes; 1061e9f7bee1STrond Myklebust unsigned int offset; 10621da177e4SLinus Torvalds int requests = 0; 1063dbae4c73STrond Myklebust int ret = 0; 1064ea2cf228SFred Isaman struct nfs_commit_info cinfo; 10651da177e4SLinus Torvalds 1066ea2cf228SFred Isaman nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); 10671da177e4SLinus Torvalds nfs_list_remove_request(req); 10686c75dc0dSFred Isaman nfs_list_add_request(req, &hdr->pages); 10691da177e4SLinus Torvalds 1070b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1071ea2cf228SFred Isaman (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || 1072b31268acSTrond Myklebust desc->pg_count > wsize)) 1073b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1074b31268acSTrond Myklebust 1075b31268acSTrond Myklebust 1076275acaafSTrond Myklebust offset = 0; 1077c76069bdSFred Isaman nbytes = desc->pg_count; 1078e9f7bee1STrond Myklebust do { 1079e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 1080e9f7bee1STrond Myklebust 10816c75dc0dSFred Isaman data = nfs_writedata_alloc(hdr, 1); 10826c75dc0dSFred Isaman if (!data) 10831da177e4SLinus Torvalds goto out_bad; 108430dd374fSFred Isaman data->pages.pagevec[0] = page; 1085ea2cf228SFred Isaman nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); 10866c75dc0dSFred Isaman list_add(&data->list, &hdr->rpc_list); 10871da177e4SLinus Torvalds requests++; 1088e9f7bee1STrond Myklebust nbytes -= len; 1089275acaafSTrond Myklebust offset += len; 1090e9f7bee1STrond Myklebust } while (nbytes != 0); 10916c75dc0dSFred Isaman desc->pg_rpc_callops = &nfs_write_common_ops; 1092dbae4c73STrond Myklebust return ret; 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds out_bad: 10956c75dc0dSFred Isaman while (!list_empty(&hdr->rpc_list)) { 10966c75dc0dSFred Isaman data = list_first_entry(&hdr->rpc_list, struct nfs_write_data, list); 10976e4efd56STrond Myklebust list_del(&data->list); 10988ccd271fSFred Isaman nfs_writedata_release(data); 10991da177e4SLinus Torvalds } 1100061ae2edSFred Isaman desc->pg_completion_ops->error_cleanup(&hdr->pages); 11011da177e4SLinus Torvalds return -ENOMEM; 11021da177e4SLinus Torvalds } 11031da177e4SLinus Torvalds 11041da177e4SLinus Torvalds /* 11051da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 11061da177e4SLinus Torvalds * The page must have been locked by the caller. 11071da177e4SLinus Torvalds * 11081da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 11091da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 11101da177e4SLinus Torvalds * that has been written but not committed. 11111da177e4SLinus Torvalds */ 11126c75dc0dSFred Isaman static int nfs_flush_one(struct nfs_pageio_descriptor *desc, 11136c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 11141da177e4SLinus Torvalds { 11151da177e4SLinus Torvalds struct nfs_page *req; 11161da177e4SLinus Torvalds struct page **pages; 11171da177e4SLinus Torvalds struct nfs_write_data *data; 1118c76069bdSFred Isaman struct list_head *head = &desc->pg_list; 11193b609184SPeng Tao int ret = 0; 1120ea2cf228SFred Isaman struct nfs_commit_info cinfo; 11211da177e4SLinus Torvalds 11226c75dc0dSFred Isaman data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base, 1123c76069bdSFred Isaman desc->pg_count)); 11246c75dc0dSFred Isaman if (!data) { 1125061ae2edSFred Isaman desc->pg_completion_ops->error_cleanup(head); 112644b83799SFred Isaman ret = -ENOMEM; 112744b83799SFred Isaman goto out; 112844b83799SFred Isaman } 11296c75dc0dSFred Isaman 1130ea2cf228SFred Isaman nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); 113130dd374fSFred Isaman pages = data->pages.pagevec; 11321da177e4SLinus Torvalds while (!list_empty(head)) { 11331da177e4SLinus Torvalds req = nfs_list_entry(head->next); 11341da177e4SLinus Torvalds nfs_list_remove_request(req); 11356c75dc0dSFred Isaman nfs_list_add_request(req, &hdr->pages); 11361da177e4SLinus Torvalds *pages++ = req->wb_page; 11371da177e4SLinus Torvalds } 11381da177e4SLinus Torvalds 1139b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1140ea2cf228SFred Isaman (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) 1141b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1142b31268acSTrond Myklebust 11431da177e4SLinus Torvalds /* Set up the argument struct */ 1144ea2cf228SFred Isaman nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); 11456c75dc0dSFred Isaman list_add(&data->list, &hdr->rpc_list); 11466c75dc0dSFred Isaman desc->pg_rpc_callops = &nfs_write_common_ops; 114744b83799SFred Isaman out: 114844b83799SFred Isaman return ret; 11491da177e4SLinus Torvalds } 11501da177e4SLinus Torvalds 11516c75dc0dSFred Isaman int nfs_generic_flush(struct nfs_pageio_descriptor *desc, 11526c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 1153dce81290STrond Myklebust { 1154dce81290STrond Myklebust if (desc->pg_bsize < PAGE_CACHE_SIZE) 11556c75dc0dSFred Isaman return nfs_flush_multi(desc, hdr); 11566c75dc0dSFred Isaman return nfs_flush_one(desc, hdr); 1157dce81290STrond Myklebust } 1158dce81290STrond Myklebust 1159dce81290STrond Myklebust static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) 11601751c363STrond Myklebust { 11616c75dc0dSFred Isaman struct nfs_write_header *whdr; 11626c75dc0dSFred Isaman struct nfs_pgio_header *hdr; 1163275acaafSTrond Myklebust int ret; 1164275acaafSTrond Myklebust 11656c75dc0dSFred Isaman whdr = nfs_writehdr_alloc(); 11666c75dc0dSFred Isaman if (!whdr) { 1167061ae2edSFred Isaman desc->pg_completion_ops->error_cleanup(&hdr->pages); 11686c75dc0dSFred Isaman return -ENOMEM; 11696c75dc0dSFred Isaman } 11706c75dc0dSFred Isaman hdr = &whdr->header; 11716c75dc0dSFred Isaman nfs_pgheader_init(desc, hdr, nfs_writehdr_free); 11726c75dc0dSFred Isaman atomic_inc(&hdr->refcnt); 11736c75dc0dSFred Isaman ret = nfs_generic_flush(desc, hdr); 1174275acaafSTrond Myklebust if (ret == 0) 11756c75dc0dSFred Isaman ret = nfs_do_multiple_writes(&hdr->rpc_list, 11766c75dc0dSFred Isaman desc->pg_rpc_callops, 1177dce81290STrond Myklebust desc->pg_ioflags); 11786c75dc0dSFred Isaman else 11796c75dc0dSFred Isaman set_bit(NFS_IOHDR_REDO, &hdr->flags); 11806c75dc0dSFred Isaman if (atomic_dec_and_test(&hdr->refcnt)) 1181061ae2edSFred Isaman hdr->completion_ops->completion(hdr); 1182275acaafSTrond Myklebust return ret; 11831751c363STrond Myklebust } 11841751c363STrond Myklebust 11851751c363STrond Myklebust static const struct nfs_pageio_ops nfs_pageio_write_ops = { 11861751c363STrond Myklebust .pg_test = nfs_generic_pg_test, 11871751c363STrond Myklebust .pg_doio = nfs_generic_pg_writepages, 11881751c363STrond Myklebust }; 11891751c363STrond Myklebust 1190e2fecb21STrond Myklebust void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, 1191061ae2edSFred Isaman struct inode *inode, int ioflags, 1192061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 11931751c363STrond Myklebust { 1194061ae2edSFred Isaman nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, compl_ops, 11951751c363STrond Myklebust NFS_SERVER(inode)->wsize, ioflags); 11961751c363STrond Myklebust } 11971751c363STrond Myklebust 1198dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1199dce81290STrond Myklebust { 1200dce81290STrond Myklebust pgio->pg_ops = &nfs_pageio_write_ops; 1201dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1202dce81290STrond Myklebust } 12031f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1204dce81290STrond Myklebust 1205*1763da12SFred Isaman void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1206061ae2edSFred Isaman struct inode *inode, int ioflags, 1207061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 12081da177e4SLinus Torvalds { 1209061ae2edSFred Isaman if (!pnfs_pageio_init_write(pgio, inode, ioflags, compl_ops)) 1210061ae2edSFred Isaman nfs_pageio_init_write_mds(pgio, inode, ioflags, compl_ops); 12111da177e4SLinus Torvalds } 12121da177e4SLinus Torvalds 1213def6ed7eSAndy Adamson void nfs_write_prepare(struct rpc_task *task, void *calldata) 1214def6ed7eSAndy Adamson { 1215def6ed7eSAndy Adamson struct nfs_write_data *data = calldata; 1216cd841605SFred Isaman NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); 1217def6ed7eSAndy Adamson } 1218def6ed7eSAndy Adamson 12190b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 12200b7c0153SFred Isaman { 12210b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 12220b7c0153SFred Isaman 12230b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 12240b7c0153SFred Isaman } 12250b7c0153SFred Isaman 12261da177e4SLinus Torvalds /* 12271da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 12281da177e4SLinus Torvalds * 12291da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 12301da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 12311da177e4SLinus Torvalds * as the page has a write request pending. 12321da177e4SLinus Torvalds */ 12336c75dc0dSFred Isaman static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) 12341da177e4SLinus Torvalds { 1235788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 12361da177e4SLinus Torvalds 1237c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1238c9d8f89dSTrond Myklebust } 1239c9d8f89dSTrond Myklebust 12406c75dc0dSFred Isaman static void nfs_writeback_release_common(void *calldata) 1241c9d8f89dSTrond Myklebust { 1242c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1243cd841605SFred Isaman struct nfs_pgio_header *hdr = data->header; 1244e2fecb21STrond Myklebust int status = data->task.tk_status; 12456c75dc0dSFred Isaman struct nfs_page *req = hdr->req; 1246788e7a89STrond Myklebust 12476c75dc0dSFred Isaman if ((status >= 0) && nfs_write_need_commit(data)) { 12486c75dc0dSFred Isaman spin_lock(&hdr->lock); 12496c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) 12506c75dc0dSFred Isaman ; /* Do nothing */ 12516c75dc0dSFred Isaman else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) 12521da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 12536c75dc0dSFred Isaman else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) 12546c75dc0dSFred Isaman set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); 12556c75dc0dSFred Isaman spin_unlock(&hdr->lock); 12561da177e4SLinus Torvalds } 1257cd841605SFred Isaman nfs_writedata_release(data); 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds 12606c75dc0dSFred Isaman static const struct rpc_call_ops nfs_write_common_ops = { 1261def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 12626c75dc0dSFred Isaman .rpc_call_done = nfs_writeback_done_common, 12636c75dc0dSFred Isaman .rpc_release = nfs_writeback_release_common, 1264788e7a89STrond Myklebust }; 1265788e7a89STrond Myklebust 1266788e7a89STrond Myklebust 12671da177e4SLinus Torvalds /* 12681da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 12691da177e4SLinus Torvalds */ 127013602896SFred Isaman void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 12711da177e4SLinus Torvalds { 12721da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 12731da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1274cd841605SFred Isaman struct inode *inode = data->header->inode; 1275788e7a89STrond Myklebust int status; 12761da177e4SLinus Torvalds 1277a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 12781da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12791da177e4SLinus Torvalds 1280f551e44fSChuck Lever /* 1281f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1282f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1283f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1284f551e44fSChuck Lever * another writer had changed the file, but some applications 1285f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1286f551e44fSChuck Lever */ 1287cd841605SFred Isaman status = NFS_PROTO(inode)->write_done(task, data); 1288788e7a89STrond Myklebust if (status != 0) 128913602896SFred Isaman return; 1290cd841605SFred Isaman nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 129191d5b470SChuck Lever 12921da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12931da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 12941da177e4SLinus Torvalds /* We tried a write call, but the server did not 12951da177e4SLinus Torvalds * commit data to stable storage even though we 12961da177e4SLinus Torvalds * requested it. 12971da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 12981da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 12991da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 13001da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 13011da177e4SLinus Torvalds */ 13021da177e4SLinus Torvalds static unsigned long complain; 13031da177e4SLinus Torvalds 1304a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 13051da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 13061da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 13071da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1308cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 13091da177e4SLinus Torvalds resp->verf->committed, argp->stable); 13101da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds } 13131da177e4SLinus Torvalds #endif 13146c75dc0dSFred Isaman if (task->tk_status < 0) 13156c75dc0dSFred Isaman nfs_set_pgio_error(data->header, task->tk_status, argp->offset); 13166c75dc0dSFred Isaman else if (resp->count < argp->count) { 13171da177e4SLinus Torvalds static unsigned long complain; 13181da177e4SLinus Torvalds 13196c75dc0dSFred Isaman /* This a short write! */ 1320cd841605SFred Isaman nfs_inc_stats(inode, NFSIOS_SHORTWRITE); 132191d5b470SChuck Lever 13221da177e4SLinus Torvalds /* Has the server at least made some progress? */ 13236c75dc0dSFred Isaman if (resp->count == 0) { 13246c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 13256c75dc0dSFred Isaman printk(KERN_WARNING 13266c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 13276c75dc0dSFred Isaman argp->count); 13286c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 13296c75dc0dSFred Isaman } 13306c75dc0dSFred Isaman nfs_set_pgio_error(data->header, -EIO, argp->offset); 13316c75dc0dSFred Isaman task->tk_status = -EIO; 13326c75dc0dSFred Isaman return; 13336c75dc0dSFred Isaman } 13341da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 13351da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 13361da177e4SLinus Torvalds /* Resend from where the server left off */ 1337a69aef14SFred Isaman data->mds_offset += resp->count; 13381da177e4SLinus Torvalds argp->offset += resp->count; 13391da177e4SLinus Torvalds argp->pgbase += resp->count; 13401da177e4SLinus Torvalds argp->count -= resp->count; 13411da177e4SLinus Torvalds } else { 13421da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 13431da177e4SLinus Torvalds * headaches in the case of a server crash. 13441da177e4SLinus Torvalds */ 13451da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 13461da177e4SLinus Torvalds } 1347d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 13481da177e4SLinus Torvalds } 13491da177e4SLinus Torvalds } 13501da177e4SLinus Torvalds 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 135371d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 135471d0a611STrond Myklebust { 1355b8413f98STrond Myklebust int ret; 1356b8413f98STrond Myklebust 135771d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 135871d0a611STrond Myklebust return 1; 1359b8413f98STrond Myklebust if (!may_wait) 136071d0a611STrond Myklebust return 0; 1361b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1362b8413f98STrond Myklebust NFS_INO_COMMIT, 1363b8413f98STrond Myklebust nfs_wait_bit_killable, 1364b8413f98STrond Myklebust TASK_KILLABLE); 1365b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 136671d0a611STrond Myklebust } 136771d0a611STrond Myklebust 1368f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 136971d0a611STrond Myklebust { 137071d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 137171d0a611STrond Myklebust smp_mb__after_clear_bit(); 137271d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 137371d0a611STrond Myklebust } 137471d0a611STrond Myklebust 13750b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 13761da177e4SLinus Torvalds { 13770b7c0153SFred Isaman put_nfs_open_context(data->context); 13780b7c0153SFred Isaman nfs_commit_free(data); 13791da177e4SLinus Torvalds } 1380e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 13811da177e4SLinus Torvalds 13820b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 13839ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 1384788e7a89STrond Myklebust int how) 13851da177e4SLinus Torvalds { 138607737691STrond Myklebust struct rpc_task *task; 13879ace33cdSFred Isaman int priority = flush_task_priority(how); 1388bdc7f021STrond Myklebust struct rpc_message msg = { 1389bdc7f021STrond Myklebust .rpc_argp = &data->args, 1390bdc7f021STrond Myklebust .rpc_resp = &data->res, 13919ace33cdSFred Isaman .rpc_cred = data->cred, 1392bdc7f021STrond Myklebust }; 139384115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 139407737691STrond Myklebust .task = &data->task, 13959ace33cdSFred Isaman .rpc_client = clnt, 1396bdc7f021STrond Myklebust .rpc_message = &msg, 13979ace33cdSFred Isaman .callback_ops = call_ops, 139884115e1cSTrond Myklebust .callback_data = data, 1399101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 14002c61be0aSTrond Myklebust .flags = RPC_TASK_ASYNC, 14013ff7576dSTrond Myklebust .priority = priority, 140284115e1cSTrond Myklebust }; 1403788e7a89STrond Myklebust /* Set up the initial task struct. */ 14049ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 14051da177e4SLinus Torvalds 1406a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1407bdc7f021STrond Myklebust 140807737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1409dbae4c73STrond Myklebust if (IS_ERR(task)) 1410dbae4c73STrond Myklebust return PTR_ERR(task); 1411d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1412d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 141307737691STrond Myklebust rpc_put_task(task); 1414dbae4c73STrond Myklebust return 0; 14151da177e4SLinus Torvalds } 1416e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 14171da177e4SLinus Torvalds 14181da177e4SLinus Torvalds /* 14199ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 14209ace33cdSFred Isaman */ 14210b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1422988b6dceSFred Isaman struct list_head *head, 1423f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1424f453a54aSFred Isaman struct nfs_commit_info *cinfo) 14259ace33cdSFred Isaman { 14269ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 14273d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 14289ace33cdSFred Isaman 14299ace33cdSFred Isaman /* Set up the RPC argument and reply structs 14309ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 14319ace33cdSFred Isaman 14329ace33cdSFred Isaman list_splice_init(head, &data->pages); 14339ace33cdSFred Isaman 14349ace33cdSFred Isaman data->inode = inode; 14359ace33cdSFred Isaman data->cred = first->wb_context->cred; 1436988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 14379ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1438f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1439b359f9d0SFred Isaman data->dreq = cinfo->dreq; 14409ace33cdSFred Isaman 14419ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 14429ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 14439ace33cdSFred Isaman data->args.offset = 0; 14449ace33cdSFred Isaman data->args.count = 0; 14450b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 14469ace33cdSFred Isaman data->res.fattr = &data->fattr; 14479ace33cdSFred Isaman data->res.verf = &data->verf; 14489ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 14499ace33cdSFred Isaman } 1450e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 14519ace33cdSFred Isaman 1452e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1453ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1454ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 145564bfeb49SFred Isaman { 145664bfeb49SFred Isaman struct nfs_page *req; 145764bfeb49SFred Isaman 145864bfeb49SFred Isaman while (!list_empty(page_list)) { 145964bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 146064bfeb49SFred Isaman nfs_list_remove_request(req); 1461ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 146256f9cd68SFred Isaman if (!cinfo->dreq) { 146364bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 146464bfeb49SFred Isaman dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 146564bfeb49SFred Isaman BDI_RECLAIMABLE); 146656f9cd68SFred Isaman } 14679994b62bSFred Isaman nfs_unlock_request(req); 146864bfeb49SFred Isaman } 146964bfeb49SFred Isaman } 1470e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 147164bfeb49SFred Isaman 14729ace33cdSFred Isaman /* 14731da177e4SLinus Torvalds * Commit dirty pages 14741da177e4SLinus Torvalds */ 14751da177e4SLinus Torvalds static int 1476ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1477ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 14781da177e4SLinus Torvalds { 14790b7c0153SFred Isaman struct nfs_commit_data *data; 14801da177e4SLinus Torvalds 1481c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 14821da177e4SLinus Torvalds 14831da177e4SLinus Torvalds if (!data) 14841da177e4SLinus Torvalds goto out_bad; 14851da177e4SLinus Torvalds 14861da177e4SLinus Torvalds /* Set up the argument struct */ 1487f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1488f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 14890b7c0153SFred Isaman return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, how); 14901da177e4SLinus Torvalds out_bad: 1491ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1492f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 14931da177e4SLinus Torvalds return -ENOMEM; 14941da177e4SLinus Torvalds } 14951da177e4SLinus Torvalds 14961da177e4SLinus Torvalds /* 14971da177e4SLinus Torvalds * COMMIT call returned 14981da177e4SLinus Torvalds */ 1499788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 15001da177e4SLinus Torvalds { 15010b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 15021da177e4SLinus Torvalds 1503a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 15041da177e4SLinus Torvalds task->tk_pid, task->tk_status); 15051da177e4SLinus Torvalds 1506788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1507c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1508c9d8f89dSTrond Myklebust } 1509c9d8f89dSTrond Myklebust 1510f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1511c9d8f89dSTrond Myklebust { 1512c9d8f89dSTrond Myklebust struct nfs_page *req; 1513c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1514f453a54aSFred Isaman struct nfs_commit_info cinfo; 1515788e7a89STrond Myklebust 15161da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 15171da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 15181da177e4SLinus Torvalds nfs_list_remove_request(req); 1519d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 15201da177e4SLinus Torvalds 152148186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 15223d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 15233d4ff43dSAl Viro (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 15241da177e4SLinus Torvalds req->wb_bytes, 15251da177e4SLinus Torvalds (long long)req_offset(req)); 1526c9d8f89dSTrond Myklebust if (status < 0) { 1527c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 15281da177e4SLinus Torvalds nfs_inode_remove_request(req); 1529c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 15301da177e4SLinus Torvalds goto next; 15311da177e4SLinus Torvalds } 15321da177e4SLinus Torvalds 15331da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 15341da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 15351da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 15361da177e4SLinus Torvalds /* We have a match */ 15371da177e4SLinus Torvalds nfs_inode_remove_request(req); 15381da177e4SLinus Torvalds dprintk(" OK\n"); 15391da177e4SLinus Torvalds goto next; 15401da177e4SLinus Torvalds } 15411da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 15421da177e4SLinus Torvalds dprintk(" mismatch\n"); 15436d884e8fSFred nfs_mark_request_dirty(req); 15441da177e4SLinus Torvalds next: 15459994b62bSFred Isaman nfs_unlock_request(req); 15461da177e4SLinus Torvalds } 1547f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1548f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1549f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 15505917ce84SFred Isaman } 15515917ce84SFred Isaman 15525917ce84SFred Isaman static void nfs_commit_release(void *calldata) 15535917ce84SFred Isaman { 15540b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 15555917ce84SFred Isaman 1556f453a54aSFred Isaman data->completion_ops->completion(data); 1557c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 15581da177e4SLinus Torvalds } 1559788e7a89STrond Myklebust 1560788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 15610b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1562788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1563788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1564788e7a89STrond Myklebust }; 15651da177e4SLinus Torvalds 1566f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1567f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1568f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1569f453a54aSFred Isaman }; 1570f453a54aSFred Isaman 1571*1763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1572ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 157384c53ab5SFred Isaman { 157484c53ab5SFred Isaman int status; 157584c53ab5SFred Isaman 1576ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 157784c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1578ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 157984c53ab5SFred Isaman return status; 158084c53ab5SFred Isaman } 158184c53ab5SFred Isaman 1582b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 15831da177e4SLinus Torvalds { 15841da177e4SLinus Torvalds LIST_HEAD(head); 1585ea2cf228SFred Isaman struct nfs_commit_info cinfo; 158671d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1587b8413f98STrond Myklebust int res; 15881da177e4SLinus Torvalds 1589b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1590b8413f98STrond Myklebust if (res <= 0) 1591c5efa5fcSTrond Myklebust goto out_mark_dirty; 1592ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1593ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 15941da177e4SLinus Torvalds if (res) { 1595a861a1e1SFred Isaman int error; 1596a861a1e1SFred Isaman 1597ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 15981da177e4SLinus Torvalds if (error < 0) 15991da177e4SLinus Torvalds return error; 1600b8413f98STrond Myklebust if (!may_wait) 1601b8413f98STrond Myklebust goto out_mark_dirty; 1602b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1603b8413f98STrond Myklebust NFS_INO_COMMIT, 160471d0a611STrond Myklebust nfs_wait_bit_killable, 160571d0a611STrond Myklebust TASK_KILLABLE); 1606b8413f98STrond Myklebust if (error < 0) 1607b8413f98STrond Myklebust return error; 160871d0a611STrond Myklebust } else 160971d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1610c5efa5fcSTrond Myklebust return res; 1611c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1612c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1613c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1614c5efa5fcSTrond Myklebust * that the data is on the disk. 1615c5efa5fcSTrond Myklebust */ 1616c5efa5fcSTrond Myklebust out_mark_dirty: 1617c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 16181da177e4SLinus Torvalds return res; 16191da177e4SLinus Torvalds } 16208fc795f7STrond Myklebust 16218fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 16228fc795f7STrond Myklebust { 1623420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1624420e3646STrond Myklebust int flags = FLUSH_SYNC; 1625420e3646STrond Myklebust int ret = 0; 16268fc795f7STrond Myklebust 16273236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1628ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 16293236c3e1SJeff Layton return ret; 16303236c3e1SJeff Layton 1631a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1632a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1633a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1634420e3646STrond Myklebust */ 1635ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1636420e3646STrond Myklebust goto out_mark_dirty; 1637420e3646STrond Myklebust 1638a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1639420e3646STrond Myklebust flags = 0; 1640a00dd6c0SJeff Layton } 1641a00dd6c0SJeff Layton 1642420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1643420e3646STrond Myklebust if (ret >= 0) { 1644420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1645420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1646420e3646STrond Myklebust wbc->nr_to_write -= ret; 1647420e3646STrond Myklebust else 1648420e3646STrond Myklebust wbc->nr_to_write = 0; 1649420e3646STrond Myklebust } 16508fc795f7STrond Myklebust return 0; 1651420e3646STrond Myklebust } 1652420e3646STrond Myklebust out_mark_dirty: 16538fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 16548fc795f7STrond Myklebust return ret; 16558fc795f7STrond Myklebust } 1656c63c7b05STrond Myklebust #else 16578fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 16588fc795f7STrond Myklebust { 16598fc795f7STrond Myklebust return 0; 16608fc795f7STrond Myklebust } 16611da177e4SLinus Torvalds #endif 16621da177e4SLinus Torvalds 16638fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 16648fc795f7STrond Myklebust { 1665863a3c6cSAndy Adamson int ret; 1666863a3c6cSAndy Adamson 1667863a3c6cSAndy Adamson ret = nfs_commit_unstable_pages(inode, wbc); 1668863a3c6cSAndy Adamson if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { 1669ef311537SAndy Adamson int status; 1670ef311537SAndy Adamson bool sync = true; 1671863a3c6cSAndy Adamson 1672846d5a09SWu Fengguang if (wbc->sync_mode == WB_SYNC_NONE) 1673ef311537SAndy Adamson sync = false; 1674863a3c6cSAndy Adamson 1675863a3c6cSAndy Adamson status = pnfs_layoutcommit_inode(inode, sync); 1676863a3c6cSAndy Adamson if (status < 0) 1677863a3c6cSAndy Adamson return status; 1678863a3c6cSAndy Adamson } 1679863a3c6cSAndy Adamson return ret; 16808fc795f7STrond Myklebust } 16818fc795f7STrond Myklebust 1682acdc53b2STrond Myklebust /* 1683acdc53b2STrond Myklebust * flush the inode to disk. 1684acdc53b2STrond Myklebust */ 1685acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 168634901f70STrond Myklebust { 168734901f70STrond Myklebust struct writeback_control wbc = { 168872cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 168934901f70STrond Myklebust .nr_to_write = LONG_MAX, 1690d7fb1207STrond Myklebust .range_start = 0, 1691d7fb1207STrond Myklebust .range_end = LLONG_MAX, 169234901f70STrond Myklebust }; 169334901f70STrond Myklebust 1694acdc53b2STrond Myklebust return sync_inode(inode, &wbc); 16951c75950bSTrond Myklebust } 16961c75950bSTrond Myklebust 16971b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 16981b3b4a1aSTrond Myklebust { 16991b3b4a1aSTrond Myklebust struct nfs_page *req; 17001b3b4a1aSTrond Myklebust int ret = 0; 17011b3b4a1aSTrond Myklebust 17021b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 17031b3b4a1aSTrond Myklebust for (;;) { 1704ba8b06e6STrond Myklebust wait_on_page_writeback(page); 17051b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 17061b3b4a1aSTrond Myklebust if (req == NULL) 17071b3b4a1aSTrond Myklebust break; 17081b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 17098dd37758STrond Myklebust nfs_clear_request_commit(req); 17101b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 17111b3b4a1aSTrond Myklebust /* 17121b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 17131b3b4a1aSTrond Myklebust * page as being dirty 17141b3b4a1aSTrond Myklebust */ 17151b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 17161b3b4a1aSTrond Myklebust nfs_unlock_request(req); 17171b3b4a1aSTrond Myklebust break; 17181b3b4a1aSTrond Myklebust } 17191b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1720c9edda71STrond Myklebust nfs_release_request(req); 17211b3b4a1aSTrond Myklebust if (ret < 0) 1722c988950eSTrond Myklebust break; 17231b3b4a1aSTrond Myklebust } 17241b3b4a1aSTrond Myklebust return ret; 17251b3b4a1aSTrond Myklebust } 17261b3b4a1aSTrond Myklebust 17271c75950bSTrond Myklebust /* 17281c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 17291c75950bSTrond Myklebust */ 17301c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 17311c75950bSTrond Myklebust { 17327f2f12d9STrond Myklebust loff_t range_start = page_offset(page); 17337f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 17347f2f12d9STrond Myklebust struct writeback_control wbc = { 17357f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 17367f2f12d9STrond Myklebust .nr_to_write = 0, 17377f2f12d9STrond Myklebust .range_start = range_start, 17387f2f12d9STrond Myklebust .range_end = range_end, 17397f2f12d9STrond Myklebust }; 17407f2f12d9STrond Myklebust int ret; 17417f2f12d9STrond Myklebust 17420522f6adSTrond Myklebust for (;;) { 1743ba8b06e6STrond Myklebust wait_on_page_writeback(page); 17447f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 17457f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 17467f2f12d9STrond Myklebust if (ret < 0) 17477f2f12d9STrond Myklebust goto out_error; 17480522f6adSTrond Myklebust continue; 17497f2f12d9STrond Myklebust } 17500522f6adSTrond Myklebust if (!PagePrivate(page)) 17510522f6adSTrond Myklebust break; 17520522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 17537f2f12d9STrond Myklebust if (ret < 0) 17547f2f12d9STrond Myklebust goto out_error; 17557f2f12d9STrond Myklebust } 17567f2f12d9STrond Myklebust return 0; 17577f2f12d9STrond Myklebust out_error: 17587f2f12d9STrond Myklebust return ret; 17591c75950bSTrond Myklebust } 17601c75950bSTrond Myklebust 1761074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1762074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1763a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1764074cc1deSTrond Myklebust { 17652da95652SJeff Layton /* 17662da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 17672da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 17682da95652SJeff Layton * 17692da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 17702da95652SJeff Layton * that we can safely release the inode reference while holding 17712da95652SJeff Layton * the page lock. 17722da95652SJeff Layton */ 17732da95652SJeff Layton if (PagePrivate(page)) 17742da95652SJeff Layton return -EBUSY; 1775074cc1deSTrond Myklebust 1776074cc1deSTrond Myklebust nfs_fscache_release_page(page, GFP_KERNEL); 1777074cc1deSTrond Myklebust 1778a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1779074cc1deSTrond Myklebust } 1780074cc1deSTrond Myklebust #endif 1781074cc1deSTrond Myklebust 1782f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 17831da177e4SLinus Torvalds { 17841da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 1785cd841605SFred Isaman sizeof(struct nfs_write_header), 17861da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 178720c2df83SPaul Mundt NULL); 17881da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 17891da177e4SLinus Torvalds return -ENOMEM; 17901da177e4SLinus Torvalds 179193d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 17921da177e4SLinus Torvalds nfs_wdata_cachep); 17931da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 17941da177e4SLinus Torvalds return -ENOMEM; 17951da177e4SLinus Torvalds 17960b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 17970b7c0153SFred Isaman sizeof(struct nfs_commit_data), 17980b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 17990b7c0153SFred Isaman NULL); 18000b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 18010b7c0153SFred Isaman return -ENOMEM; 18020b7c0153SFred Isaman 180393d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 18041da177e4SLinus Torvalds nfs_wdata_cachep); 18051da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 18061da177e4SLinus Torvalds return -ENOMEM; 18071da177e4SLinus Torvalds 180889a09141SPeter Zijlstra /* 180989a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 181089a09141SPeter Zijlstra * 181189a09141SPeter Zijlstra * 64MB: 8192k 181289a09141SPeter Zijlstra * 128MB: 11585k 181389a09141SPeter Zijlstra * 256MB: 16384k 181489a09141SPeter Zijlstra * 512MB: 23170k 181589a09141SPeter Zijlstra * 1GB: 32768k 181689a09141SPeter Zijlstra * 2GB: 46340k 181789a09141SPeter Zijlstra * 4GB: 65536k 181889a09141SPeter Zijlstra * 8GB: 92681k 181989a09141SPeter Zijlstra * 16GB: 131072k 182089a09141SPeter Zijlstra * 182189a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 182289a09141SPeter Zijlstra * Limit the default to 256M 182389a09141SPeter Zijlstra */ 182489a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 182589a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 182689a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 182789a09141SPeter Zijlstra 18281da177e4SLinus Torvalds return 0; 18291da177e4SLinus Torvalds } 18301da177e4SLinus Torvalds 1831266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 18321da177e4SLinus Torvalds { 18331da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 18341da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 18351a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 18361da177e4SLinus Torvalds } 18371da177e4SLinus Torvalds 1838