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 34f4ce1299STrond Myklebust #include "nfstrace.h" 35f4ce1299STrond Myklebust 361da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 391da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds /* 421da177e4SLinus Torvalds * Local function declarations 431da177e4SLinus Torvalds */ 44f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 456c75dc0dSFred Isaman static const struct rpc_call_ops nfs_write_common_ops; 46788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 47061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 48f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 491da177e4SLinus Torvalds 50e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 513feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 520b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 531da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 541da177e4SLinus Torvalds 550b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 561da177e4SLinus Torvalds { 57192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 5840859d7eSChuck Lever 591da177e4SLinus Torvalds if (p) { 601da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 611da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 621da177e4SLinus Torvalds } 631da177e4SLinus Torvalds return p; 641da177e4SLinus Torvalds } 65e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 661da177e4SLinus Torvalds 670b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 681da177e4SLinus Torvalds { 691da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 701da177e4SLinus Torvalds } 71e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 721da177e4SLinus Torvalds 736c75dc0dSFred Isaman struct nfs_write_header *nfs_writehdr_alloc(void) 743feb2d49STrond Myklebust { 75192e501bSMel Gorman struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 763feb2d49STrond Myklebust 773feb2d49STrond Myklebust if (p) { 78cd841605SFred Isaman struct nfs_pgio_header *hdr = &p->header; 79cd841605SFred Isaman 803feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 81cd841605SFred Isaman INIT_LIST_HEAD(&hdr->pages); 826c75dc0dSFred Isaman INIT_LIST_HEAD(&hdr->rpc_list); 836c75dc0dSFred Isaman spin_lock_init(&hdr->lock); 846c75dc0dSFred Isaman atomic_set(&hdr->refcnt, 0); 859bce008bSTrond Myklebust hdr->verf = &p->verf; 863feb2d49STrond Myklebust } 873feb2d49STrond Myklebust return p; 883feb2d49STrond Myklebust } 8989d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_writehdr_alloc); 903feb2d49STrond Myklebust 911763da12SFred Isaman static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr, 926c75dc0dSFred Isaman unsigned int pagecount) 936c75dc0dSFred Isaman { 946c75dc0dSFred Isaman struct nfs_write_data *data, *prealloc; 956c75dc0dSFred Isaman 966c75dc0dSFred Isaman prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data; 976c75dc0dSFred Isaman if (prealloc->header == NULL) 986c75dc0dSFred Isaman data = prealloc; 996c75dc0dSFred Isaman else 1006c75dc0dSFred Isaman data = kzalloc(sizeof(*data), GFP_KERNEL); 1016c75dc0dSFred Isaman if (!data) 1026c75dc0dSFred Isaman goto out; 1036c75dc0dSFred Isaman 1046c75dc0dSFred Isaman if (nfs_pgarray_set(&data->pages, pagecount)) { 1056c75dc0dSFred Isaman data->header = hdr; 1066c75dc0dSFred Isaman atomic_inc(&hdr->refcnt); 1076c75dc0dSFred Isaman } else { 1086c75dc0dSFred Isaman if (data != prealloc) 1096c75dc0dSFred Isaman kfree(data); 1106c75dc0dSFred Isaman data = NULL; 1116c75dc0dSFred Isaman } 1126c75dc0dSFred Isaman out: 1136c75dc0dSFred Isaman return data; 1146c75dc0dSFred Isaman } 1156c75dc0dSFred Isaman 116cd841605SFred Isaman void nfs_writehdr_free(struct nfs_pgio_header *hdr) 1173feb2d49STrond Myklebust { 118cd841605SFred Isaman struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header); 119cd841605SFred Isaman mempool_free(whdr, nfs_wdata_mempool); 1203feb2d49STrond Myklebust } 12189d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_writehdr_free); 1223feb2d49STrond Myklebust 123dce81290STrond Myklebust void nfs_writedata_release(struct nfs_write_data *wdata) 1241da177e4SLinus Torvalds { 1256c75dc0dSFred Isaman struct nfs_pgio_header *hdr = wdata->header; 1266c75dc0dSFred Isaman struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header); 1276c75dc0dSFred Isaman 128383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 12930dd374fSFred Isaman if (wdata->pages.pagevec != wdata->pages.page_array) 13030dd374fSFred Isaman kfree(wdata->pages.pagevec); 1316db6dd7dSTrond Myklebust if (wdata == &write_header->rpc_data) { 1326c75dc0dSFred Isaman wdata->header = NULL; 1336db6dd7dSTrond Myklebust wdata = NULL; 1346db6dd7dSTrond Myklebust } 1356c75dc0dSFred Isaman if (atomic_dec_and_test(&hdr->refcnt)) 136061ae2edSFred Isaman hdr->completion_ops->completion(hdr); 1376db6dd7dSTrond Myklebust /* Note: we only free the rpc_task after callbacks are done. 1386db6dd7dSTrond Myklebust * See the comment in rpc_free_task() for why 1396db6dd7dSTrond Myklebust */ 1406db6dd7dSTrond Myklebust kfree(wdata); 1411da177e4SLinus Torvalds } 14289d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_writedata_release); 1431da177e4SLinus Torvalds 1447b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1457b159fc1STrond Myklebust { 1467b159fc1STrond Myklebust ctx->error = error; 1477b159fc1STrond Myklebust smp_wmb(); 1487b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1497b159fc1STrond Myklebust } 1507b159fc1STrond Myklebust 15129418aa4SMel Gorman static struct nfs_page * 15229418aa4SMel Gorman nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page) 153277459d2STrond Myklebust { 154277459d2STrond Myklebust struct nfs_page *req = NULL; 155277459d2STrond Myklebust 15629418aa4SMel Gorman if (PagePrivate(page)) 157277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 15829418aa4SMel Gorman else if (unlikely(PageSwapCache(page))) { 15929418aa4SMel Gorman struct nfs_page *freq, *t; 16029418aa4SMel Gorman 16129418aa4SMel Gorman /* Linearly search the commit list for the correct req */ 16229418aa4SMel Gorman list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { 16329418aa4SMel Gorman if (freq->wb_page == page) { 16429418aa4SMel Gorman req = freq; 16529418aa4SMel Gorman break; 166277459d2STrond Myklebust } 16729418aa4SMel Gorman } 16829418aa4SMel Gorman } 16929418aa4SMel Gorman 17029418aa4SMel Gorman if (req) 17129418aa4SMel Gorman kref_get(&req->wb_kref); 17229418aa4SMel Gorman 173277459d2STrond Myklebust return req; 174277459d2STrond Myklebust } 175277459d2STrond Myklebust 176277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 177277459d2STrond Myklebust { 178d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 179277459d2STrond Myklebust struct nfs_page *req = NULL; 180277459d2STrond Myklebust 181587142f8STrond Myklebust spin_lock(&inode->i_lock); 18229418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 183587142f8STrond Myklebust spin_unlock(&inode->i_lock); 184277459d2STrond Myklebust return req; 185277459d2STrond Myklebust } 186277459d2STrond Myklebust 1871da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1881da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1891da177e4SLinus Torvalds { 190d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 191a3d01454STrond Myklebust loff_t end, i_size; 192a3d01454STrond Myklebust pgoff_t end_index; 1931da177e4SLinus Torvalds 194a3d01454STrond Myklebust spin_lock(&inode->i_lock); 195a3d01454STrond Myklebust i_size = i_size_read(inode); 196a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 197d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 198a3d01454STrond Myklebust goto out; 199d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2001da177e4SLinus Torvalds if (i_size >= end) 201a3d01454STrond Myklebust goto out; 2021da177e4SLinus Torvalds i_size_write(inode, end); 203a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 204a3d01454STrond Myklebust out: 205a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2061da177e4SLinus Torvalds } 2071da177e4SLinus Torvalds 208a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 209a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 210a301b777STrond Myklebust { 211d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 212a301b777STrond Myklebust } 213a301b777STrond Myklebust 2141da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2151da177e4SLinus Torvalds * covers the full page. 2161da177e4SLinus Torvalds */ 2171da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 2181da177e4SLinus Torvalds { 2191da177e4SLinus Torvalds if (PageUptodate(page)) 2201da177e4SLinus Torvalds return; 2211da177e4SLinus Torvalds if (base != 0) 2221da177e4SLinus Torvalds return; 22349a70f27STrond Myklebust if (count != nfs_page_length(page)) 2241da177e4SLinus Torvalds return; 2251da177e4SLinus Torvalds SetPageUptodate(page); 2261da177e4SLinus Torvalds } 2271da177e4SLinus Torvalds 2281da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2291da177e4SLinus Torvalds { 2301da177e4SLinus Torvalds if (wbc->for_reclaim) 231c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 232b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 233b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 234b31268acSTrond Myklebust return FLUSH_COND_STABLE; 2351da177e4SLinus Torvalds } 2361da177e4SLinus Torvalds 2371da177e4SLinus Torvalds /* 23889a09141SPeter Zijlstra * NFS congestion control 23989a09141SPeter Zijlstra */ 24089a09141SPeter Zijlstra 24189a09141SPeter Zijlstra int nfs_congestion_kb; 24289a09141SPeter Zijlstra 24389a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 24489a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 24589a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 24689a09141SPeter Zijlstra 247deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 24889a09141SPeter Zijlstra { 249deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2505a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2515a6d41b3STrond Myklebust 252deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 25389a09141SPeter Zijlstra 254277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2558aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2568aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2578aa7e847SJens Axboe BLK_RW_ASYNC); 2588aa7e847SJens Axboe } 25989a09141SPeter Zijlstra } 26089a09141SPeter Zijlstra 26189a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 26289a09141SPeter Zijlstra { 263d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 26489a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 26589a09141SPeter Zijlstra 26689a09141SPeter Zijlstra end_page_writeback(page); 267c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2688aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 26989a09141SPeter Zijlstra } 27089a09141SPeter Zijlstra 271cfb506e1STrond Myklebust static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) 272e261f51fSTrond Myklebust { 273d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 274e261f51fSTrond Myklebust struct nfs_page *req; 275e261f51fSTrond Myklebust int ret; 276e261f51fSTrond Myklebust 277587142f8STrond Myklebust spin_lock(&inode->i_lock); 278e261f51fSTrond Myklebust for (;;) { 27929418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 280074cc1deSTrond Myklebust if (req == NULL) 281074cc1deSTrond Myklebust break; 2827ad84aa9STrond Myklebust if (nfs_lock_request(req)) 283e261f51fSTrond Myklebust break; 284e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 2857ad84aa9STrond Myklebust * then the call to nfs_lock_request() will always 286e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 287e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 288e261f51fSTrond Myklebust */ 289587142f8STrond Myklebust spin_unlock(&inode->i_lock); 290cfb506e1STrond Myklebust if (!nonblock) 291e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 292cfb506e1STrond Myklebust else 293cfb506e1STrond Myklebust ret = -EAGAIN; 294e261f51fSTrond Myklebust nfs_release_request(req); 295e261f51fSTrond Myklebust if (ret != 0) 296074cc1deSTrond Myklebust return ERR_PTR(ret); 297587142f8STrond Myklebust spin_lock(&inode->i_lock); 298e261f51fSTrond Myklebust } 299587142f8STrond Myklebust spin_unlock(&inode->i_lock); 300074cc1deSTrond Myklebust return req; 301612c9384STrond Myklebust } 302074cc1deSTrond Myklebust 303074cc1deSTrond Myklebust /* 304074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 305074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 306074cc1deSTrond Myklebust */ 307074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 308cfb506e1STrond Myklebust struct page *page, bool nonblock) 309074cc1deSTrond Myklebust { 310074cc1deSTrond Myklebust struct nfs_page *req; 311074cc1deSTrond Myklebust int ret = 0; 312074cc1deSTrond Myklebust 313cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, nonblock); 314074cc1deSTrond Myklebust if (!req) 315074cc1deSTrond Myklebust goto out; 316074cc1deSTrond Myklebust ret = PTR_ERR(req); 317074cc1deSTrond Myklebust if (IS_ERR(req)) 318074cc1deSTrond Myklebust goto out; 319074cc1deSTrond Myklebust 320deed85e7STrond Myklebust nfs_set_page_writeback(page); 321deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 322074cc1deSTrond Myklebust 323deed85e7STrond Myklebust ret = 0; 324f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 325f8512ad0SFred Isaman nfs_redirty_request(req); 326074cc1deSTrond Myklebust ret = pgio->pg_error; 327f8512ad0SFred Isaman } 328074cc1deSTrond Myklebust out: 329074cc1deSTrond Myklebust return ret; 330e261f51fSTrond Myklebust } 331e261f51fSTrond Myklebust 332f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 333f758c885STrond Myklebust { 334d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 335cfb506e1STrond Myklebust int ret; 336f758c885STrond Myklebust 337f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 338f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 339f758c885STrond Myklebust 340d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 3411b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 342cfb506e1STrond Myklebust if (ret == -EAGAIN) { 343cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 344cfb506e1STrond Myklebust ret = 0; 345cfb506e1STrond Myklebust } 346cfb506e1STrond Myklebust return ret; 347f758c885STrond Myklebust } 348f758c885STrond Myklebust 349e261f51fSTrond Myklebust /* 3501da177e4SLinus Torvalds * Write an mmapped page to the server. 3511da177e4SLinus Torvalds */ 3524d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3531da177e4SLinus Torvalds { 354f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 355e261f51fSTrond Myklebust int err; 3561da177e4SLinus Torvalds 357a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 358a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 359f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 360f758c885STrond Myklebust nfs_pageio_complete(&pgio); 361f758c885STrond Myklebust if (err < 0) 3624d770ccfSTrond Myklebust return err; 363f758c885STrond Myklebust if (pgio.pg_error < 0) 364f758c885STrond Myklebust return pgio.pg_error; 365f758c885STrond Myklebust return 0; 3664d770ccfSTrond Myklebust } 3674d770ccfSTrond Myklebust 3684d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3694d770ccfSTrond Myklebust { 370f758c885STrond Myklebust int ret; 3714d770ccfSTrond Myklebust 372f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3731da177e4SLinus Torvalds unlock_page(page); 374f758c885STrond Myklebust return ret; 375f758c885STrond Myklebust } 376f758c885STrond Myklebust 377f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 378f758c885STrond Myklebust { 379f758c885STrond Myklebust int ret; 380f758c885STrond Myklebust 381f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 382f758c885STrond Myklebust unlock_page(page); 383f758c885STrond Myklebust return ret; 3841da177e4SLinus Torvalds } 3851da177e4SLinus Torvalds 3861da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3871da177e4SLinus Torvalds { 3881da177e4SLinus Torvalds struct inode *inode = mapping->host; 38972cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 390c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3911da177e4SLinus Torvalds int err; 3921da177e4SLinus Torvalds 39372cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 39472cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 39572cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 39672cb77f4STrond Myklebust if (err) 39772cb77f4STrond Myklebust goto out_err; 39872cb77f4STrond Myklebust 39991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 40091d5b470SChuck Lever 401a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 402a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 403f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 404c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 40572cb77f4STrond Myklebust 40672cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 40772cb77f4STrond Myklebust smp_mb__after_clear_bit(); 40872cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 40972cb77f4STrond Myklebust 410f758c885STrond Myklebust if (err < 0) 41172cb77f4STrond Myklebust goto out_err; 41272cb77f4STrond Myklebust err = pgio.pg_error; 41372cb77f4STrond Myklebust if (err < 0) 41472cb77f4STrond Myklebust goto out_err; 415c63c7b05STrond Myklebust return 0; 41672cb77f4STrond Myklebust out_err: 41772cb77f4STrond Myklebust return err; 4181da177e4SLinus Torvalds } 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds /* 4211da177e4SLinus Torvalds * Insert a write request into an inode 4221da177e4SLinus Torvalds */ 423d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 4241da177e4SLinus Torvalds { 4251da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 426e7d39069STrond Myklebust 427e7d39069STrond Myklebust /* Lock the request! */ 4287ad84aa9STrond Myklebust nfs_lock_request(req); 429e7d39069STrond Myklebust 430e7d39069STrond Myklebust spin_lock(&inode->i_lock); 431011e2a7fSBryan Schumaker if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 432a9a4a87aSTrond Myklebust inode->i_version++; 43329418aa4SMel Gorman /* 43429418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 43529418aa4SMel Gorman * with invalidate/truncate. 43629418aa4SMel Gorman */ 43729418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 4382df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 439deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 440277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 44129418aa4SMel Gorman } 4421da177e4SLinus Torvalds nfsi->npages++; 443c03b4024STrond Myklebust kref_get(&req->wb_kref); 444e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 4451da177e4SLinus Torvalds } 4461da177e4SLinus Torvalds 4471da177e4SLinus Torvalds /* 44889a09141SPeter Zijlstra * Remove a write request from an inode 4491da177e4SLinus Torvalds */ 4501da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4511da177e4SLinus Torvalds { 4523d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 4531da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4541da177e4SLinus Torvalds 455587142f8STrond Myklebust spin_lock(&inode->i_lock); 45629418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 457277459d2STrond Myklebust set_page_private(req->wb_page, 0); 458deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4592df485a7STrond Myklebust clear_bit(PG_MAPPED, &req->wb_flags); 46029418aa4SMel Gorman } 4611da177e4SLinus Torvalds nfsi->npages--; 462587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4631da177e4SLinus Torvalds nfs_release_request(req); 4641da177e4SLinus Torvalds } 4651da177e4SLinus Torvalds 46661822ab5STrond Myklebust static void 4676d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 46861822ab5STrond Myklebust { 46961822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 47061822ab5STrond Myklebust } 47161822ab5STrond Myklebust 47289d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 4738dd37758STrond Myklebust /** 4748dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 4758dd37758STrond Myklebust * @req: pointer to a struct nfs_page 476ea2cf228SFred Isaman * @dst: commit list head 477ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 4788dd37758STrond Myklebust * 479ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 4808dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 4818dd37758STrond Myklebust * the MM page stats. 4828dd37758STrond Myklebust * 483ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 4848dd37758STrond Myklebust * holding the nfs_page lock. 4858dd37758STrond Myklebust */ 4868dd37758STrond Myklebust void 487ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 488ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 4898dd37758STrond Myklebust { 4908dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 491ea2cf228SFred Isaman spin_lock(cinfo->lock); 492ea2cf228SFred Isaman nfs_list_add_request(req, dst); 493ea2cf228SFred Isaman cinfo->mds->ncommit++; 494ea2cf228SFred Isaman spin_unlock(cinfo->lock); 49556f9cd68SFred Isaman if (!cinfo->dreq) { 4968dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 497d56b4ddfSMel Gorman inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 49856f9cd68SFred Isaman BDI_RECLAIMABLE); 49956f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 50056f9cd68SFred Isaman I_DIRTY_DATASYNC); 50156f9cd68SFred Isaman } 5028dd37758STrond Myklebust } 5038dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 5048dd37758STrond Myklebust 5058dd37758STrond Myklebust /** 5068dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 5078dd37758STrond Myklebust * @req: pointer to a nfs_page 508ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 5098dd37758STrond Myklebust * 510ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 5118dd37758STrond Myklebust * number of outstanding requests requiring a commit 5128dd37758STrond Myklebust * It does not update the MM page stats. 5138dd37758STrond Myklebust * 514ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 5158dd37758STrond Myklebust */ 5168dd37758STrond Myklebust void 517ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 518ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5198dd37758STrond Myklebust { 5208dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 5218dd37758STrond Myklebust return; 5228dd37758STrond Myklebust nfs_list_remove_request(req); 523ea2cf228SFred Isaman cinfo->mds->ncommit--; 5248dd37758STrond Myklebust } 5258dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 5268dd37758STrond Myklebust 527ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 528ea2cf228SFred Isaman struct inode *inode) 529ea2cf228SFred Isaman { 530ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 531ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 532ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 533b359f9d0SFred Isaman cinfo->dreq = NULL; 534f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 535ea2cf228SFred Isaman } 536ea2cf228SFred Isaman 537ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 538ea2cf228SFred Isaman struct inode *inode, 539ea2cf228SFred Isaman struct nfs_direct_req *dreq) 540ea2cf228SFred Isaman { 5411763da12SFred Isaman if (dreq) 5421763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 5431763da12SFred Isaman else 544ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 545ea2cf228SFred Isaman } 546ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 5478dd37758STrond Myklebust 5481da177e4SLinus Torvalds /* 5491da177e4SLinus Torvalds * Add a request to the inode's commit list. 5501da177e4SLinus Torvalds */ 5511763da12SFred Isaman void 552ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 553ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5541da177e4SLinus Torvalds { 555ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 5568dd37758STrond Myklebust return; 557ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 5581da177e4SLinus Torvalds } 5598e821cadSTrond Myklebust 560d6d6dc7cSFred Isaman static void 561d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 562e468bae9STrond Myklebust { 563e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 564d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); 565e468bae9STrond Myklebust } 566d6d6dc7cSFred Isaman 5678dd37758STrond Myklebust static void 568d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 569d6d6dc7cSFred Isaman { 5708dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 5718dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 572ea2cf228SFred Isaman struct nfs_commit_info cinfo; 573d6d6dc7cSFred Isaman 574ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 575ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 576ea2cf228SFred Isaman spin_lock(cinfo.lock); 577ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 578ea2cf228SFred Isaman spin_unlock(cinfo.lock); 579d6d6dc7cSFred Isaman } 5808dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 5818dd37758STrond Myklebust } 582e468bae9STrond Myklebust } 583e468bae9STrond Myklebust 5848e821cadSTrond Myklebust static inline 5858e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 5868e821cadSTrond Myklebust { 587465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 588cd841605SFred Isaman return data->header->lseg == NULL; 5898e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 5908e821cadSTrond Myklebust } 5918e821cadSTrond Myklebust 5928e821cadSTrond Myklebust #else 59368cd6fa4SBryan Schumaker static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 59468cd6fa4SBryan Schumaker struct inode *inode) 59568cd6fa4SBryan Schumaker { 59668cd6fa4SBryan Schumaker } 59768cd6fa4SBryan Schumaker 59868cd6fa4SBryan Schumaker void nfs_init_cinfo(struct nfs_commit_info *cinfo, 59968cd6fa4SBryan Schumaker struct inode *inode, 60068cd6fa4SBryan Schumaker struct nfs_direct_req *dreq) 60168cd6fa4SBryan Schumaker { 60268cd6fa4SBryan Schumaker } 60368cd6fa4SBryan Schumaker 6041763da12SFred Isaman void 605ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 606ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 6078e821cadSTrond Myklebust { 6088e821cadSTrond Myklebust } 6098e821cadSTrond Myklebust 6108dd37758STrond Myklebust static void 611e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 612e468bae9STrond Myklebust { 613e468bae9STrond Myklebust } 614e468bae9STrond Myklebust 6158e821cadSTrond Myklebust static inline 6168e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 6178e821cadSTrond Myklebust { 6188e821cadSTrond Myklebust return 0; 6198e821cadSTrond Myklebust } 6208e821cadSTrond Myklebust 6211da177e4SLinus Torvalds #endif 6221da177e4SLinus Torvalds 623061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 6246c75dc0dSFred Isaman { 625ea2cf228SFred Isaman struct nfs_commit_info cinfo; 6266c75dc0dSFred Isaman unsigned long bytes = 0; 6276c75dc0dSFred Isaman 6286c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 6296c75dc0dSFred Isaman goto out; 630ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 6316c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 6326c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 6336c75dc0dSFred Isaman 6346c75dc0dSFred Isaman bytes += req->wb_bytes; 6356c75dc0dSFred Isaman nfs_list_remove_request(req); 6366c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 6376c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 638d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 6396c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 6406c75dc0dSFred Isaman goto remove_req; 6416c75dc0dSFred Isaman } 6426c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { 6436c75dc0dSFred Isaman nfs_mark_request_dirty(req); 6446c75dc0dSFred Isaman goto next; 6456c75dc0dSFred Isaman } 6466c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { 6472f2c63bcSTrond Myklebust memcpy(&req->wb_verf, &hdr->verf->verifier, sizeof(req->wb_verf)); 648ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 6496c75dc0dSFred Isaman goto next; 6506c75dc0dSFred Isaman } 6516c75dc0dSFred Isaman remove_req: 6526c75dc0dSFred Isaman nfs_inode_remove_request(req); 6536c75dc0dSFred Isaman next: 6541d1afcbcSTrond Myklebust nfs_unlock_request(req); 655d1182b33STrond Myklebust nfs_end_page_writeback(req->wb_page); 6563aff4ebbSTrond Myklebust nfs_release_request(req); 6576c75dc0dSFred Isaman } 6586c75dc0dSFred Isaman out: 6596c75dc0dSFred Isaman hdr->release(hdr); 6606c75dc0dSFred Isaman } 6616c75dc0dSFred Isaman 66289d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 663ea2cf228SFred Isaman static unsigned long 664ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 665fb8a1f11STrond Myklebust { 666ea2cf228SFred Isaman return cinfo->mds->ncommit; 667fb8a1f11STrond Myklebust } 668fb8a1f11STrond Myklebust 669ea2cf228SFred Isaman /* cinfo->lock held by caller */ 6701763da12SFred Isaman int 671ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 672ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 673d6d6dc7cSFred Isaman { 674d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 675d6d6dc7cSFred Isaman int ret = 0; 676d6d6dc7cSFred Isaman 677d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 6788dd37758STrond Myklebust if (!nfs_lock_request(req)) 6798dd37758STrond Myklebust continue; 6807ad84aa9STrond Myklebust kref_get(&req->wb_kref); 681ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 6823b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 683ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 6848dd37758STrond Myklebust nfs_list_add_request(req, dst); 685d6d6dc7cSFred Isaman ret++; 6861763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 687d6d6dc7cSFred Isaman break; 688d6d6dc7cSFred Isaman } 689d6d6dc7cSFred Isaman return ret; 690d6d6dc7cSFred Isaman } 691d6d6dc7cSFred Isaman 6921da177e4SLinus Torvalds /* 6931da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 6941da177e4SLinus Torvalds * @inode: NFS inode to scan 695ea2cf228SFred Isaman * @dst: mds destination list 696ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 6971da177e4SLinus Torvalds * 6981da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 6991da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 7001da177e4SLinus Torvalds */ 7011763da12SFred Isaman int 702ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 703ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7041da177e4SLinus Torvalds { 705d6d6dc7cSFred Isaman int ret = 0; 706fb8a1f11STrond Myklebust 707ea2cf228SFred Isaman spin_lock(cinfo->lock); 708ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 7098dd37758STrond Myklebust const int max = INT_MAX; 710d6d6dc7cSFred Isaman 711ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 712ea2cf228SFred Isaman cinfo, max); 713ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 714d6d6dc7cSFred Isaman } 715ea2cf228SFred Isaman spin_unlock(cinfo->lock); 716ff778d02STrond Myklebust return ret; 7171da177e4SLinus Torvalds } 718d6d6dc7cSFred Isaman 719c42de9ddSTrond Myklebust #else 720ea2cf228SFred Isaman static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 721fb8a1f11STrond Myklebust { 722fb8a1f11STrond Myklebust return 0; 723fb8a1f11STrond Myklebust } 724fb8a1f11STrond Myklebust 7251763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 726ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 727c42de9ddSTrond Myklebust { 728c42de9ddSTrond Myklebust return 0; 729c42de9ddSTrond Myklebust } 7301da177e4SLinus Torvalds #endif 7311da177e4SLinus Torvalds 7321da177e4SLinus Torvalds /* 733e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 734e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 7351da177e4SLinus Torvalds * 736e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 737e7d39069STrond Myklebust * to disk. 7381da177e4SLinus Torvalds */ 739e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 740e7d39069STrond Myklebust struct page *page, 741e7d39069STrond Myklebust unsigned int offset, 742e7d39069STrond Myklebust unsigned int bytes) 7431da177e4SLinus Torvalds { 744e7d39069STrond Myklebust struct nfs_page *req; 745e7d39069STrond Myklebust unsigned int rqend; 746e7d39069STrond Myklebust unsigned int end; 7471da177e4SLinus Torvalds int error; 748277459d2STrond Myklebust 749e7d39069STrond Myklebust if (!PagePrivate(page)) 750e7d39069STrond Myklebust return NULL; 751e7d39069STrond Myklebust 752e7d39069STrond Myklebust end = offset + bytes; 753e7d39069STrond Myklebust spin_lock(&inode->i_lock); 754e7d39069STrond Myklebust 755e7d39069STrond Myklebust for (;;) { 75629418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 757e7d39069STrond Myklebust if (req == NULL) 758e7d39069STrond Myklebust goto out_unlock; 759e7d39069STrond Myklebust 760e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 761e7d39069STrond Myklebust /* 762e7d39069STrond Myklebust * Tell the caller to flush out the request if 763e7d39069STrond Myklebust * the offsets are non-contiguous. 764e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 765e7d39069STrond Myklebust * have flushed out requests having wrong owners. 766e7d39069STrond Myklebust */ 767e468bae9STrond Myklebust if (offset > rqend 768e7d39069STrond Myklebust || end < req->wb_offset) 769e7d39069STrond Myklebust goto out_flushme; 770e7d39069STrond Myklebust 7717ad84aa9STrond Myklebust if (nfs_lock_request(req)) 772e7d39069STrond Myklebust break; 773e7d39069STrond Myklebust 774e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 775587142f8STrond Myklebust spin_unlock(&inode->i_lock); 7761da177e4SLinus Torvalds error = nfs_wait_on_request(req); 7771da177e4SLinus Torvalds nfs_release_request(req); 778e7d39069STrond Myklebust if (error != 0) 779e7d39069STrond Myklebust goto out_err; 780e7d39069STrond Myklebust spin_lock(&inode->i_lock); 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds 7831da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7841da177e4SLinus Torvalds if (offset < req->wb_offset) { 7851da177e4SLinus Torvalds req->wb_offset = offset; 7861da177e4SLinus Torvalds req->wb_pgbase = offset; 7871da177e4SLinus Torvalds } 7881da177e4SLinus Torvalds if (end > rqend) 7891da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 790e7d39069STrond Myklebust else 791e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 792e7d39069STrond Myklebust out_unlock: 793e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 794ca138f36SFred Isaman if (req) 7958dd37758STrond Myklebust nfs_clear_request_commit(req); 796e7d39069STrond Myklebust return req; 797e7d39069STrond Myklebust out_flushme: 798e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 799e7d39069STrond Myklebust nfs_release_request(req); 800e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 801e7d39069STrond Myklebust out_err: 802e7d39069STrond Myklebust return ERR_PTR(error); 803e7d39069STrond Myklebust } 8041da177e4SLinus Torvalds 805e7d39069STrond Myklebust /* 806e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 807e7d39069STrond Myklebust * 808e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 809e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 810e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 811e7d39069STrond Myklebust */ 812e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 813e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 814e7d39069STrond Myklebust { 815d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 816e7d39069STrond Myklebust struct nfs_page *req; 817e7d39069STrond Myklebust 818e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 819e7d39069STrond Myklebust if (req != NULL) 820e7d39069STrond Myklebust goto out; 821e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 822e7d39069STrond Myklebust if (IS_ERR(req)) 823e7d39069STrond Myklebust goto out; 824d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 825efc91ed0STrond Myklebust out: 82661e930a9STrond Myklebust return req; 8271da177e4SLinus Torvalds } 8281da177e4SLinus Torvalds 829e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 830e7d39069STrond Myklebust unsigned int offset, unsigned int count) 831e7d39069STrond Myklebust { 832e7d39069STrond Myklebust struct nfs_page *req; 833e7d39069STrond Myklebust 834e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 835e7d39069STrond Myklebust if (IS_ERR(req)) 836e7d39069STrond Myklebust return PTR_ERR(req); 837e7d39069STrond Myklebust /* Update file length */ 838e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 839e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 840a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 8411d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 842e7d39069STrond Myklebust return 0; 843e7d39069STrond Myklebust } 844e7d39069STrond Myklebust 8451da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 8461da177e4SLinus Torvalds { 847cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 8482a369153STrond Myklebust struct nfs_lock_context *l_ctx; 8491da177e4SLinus Torvalds struct nfs_page *req; 8501a54533eSTrond Myklebust int do_flush, status; 8511da177e4SLinus Torvalds /* 8521da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 8531da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 8541da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 8551da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 8561da177e4SLinus Torvalds * Also do the same if we find a request from an existing 8571da177e4SLinus Torvalds * dropped page. 8581da177e4SLinus Torvalds */ 8591a54533eSTrond Myklebust do { 860277459d2STrond Myklebust req = nfs_page_find_request(page); 8611a54533eSTrond Myklebust if (req == NULL) 8621a54533eSTrond Myklebust return 0; 8632a369153STrond Myklebust l_ctx = req->wb_lock_context; 8642a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 8650f1d2605STrond Myklebust if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { 8662a369153STrond Myklebust do_flush |= l_ctx->lockowner.l_owner != current->files 8672a369153STrond Myklebust || l_ctx->lockowner.l_pid != current->tgid; 8682a369153STrond Myklebust } 8691da177e4SLinus Torvalds nfs_release_request(req); 8701a54533eSTrond Myklebust if (!do_flush) 8711a54533eSTrond Myklebust return 0; 872d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 8731a54533eSTrond Myklebust } while (status == 0); 8741a54533eSTrond Myklebust return status; 8751da177e4SLinus Torvalds } 8761da177e4SLinus Torvalds 8771da177e4SLinus Torvalds /* 878dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 879dc24826bSAndy Adamson * expire soon. 880dc24826bSAndy Adamson * 881dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 882dc24826bSAndy Adamson * 883dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 884dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 885dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 886dc24826bSAndy Adamson */ 887dc24826bSAndy Adamson int 888dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 889dc24826bSAndy Adamson { 890dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 891dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 892dc24826bSAndy Adamson 893dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 894dc24826bSAndy Adamson } 895dc24826bSAndy Adamson 896dc24826bSAndy Adamson /* 897dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 898dc24826bSAndy Adamson */ 899dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 900dc24826bSAndy Adamson { 901dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 902dc24826bSAndy Adamson } 903dc24826bSAndy Adamson 904dc24826bSAndy Adamson /* 9055d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 9065d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 9075d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 9085d47a356STrond Myklebust */ 9098d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 9105d47a356STrond Myklebust { 911d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 912d529ef83SJeff Layton 9138d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 9148d197a56STrond Myklebust goto out; 915d529ef83SJeff Layton if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE)) 916d529ef83SJeff Layton return false; 9174db72b40SJeff Layton smp_rmb(); 918d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 9198d197a56STrond Myklebust return false; 9208d197a56STrond Myklebust out: 9218d197a56STrond Myklebust return PageUptodate(page) != 0; 9225d47a356STrond Myklebust } 9235d47a356STrond Myklebust 924c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 925c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 926c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 927c7559663SScott Mayhew * inefficiencies. 928c7559663SScott Mayhew * 929263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 930263b4509SScott Mayhew * of the checks. 931c7559663SScott Mayhew */ 932c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 933c7559663SScott Mayhew { 934c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 935c7559663SScott Mayhew return 0; 936263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 937263b4509SScott Mayhew return 0; 938c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 939c7559663SScott Mayhew return 1; 940263b4509SScott Mayhew if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && 941c7559663SScott Mayhew inode->i_flock->fl_end == OFFSET_MAX && 942263b4509SScott Mayhew inode->i_flock->fl_type != F_RDLCK)) 943c7559663SScott Mayhew return 1; 944c7559663SScott Mayhew return 0; 945c7559663SScott Mayhew } 946c7559663SScott Mayhew 9475d47a356STrond Myklebust /* 9481da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 9491da177e4SLinus Torvalds * 9501da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 9511da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 9521da177e4SLinus Torvalds */ 9531da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 9541da177e4SLinus Torvalds unsigned int offset, unsigned int count) 9551da177e4SLinus Torvalds { 956cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 957d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 9581da177e4SLinus Torvalds int status = 0; 9591da177e4SLinus Torvalds 96091d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 96191d5b470SChuck Lever 9626de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 9636de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 9641da177e4SLinus Torvalds 965c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 96649a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 9671da177e4SLinus Torvalds offset = 0; 9681da177e4SLinus Torvalds } 9691da177e4SLinus Torvalds 970e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 97103fa9e84STrond Myklebust if (status < 0) 97203fa9e84STrond Myklebust nfs_set_pageerror(page); 97359b7c05fSTrond Myklebust else 97459b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 9751da177e4SLinus Torvalds 97648186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 9771da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 9781da177e4SLinus Torvalds return status; 9791da177e4SLinus Torvalds } 9801da177e4SLinus Torvalds 9813ff7576dSTrond Myklebust static int flush_task_priority(int how) 9821da177e4SLinus Torvalds { 9831da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 9841da177e4SLinus Torvalds case FLUSH_HIGHPRI: 9851da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 9861da177e4SLinus Torvalds case FLUSH_LOWPRI: 9871da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 9881da177e4SLinus Torvalds } 9891da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 9901da177e4SLinus Torvalds } 9911da177e4SLinus Torvalds 992c5996c4eSFred Isaman int nfs_initiate_write(struct rpc_clnt *clnt, 993c5996c4eSFred Isaman struct nfs_write_data *data, 994788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 9959f0ec176SAndy Adamson int how, int flags) 9961da177e4SLinus Torvalds { 997cd841605SFred Isaman struct inode *inode = data->header->inode; 9983ff7576dSTrond Myklebust int priority = flush_task_priority(how); 99907737691STrond Myklebust struct rpc_task *task; 1000bdc7f021STrond Myklebust struct rpc_message msg = { 1001bdc7f021STrond Myklebust .rpc_argp = &data->args, 1002bdc7f021STrond Myklebust .rpc_resp = &data->res, 1003cd841605SFred Isaman .rpc_cred = data->header->cred, 1004bdc7f021STrond Myklebust }; 100584115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 1006d138d5d1SAndy Adamson .rpc_client = clnt, 100707737691STrond Myklebust .task = &data->task, 1008bdc7f021STrond Myklebust .rpc_message = &msg, 100984115e1cSTrond Myklebust .callback_ops = call_ops, 101084115e1cSTrond Myklebust .callback_data = data, 1011101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 10129f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 10133ff7576dSTrond Myklebust .priority = priority, 101484115e1cSTrond Myklebust }; 10152c61be0aSTrond Myklebust int ret = 0; 10161da177e4SLinus Torvalds 1017d138d5d1SAndy Adamson /* Set up the initial task struct. */ 1018d138d5d1SAndy Adamson NFS_PROTO(inode)->write_setup(data, &msg); 1019d138d5d1SAndy Adamson 1020d138d5d1SAndy Adamson dprintk("NFS: %5u initiated write call " 10211e8968c5SNiels de Vos "(req %s/%llu, %u bytes @ offset %llu)\n", 1022d138d5d1SAndy Adamson data->task.tk_pid, 1023d138d5d1SAndy Adamson inode->i_sb->s_id, 10241e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(inode), 1025d138d5d1SAndy Adamson data->args.count, 1026d138d5d1SAndy Adamson (unsigned long long)data->args.offset); 1027d138d5d1SAndy Adamson 10288c21c62cSWeston Andros Adamson nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 10298c21c62cSWeston Andros Adamson &task_setup_data.rpc_client, &msg, data); 10308c21c62cSWeston Andros Adamson 1031d138d5d1SAndy Adamson task = rpc_run_task(&task_setup_data); 1032d138d5d1SAndy Adamson if (IS_ERR(task)) { 1033d138d5d1SAndy Adamson ret = PTR_ERR(task); 1034d138d5d1SAndy Adamson goto out; 1035d138d5d1SAndy Adamson } 1036d138d5d1SAndy Adamson if (how & FLUSH_SYNC) { 1037d138d5d1SAndy Adamson ret = rpc_wait_for_completion_task(task); 1038d138d5d1SAndy Adamson if (ret == 0) 1039d138d5d1SAndy Adamson ret = task->tk_status; 1040d138d5d1SAndy Adamson } 1041d138d5d1SAndy Adamson rpc_put_task(task); 1042d138d5d1SAndy Adamson out: 1043d138d5d1SAndy Adamson return ret; 1044d138d5d1SAndy Adamson } 1045a69aef14SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_write); 1046d138d5d1SAndy Adamson 1047d138d5d1SAndy Adamson /* 1048d138d5d1SAndy Adamson * Set up the argument/result storage required for the RPC call. 1049d138d5d1SAndy Adamson */ 10506c75dc0dSFred Isaman static void nfs_write_rpcsetup(struct nfs_write_data *data, 1051d138d5d1SAndy Adamson unsigned int count, unsigned int offset, 1052ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 1053d138d5d1SAndy Adamson { 10546c75dc0dSFred Isaman struct nfs_page *req = data->header->req; 1055d138d5d1SAndy Adamson 10561da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 10571da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 10581da177e4SLinus Torvalds 10596c75dc0dSFred Isaman data->args.fh = NFS_FH(data->header->inode); 10601da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 10612bea038cSBoaz Harrosh /* pnfs_set_layoutcommit needs this */ 10622bea038cSBoaz Harrosh data->mds_offset = data->args.offset; 10631da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 106430dd374fSFred Isaman data->args.pages = data->pages.pagevec; 10651da177e4SLinus Torvalds data->args.count = count; 1066383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 1067f11ac8dbSTrond Myklebust data->args.lock_context = req->wb_lock_context; 1068bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 106987ed5eb4STrond Myklebust switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { 107087ed5eb4STrond Myklebust case 0: 107187ed5eb4STrond Myklebust break; 107287ed5eb4STrond Myklebust case FLUSH_COND_STABLE: 1073ea2cf228SFred Isaman if (nfs_reqs_to_commit(cinfo)) 107487ed5eb4STrond Myklebust break; 107587ed5eb4STrond Myklebust default: 1076bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 1077bdc7f021STrond Myklebust } 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds data->res.fattr = &data->fattr; 10801da177e4SLinus Torvalds data->res.count = count; 10811da177e4SLinus Torvalds data->res.verf = &data->verf; 10820e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 10836e4efd56STrond Myklebust } 10841da177e4SLinus Torvalds 10856e4efd56STrond Myklebust static int nfs_do_write(struct nfs_write_data *data, 10866e4efd56STrond Myklebust const struct rpc_call_ops *call_ops, 10876e4efd56STrond Myklebust int how) 10886e4efd56STrond Myklebust { 1089cd841605SFred Isaman struct inode *inode = data->header->inode; 10900382b744SAndy Adamson 10919f0ec176SAndy Adamson return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how, 0); 10921da177e4SLinus Torvalds } 10931da177e4SLinus Torvalds 1094275acaafSTrond Myklebust static int nfs_do_multiple_writes(struct list_head *head, 1095275acaafSTrond Myklebust const struct rpc_call_ops *call_ops, 1096275acaafSTrond Myklebust int how) 1097275acaafSTrond Myklebust { 1098275acaafSTrond Myklebust struct nfs_write_data *data; 1099275acaafSTrond Myklebust int ret = 0; 1100275acaafSTrond Myklebust 1101275acaafSTrond Myklebust while (!list_empty(head)) { 1102275acaafSTrond Myklebust int ret2; 1103275acaafSTrond Myklebust 11046c75dc0dSFred Isaman data = list_first_entry(head, struct nfs_write_data, list); 1105275acaafSTrond Myklebust list_del_init(&data->list); 1106275acaafSTrond Myklebust 1107dce81290STrond Myklebust ret2 = nfs_do_write(data, call_ops, how); 1108275acaafSTrond Myklebust if (ret == 0) 1109275acaafSTrond Myklebust ret = ret2; 1110275acaafSTrond Myklebust } 1111275acaafSTrond Myklebust return ret; 1112275acaafSTrond Myklebust } 1113275acaafSTrond Myklebust 11146d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 11156d884e8fSFred * call this on each, which will prepare them to be retried on next 11166d884e8fSFred * writeback using standard nfs. 11176d884e8fSFred */ 11186d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 11196d884e8fSFred { 11206d884e8fSFred nfs_mark_request_dirty(req); 11211d1afcbcSTrond Myklebust nfs_unlock_request(req); 1122d1182b33STrond Myklebust nfs_end_page_writeback(req->wb_page); 11233aff4ebbSTrond Myklebust nfs_release_request(req); 11246d884e8fSFred } 11256d884e8fSFred 1126061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 11276c75dc0dSFred Isaman { 11286c75dc0dSFred Isaman struct nfs_page *req; 11296c75dc0dSFred Isaman 11306c75dc0dSFred Isaman while (!list_empty(head)) { 11316c75dc0dSFred Isaman req = nfs_list_entry(head->next); 11326c75dc0dSFred Isaman nfs_list_remove_request(req); 11336c75dc0dSFred Isaman nfs_redirty_request(req); 11346c75dc0dSFred Isaman } 11356c75dc0dSFred Isaman } 11366c75dc0dSFred Isaman 1137061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1138061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1139061ae2edSFred Isaman .completion = nfs_write_completion, 1140061ae2edSFred Isaman }; 1141061ae2edSFred Isaman 114225b11dcdSTrond Myklebust static void nfs_flush_error(struct nfs_pageio_descriptor *desc, 114325b11dcdSTrond Myklebust struct nfs_pgio_header *hdr) 114425b11dcdSTrond Myklebust { 114525b11dcdSTrond Myklebust set_bit(NFS_IOHDR_REDO, &hdr->flags); 114625b11dcdSTrond Myklebust while (!list_empty(&hdr->rpc_list)) { 114725b11dcdSTrond Myklebust struct nfs_write_data *data = list_first_entry(&hdr->rpc_list, 114825b11dcdSTrond Myklebust struct nfs_write_data, list); 114925b11dcdSTrond Myklebust list_del(&data->list); 115025b11dcdSTrond Myklebust nfs_writedata_release(data); 115125b11dcdSTrond Myklebust } 115225b11dcdSTrond Myklebust desc->pg_completion_ops->error_cleanup(&desc->pg_list); 115325b11dcdSTrond Myklebust } 115425b11dcdSTrond Myklebust 11551da177e4SLinus Torvalds /* 11561da177e4SLinus Torvalds * Generate multiple small requests to write out a single 11571da177e4SLinus Torvalds * contiguous dirty area on one page. 11581da177e4SLinus Torvalds */ 11596c75dc0dSFred Isaman static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, 11606c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 11611da177e4SLinus Torvalds { 11626c75dc0dSFred Isaman struct nfs_page *req = hdr->req; 11631da177e4SLinus Torvalds struct page *page = req->wb_page; 11641da177e4SLinus Torvalds struct nfs_write_data *data; 1165d097971dSTrond Myklebust size_t wsize = desc->pg_bsize, nbytes; 1166e9f7bee1STrond Myklebust unsigned int offset; 11671da177e4SLinus Torvalds int requests = 0; 1168ea2cf228SFred Isaman struct nfs_commit_info cinfo; 11691da177e4SLinus Torvalds 1170ea2cf228SFred Isaman nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); 11711da177e4SLinus Torvalds 1172b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1173ea2cf228SFred Isaman (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || 1174b31268acSTrond Myklebust desc->pg_count > wsize)) 1175b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1176b31268acSTrond Myklebust 1177b31268acSTrond Myklebust 1178275acaafSTrond Myklebust offset = 0; 1179c76069bdSFred Isaman nbytes = desc->pg_count; 1180e9f7bee1STrond Myklebust do { 1181e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 1182e9f7bee1STrond Myklebust 11836c75dc0dSFred Isaman data = nfs_writedata_alloc(hdr, 1); 118425b11dcdSTrond Myklebust if (!data) { 118525b11dcdSTrond Myklebust nfs_flush_error(desc, hdr); 118625b11dcdSTrond Myklebust return -ENOMEM; 118725b11dcdSTrond Myklebust } 118830dd374fSFred Isaman data->pages.pagevec[0] = page; 1189ea2cf228SFred Isaman nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); 11906c75dc0dSFred Isaman list_add(&data->list, &hdr->rpc_list); 11911da177e4SLinus Torvalds requests++; 1192e9f7bee1STrond Myklebust nbytes -= len; 1193275acaafSTrond Myklebust offset += len; 1194e9f7bee1STrond Myklebust } while (nbytes != 0); 119525b11dcdSTrond Myklebust nfs_list_remove_request(req); 119625b11dcdSTrond Myklebust nfs_list_add_request(req, &hdr->pages); 11976c75dc0dSFred Isaman desc->pg_rpc_callops = &nfs_write_common_ops; 119825b11dcdSTrond Myklebust return 0; 11991da177e4SLinus Torvalds } 12001da177e4SLinus Torvalds 12011da177e4SLinus Torvalds /* 12021da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 12031da177e4SLinus Torvalds * The page must have been locked by the caller. 12041da177e4SLinus Torvalds * 12051da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 12061da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 12071da177e4SLinus Torvalds * that has been written but not committed. 12081da177e4SLinus Torvalds */ 12096c75dc0dSFred Isaman static int nfs_flush_one(struct nfs_pageio_descriptor *desc, 12106c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 12111da177e4SLinus Torvalds { 12121da177e4SLinus Torvalds struct nfs_page *req; 12131da177e4SLinus Torvalds struct page **pages; 12141da177e4SLinus Torvalds struct nfs_write_data *data; 1215c76069bdSFred Isaman struct list_head *head = &desc->pg_list; 1216ea2cf228SFred Isaman struct nfs_commit_info cinfo; 12171da177e4SLinus Torvalds 12186c75dc0dSFred Isaman data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base, 1219c76069bdSFred Isaman desc->pg_count)); 12206c75dc0dSFred Isaman if (!data) { 122125b11dcdSTrond Myklebust nfs_flush_error(desc, hdr); 122225b11dcdSTrond Myklebust return -ENOMEM; 122344b83799SFred Isaman } 12246c75dc0dSFred Isaman 1225ea2cf228SFred Isaman nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); 122630dd374fSFred Isaman pages = data->pages.pagevec; 12271da177e4SLinus Torvalds while (!list_empty(head)) { 12281da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12291da177e4SLinus Torvalds nfs_list_remove_request(req); 12306c75dc0dSFred Isaman nfs_list_add_request(req, &hdr->pages); 12311da177e4SLinus Torvalds *pages++ = req->wb_page; 12321da177e4SLinus Torvalds } 12331da177e4SLinus Torvalds 1234b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1235ea2cf228SFred Isaman (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) 1236b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1237b31268acSTrond Myklebust 12381da177e4SLinus Torvalds /* Set up the argument struct */ 1239ea2cf228SFred Isaman nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); 12406c75dc0dSFred Isaman list_add(&data->list, &hdr->rpc_list); 12416c75dc0dSFred Isaman desc->pg_rpc_callops = &nfs_write_common_ops; 124225b11dcdSTrond Myklebust return 0; 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds 12456c75dc0dSFred Isaman int nfs_generic_flush(struct nfs_pageio_descriptor *desc, 12466c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 1247dce81290STrond Myklebust { 1248dce81290STrond Myklebust if (desc->pg_bsize < PAGE_CACHE_SIZE) 12496c75dc0dSFred Isaman return nfs_flush_multi(desc, hdr); 12506c75dc0dSFred Isaman return nfs_flush_one(desc, hdr); 1251dce81290STrond Myklebust } 125289d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_generic_flush); 1253dce81290STrond Myklebust 1254dce81290STrond Myklebust static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) 12551751c363STrond Myklebust { 12566c75dc0dSFred Isaman struct nfs_write_header *whdr; 12576c75dc0dSFred Isaman struct nfs_pgio_header *hdr; 1258275acaafSTrond Myklebust int ret; 1259275acaafSTrond Myklebust 12606c75dc0dSFred Isaman whdr = nfs_writehdr_alloc(); 12616c75dc0dSFred Isaman if (!whdr) { 12629b5415b5STrond Myklebust desc->pg_completion_ops->error_cleanup(&desc->pg_list); 12636c75dc0dSFred Isaman return -ENOMEM; 12646c75dc0dSFred Isaman } 12656c75dc0dSFred Isaman hdr = &whdr->header; 12666c75dc0dSFred Isaman nfs_pgheader_init(desc, hdr, nfs_writehdr_free); 12676c75dc0dSFred Isaman atomic_inc(&hdr->refcnt); 12686c75dc0dSFred Isaman ret = nfs_generic_flush(desc, hdr); 1269275acaafSTrond Myklebust if (ret == 0) 12706c75dc0dSFred Isaman ret = nfs_do_multiple_writes(&hdr->rpc_list, 12716c75dc0dSFred Isaman desc->pg_rpc_callops, 1272dce81290STrond Myklebust desc->pg_ioflags); 12736c75dc0dSFred Isaman if (atomic_dec_and_test(&hdr->refcnt)) 1274061ae2edSFred Isaman hdr->completion_ops->completion(hdr); 1275275acaafSTrond Myklebust return ret; 12761751c363STrond Myklebust } 12771751c363STrond Myklebust 12781751c363STrond Myklebust static const struct nfs_pageio_ops nfs_pageio_write_ops = { 12791751c363STrond Myklebust .pg_test = nfs_generic_pg_test, 12801751c363STrond Myklebust .pg_doio = nfs_generic_pg_writepages, 12811751c363STrond Myklebust }; 12821751c363STrond Myklebust 128357208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1284a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1285061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 12861751c363STrond Myklebust { 1287a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 1288a20c93e3SChristoph Hellwig const struct nfs_pageio_ops *pg_ops = &nfs_pageio_write_ops; 1289a20c93e3SChristoph Hellwig 1290a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1291a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1292a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1293a20c93e3SChristoph Hellwig #endif 1294a20c93e3SChristoph Hellwig nfs_pageio_init(pgio, inode, pg_ops, compl_ops, server->wsize, ioflags); 12951751c363STrond Myklebust } 1296ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 12971751c363STrond Myklebust 1298dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1299dce81290STrond Myklebust { 1300dce81290STrond Myklebust pgio->pg_ops = &nfs_pageio_write_ops; 1301dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1302dce81290STrond Myklebust } 13031f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1304dce81290STrond Myklebust 13051da177e4SLinus Torvalds 1306def6ed7eSAndy Adamson void nfs_write_prepare(struct rpc_task *task, void *calldata) 1307def6ed7eSAndy Adamson { 1308def6ed7eSAndy Adamson struct nfs_write_data *data = calldata; 1309ef1820f9SNeilBrown int err; 1310ef1820f9SNeilBrown err = NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); 1311ef1820f9SNeilBrown if (err) 1312ef1820f9SNeilBrown rpc_exit(task, err); 1313def6ed7eSAndy Adamson } 1314def6ed7eSAndy Adamson 13150b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13160b7c0153SFred Isaman { 13170b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13180b7c0153SFred Isaman 13190b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13200b7c0153SFred Isaman } 13210b7c0153SFred Isaman 13221da177e4SLinus Torvalds /* 13231da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 13241da177e4SLinus Torvalds * 13251da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 13261da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 13271da177e4SLinus Torvalds * as the page has a write request pending. 13281da177e4SLinus Torvalds */ 13296c75dc0dSFred Isaman static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) 13301da177e4SLinus Torvalds { 1331788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 13321da177e4SLinus Torvalds 1333c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1334c9d8f89dSTrond Myklebust } 1335c9d8f89dSTrond Myklebust 13366c75dc0dSFred Isaman static void nfs_writeback_release_common(void *calldata) 1337c9d8f89dSTrond Myklebust { 1338c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1339cd841605SFred Isaman struct nfs_pgio_header *hdr = data->header; 1340e2fecb21STrond Myklebust int status = data->task.tk_status; 1341788e7a89STrond Myklebust 13426c75dc0dSFred Isaman if ((status >= 0) && nfs_write_need_commit(data)) { 13436c75dc0dSFred Isaman spin_lock(&hdr->lock); 13446c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) 13456c75dc0dSFred Isaman ; /* Do nothing */ 13466c75dc0dSFred Isaman else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) 13479bce008bSTrond Myklebust memcpy(hdr->verf, &data->verf, sizeof(*hdr->verf)); 13489bce008bSTrond Myklebust else if (memcmp(hdr->verf, &data->verf, sizeof(*hdr->verf))) 13496c75dc0dSFred Isaman set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); 13506c75dc0dSFred Isaman spin_unlock(&hdr->lock); 13511da177e4SLinus Torvalds } 1352cd841605SFred Isaman nfs_writedata_release(data); 13531da177e4SLinus Torvalds } 13541da177e4SLinus Torvalds 13556c75dc0dSFred Isaman static const struct rpc_call_ops nfs_write_common_ops = { 1356def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 13576c75dc0dSFred Isaman .rpc_call_done = nfs_writeback_done_common, 13586c75dc0dSFred Isaman .rpc_release = nfs_writeback_release_common, 1359788e7a89STrond Myklebust }; 1360788e7a89STrond Myklebust 13611f2edbe3STrond Myklebust /* 13621f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 13631f2edbe3STrond Myklebust */ 13641f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 13651f2edbe3STrond Myklebust { 13661f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 13671f2edbe3STrond Myklebust int kill = 0; 13681f2edbe3STrond Myklebust 13691f2edbe3STrond Myklebust /* suid always must be killed */ 13701f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 13711f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 13721f2edbe3STrond Myklebust 13731f2edbe3STrond Myklebust /* 13741f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 13751f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 13761f2edbe3STrond Myklebust */ 13771f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 13781f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 13791f2edbe3STrond Myklebust 13801f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 13811f2edbe3STrond Myklebust return kill; 13821f2edbe3STrond Myklebust 13831f2edbe3STrond Myklebust return 0; 13841f2edbe3STrond Myklebust } 1385788e7a89STrond Myklebust 13861da177e4SLinus Torvalds /* 13871da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 13881da177e4SLinus Torvalds */ 138913602896SFred Isaman void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 13901da177e4SLinus Torvalds { 1391*3c6b899cSAnna Schumaker struct nfs_pgio_args *argp = &data->args; 13921da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1393cd841605SFred Isaman struct inode *inode = data->header->inode; 1394788e7a89STrond Myklebust int status; 13951da177e4SLinus Torvalds 1396a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 13971da177e4SLinus Torvalds task->tk_pid, task->tk_status); 13981da177e4SLinus Torvalds 1399f551e44fSChuck Lever /* 1400f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1401f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1402f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1403f551e44fSChuck Lever * another writer had changed the file, but some applications 1404f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1405f551e44fSChuck Lever */ 1406cd841605SFred Isaman status = NFS_PROTO(inode)->write_done(task, data); 1407788e7a89STrond Myklebust if (status != 0) 140813602896SFred Isaman return; 1409cd841605SFred Isaman nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 141091d5b470SChuck Lever 141189d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 14121da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 14131da177e4SLinus Torvalds /* We tried a write call, but the server did not 14141da177e4SLinus Torvalds * commit data to stable storage even though we 14151da177e4SLinus Torvalds * requested it. 14161da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 14171da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 14181da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 14191da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 14201da177e4SLinus Torvalds */ 14211da177e4SLinus Torvalds static unsigned long complain; 14221da177e4SLinus Torvalds 1423a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 14241da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 14251da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 14261da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1427cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 14281da177e4SLinus Torvalds resp->verf->committed, argp->stable); 14291da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 14301da177e4SLinus Torvalds } 14311da177e4SLinus Torvalds } 14321da177e4SLinus Torvalds #endif 14331f2edbe3STrond Myklebust if (task->tk_status < 0) { 14346c75dc0dSFred Isaman nfs_set_pgio_error(data->header, task->tk_status, argp->offset); 14351f2edbe3STrond Myklebust return; 14361f2edbe3STrond Myklebust } 14371f2edbe3STrond Myklebust 14381f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 14391f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 14401f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 14411f2edbe3STrond Myklebust 14421f2edbe3STrond Myklebust if (resp->count < argp->count) { 14431da177e4SLinus Torvalds static unsigned long complain; 14441da177e4SLinus Torvalds 14456c75dc0dSFred Isaman /* This a short write! */ 1446cd841605SFred Isaman nfs_inc_stats(inode, NFSIOS_SHORTWRITE); 144791d5b470SChuck Lever 14481da177e4SLinus Torvalds /* Has the server at least made some progress? */ 14496c75dc0dSFred Isaman if (resp->count == 0) { 14506c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 14516c75dc0dSFred Isaman printk(KERN_WARNING 14526c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 14536c75dc0dSFred Isaman argp->count); 14546c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 14556c75dc0dSFred Isaman } 14566c75dc0dSFred Isaman nfs_set_pgio_error(data->header, -EIO, argp->offset); 14576c75dc0dSFred Isaman task->tk_status = -EIO; 14586c75dc0dSFred Isaman return; 14596c75dc0dSFred Isaman } 14601da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 14611da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 14621da177e4SLinus Torvalds /* Resend from where the server left off */ 1463a69aef14SFred Isaman data->mds_offset += resp->count; 14641da177e4SLinus Torvalds argp->offset += resp->count; 14651da177e4SLinus Torvalds argp->pgbase += resp->count; 14661da177e4SLinus Torvalds argp->count -= resp->count; 14671da177e4SLinus Torvalds } else { 14681da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 14691da177e4SLinus Torvalds * headaches in the case of a server crash. 14701da177e4SLinus Torvalds */ 14711da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 14721da177e4SLinus Torvalds } 1473d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 14741da177e4SLinus Torvalds } 14751da177e4SLinus Torvalds } 14761da177e4SLinus Torvalds 14771da177e4SLinus Torvalds 147889d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 147971d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 148071d0a611STrond Myklebust { 1481b8413f98STrond Myklebust int ret; 1482b8413f98STrond Myklebust 148371d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 148471d0a611STrond Myklebust return 1; 1485b8413f98STrond Myklebust if (!may_wait) 148671d0a611STrond Myklebust return 0; 1487b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1488b8413f98STrond Myklebust NFS_INO_COMMIT, 1489b8413f98STrond Myklebust nfs_wait_bit_killable, 1490b8413f98STrond Myklebust TASK_KILLABLE); 1491b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 149271d0a611STrond Myklebust } 149371d0a611STrond Myklebust 1494f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 149571d0a611STrond Myklebust { 149671d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 149771d0a611STrond Myklebust smp_mb__after_clear_bit(); 149871d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 149971d0a611STrond Myklebust } 150071d0a611STrond Myklebust 15010b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 15021da177e4SLinus Torvalds { 15030b7c0153SFred Isaman put_nfs_open_context(data->context); 15040b7c0153SFred Isaman nfs_commit_free(data); 15051da177e4SLinus Torvalds } 1506e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15071da177e4SLinus Torvalds 15080b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 15099ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15109f0ec176SAndy Adamson int how, int flags) 15111da177e4SLinus Torvalds { 151207737691STrond Myklebust struct rpc_task *task; 15139ace33cdSFred Isaman int priority = flush_task_priority(how); 1514bdc7f021STrond Myklebust struct rpc_message msg = { 1515bdc7f021STrond Myklebust .rpc_argp = &data->args, 1516bdc7f021STrond Myklebust .rpc_resp = &data->res, 15179ace33cdSFred Isaman .rpc_cred = data->cred, 1518bdc7f021STrond Myklebust }; 151984115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 152007737691STrond Myklebust .task = &data->task, 15219ace33cdSFred Isaman .rpc_client = clnt, 1522bdc7f021STrond Myklebust .rpc_message = &msg, 15239ace33cdSFred Isaman .callback_ops = call_ops, 152484115e1cSTrond Myklebust .callback_data = data, 1525101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 15269f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 15273ff7576dSTrond Myklebust .priority = priority, 152884115e1cSTrond Myklebust }; 1529788e7a89STrond Myklebust /* Set up the initial task struct. */ 15309ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 15311da177e4SLinus Torvalds 1532a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1533bdc7f021STrond Myklebust 15348c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 15358c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 15368c21c62cSWeston Andros Adamson 153707737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1538dbae4c73STrond Myklebust if (IS_ERR(task)) 1539dbae4c73STrond Myklebust return PTR_ERR(task); 1540d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1541d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 154207737691STrond Myklebust rpc_put_task(task); 1543dbae4c73STrond Myklebust return 0; 15441da177e4SLinus Torvalds } 1545e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 15461da177e4SLinus Torvalds 15471da177e4SLinus Torvalds /* 15489ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 15499ace33cdSFred Isaman */ 15500b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1551988b6dceSFred Isaman struct list_head *head, 1552f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1553f453a54aSFred Isaman struct nfs_commit_info *cinfo) 15549ace33cdSFred Isaman { 15559ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 15563d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 15579ace33cdSFred Isaman 15589ace33cdSFred Isaman /* Set up the RPC argument and reply structs 15599ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 15609ace33cdSFred Isaman 15619ace33cdSFred Isaman list_splice_init(head, &data->pages); 15629ace33cdSFred Isaman 15639ace33cdSFred Isaman data->inode = inode; 15649ace33cdSFred Isaman data->cred = first->wb_context->cred; 1565988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 15669ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1567f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1568b359f9d0SFred Isaman data->dreq = cinfo->dreq; 15699ace33cdSFred Isaman 15709ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 15719ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 15729ace33cdSFred Isaman data->args.offset = 0; 15739ace33cdSFred Isaman data->args.count = 0; 15740b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 15759ace33cdSFred Isaman data->res.fattr = &data->fattr; 15769ace33cdSFred Isaman data->res.verf = &data->verf; 15779ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 15789ace33cdSFred Isaman } 1579e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 15809ace33cdSFred Isaman 1581e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1582ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1583ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 158464bfeb49SFred Isaman { 158564bfeb49SFred Isaman struct nfs_page *req; 158664bfeb49SFred Isaman 158764bfeb49SFred Isaman while (!list_empty(page_list)) { 158864bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 158964bfeb49SFred Isaman nfs_list_remove_request(req); 1590ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 159156f9cd68SFred Isaman if (!cinfo->dreq) { 159264bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1593d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 159464bfeb49SFred Isaman BDI_RECLAIMABLE); 159556f9cd68SFred Isaman } 15961d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 159764bfeb49SFred Isaman } 159864bfeb49SFred Isaman } 1599e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 160064bfeb49SFred Isaman 16019ace33cdSFred Isaman /* 16021da177e4SLinus Torvalds * Commit dirty pages 16031da177e4SLinus Torvalds */ 16041da177e4SLinus Torvalds static int 1605ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1606ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 16071da177e4SLinus Torvalds { 16080b7c0153SFred Isaman struct nfs_commit_data *data; 16091da177e4SLinus Torvalds 1610c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 16111da177e4SLinus Torvalds 16121da177e4SLinus Torvalds if (!data) 16131da177e4SLinus Torvalds goto out_bad; 16141da177e4SLinus Torvalds 16151da177e4SLinus Torvalds /* Set up the argument struct */ 1616f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1617f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 16189f0ec176SAndy Adamson return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 16199f0ec176SAndy Adamson how, 0); 16201da177e4SLinus Torvalds out_bad: 1621ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1622f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 16231da177e4SLinus Torvalds return -ENOMEM; 16241da177e4SLinus Torvalds } 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds /* 16271da177e4SLinus Torvalds * COMMIT call returned 16281da177e4SLinus Torvalds */ 1629788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 16301da177e4SLinus Torvalds { 16310b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16321da177e4SLinus Torvalds 1633a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 16341da177e4SLinus Torvalds task->tk_pid, task->tk_status); 16351da177e4SLinus Torvalds 1636788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1637c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1638c9d8f89dSTrond Myklebust } 1639c9d8f89dSTrond Myklebust 1640f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1641c9d8f89dSTrond Myklebust { 1642c9d8f89dSTrond Myklebust struct nfs_page *req; 1643c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1644f453a54aSFred Isaman struct nfs_commit_info cinfo; 1645788e7a89STrond Myklebust 16461da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 16471da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 16481da177e4SLinus Torvalds nfs_list_remove_request(req); 1649d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 16501da177e4SLinus Torvalds 16511e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 16523d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 16531e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), 16541da177e4SLinus Torvalds req->wb_bytes, 16551da177e4SLinus Torvalds (long long)req_offset(req)); 1656c9d8f89dSTrond Myklebust if (status < 0) { 1657c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 16581da177e4SLinus Torvalds nfs_inode_remove_request(req); 1659c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 16601da177e4SLinus Torvalds goto next; 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 16641da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 16652f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 16661da177e4SLinus Torvalds /* We have a match */ 16671da177e4SLinus Torvalds nfs_inode_remove_request(req); 16681da177e4SLinus Torvalds dprintk(" OK\n"); 16691da177e4SLinus Torvalds goto next; 16701da177e4SLinus Torvalds } 16711da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 16721da177e4SLinus Torvalds dprintk(" mismatch\n"); 16736d884e8fSFred nfs_mark_request_dirty(req); 167405990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 16751da177e4SLinus Torvalds next: 16761d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 16771da177e4SLinus Torvalds } 1678f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1679f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1680f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 16815917ce84SFred Isaman } 16825917ce84SFred Isaman 16835917ce84SFred Isaman static void nfs_commit_release(void *calldata) 16845917ce84SFred Isaman { 16850b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16865917ce84SFred Isaman 1687f453a54aSFred Isaman data->completion_ops->completion(data); 1688c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 16891da177e4SLinus Torvalds } 1690788e7a89STrond Myklebust 1691788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 16920b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1693788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1694788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1695788e7a89STrond Myklebust }; 16961da177e4SLinus Torvalds 1697f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1698f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1699f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1700f453a54aSFred Isaman }; 1701f453a54aSFred Isaman 17021763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1703ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 170484c53ab5SFred Isaman { 170584c53ab5SFred Isaman int status; 170684c53ab5SFred Isaman 1707ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 170884c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1709ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 171084c53ab5SFred Isaman return status; 171184c53ab5SFred Isaman } 171284c53ab5SFred Isaman 1713b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 17141da177e4SLinus Torvalds { 17151da177e4SLinus Torvalds LIST_HEAD(head); 1716ea2cf228SFred Isaman struct nfs_commit_info cinfo; 171771d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1718b8413f98STrond Myklebust int res; 17191da177e4SLinus Torvalds 1720b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1721b8413f98STrond Myklebust if (res <= 0) 1722c5efa5fcSTrond Myklebust goto out_mark_dirty; 1723ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1724ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 17251da177e4SLinus Torvalds if (res) { 1726a861a1e1SFred Isaman int error; 1727a861a1e1SFred Isaman 1728ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 17291da177e4SLinus Torvalds if (error < 0) 17301da177e4SLinus Torvalds return error; 1731b8413f98STrond Myklebust if (!may_wait) 1732b8413f98STrond Myklebust goto out_mark_dirty; 1733b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1734b8413f98STrond Myklebust NFS_INO_COMMIT, 173571d0a611STrond Myklebust nfs_wait_bit_killable, 173671d0a611STrond Myklebust TASK_KILLABLE); 1737b8413f98STrond Myklebust if (error < 0) 1738b8413f98STrond Myklebust return error; 173971d0a611STrond Myklebust } else 174071d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1741c5efa5fcSTrond Myklebust return res; 1742c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1743c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1744c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1745c5efa5fcSTrond Myklebust * that the data is on the disk. 1746c5efa5fcSTrond Myklebust */ 1747c5efa5fcSTrond Myklebust out_mark_dirty: 1748c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17491da177e4SLinus Torvalds return res; 17501da177e4SLinus Torvalds } 17518fc795f7STrond Myklebust 17528fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17538fc795f7STrond Myklebust { 1754420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1755420e3646STrond Myklebust int flags = FLUSH_SYNC; 1756420e3646STrond Myklebust int ret = 0; 17578fc795f7STrond Myklebust 17583236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1759ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 17603236c3e1SJeff Layton return ret; 17613236c3e1SJeff Layton 1762a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1763a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1764a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1765420e3646STrond Myklebust */ 1766ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1767420e3646STrond Myklebust goto out_mark_dirty; 1768420e3646STrond Myklebust 1769a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1770420e3646STrond Myklebust flags = 0; 1771a00dd6c0SJeff Layton } 1772a00dd6c0SJeff Layton 1773420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1774420e3646STrond Myklebust if (ret >= 0) { 1775420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1776420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1777420e3646STrond Myklebust wbc->nr_to_write -= ret; 1778420e3646STrond Myklebust else 1779420e3646STrond Myklebust wbc->nr_to_write = 0; 1780420e3646STrond Myklebust } 17818fc795f7STrond Myklebust return 0; 1782420e3646STrond Myklebust } 1783420e3646STrond Myklebust out_mark_dirty: 17848fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17858fc795f7STrond Myklebust return ret; 17868fc795f7STrond Myklebust } 1787c63c7b05STrond Myklebust #else 17888fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17898fc795f7STrond Myklebust { 17908fc795f7STrond Myklebust return 0; 17918fc795f7STrond Myklebust } 17921da177e4SLinus Torvalds #endif 17931da177e4SLinus Torvalds 17948fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 17958fc795f7STrond Myklebust { 1796a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1797a8d8f02cSBryan Schumaker } 179889d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1799863a3c6cSAndy Adamson 1800acdc53b2STrond Myklebust /* 1801acdc53b2STrond Myklebust * flush the inode to disk. 1802acdc53b2STrond Myklebust */ 1803acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 180434901f70STrond Myklebust { 180534901f70STrond Myklebust struct writeback_control wbc = { 180672cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 180734901f70STrond Myklebust .nr_to_write = LONG_MAX, 1808d7fb1207STrond Myklebust .range_start = 0, 1809d7fb1207STrond Myklebust .range_end = LLONG_MAX, 181034901f70STrond Myklebust }; 1811f4ce1299STrond Myklebust int ret; 181234901f70STrond Myklebust 1813f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1814f4ce1299STrond Myklebust 1815f4ce1299STrond Myklebust ret = sync_inode(inode, &wbc); 1816f4ce1299STrond Myklebust 1817f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1818f4ce1299STrond Myklebust return ret; 18191c75950bSTrond Myklebust } 1820ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 18211c75950bSTrond Myklebust 18221b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 18231b3b4a1aSTrond Myklebust { 18241b3b4a1aSTrond Myklebust struct nfs_page *req; 18251b3b4a1aSTrond Myklebust int ret = 0; 18261b3b4a1aSTrond Myklebust 18271b3b4a1aSTrond Myklebust for (;;) { 1828ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18291b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 18301b3b4a1aSTrond Myklebust if (req == NULL) 18311b3b4a1aSTrond Myklebust break; 18327ad84aa9STrond Myklebust if (nfs_lock_request(req)) { 18338dd37758STrond Myklebust nfs_clear_request_commit(req); 18341b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 18351b3b4a1aSTrond Myklebust /* 18361b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 18371b3b4a1aSTrond Myklebust * page as being dirty 18381b3b4a1aSTrond Myklebust */ 18391b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 18401d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18411b3b4a1aSTrond Myklebust break; 18421b3b4a1aSTrond Myklebust } 18431b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1844c9edda71STrond Myklebust nfs_release_request(req); 18451b3b4a1aSTrond Myklebust if (ret < 0) 1846c988950eSTrond Myklebust break; 18471b3b4a1aSTrond Myklebust } 18481b3b4a1aSTrond Myklebust return ret; 18491b3b4a1aSTrond Myklebust } 18501b3b4a1aSTrond Myklebust 18511c75950bSTrond Myklebust /* 18521c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 18531c75950bSTrond Myklebust */ 18541c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 18551c75950bSTrond Myklebust { 185629418aa4SMel Gorman loff_t range_start = page_file_offset(page); 18577f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 18587f2f12d9STrond Myklebust struct writeback_control wbc = { 18597f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 18607f2f12d9STrond Myklebust .nr_to_write = 0, 18617f2f12d9STrond Myklebust .range_start = range_start, 18627f2f12d9STrond Myklebust .range_end = range_end, 18637f2f12d9STrond Myklebust }; 18647f2f12d9STrond Myklebust int ret; 18657f2f12d9STrond Myklebust 1866f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1867f4ce1299STrond Myklebust 18680522f6adSTrond Myklebust for (;;) { 1869ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18707f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 18717f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 18727f2f12d9STrond Myklebust if (ret < 0) 18737f2f12d9STrond Myklebust goto out_error; 18740522f6adSTrond Myklebust continue; 18757f2f12d9STrond Myklebust } 1876f4ce1299STrond Myklebust ret = 0; 18770522f6adSTrond Myklebust if (!PagePrivate(page)) 18780522f6adSTrond Myklebust break; 18790522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 18807f2f12d9STrond Myklebust if (ret < 0) 18817f2f12d9STrond Myklebust goto out_error; 18827f2f12d9STrond Myklebust } 18837f2f12d9STrond Myklebust out_error: 1884f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 18857f2f12d9STrond Myklebust return ret; 18861c75950bSTrond Myklebust } 18871c75950bSTrond Myklebust 1888074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1889074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1890a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1891074cc1deSTrond Myklebust { 18922da95652SJeff Layton /* 18932da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 18942da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 18952da95652SJeff Layton * 18962da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 18972da95652SJeff Layton * that we can safely release the inode reference while holding 18982da95652SJeff Layton * the page lock. 18992da95652SJeff Layton */ 19002da95652SJeff Layton if (PagePrivate(page)) 19012da95652SJeff Layton return -EBUSY; 1902074cc1deSTrond Myklebust 19038c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 19048c209ce7SDavid Howells return -EBUSY; 1905074cc1deSTrond Myklebust 1906a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1907074cc1deSTrond Myklebust } 1908074cc1deSTrond Myklebust #endif 1909074cc1deSTrond Myklebust 1910f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 19111da177e4SLinus Torvalds { 19121da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 1913cd841605SFred Isaman sizeof(struct nfs_write_header), 19141da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 191520c2df83SPaul Mundt NULL); 19161da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 19171da177e4SLinus Torvalds return -ENOMEM; 19181da177e4SLinus Torvalds 191993d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 19201da177e4SLinus Torvalds nfs_wdata_cachep); 19211da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 19223dd4765fSJeff Layton goto out_destroy_write_cache; 19231da177e4SLinus Torvalds 19240b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 19250b7c0153SFred Isaman sizeof(struct nfs_commit_data), 19260b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 19270b7c0153SFred Isaman NULL); 19280b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 19293dd4765fSJeff Layton goto out_destroy_write_mempool; 19300b7c0153SFred Isaman 193193d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 19324c100210SYanchuan Nian nfs_cdata_cachep); 19331da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 19343dd4765fSJeff Layton goto out_destroy_commit_cache; 19351da177e4SLinus Torvalds 193689a09141SPeter Zijlstra /* 193789a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 193889a09141SPeter Zijlstra * 193989a09141SPeter Zijlstra * 64MB: 8192k 194089a09141SPeter Zijlstra * 128MB: 11585k 194189a09141SPeter Zijlstra * 256MB: 16384k 194289a09141SPeter Zijlstra * 512MB: 23170k 194389a09141SPeter Zijlstra * 1GB: 32768k 194489a09141SPeter Zijlstra * 2GB: 46340k 194589a09141SPeter Zijlstra * 4GB: 65536k 194689a09141SPeter Zijlstra * 8GB: 92681k 194789a09141SPeter Zijlstra * 16GB: 131072k 194889a09141SPeter Zijlstra * 194989a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 195089a09141SPeter Zijlstra * Limit the default to 256M 195189a09141SPeter Zijlstra */ 195289a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 195389a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 195489a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 195589a09141SPeter Zijlstra 19561da177e4SLinus Torvalds return 0; 19573dd4765fSJeff Layton 19583dd4765fSJeff Layton out_destroy_commit_cache: 19593dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19603dd4765fSJeff Layton out_destroy_write_mempool: 19613dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 19623dd4765fSJeff Layton out_destroy_write_cache: 19633dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 19643dd4765fSJeff Layton return -ENOMEM; 19651da177e4SLinus Torvalds } 19661da177e4SLinus Torvalds 1967266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 19681da177e4SLinus Torvalds { 19691da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 19703dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19711da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 19721a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 19731da177e4SLinus Torvalds } 19741da177e4SLinus Torvalds 1975