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 73c0752cdfSAnna Schumaker struct nfs_rw_header *nfs_writehdr_alloc(void) 743feb2d49STrond Myklebust { 75c0752cdfSAnna Schumaker struct nfs_rw_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); 853feb2d49STrond Myklebust } 863feb2d49STrond Myklebust return p; 873feb2d49STrond Myklebust } 8889d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_writehdr_alloc); 893feb2d49STrond Myklebust 90cd841605SFred Isaman void nfs_writehdr_free(struct nfs_pgio_header *hdr) 913feb2d49STrond Myklebust { 92c0752cdfSAnna Schumaker struct nfs_rw_header *whdr = container_of(hdr, struct nfs_rw_header, header); 93cd841605SFred Isaman mempool_free(whdr, nfs_wdata_mempool); 943feb2d49STrond Myklebust } 9589d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_writehdr_free); 963feb2d49STrond Myklebust 977b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 987b159fc1STrond Myklebust { 997b159fc1STrond Myklebust ctx->error = error; 1007b159fc1STrond Myklebust smp_wmb(); 1017b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1027b159fc1STrond Myklebust } 1037b159fc1STrond Myklebust 10429418aa4SMel Gorman static struct nfs_page * 10529418aa4SMel Gorman nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page) 106277459d2STrond Myklebust { 107277459d2STrond Myklebust struct nfs_page *req = NULL; 108277459d2STrond Myklebust 10929418aa4SMel Gorman if (PagePrivate(page)) 110277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 11129418aa4SMel Gorman else if (unlikely(PageSwapCache(page))) { 11229418aa4SMel Gorman struct nfs_page *freq, *t; 11329418aa4SMel Gorman 11429418aa4SMel Gorman /* Linearly search the commit list for the correct req */ 11529418aa4SMel Gorman list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { 11629418aa4SMel Gorman if (freq->wb_page == page) { 11729418aa4SMel Gorman req = freq; 11829418aa4SMel Gorman break; 119277459d2STrond Myklebust } 12029418aa4SMel Gorman } 12129418aa4SMel Gorman } 12229418aa4SMel Gorman 12329418aa4SMel Gorman if (req) 12429418aa4SMel Gorman kref_get(&req->wb_kref); 12529418aa4SMel Gorman 126277459d2STrond Myklebust return req; 127277459d2STrond Myklebust } 128277459d2STrond Myklebust 129277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 130277459d2STrond Myklebust { 131d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 132277459d2STrond Myklebust struct nfs_page *req = NULL; 133277459d2STrond Myklebust 134587142f8STrond Myklebust spin_lock(&inode->i_lock); 13529418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 136587142f8STrond Myklebust spin_unlock(&inode->i_lock); 137277459d2STrond Myklebust return req; 138277459d2STrond Myklebust } 139277459d2STrond Myklebust 1401da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1411da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1421da177e4SLinus Torvalds { 143d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 144a3d01454STrond Myklebust loff_t end, i_size; 145a3d01454STrond Myklebust pgoff_t end_index; 1461da177e4SLinus Torvalds 147a3d01454STrond Myklebust spin_lock(&inode->i_lock); 148a3d01454STrond Myklebust i_size = i_size_read(inode); 149a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 150d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 151a3d01454STrond Myklebust goto out; 152d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1531da177e4SLinus Torvalds if (i_size >= end) 154a3d01454STrond Myklebust goto out; 1551da177e4SLinus Torvalds i_size_write(inode, end); 156a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 157a3d01454STrond Myklebust out: 158a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1591da177e4SLinus Torvalds } 1601da177e4SLinus Torvalds 161a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 162a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 163a301b777STrond Myklebust { 164d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 165a301b777STrond Myklebust } 166a301b777STrond Myklebust 1671da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1681da177e4SLinus Torvalds * covers the full page. 1691da177e4SLinus Torvalds */ 1701da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1711da177e4SLinus Torvalds { 1721da177e4SLinus Torvalds if (PageUptodate(page)) 1731da177e4SLinus Torvalds return; 1741da177e4SLinus Torvalds if (base != 0) 1751da177e4SLinus Torvalds return; 17649a70f27STrond Myklebust if (count != nfs_page_length(page)) 1771da177e4SLinus Torvalds return; 1781da177e4SLinus Torvalds SetPageUptodate(page); 1791da177e4SLinus Torvalds } 1801da177e4SLinus Torvalds 1811da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1821da177e4SLinus Torvalds { 1831da177e4SLinus Torvalds if (wbc->for_reclaim) 184c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 185b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 186b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 187b31268acSTrond Myklebust return FLUSH_COND_STABLE; 1881da177e4SLinus Torvalds } 1891da177e4SLinus Torvalds 1901da177e4SLinus Torvalds /* 19189a09141SPeter Zijlstra * NFS congestion control 19289a09141SPeter Zijlstra */ 19389a09141SPeter Zijlstra 19489a09141SPeter Zijlstra int nfs_congestion_kb; 19589a09141SPeter Zijlstra 19689a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 19789a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 19889a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 19989a09141SPeter Zijlstra 200deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 20189a09141SPeter Zijlstra { 202deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2035a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2045a6d41b3STrond Myklebust 205deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 20689a09141SPeter Zijlstra 207277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2088aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2098aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2108aa7e847SJens Axboe BLK_RW_ASYNC); 2118aa7e847SJens Axboe } 21289a09141SPeter Zijlstra } 21389a09141SPeter Zijlstra 21489a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 21589a09141SPeter Zijlstra { 216d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 21789a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 21889a09141SPeter Zijlstra 21989a09141SPeter Zijlstra end_page_writeback(page); 220c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2218aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 22289a09141SPeter Zijlstra } 22389a09141SPeter Zijlstra 224cfb506e1STrond Myklebust static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) 225e261f51fSTrond Myklebust { 226d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 227e261f51fSTrond Myklebust struct nfs_page *req; 228e261f51fSTrond Myklebust int ret; 229e261f51fSTrond Myklebust 230587142f8STrond Myklebust spin_lock(&inode->i_lock); 231e261f51fSTrond Myklebust for (;;) { 23229418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 233074cc1deSTrond Myklebust if (req == NULL) 234074cc1deSTrond Myklebust break; 2357ad84aa9STrond Myklebust if (nfs_lock_request(req)) 236e261f51fSTrond Myklebust break; 237e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 2387ad84aa9STrond Myklebust * then the call to nfs_lock_request() will always 239e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 240e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 241e261f51fSTrond Myklebust */ 242587142f8STrond Myklebust spin_unlock(&inode->i_lock); 243cfb506e1STrond Myklebust if (!nonblock) 244e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 245cfb506e1STrond Myklebust else 246cfb506e1STrond Myklebust ret = -EAGAIN; 247e261f51fSTrond Myklebust nfs_release_request(req); 248e261f51fSTrond Myklebust if (ret != 0) 249074cc1deSTrond Myklebust return ERR_PTR(ret); 250587142f8STrond Myklebust spin_lock(&inode->i_lock); 251e261f51fSTrond Myklebust } 252587142f8STrond Myklebust spin_unlock(&inode->i_lock); 253074cc1deSTrond Myklebust return req; 254612c9384STrond Myklebust } 255074cc1deSTrond Myklebust 256074cc1deSTrond Myklebust /* 257074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 258074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 259074cc1deSTrond Myklebust */ 260074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 261cfb506e1STrond Myklebust struct page *page, bool nonblock) 262074cc1deSTrond Myklebust { 263074cc1deSTrond Myklebust struct nfs_page *req; 264074cc1deSTrond Myklebust int ret = 0; 265074cc1deSTrond Myklebust 266cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, nonblock); 267074cc1deSTrond Myklebust if (!req) 268074cc1deSTrond Myklebust goto out; 269074cc1deSTrond Myklebust ret = PTR_ERR(req); 270074cc1deSTrond Myklebust if (IS_ERR(req)) 271074cc1deSTrond Myklebust goto out; 272074cc1deSTrond Myklebust 273deed85e7STrond Myklebust nfs_set_page_writeback(page); 274deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 275074cc1deSTrond Myklebust 276deed85e7STrond Myklebust ret = 0; 277f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 278f8512ad0SFred Isaman nfs_redirty_request(req); 279074cc1deSTrond Myklebust ret = pgio->pg_error; 280f8512ad0SFred Isaman } 281074cc1deSTrond Myklebust out: 282074cc1deSTrond Myklebust return ret; 283e261f51fSTrond Myklebust } 284e261f51fSTrond Myklebust 285f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 286f758c885STrond Myklebust { 287d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 288cfb506e1STrond Myklebust int ret; 289f758c885STrond Myklebust 290f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 291f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 292f758c885STrond Myklebust 293d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 2941b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 295cfb506e1STrond Myklebust if (ret == -EAGAIN) { 296cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 297cfb506e1STrond Myklebust ret = 0; 298cfb506e1STrond Myklebust } 299cfb506e1STrond Myklebust return ret; 300f758c885STrond Myklebust } 301f758c885STrond Myklebust 302e261f51fSTrond Myklebust /* 3031da177e4SLinus Torvalds * Write an mmapped page to the server. 3041da177e4SLinus Torvalds */ 3054d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3061da177e4SLinus Torvalds { 307f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 308e261f51fSTrond Myklebust int err; 3091da177e4SLinus Torvalds 310a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 311a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 312f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 313f758c885STrond Myklebust nfs_pageio_complete(&pgio); 314f758c885STrond Myklebust if (err < 0) 3154d770ccfSTrond Myklebust return err; 316f758c885STrond Myklebust if (pgio.pg_error < 0) 317f758c885STrond Myklebust return pgio.pg_error; 318f758c885STrond Myklebust return 0; 3194d770ccfSTrond Myklebust } 3204d770ccfSTrond Myklebust 3214d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3224d770ccfSTrond Myklebust { 323f758c885STrond Myklebust int ret; 3244d770ccfSTrond Myklebust 325f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3261da177e4SLinus Torvalds unlock_page(page); 327f758c885STrond Myklebust return ret; 328f758c885STrond Myklebust } 329f758c885STrond Myklebust 330f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 331f758c885STrond Myklebust { 332f758c885STrond Myklebust int ret; 333f758c885STrond Myklebust 334f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 335f758c885STrond Myklebust unlock_page(page); 336f758c885STrond Myklebust return ret; 3371da177e4SLinus Torvalds } 3381da177e4SLinus Torvalds 3391da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3401da177e4SLinus Torvalds { 3411da177e4SLinus Torvalds struct inode *inode = mapping->host; 34272cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 343c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3441da177e4SLinus Torvalds int err; 3451da177e4SLinus Torvalds 34672cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 34772cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 34872cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 34972cb77f4STrond Myklebust if (err) 35072cb77f4STrond Myklebust goto out_err; 35172cb77f4STrond Myklebust 35291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 35391d5b470SChuck Lever 354a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 355a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 356f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 357c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 35872cb77f4STrond Myklebust 35972cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 36072cb77f4STrond Myklebust smp_mb__after_clear_bit(); 36172cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 36272cb77f4STrond Myklebust 363f758c885STrond Myklebust if (err < 0) 36472cb77f4STrond Myklebust goto out_err; 36572cb77f4STrond Myklebust err = pgio.pg_error; 36672cb77f4STrond Myklebust if (err < 0) 36772cb77f4STrond Myklebust goto out_err; 368c63c7b05STrond Myklebust return 0; 36972cb77f4STrond Myklebust out_err: 37072cb77f4STrond Myklebust return err; 3711da177e4SLinus Torvalds } 3721da177e4SLinus Torvalds 3731da177e4SLinus Torvalds /* 3741da177e4SLinus Torvalds * Insert a write request into an inode 3751da177e4SLinus Torvalds */ 376d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3771da177e4SLinus Torvalds { 3781da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 379e7d39069STrond Myklebust 380e7d39069STrond Myklebust /* Lock the request! */ 3817ad84aa9STrond Myklebust nfs_lock_request(req); 382e7d39069STrond Myklebust 383e7d39069STrond Myklebust spin_lock(&inode->i_lock); 384011e2a7fSBryan Schumaker if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 385a9a4a87aSTrond Myklebust inode->i_version++; 38629418aa4SMel Gorman /* 38729418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 38829418aa4SMel Gorman * with invalidate/truncate. 38929418aa4SMel Gorman */ 39029418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 3912df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 392deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 393277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 39429418aa4SMel Gorman } 3951da177e4SLinus Torvalds nfsi->npages++; 396c03b4024STrond Myklebust kref_get(&req->wb_kref); 397e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds 4001da177e4SLinus Torvalds /* 40189a09141SPeter Zijlstra * Remove a write request from an inode 4021da177e4SLinus Torvalds */ 4031da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4041da177e4SLinus Torvalds { 4053d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 4061da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4071da177e4SLinus Torvalds 408587142f8STrond Myklebust spin_lock(&inode->i_lock); 40929418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 410277459d2STrond Myklebust set_page_private(req->wb_page, 0); 411deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4122df485a7STrond Myklebust clear_bit(PG_MAPPED, &req->wb_flags); 41329418aa4SMel Gorman } 4141da177e4SLinus Torvalds nfsi->npages--; 415587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4161da177e4SLinus Torvalds nfs_release_request(req); 4171da177e4SLinus Torvalds } 4181da177e4SLinus Torvalds 41961822ab5STrond Myklebust static void 4206d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 42161822ab5STrond Myklebust { 42261822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 42361822ab5STrond Myklebust } 42461822ab5STrond Myklebust 42589d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 4268dd37758STrond Myklebust /** 4278dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 4288dd37758STrond Myklebust * @req: pointer to a struct nfs_page 429ea2cf228SFred Isaman * @dst: commit list head 430ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 4318dd37758STrond Myklebust * 432ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 4338dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 4348dd37758STrond Myklebust * the MM page stats. 4358dd37758STrond Myklebust * 436ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 4378dd37758STrond Myklebust * holding the nfs_page lock. 4388dd37758STrond Myklebust */ 4398dd37758STrond Myklebust void 440ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 441ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 4428dd37758STrond Myklebust { 4438dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 444ea2cf228SFred Isaman spin_lock(cinfo->lock); 445ea2cf228SFred Isaman nfs_list_add_request(req, dst); 446ea2cf228SFred Isaman cinfo->mds->ncommit++; 447ea2cf228SFred Isaman spin_unlock(cinfo->lock); 44856f9cd68SFred Isaman if (!cinfo->dreq) { 4498dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 450d56b4ddfSMel Gorman inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 45156f9cd68SFred Isaman BDI_RECLAIMABLE); 45256f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 45356f9cd68SFred Isaman I_DIRTY_DATASYNC); 45456f9cd68SFred Isaman } 4558dd37758STrond Myklebust } 4568dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 4578dd37758STrond Myklebust 4588dd37758STrond Myklebust /** 4598dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 4608dd37758STrond Myklebust * @req: pointer to a nfs_page 461ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 4628dd37758STrond Myklebust * 463ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 4648dd37758STrond Myklebust * number of outstanding requests requiring a commit 4658dd37758STrond Myklebust * It does not update the MM page stats. 4668dd37758STrond Myklebust * 467ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 4688dd37758STrond Myklebust */ 4698dd37758STrond Myklebust void 470ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 471ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 4728dd37758STrond Myklebust { 4738dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 4748dd37758STrond Myklebust return; 4758dd37758STrond Myklebust nfs_list_remove_request(req); 476ea2cf228SFred Isaman cinfo->mds->ncommit--; 4778dd37758STrond Myklebust } 4788dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 4798dd37758STrond Myklebust 480ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 481ea2cf228SFred Isaman struct inode *inode) 482ea2cf228SFred Isaman { 483ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 484ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 485ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 486b359f9d0SFred Isaman cinfo->dreq = NULL; 487f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 488ea2cf228SFred Isaman } 489ea2cf228SFred Isaman 490ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 491ea2cf228SFred Isaman struct inode *inode, 492ea2cf228SFred Isaman struct nfs_direct_req *dreq) 493ea2cf228SFred Isaman { 4941763da12SFred Isaman if (dreq) 4951763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 4961763da12SFred Isaman else 497ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 498ea2cf228SFred Isaman } 499ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 5008dd37758STrond Myklebust 5011da177e4SLinus Torvalds /* 5021da177e4SLinus Torvalds * Add a request to the inode's commit list. 5031da177e4SLinus Torvalds */ 5041763da12SFred Isaman void 505ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 506ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5071da177e4SLinus Torvalds { 508ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 5098dd37758STrond Myklebust return; 510ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 5111da177e4SLinus Torvalds } 5128e821cadSTrond Myklebust 513d6d6dc7cSFred Isaman static void 514d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 515e468bae9STrond Myklebust { 516e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 517d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); 518e468bae9STrond Myklebust } 519d6d6dc7cSFred Isaman 5208dd37758STrond Myklebust static void 521d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 522d6d6dc7cSFred Isaman { 5238dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 5248dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 525ea2cf228SFred Isaman struct nfs_commit_info cinfo; 526d6d6dc7cSFred Isaman 527ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 528ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 529ea2cf228SFred Isaman spin_lock(cinfo.lock); 530ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 531ea2cf228SFred Isaman spin_unlock(cinfo.lock); 532d6d6dc7cSFred Isaman } 5338dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 5348dd37758STrond Myklebust } 535e468bae9STrond Myklebust } 536e468bae9STrond Myklebust 5378e821cadSTrond Myklebust static inline 5389c7e1b3dSAnna Schumaker int nfs_write_need_commit(struct nfs_pgio_data *data) 5398e821cadSTrond Myklebust { 540465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 541cd841605SFred Isaman return data->header->lseg == NULL; 5428e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 5438e821cadSTrond Myklebust } 5448e821cadSTrond Myklebust 5458e821cadSTrond Myklebust #else 54668cd6fa4SBryan Schumaker static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 54768cd6fa4SBryan Schumaker struct inode *inode) 54868cd6fa4SBryan Schumaker { 54968cd6fa4SBryan Schumaker } 55068cd6fa4SBryan Schumaker 55168cd6fa4SBryan Schumaker void nfs_init_cinfo(struct nfs_commit_info *cinfo, 55268cd6fa4SBryan Schumaker struct inode *inode, 55368cd6fa4SBryan Schumaker struct nfs_direct_req *dreq) 55468cd6fa4SBryan Schumaker { 55568cd6fa4SBryan Schumaker } 55668cd6fa4SBryan Schumaker 5571763da12SFred Isaman void 558ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 559ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5608e821cadSTrond Myklebust { 5618e821cadSTrond Myklebust } 5628e821cadSTrond Myklebust 5638dd37758STrond Myklebust static void 564e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 565e468bae9STrond Myklebust { 566e468bae9STrond Myklebust } 567e468bae9STrond Myklebust 5688e821cadSTrond Myklebust static inline 5699c7e1b3dSAnna Schumaker int nfs_write_need_commit(struct nfs_pgio_data *data) 5708e821cadSTrond Myklebust { 5718e821cadSTrond Myklebust return 0; 5728e821cadSTrond Myklebust } 5738e821cadSTrond Myklebust 5741da177e4SLinus Torvalds #endif 5751da177e4SLinus Torvalds 576061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 5776c75dc0dSFred Isaman { 578ea2cf228SFred Isaman struct nfs_commit_info cinfo; 5796c75dc0dSFred Isaman unsigned long bytes = 0; 5806c75dc0dSFred Isaman 5816c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 5826c75dc0dSFred Isaman goto out; 583ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 5846c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 5856c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 5866c75dc0dSFred Isaman 5876c75dc0dSFred Isaman bytes += req->wb_bytes; 5886c75dc0dSFred Isaman nfs_list_remove_request(req); 5896c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 5906c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 591d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 5926c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 5936c75dc0dSFred Isaman goto remove_req; 5946c75dc0dSFred Isaman } 5956c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { 5966c75dc0dSFred Isaman nfs_mark_request_dirty(req); 5976c75dc0dSFred Isaman goto next; 5986c75dc0dSFred Isaman } 5996c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { 600f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 601ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 6026c75dc0dSFred Isaman goto next; 6036c75dc0dSFred Isaman } 6046c75dc0dSFred Isaman remove_req: 6056c75dc0dSFred Isaman nfs_inode_remove_request(req); 6066c75dc0dSFred Isaman next: 6071d1afcbcSTrond Myklebust nfs_unlock_request(req); 608d1182b33STrond Myklebust nfs_end_page_writeback(req->wb_page); 6093aff4ebbSTrond Myklebust nfs_release_request(req); 6106c75dc0dSFred Isaman } 6116c75dc0dSFred Isaman out: 6126c75dc0dSFred Isaman hdr->release(hdr); 6136c75dc0dSFred Isaman } 6146c75dc0dSFred Isaman 61589d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 616ea2cf228SFred Isaman static unsigned long 617ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 618fb8a1f11STrond Myklebust { 619ea2cf228SFred Isaman return cinfo->mds->ncommit; 620fb8a1f11STrond Myklebust } 621fb8a1f11STrond Myklebust 622ea2cf228SFred Isaman /* cinfo->lock held by caller */ 6231763da12SFred Isaman int 624ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 625ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 626d6d6dc7cSFred Isaman { 627d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 628d6d6dc7cSFred Isaman int ret = 0; 629d6d6dc7cSFred Isaman 630d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 6318dd37758STrond Myklebust if (!nfs_lock_request(req)) 6328dd37758STrond Myklebust continue; 6337ad84aa9STrond Myklebust kref_get(&req->wb_kref); 634ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 6353b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 636ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 6378dd37758STrond Myklebust nfs_list_add_request(req, dst); 638d6d6dc7cSFred Isaman ret++; 6391763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 640d6d6dc7cSFred Isaman break; 641d6d6dc7cSFred Isaman } 642d6d6dc7cSFred Isaman return ret; 643d6d6dc7cSFred Isaman } 644d6d6dc7cSFred Isaman 6451da177e4SLinus Torvalds /* 6461da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 6471da177e4SLinus Torvalds * @inode: NFS inode to scan 648ea2cf228SFred Isaman * @dst: mds destination list 649ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 6501da177e4SLinus Torvalds * 6511da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 6521da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6531da177e4SLinus Torvalds */ 6541763da12SFred Isaman int 655ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 656ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 6571da177e4SLinus Torvalds { 658d6d6dc7cSFred Isaman int ret = 0; 659fb8a1f11STrond Myklebust 660ea2cf228SFred Isaman spin_lock(cinfo->lock); 661ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 6628dd37758STrond Myklebust const int max = INT_MAX; 663d6d6dc7cSFred Isaman 664ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 665ea2cf228SFred Isaman cinfo, max); 666ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 667d6d6dc7cSFred Isaman } 668ea2cf228SFred Isaman spin_unlock(cinfo->lock); 669ff778d02STrond Myklebust return ret; 6701da177e4SLinus Torvalds } 671d6d6dc7cSFred Isaman 672c42de9ddSTrond Myklebust #else 673ea2cf228SFred Isaman static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 674fb8a1f11STrond Myklebust { 675fb8a1f11STrond Myklebust return 0; 676fb8a1f11STrond Myklebust } 677fb8a1f11STrond Myklebust 6781763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 679ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 680c42de9ddSTrond Myklebust { 681c42de9ddSTrond Myklebust return 0; 682c42de9ddSTrond Myklebust } 6831da177e4SLinus Torvalds #endif 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds /* 686e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 687e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 6881da177e4SLinus Torvalds * 689e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 690e7d39069STrond Myklebust * to disk. 6911da177e4SLinus Torvalds */ 692e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 693e7d39069STrond Myklebust struct page *page, 694e7d39069STrond Myklebust unsigned int offset, 695e7d39069STrond Myklebust unsigned int bytes) 6961da177e4SLinus Torvalds { 697e7d39069STrond Myklebust struct nfs_page *req; 698e7d39069STrond Myklebust unsigned int rqend; 699e7d39069STrond Myklebust unsigned int end; 7001da177e4SLinus Torvalds int error; 701277459d2STrond Myklebust 702e7d39069STrond Myklebust if (!PagePrivate(page)) 703e7d39069STrond Myklebust return NULL; 704e7d39069STrond Myklebust 705e7d39069STrond Myklebust end = offset + bytes; 706e7d39069STrond Myklebust spin_lock(&inode->i_lock); 707e7d39069STrond Myklebust 708e7d39069STrond Myklebust for (;;) { 70929418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 710e7d39069STrond Myklebust if (req == NULL) 711e7d39069STrond Myklebust goto out_unlock; 712e7d39069STrond Myklebust 713e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 714e7d39069STrond Myklebust /* 715e7d39069STrond Myklebust * Tell the caller to flush out the request if 716e7d39069STrond Myklebust * the offsets are non-contiguous. 717e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 718e7d39069STrond Myklebust * have flushed out requests having wrong owners. 719e7d39069STrond Myklebust */ 720e468bae9STrond Myklebust if (offset > rqend 721e7d39069STrond Myklebust || end < req->wb_offset) 722e7d39069STrond Myklebust goto out_flushme; 723e7d39069STrond Myklebust 7247ad84aa9STrond Myklebust if (nfs_lock_request(req)) 725e7d39069STrond Myklebust break; 726e7d39069STrond Myklebust 727e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 728587142f8STrond Myklebust spin_unlock(&inode->i_lock); 7291da177e4SLinus Torvalds error = nfs_wait_on_request(req); 7301da177e4SLinus Torvalds nfs_release_request(req); 731e7d39069STrond Myklebust if (error != 0) 732e7d39069STrond Myklebust goto out_err; 733e7d39069STrond Myklebust spin_lock(&inode->i_lock); 7341da177e4SLinus Torvalds } 7351da177e4SLinus Torvalds 7361da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7371da177e4SLinus Torvalds if (offset < req->wb_offset) { 7381da177e4SLinus Torvalds req->wb_offset = offset; 7391da177e4SLinus Torvalds req->wb_pgbase = offset; 7401da177e4SLinus Torvalds } 7411da177e4SLinus Torvalds if (end > rqend) 7421da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 743e7d39069STrond Myklebust else 744e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 745e7d39069STrond Myklebust out_unlock: 746e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 747ca138f36SFred Isaman if (req) 7488dd37758STrond Myklebust nfs_clear_request_commit(req); 749e7d39069STrond Myklebust return req; 750e7d39069STrond Myklebust out_flushme: 751e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 752e7d39069STrond Myklebust nfs_release_request(req); 753e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 754e7d39069STrond Myklebust out_err: 755e7d39069STrond Myklebust return ERR_PTR(error); 756e7d39069STrond Myklebust } 7571da177e4SLinus Torvalds 758e7d39069STrond Myklebust /* 759e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 760e7d39069STrond Myklebust * 761e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 762e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 763e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 764e7d39069STrond Myklebust */ 765e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 766e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 767e7d39069STrond Myklebust { 768d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 769e7d39069STrond Myklebust struct nfs_page *req; 770e7d39069STrond Myklebust 771e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 772e7d39069STrond Myklebust if (req != NULL) 773e7d39069STrond Myklebust goto out; 774e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 775e7d39069STrond Myklebust if (IS_ERR(req)) 776e7d39069STrond Myklebust goto out; 777d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 778efc91ed0STrond Myklebust out: 77961e930a9STrond Myklebust return req; 7801da177e4SLinus Torvalds } 7811da177e4SLinus Torvalds 782e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 783e7d39069STrond Myklebust unsigned int offset, unsigned int count) 784e7d39069STrond Myklebust { 785e7d39069STrond Myklebust struct nfs_page *req; 786e7d39069STrond Myklebust 787e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 788e7d39069STrond Myklebust if (IS_ERR(req)) 789e7d39069STrond Myklebust return PTR_ERR(req); 790e7d39069STrond Myklebust /* Update file length */ 791e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 792e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 793a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 7941d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 795e7d39069STrond Myklebust return 0; 796e7d39069STrond Myklebust } 797e7d39069STrond Myklebust 7981da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7991da177e4SLinus Torvalds { 800cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 8012a369153STrond Myklebust struct nfs_lock_context *l_ctx; 8021da177e4SLinus Torvalds struct nfs_page *req; 8031a54533eSTrond Myklebust int do_flush, status; 8041da177e4SLinus Torvalds /* 8051da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 8061da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 8071da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 8081da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 8091da177e4SLinus Torvalds * Also do the same if we find a request from an existing 8101da177e4SLinus Torvalds * dropped page. 8111da177e4SLinus Torvalds */ 8121a54533eSTrond Myklebust do { 813277459d2STrond Myklebust req = nfs_page_find_request(page); 8141a54533eSTrond Myklebust if (req == NULL) 8151a54533eSTrond Myklebust return 0; 8162a369153STrond Myklebust l_ctx = req->wb_lock_context; 8172a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 8180f1d2605STrond Myklebust if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { 8192a369153STrond Myklebust do_flush |= l_ctx->lockowner.l_owner != current->files 8202a369153STrond Myklebust || l_ctx->lockowner.l_pid != current->tgid; 8212a369153STrond Myklebust } 8221da177e4SLinus Torvalds nfs_release_request(req); 8231a54533eSTrond Myklebust if (!do_flush) 8241a54533eSTrond Myklebust return 0; 825d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 8261a54533eSTrond Myklebust } while (status == 0); 8271a54533eSTrond Myklebust return status; 8281da177e4SLinus Torvalds } 8291da177e4SLinus Torvalds 8301da177e4SLinus Torvalds /* 831dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 832dc24826bSAndy Adamson * expire soon. 833dc24826bSAndy Adamson * 834dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 835dc24826bSAndy Adamson * 836dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 837dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 838dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 839dc24826bSAndy Adamson */ 840dc24826bSAndy Adamson int 841dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 842dc24826bSAndy Adamson { 843dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 844dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 845dc24826bSAndy Adamson 846dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 847dc24826bSAndy Adamson } 848dc24826bSAndy Adamson 849dc24826bSAndy Adamson /* 850dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 851dc24826bSAndy Adamson */ 852dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 853dc24826bSAndy Adamson { 854dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 855dc24826bSAndy Adamson } 856dc24826bSAndy Adamson 857dc24826bSAndy Adamson /* 8585d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 8595d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 8605d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 8615d47a356STrond Myklebust */ 8628d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 8635d47a356STrond Myklebust { 864d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 865d529ef83SJeff Layton 8668d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 8678d197a56STrond Myklebust goto out; 868d529ef83SJeff Layton if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE)) 869d529ef83SJeff Layton return false; 8704db72b40SJeff Layton smp_rmb(); 871d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 8728d197a56STrond Myklebust return false; 8738d197a56STrond Myklebust out: 8748d197a56STrond Myklebust return PageUptodate(page) != 0; 8755d47a356STrond Myklebust } 8765d47a356STrond Myklebust 877c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 878c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 879c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 880c7559663SScott Mayhew * inefficiencies. 881c7559663SScott Mayhew * 882263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 883263b4509SScott Mayhew * of the checks. 884c7559663SScott Mayhew */ 885c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 886c7559663SScott Mayhew { 887c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 888c7559663SScott Mayhew return 0; 889263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 890263b4509SScott Mayhew return 0; 891c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 892c7559663SScott Mayhew return 1; 893263b4509SScott Mayhew if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && 894c7559663SScott Mayhew inode->i_flock->fl_end == OFFSET_MAX && 895263b4509SScott Mayhew inode->i_flock->fl_type != F_RDLCK)) 896c7559663SScott Mayhew return 1; 897c7559663SScott Mayhew return 0; 898c7559663SScott Mayhew } 899c7559663SScott Mayhew 9005d47a356STrond Myklebust /* 9011da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 9021da177e4SLinus Torvalds * 9031da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 9041da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 9051da177e4SLinus Torvalds */ 9061da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 9071da177e4SLinus Torvalds unsigned int offset, unsigned int count) 9081da177e4SLinus Torvalds { 909cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 910d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 9111da177e4SLinus Torvalds int status = 0; 9121da177e4SLinus Torvalds 91391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 91491d5b470SChuck Lever 9156de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 9166de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 9171da177e4SLinus Torvalds 918c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 91949a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 9201da177e4SLinus Torvalds offset = 0; 9211da177e4SLinus Torvalds } 9221da177e4SLinus Torvalds 923e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 92403fa9e84STrond Myklebust if (status < 0) 92503fa9e84STrond Myklebust nfs_set_pageerror(page); 92659b7c05fSTrond Myklebust else 92759b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 9281da177e4SLinus Torvalds 92948186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 9301da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 9311da177e4SLinus Torvalds return status; 9321da177e4SLinus Torvalds } 9331da177e4SLinus Torvalds 9343ff7576dSTrond Myklebust static int flush_task_priority(int how) 9351da177e4SLinus Torvalds { 9361da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 9371da177e4SLinus Torvalds case FLUSH_HIGHPRI: 9381da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 9391da177e4SLinus Torvalds case FLUSH_LOWPRI: 9401da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 9431da177e4SLinus Torvalds } 9441da177e4SLinus Torvalds 945c5996c4eSFred Isaman int nfs_initiate_write(struct rpc_clnt *clnt, 9469c7e1b3dSAnna Schumaker struct nfs_pgio_data *data, 947788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 9489f0ec176SAndy Adamson int how, int flags) 9491da177e4SLinus Torvalds { 950cd841605SFred Isaman struct inode *inode = data->header->inode; 9513ff7576dSTrond Myklebust int priority = flush_task_priority(how); 95207737691STrond Myklebust struct rpc_task *task; 953bdc7f021STrond Myklebust struct rpc_message msg = { 954bdc7f021STrond Myklebust .rpc_argp = &data->args, 955bdc7f021STrond Myklebust .rpc_resp = &data->res, 956cd841605SFred Isaman .rpc_cred = data->header->cred, 957bdc7f021STrond Myklebust }; 95884115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 959d138d5d1SAndy Adamson .rpc_client = clnt, 96007737691STrond Myklebust .task = &data->task, 961bdc7f021STrond Myklebust .rpc_message = &msg, 96284115e1cSTrond Myklebust .callback_ops = call_ops, 96384115e1cSTrond Myklebust .callback_data = data, 964101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 9659f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 9663ff7576dSTrond Myklebust .priority = priority, 96784115e1cSTrond Myklebust }; 9682c61be0aSTrond Myklebust int ret = 0; 9691da177e4SLinus Torvalds 970d138d5d1SAndy Adamson /* Set up the initial task struct. */ 971d138d5d1SAndy Adamson NFS_PROTO(inode)->write_setup(data, &msg); 972d138d5d1SAndy Adamson 973d138d5d1SAndy Adamson dprintk("NFS: %5u initiated write call " 9741e8968c5SNiels de Vos "(req %s/%llu, %u bytes @ offset %llu)\n", 975d138d5d1SAndy Adamson data->task.tk_pid, 976d138d5d1SAndy Adamson inode->i_sb->s_id, 9771e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(inode), 978d138d5d1SAndy Adamson data->args.count, 979d138d5d1SAndy Adamson (unsigned long long)data->args.offset); 980d138d5d1SAndy Adamson 9818c21c62cSWeston Andros Adamson nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 9828c21c62cSWeston Andros Adamson &task_setup_data.rpc_client, &msg, data); 9838c21c62cSWeston Andros Adamson 984d138d5d1SAndy Adamson task = rpc_run_task(&task_setup_data); 985d138d5d1SAndy Adamson if (IS_ERR(task)) { 986d138d5d1SAndy Adamson ret = PTR_ERR(task); 987d138d5d1SAndy Adamson goto out; 988d138d5d1SAndy Adamson } 989d138d5d1SAndy Adamson if (how & FLUSH_SYNC) { 990d138d5d1SAndy Adamson ret = rpc_wait_for_completion_task(task); 991d138d5d1SAndy Adamson if (ret == 0) 992d138d5d1SAndy Adamson ret = task->tk_status; 993d138d5d1SAndy Adamson } 994d138d5d1SAndy Adamson rpc_put_task(task); 995d138d5d1SAndy Adamson out: 996d138d5d1SAndy Adamson return ret; 997d138d5d1SAndy Adamson } 998a69aef14SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_write); 999d138d5d1SAndy Adamson 1000d138d5d1SAndy Adamson /* 1001d138d5d1SAndy Adamson * Set up the argument/result storage required for the RPC call. 1002d138d5d1SAndy Adamson */ 10039c7e1b3dSAnna Schumaker static void nfs_write_rpcsetup(struct nfs_pgio_data *data, 1004d138d5d1SAndy Adamson unsigned int count, unsigned int offset, 1005ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 1006d138d5d1SAndy Adamson { 10076c75dc0dSFred Isaman struct nfs_page *req = data->header->req; 1008d138d5d1SAndy Adamson 10091da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 10101da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 10111da177e4SLinus Torvalds 10126c75dc0dSFred Isaman data->args.fh = NFS_FH(data->header->inode); 10131da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 10142bea038cSBoaz Harrosh /* pnfs_set_layoutcommit needs this */ 10152bea038cSBoaz Harrosh data->mds_offset = data->args.offset; 10161da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 101730dd374fSFred Isaman data->args.pages = data->pages.pagevec; 10181da177e4SLinus Torvalds data->args.count = count; 1019383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 1020f11ac8dbSTrond Myklebust data->args.lock_context = req->wb_lock_context; 1021bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 102287ed5eb4STrond Myklebust switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { 102387ed5eb4STrond Myklebust case 0: 102487ed5eb4STrond Myklebust break; 102587ed5eb4STrond Myklebust case FLUSH_COND_STABLE: 1026ea2cf228SFred Isaman if (nfs_reqs_to_commit(cinfo)) 102787ed5eb4STrond Myklebust break; 102887ed5eb4STrond Myklebust default: 1029bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 1030bdc7f021STrond Myklebust } 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds data->res.fattr = &data->fattr; 10331da177e4SLinus Torvalds data->res.count = count; 10341da177e4SLinus Torvalds data->res.verf = &data->verf; 10350e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 10366e4efd56STrond Myklebust } 10371da177e4SLinus Torvalds 10389c7e1b3dSAnna Schumaker static int nfs_do_write(struct nfs_pgio_data *data, 10396e4efd56STrond Myklebust const struct rpc_call_ops *call_ops, 10406e4efd56STrond Myklebust int how) 10416e4efd56STrond Myklebust { 1042cd841605SFred Isaman struct inode *inode = data->header->inode; 10430382b744SAndy Adamson 10449f0ec176SAndy Adamson return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how, 0); 10451da177e4SLinus Torvalds } 10461da177e4SLinus Torvalds 1047275acaafSTrond Myklebust static int nfs_do_multiple_writes(struct list_head *head, 1048275acaafSTrond Myklebust const struct rpc_call_ops *call_ops, 1049275acaafSTrond Myklebust int how) 1050275acaafSTrond Myklebust { 10519c7e1b3dSAnna Schumaker struct nfs_pgio_data *data; 1052275acaafSTrond Myklebust int ret = 0; 1053275acaafSTrond Myklebust 1054275acaafSTrond Myklebust while (!list_empty(head)) { 1055275acaafSTrond Myklebust int ret2; 1056275acaafSTrond Myklebust 10579c7e1b3dSAnna Schumaker data = list_first_entry(head, struct nfs_pgio_data, list); 1058275acaafSTrond Myklebust list_del_init(&data->list); 1059275acaafSTrond Myklebust 1060dce81290STrond Myklebust ret2 = nfs_do_write(data, call_ops, how); 1061275acaafSTrond Myklebust if (ret == 0) 1062275acaafSTrond Myklebust ret = ret2; 1063275acaafSTrond Myklebust } 1064275acaafSTrond Myklebust return ret; 1065275acaafSTrond Myklebust } 1066275acaafSTrond Myklebust 10676d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 10686d884e8fSFred * call this on each, which will prepare them to be retried on next 10696d884e8fSFred * writeback using standard nfs. 10706d884e8fSFred */ 10716d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 10726d884e8fSFred { 10736d884e8fSFred nfs_mark_request_dirty(req); 10741d1afcbcSTrond Myklebust nfs_unlock_request(req); 1075d1182b33STrond Myklebust nfs_end_page_writeback(req->wb_page); 10763aff4ebbSTrond Myklebust nfs_release_request(req); 10776d884e8fSFred } 10786d884e8fSFred 1079061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 10806c75dc0dSFred Isaman { 10816c75dc0dSFred Isaman struct nfs_page *req; 10826c75dc0dSFred Isaman 10836c75dc0dSFred Isaman while (!list_empty(head)) { 10846c75dc0dSFred Isaman req = nfs_list_entry(head->next); 10856c75dc0dSFred Isaman nfs_list_remove_request(req); 10866c75dc0dSFred Isaman nfs_redirty_request(req); 10876c75dc0dSFred Isaman } 10886c75dc0dSFred Isaman } 10896c75dc0dSFred Isaman 1090061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1091061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1092061ae2edSFred Isaman .completion = nfs_write_completion, 1093061ae2edSFred Isaman }; 1094061ae2edSFred Isaman 109525b11dcdSTrond Myklebust static void nfs_flush_error(struct nfs_pageio_descriptor *desc, 109625b11dcdSTrond Myklebust struct nfs_pgio_header *hdr) 109725b11dcdSTrond Myklebust { 109825b11dcdSTrond Myklebust set_bit(NFS_IOHDR_REDO, &hdr->flags); 109925b11dcdSTrond Myklebust while (!list_empty(&hdr->rpc_list)) { 11009c7e1b3dSAnna Schumaker struct nfs_pgio_data *data = list_first_entry(&hdr->rpc_list, 11019c7e1b3dSAnna Schumaker struct nfs_pgio_data, list); 110225b11dcdSTrond Myklebust list_del(&data->list); 1103*00bfa30aSAnna Schumaker nfs_pgio_data_release(data); 110425b11dcdSTrond Myklebust } 110525b11dcdSTrond Myklebust desc->pg_completion_ops->error_cleanup(&desc->pg_list); 110625b11dcdSTrond Myklebust } 110725b11dcdSTrond Myklebust 11081da177e4SLinus Torvalds /* 11091da177e4SLinus Torvalds * Generate multiple small requests to write out a single 11101da177e4SLinus Torvalds * contiguous dirty area on one page. 11111da177e4SLinus Torvalds */ 11126c75dc0dSFred Isaman static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, 11136c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 11141da177e4SLinus Torvalds { 11156c75dc0dSFred Isaman struct nfs_page *req = hdr->req; 11161da177e4SLinus Torvalds struct page *page = req->wb_page; 11179c7e1b3dSAnna Schumaker struct nfs_pgio_data *data; 1118d097971dSTrond Myklebust size_t wsize = desc->pg_bsize, nbytes; 1119e9f7bee1STrond Myklebust unsigned int offset; 11201da177e4SLinus Torvalds int requests = 0; 1121ea2cf228SFred Isaman struct nfs_commit_info cinfo; 11221da177e4SLinus Torvalds 1123ea2cf228SFred Isaman nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); 11241da177e4SLinus Torvalds 1125b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1126ea2cf228SFred Isaman (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || 1127b31268acSTrond Myklebust desc->pg_count > wsize)) 1128b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1129b31268acSTrond Myklebust 1130b31268acSTrond Myklebust 1131275acaafSTrond Myklebust offset = 0; 1132c76069bdSFred Isaman nbytes = desc->pg_count; 1133e9f7bee1STrond Myklebust do { 1134e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 1135e9f7bee1STrond Myklebust 1136*00bfa30aSAnna Schumaker data = nfs_pgio_data_alloc(hdr, 1); 113725b11dcdSTrond Myklebust if (!data) { 113825b11dcdSTrond Myklebust nfs_flush_error(desc, hdr); 113925b11dcdSTrond Myklebust return -ENOMEM; 114025b11dcdSTrond Myklebust } 114130dd374fSFred Isaman data->pages.pagevec[0] = page; 1142ea2cf228SFred Isaman nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); 11436c75dc0dSFred Isaman list_add(&data->list, &hdr->rpc_list); 11441da177e4SLinus Torvalds requests++; 1145e9f7bee1STrond Myklebust nbytes -= len; 1146275acaafSTrond Myklebust offset += len; 1147e9f7bee1STrond Myklebust } while (nbytes != 0); 114825b11dcdSTrond Myklebust nfs_list_remove_request(req); 114925b11dcdSTrond Myklebust nfs_list_add_request(req, &hdr->pages); 11506c75dc0dSFred Isaman desc->pg_rpc_callops = &nfs_write_common_ops; 115125b11dcdSTrond Myklebust return 0; 11521da177e4SLinus Torvalds } 11531da177e4SLinus Torvalds 11541da177e4SLinus Torvalds /* 11551da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 11561da177e4SLinus Torvalds * The page must have been locked by the caller. 11571da177e4SLinus Torvalds * 11581da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 11591da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 11601da177e4SLinus Torvalds * that has been written but not committed. 11611da177e4SLinus Torvalds */ 11626c75dc0dSFred Isaman static int nfs_flush_one(struct nfs_pageio_descriptor *desc, 11636c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 11641da177e4SLinus Torvalds { 11651da177e4SLinus Torvalds struct nfs_page *req; 11661da177e4SLinus Torvalds struct page **pages; 11679c7e1b3dSAnna Schumaker struct nfs_pgio_data *data; 1168c76069bdSFred Isaman struct list_head *head = &desc->pg_list; 1169ea2cf228SFred Isaman struct nfs_commit_info cinfo; 11701da177e4SLinus Torvalds 1171*00bfa30aSAnna Schumaker data = nfs_pgio_data_alloc(hdr, nfs_page_array_len(desc->pg_base, 1172c76069bdSFred Isaman desc->pg_count)); 11736c75dc0dSFred Isaman if (!data) { 117425b11dcdSTrond Myklebust nfs_flush_error(desc, hdr); 117525b11dcdSTrond Myklebust return -ENOMEM; 117644b83799SFred Isaman } 11776c75dc0dSFred Isaman 1178ea2cf228SFred Isaman nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); 117930dd374fSFred Isaman pages = data->pages.pagevec; 11801da177e4SLinus Torvalds while (!list_empty(head)) { 11811da177e4SLinus Torvalds req = nfs_list_entry(head->next); 11821da177e4SLinus Torvalds nfs_list_remove_request(req); 11836c75dc0dSFred Isaman nfs_list_add_request(req, &hdr->pages); 11841da177e4SLinus Torvalds *pages++ = req->wb_page; 11851da177e4SLinus Torvalds } 11861da177e4SLinus Torvalds 1187b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1188ea2cf228SFred Isaman (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) 1189b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1190b31268acSTrond Myklebust 11911da177e4SLinus Torvalds /* Set up the argument struct */ 1192ea2cf228SFred Isaman nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); 11936c75dc0dSFred Isaman list_add(&data->list, &hdr->rpc_list); 11946c75dc0dSFred Isaman desc->pg_rpc_callops = &nfs_write_common_ops; 119525b11dcdSTrond Myklebust return 0; 11961da177e4SLinus Torvalds } 11971da177e4SLinus Torvalds 11986c75dc0dSFred Isaman int nfs_generic_flush(struct nfs_pageio_descriptor *desc, 11996c75dc0dSFred Isaman struct nfs_pgio_header *hdr) 1200dce81290STrond Myklebust { 1201dce81290STrond Myklebust if (desc->pg_bsize < PAGE_CACHE_SIZE) 12026c75dc0dSFred Isaman return nfs_flush_multi(desc, hdr); 12036c75dc0dSFred Isaman return nfs_flush_one(desc, hdr); 1204dce81290STrond Myklebust } 120589d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_generic_flush); 1206dce81290STrond Myklebust 1207dce81290STrond Myklebust static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) 12081751c363STrond Myklebust { 1209c0752cdfSAnna Schumaker struct nfs_rw_header *whdr; 12106c75dc0dSFred Isaman struct nfs_pgio_header *hdr; 1211275acaafSTrond Myklebust int ret; 1212275acaafSTrond Myklebust 12136c75dc0dSFred Isaman whdr = nfs_writehdr_alloc(); 12146c75dc0dSFred Isaman if (!whdr) { 12159b5415b5STrond Myklebust desc->pg_completion_ops->error_cleanup(&desc->pg_list); 12166c75dc0dSFred Isaman return -ENOMEM; 12176c75dc0dSFred Isaman } 12186c75dc0dSFred Isaman hdr = &whdr->header; 12196c75dc0dSFred Isaman nfs_pgheader_init(desc, hdr, nfs_writehdr_free); 12206c75dc0dSFred Isaman atomic_inc(&hdr->refcnt); 12216c75dc0dSFred Isaman ret = nfs_generic_flush(desc, hdr); 1222275acaafSTrond Myklebust if (ret == 0) 12236c75dc0dSFred Isaman ret = nfs_do_multiple_writes(&hdr->rpc_list, 12246c75dc0dSFred Isaman desc->pg_rpc_callops, 1225dce81290STrond Myklebust desc->pg_ioflags); 12266c75dc0dSFred Isaman if (atomic_dec_and_test(&hdr->refcnt)) 1227061ae2edSFred Isaman hdr->completion_ops->completion(hdr); 1228275acaafSTrond Myklebust return ret; 12291751c363STrond Myklebust } 12301751c363STrond Myklebust 12311751c363STrond Myklebust static const struct nfs_pageio_ops nfs_pageio_write_ops = { 12321751c363STrond Myklebust .pg_test = nfs_generic_pg_test, 12331751c363STrond Myklebust .pg_doio = nfs_generic_pg_writepages, 12341751c363STrond Myklebust }; 12351751c363STrond Myklebust 123657208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1237a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1238061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 12391751c363STrond Myklebust { 1240a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 1241a20c93e3SChristoph Hellwig const struct nfs_pageio_ops *pg_ops = &nfs_pageio_write_ops; 1242a20c93e3SChristoph Hellwig 1243a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1244a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1245a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1246a20c93e3SChristoph Hellwig #endif 1247a20c93e3SChristoph Hellwig nfs_pageio_init(pgio, inode, pg_ops, compl_ops, server->wsize, ioflags); 12481751c363STrond Myklebust } 1249ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 12501751c363STrond Myklebust 1251dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1252dce81290STrond Myklebust { 1253dce81290STrond Myklebust pgio->pg_ops = &nfs_pageio_write_ops; 1254dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1255dce81290STrond Myklebust } 12561f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1257dce81290STrond Myklebust 12581da177e4SLinus Torvalds 1259def6ed7eSAndy Adamson void nfs_write_prepare(struct rpc_task *task, void *calldata) 1260def6ed7eSAndy Adamson { 12619c7e1b3dSAnna Schumaker struct nfs_pgio_data *data = calldata; 1262ef1820f9SNeilBrown int err; 1263ef1820f9SNeilBrown err = NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); 1264ef1820f9SNeilBrown if (err) 1265ef1820f9SNeilBrown rpc_exit(task, err); 1266def6ed7eSAndy Adamson } 1267def6ed7eSAndy Adamson 12680b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 12690b7c0153SFred Isaman { 12700b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 12710b7c0153SFred Isaman 12720b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 12730b7c0153SFred Isaman } 12740b7c0153SFred Isaman 12751da177e4SLinus Torvalds /* 12761da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 12771da177e4SLinus Torvalds * 12781da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 12791da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 12801da177e4SLinus Torvalds * as the page has a write request pending. 12811da177e4SLinus Torvalds */ 12826c75dc0dSFred Isaman static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) 12831da177e4SLinus Torvalds { 12849c7e1b3dSAnna Schumaker struct nfs_pgio_data *data = calldata; 12851da177e4SLinus Torvalds 1286c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1287c9d8f89dSTrond Myklebust } 1288c9d8f89dSTrond Myklebust 12896c75dc0dSFred Isaman static void nfs_writeback_release_common(void *calldata) 1290c9d8f89dSTrond Myklebust { 12919c7e1b3dSAnna Schumaker struct nfs_pgio_data *data = calldata; 1292cd841605SFred Isaman struct nfs_pgio_header *hdr = data->header; 1293e2fecb21STrond Myklebust int status = data->task.tk_status; 1294788e7a89STrond Myklebust 12956c75dc0dSFred Isaman if ((status >= 0) && nfs_write_need_commit(data)) { 12966c75dc0dSFred Isaman spin_lock(&hdr->lock); 12976c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) 12986c75dc0dSFred Isaman ; /* Do nothing */ 12996c75dc0dSFred Isaman else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) 1300f79d06f5SAnna Schumaker memcpy(&hdr->verf, &data->verf, sizeof(hdr->verf)); 1301f79d06f5SAnna Schumaker else if (memcmp(&hdr->verf, &data->verf, sizeof(hdr->verf))) 13026c75dc0dSFred Isaman set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); 13036c75dc0dSFred Isaman spin_unlock(&hdr->lock); 13041da177e4SLinus Torvalds } 1305*00bfa30aSAnna Schumaker nfs_pgio_data_release(data); 13061da177e4SLinus Torvalds } 13071da177e4SLinus Torvalds 13086c75dc0dSFred Isaman static const struct rpc_call_ops nfs_write_common_ops = { 1309def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 13106c75dc0dSFred Isaman .rpc_call_done = nfs_writeback_done_common, 13116c75dc0dSFred Isaman .rpc_release = nfs_writeback_release_common, 1312788e7a89STrond Myklebust }; 1313788e7a89STrond Myklebust 13141f2edbe3STrond Myklebust /* 13151f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 13161f2edbe3STrond Myklebust */ 13171f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 13181f2edbe3STrond Myklebust { 13191f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 13201f2edbe3STrond Myklebust int kill = 0; 13211f2edbe3STrond Myklebust 13221f2edbe3STrond Myklebust /* suid always must be killed */ 13231f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 13241f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 13251f2edbe3STrond Myklebust 13261f2edbe3STrond Myklebust /* 13271f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 13281f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 13291f2edbe3STrond Myklebust */ 13301f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 13311f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 13321f2edbe3STrond Myklebust 13331f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 13341f2edbe3STrond Myklebust return kill; 13351f2edbe3STrond Myklebust 13361f2edbe3STrond Myklebust return 0; 13371f2edbe3STrond Myklebust } 1338788e7a89STrond Myklebust 13391da177e4SLinus Torvalds /* 13401da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 13411da177e4SLinus Torvalds */ 13429c7e1b3dSAnna Schumaker void nfs_writeback_done(struct rpc_task *task, struct nfs_pgio_data *data) 13431da177e4SLinus Torvalds { 13443c6b899cSAnna Schumaker struct nfs_pgio_args *argp = &data->args; 13459137bdf3SAnna Schumaker struct nfs_pgio_res *resp = &data->res; 1346cd841605SFred Isaman struct inode *inode = data->header->inode; 1347788e7a89STrond Myklebust int status; 13481da177e4SLinus Torvalds 1349a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 13501da177e4SLinus Torvalds task->tk_pid, task->tk_status); 13511da177e4SLinus Torvalds 1352f551e44fSChuck Lever /* 1353f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1354f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1355f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1356f551e44fSChuck Lever * another writer had changed the file, but some applications 1357f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1358f551e44fSChuck Lever */ 1359cd841605SFred Isaman status = NFS_PROTO(inode)->write_done(task, data); 1360788e7a89STrond Myklebust if (status != 0) 136113602896SFred Isaman return; 1362cd841605SFred Isaman nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 136391d5b470SChuck Lever 136489d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 13651da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 13661da177e4SLinus Torvalds /* We tried a write call, but the server did not 13671da177e4SLinus Torvalds * commit data to stable storage even though we 13681da177e4SLinus Torvalds * requested it. 13691da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 13701da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 13711da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 13721da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 13731da177e4SLinus Torvalds */ 13741da177e4SLinus Torvalds static unsigned long complain; 13751da177e4SLinus Torvalds 1376a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 13771da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 13781da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 13791da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1380cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 13811da177e4SLinus Torvalds resp->verf->committed, argp->stable); 13821da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds #endif 13861f2edbe3STrond Myklebust if (task->tk_status < 0) { 13876c75dc0dSFred Isaman nfs_set_pgio_error(data->header, task->tk_status, argp->offset); 13881f2edbe3STrond Myklebust return; 13891f2edbe3STrond Myklebust } 13901f2edbe3STrond Myklebust 13911f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 13921f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 13931f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 13941f2edbe3STrond Myklebust 13951f2edbe3STrond Myklebust if (resp->count < argp->count) { 13961da177e4SLinus Torvalds static unsigned long complain; 13971da177e4SLinus Torvalds 13986c75dc0dSFred Isaman /* This a short write! */ 1399cd841605SFred Isaman nfs_inc_stats(inode, NFSIOS_SHORTWRITE); 140091d5b470SChuck Lever 14011da177e4SLinus Torvalds /* Has the server at least made some progress? */ 14026c75dc0dSFred Isaman if (resp->count == 0) { 14036c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 14046c75dc0dSFred Isaman printk(KERN_WARNING 14056c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 14066c75dc0dSFred Isaman argp->count); 14076c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 14086c75dc0dSFred Isaman } 14096c75dc0dSFred Isaman nfs_set_pgio_error(data->header, -EIO, argp->offset); 14106c75dc0dSFred Isaman task->tk_status = -EIO; 14116c75dc0dSFred Isaman return; 14126c75dc0dSFred Isaman } 14131da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 14141da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 14151da177e4SLinus Torvalds /* Resend from where the server left off */ 1416a69aef14SFred Isaman data->mds_offset += resp->count; 14171da177e4SLinus Torvalds argp->offset += resp->count; 14181da177e4SLinus Torvalds argp->pgbase += resp->count; 14191da177e4SLinus Torvalds argp->count -= resp->count; 14201da177e4SLinus Torvalds } else { 14211da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 14221da177e4SLinus Torvalds * headaches in the case of a server crash. 14231da177e4SLinus Torvalds */ 14241da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 14251da177e4SLinus Torvalds } 1426d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 14271da177e4SLinus Torvalds } 14281da177e4SLinus Torvalds } 14291da177e4SLinus Torvalds 14301da177e4SLinus Torvalds 143189d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 143271d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 143371d0a611STrond Myklebust { 1434b8413f98STrond Myklebust int ret; 1435b8413f98STrond Myklebust 143671d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 143771d0a611STrond Myklebust return 1; 1438b8413f98STrond Myklebust if (!may_wait) 143971d0a611STrond Myklebust return 0; 1440b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1441b8413f98STrond Myklebust NFS_INO_COMMIT, 1442b8413f98STrond Myklebust nfs_wait_bit_killable, 1443b8413f98STrond Myklebust TASK_KILLABLE); 1444b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 144571d0a611STrond Myklebust } 144671d0a611STrond Myklebust 1447f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 144871d0a611STrond Myklebust { 144971d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 145071d0a611STrond Myklebust smp_mb__after_clear_bit(); 145171d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 145271d0a611STrond Myklebust } 145371d0a611STrond Myklebust 14540b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 14551da177e4SLinus Torvalds { 14560b7c0153SFred Isaman put_nfs_open_context(data->context); 14570b7c0153SFred Isaman nfs_commit_free(data); 14581da177e4SLinus Torvalds } 1459e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 14601da177e4SLinus Torvalds 14610b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 14629ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 14639f0ec176SAndy Adamson int how, int flags) 14641da177e4SLinus Torvalds { 146507737691STrond Myklebust struct rpc_task *task; 14669ace33cdSFred Isaman int priority = flush_task_priority(how); 1467bdc7f021STrond Myklebust struct rpc_message msg = { 1468bdc7f021STrond Myklebust .rpc_argp = &data->args, 1469bdc7f021STrond Myklebust .rpc_resp = &data->res, 14709ace33cdSFred Isaman .rpc_cred = data->cred, 1471bdc7f021STrond Myklebust }; 147284115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 147307737691STrond Myklebust .task = &data->task, 14749ace33cdSFred Isaman .rpc_client = clnt, 1475bdc7f021STrond Myklebust .rpc_message = &msg, 14769ace33cdSFred Isaman .callback_ops = call_ops, 147784115e1cSTrond Myklebust .callback_data = data, 1478101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 14799f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 14803ff7576dSTrond Myklebust .priority = priority, 148184115e1cSTrond Myklebust }; 1482788e7a89STrond Myklebust /* Set up the initial task struct. */ 14839ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 14841da177e4SLinus Torvalds 1485a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1486bdc7f021STrond Myklebust 14878c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 14888c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 14898c21c62cSWeston Andros Adamson 149007737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1491dbae4c73STrond Myklebust if (IS_ERR(task)) 1492dbae4c73STrond Myklebust return PTR_ERR(task); 1493d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1494d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 149507737691STrond Myklebust rpc_put_task(task); 1496dbae4c73STrond Myklebust return 0; 14971da177e4SLinus Torvalds } 1498e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 14991da177e4SLinus Torvalds 15001da177e4SLinus Torvalds /* 15019ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 15029ace33cdSFred Isaman */ 15030b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1504988b6dceSFred Isaman struct list_head *head, 1505f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1506f453a54aSFred Isaman struct nfs_commit_info *cinfo) 15079ace33cdSFred Isaman { 15089ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 15093d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 15109ace33cdSFred Isaman 15119ace33cdSFred Isaman /* Set up the RPC argument and reply structs 15129ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 15139ace33cdSFred Isaman 15149ace33cdSFred Isaman list_splice_init(head, &data->pages); 15159ace33cdSFred Isaman 15169ace33cdSFred Isaman data->inode = inode; 15179ace33cdSFred Isaman data->cred = first->wb_context->cred; 1518988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 15199ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1520f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1521b359f9d0SFred Isaman data->dreq = cinfo->dreq; 15229ace33cdSFred Isaman 15239ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 15249ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 15259ace33cdSFred Isaman data->args.offset = 0; 15269ace33cdSFred Isaman data->args.count = 0; 15270b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 15289ace33cdSFred Isaman data->res.fattr = &data->fattr; 15299ace33cdSFred Isaman data->res.verf = &data->verf; 15309ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 15319ace33cdSFred Isaman } 1532e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 15339ace33cdSFred Isaman 1534e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1535ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1536ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 153764bfeb49SFred Isaman { 153864bfeb49SFred Isaman struct nfs_page *req; 153964bfeb49SFred Isaman 154064bfeb49SFred Isaman while (!list_empty(page_list)) { 154164bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 154264bfeb49SFred Isaman nfs_list_remove_request(req); 1543ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 154456f9cd68SFred Isaman if (!cinfo->dreq) { 154564bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1546d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 154764bfeb49SFred Isaman BDI_RECLAIMABLE); 154856f9cd68SFred Isaman } 15491d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 155064bfeb49SFred Isaman } 155164bfeb49SFred Isaman } 1552e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 155364bfeb49SFred Isaman 15549ace33cdSFred Isaman /* 15551da177e4SLinus Torvalds * Commit dirty pages 15561da177e4SLinus Torvalds */ 15571da177e4SLinus Torvalds static int 1558ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1559ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 15601da177e4SLinus Torvalds { 15610b7c0153SFred Isaman struct nfs_commit_data *data; 15621da177e4SLinus Torvalds 1563c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 15641da177e4SLinus Torvalds 15651da177e4SLinus Torvalds if (!data) 15661da177e4SLinus Torvalds goto out_bad; 15671da177e4SLinus Torvalds 15681da177e4SLinus Torvalds /* Set up the argument struct */ 1569f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1570f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 15719f0ec176SAndy Adamson return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 15729f0ec176SAndy Adamson how, 0); 15731da177e4SLinus Torvalds out_bad: 1574ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1575f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 15761da177e4SLinus Torvalds return -ENOMEM; 15771da177e4SLinus Torvalds } 15781da177e4SLinus Torvalds 15791da177e4SLinus Torvalds /* 15801da177e4SLinus Torvalds * COMMIT call returned 15811da177e4SLinus Torvalds */ 1582788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 15831da177e4SLinus Torvalds { 15840b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 15851da177e4SLinus Torvalds 1586a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 15871da177e4SLinus Torvalds task->tk_pid, task->tk_status); 15881da177e4SLinus Torvalds 1589788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1590c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1591c9d8f89dSTrond Myklebust } 1592c9d8f89dSTrond Myklebust 1593f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1594c9d8f89dSTrond Myklebust { 1595c9d8f89dSTrond Myklebust struct nfs_page *req; 1596c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1597f453a54aSFred Isaman struct nfs_commit_info cinfo; 1598788e7a89STrond Myklebust 15991da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 16001da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 16011da177e4SLinus Torvalds nfs_list_remove_request(req); 1602d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 16031da177e4SLinus Torvalds 16041e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 16053d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 16061e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), 16071da177e4SLinus Torvalds req->wb_bytes, 16081da177e4SLinus Torvalds (long long)req_offset(req)); 1609c9d8f89dSTrond Myklebust if (status < 0) { 1610c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 16111da177e4SLinus Torvalds nfs_inode_remove_request(req); 1612c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 16131da177e4SLinus Torvalds goto next; 16141da177e4SLinus Torvalds } 16151da177e4SLinus Torvalds 16161da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 16171da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 16182f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 16191da177e4SLinus Torvalds /* We have a match */ 16201da177e4SLinus Torvalds nfs_inode_remove_request(req); 16211da177e4SLinus Torvalds dprintk(" OK\n"); 16221da177e4SLinus Torvalds goto next; 16231da177e4SLinus Torvalds } 16241da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 16251da177e4SLinus Torvalds dprintk(" mismatch\n"); 16266d884e8fSFred nfs_mark_request_dirty(req); 162705990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 16281da177e4SLinus Torvalds next: 16291d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 16301da177e4SLinus Torvalds } 1631f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1632f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1633f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 16345917ce84SFred Isaman } 16355917ce84SFred Isaman 16365917ce84SFred Isaman static void nfs_commit_release(void *calldata) 16375917ce84SFred Isaman { 16380b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16395917ce84SFred Isaman 1640f453a54aSFred Isaman data->completion_ops->completion(data); 1641c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 16421da177e4SLinus Torvalds } 1643788e7a89STrond Myklebust 1644788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 16450b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1646788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1647788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1648788e7a89STrond Myklebust }; 16491da177e4SLinus Torvalds 1650f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1651f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1652f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1653f453a54aSFred Isaman }; 1654f453a54aSFred Isaman 16551763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1656ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 165784c53ab5SFred Isaman { 165884c53ab5SFred Isaman int status; 165984c53ab5SFred Isaman 1660ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 166184c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1662ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 166384c53ab5SFred Isaman return status; 166484c53ab5SFred Isaman } 166584c53ab5SFred Isaman 1666b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 16671da177e4SLinus Torvalds { 16681da177e4SLinus Torvalds LIST_HEAD(head); 1669ea2cf228SFred Isaman struct nfs_commit_info cinfo; 167071d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1671b8413f98STrond Myklebust int res; 16721da177e4SLinus Torvalds 1673b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1674b8413f98STrond Myklebust if (res <= 0) 1675c5efa5fcSTrond Myklebust goto out_mark_dirty; 1676ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1677ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 16781da177e4SLinus Torvalds if (res) { 1679a861a1e1SFred Isaman int error; 1680a861a1e1SFred Isaman 1681ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 16821da177e4SLinus Torvalds if (error < 0) 16831da177e4SLinus Torvalds return error; 1684b8413f98STrond Myklebust if (!may_wait) 1685b8413f98STrond Myklebust goto out_mark_dirty; 1686b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1687b8413f98STrond Myklebust NFS_INO_COMMIT, 168871d0a611STrond Myklebust nfs_wait_bit_killable, 168971d0a611STrond Myklebust TASK_KILLABLE); 1690b8413f98STrond Myklebust if (error < 0) 1691b8413f98STrond Myklebust return error; 169271d0a611STrond Myklebust } else 169371d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1694c5efa5fcSTrond Myklebust return res; 1695c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1696c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1697c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1698c5efa5fcSTrond Myklebust * that the data is on the disk. 1699c5efa5fcSTrond Myklebust */ 1700c5efa5fcSTrond Myklebust out_mark_dirty: 1701c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17021da177e4SLinus Torvalds return res; 17031da177e4SLinus Torvalds } 17048fc795f7STrond Myklebust 17058fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17068fc795f7STrond Myklebust { 1707420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1708420e3646STrond Myklebust int flags = FLUSH_SYNC; 1709420e3646STrond Myklebust int ret = 0; 17108fc795f7STrond Myklebust 17113236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1712ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 17133236c3e1SJeff Layton return ret; 17143236c3e1SJeff Layton 1715a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1716a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1717a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1718420e3646STrond Myklebust */ 1719ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1720420e3646STrond Myklebust goto out_mark_dirty; 1721420e3646STrond Myklebust 1722a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1723420e3646STrond Myklebust flags = 0; 1724a00dd6c0SJeff Layton } 1725a00dd6c0SJeff Layton 1726420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1727420e3646STrond Myklebust if (ret >= 0) { 1728420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1729420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1730420e3646STrond Myklebust wbc->nr_to_write -= ret; 1731420e3646STrond Myklebust else 1732420e3646STrond Myklebust wbc->nr_to_write = 0; 1733420e3646STrond Myklebust } 17348fc795f7STrond Myklebust return 0; 1735420e3646STrond Myklebust } 1736420e3646STrond Myklebust out_mark_dirty: 17378fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17388fc795f7STrond Myklebust return ret; 17398fc795f7STrond Myklebust } 1740c63c7b05STrond Myklebust #else 17418fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17428fc795f7STrond Myklebust { 17438fc795f7STrond Myklebust return 0; 17448fc795f7STrond Myklebust } 17451da177e4SLinus Torvalds #endif 17461da177e4SLinus Torvalds 17478fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 17488fc795f7STrond Myklebust { 1749a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1750a8d8f02cSBryan Schumaker } 175189d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1752863a3c6cSAndy Adamson 1753acdc53b2STrond Myklebust /* 1754acdc53b2STrond Myklebust * flush the inode to disk. 1755acdc53b2STrond Myklebust */ 1756acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 175734901f70STrond Myklebust { 175834901f70STrond Myklebust struct writeback_control wbc = { 175972cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 176034901f70STrond Myklebust .nr_to_write = LONG_MAX, 1761d7fb1207STrond Myklebust .range_start = 0, 1762d7fb1207STrond Myklebust .range_end = LLONG_MAX, 176334901f70STrond Myklebust }; 1764f4ce1299STrond Myklebust int ret; 176534901f70STrond Myklebust 1766f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1767f4ce1299STrond Myklebust 1768f4ce1299STrond Myklebust ret = sync_inode(inode, &wbc); 1769f4ce1299STrond Myklebust 1770f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1771f4ce1299STrond Myklebust return ret; 17721c75950bSTrond Myklebust } 1773ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 17741c75950bSTrond Myklebust 17751b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 17761b3b4a1aSTrond Myklebust { 17771b3b4a1aSTrond Myklebust struct nfs_page *req; 17781b3b4a1aSTrond Myklebust int ret = 0; 17791b3b4a1aSTrond Myklebust 17801b3b4a1aSTrond Myklebust for (;;) { 1781ba8b06e6STrond Myklebust wait_on_page_writeback(page); 17821b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 17831b3b4a1aSTrond Myklebust if (req == NULL) 17841b3b4a1aSTrond Myklebust break; 17857ad84aa9STrond Myklebust if (nfs_lock_request(req)) { 17868dd37758STrond Myklebust nfs_clear_request_commit(req); 17871b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 17881b3b4a1aSTrond Myklebust /* 17891b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 17901b3b4a1aSTrond Myklebust * page as being dirty 17911b3b4a1aSTrond Myklebust */ 17921b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 17931d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 17941b3b4a1aSTrond Myklebust break; 17951b3b4a1aSTrond Myklebust } 17961b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1797c9edda71STrond Myklebust nfs_release_request(req); 17981b3b4a1aSTrond Myklebust if (ret < 0) 1799c988950eSTrond Myklebust break; 18001b3b4a1aSTrond Myklebust } 18011b3b4a1aSTrond Myklebust return ret; 18021b3b4a1aSTrond Myklebust } 18031b3b4a1aSTrond Myklebust 18041c75950bSTrond Myklebust /* 18051c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 18061c75950bSTrond Myklebust */ 18071c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 18081c75950bSTrond Myklebust { 180929418aa4SMel Gorman loff_t range_start = page_file_offset(page); 18107f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 18117f2f12d9STrond Myklebust struct writeback_control wbc = { 18127f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 18137f2f12d9STrond Myklebust .nr_to_write = 0, 18147f2f12d9STrond Myklebust .range_start = range_start, 18157f2f12d9STrond Myklebust .range_end = range_end, 18167f2f12d9STrond Myklebust }; 18177f2f12d9STrond Myklebust int ret; 18187f2f12d9STrond Myklebust 1819f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1820f4ce1299STrond Myklebust 18210522f6adSTrond Myklebust for (;;) { 1822ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18237f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 18247f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 18257f2f12d9STrond Myklebust if (ret < 0) 18267f2f12d9STrond Myklebust goto out_error; 18270522f6adSTrond Myklebust continue; 18287f2f12d9STrond Myklebust } 1829f4ce1299STrond Myklebust ret = 0; 18300522f6adSTrond Myklebust if (!PagePrivate(page)) 18310522f6adSTrond Myklebust break; 18320522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 18337f2f12d9STrond Myklebust if (ret < 0) 18347f2f12d9STrond Myklebust goto out_error; 18357f2f12d9STrond Myklebust } 18367f2f12d9STrond Myklebust out_error: 1837f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 18387f2f12d9STrond Myklebust return ret; 18391c75950bSTrond Myklebust } 18401c75950bSTrond Myklebust 1841074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1842074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1843a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1844074cc1deSTrond Myklebust { 18452da95652SJeff Layton /* 18462da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 18472da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 18482da95652SJeff Layton * 18492da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 18502da95652SJeff Layton * that we can safely release the inode reference while holding 18512da95652SJeff Layton * the page lock. 18522da95652SJeff Layton */ 18532da95652SJeff Layton if (PagePrivate(page)) 18542da95652SJeff Layton return -EBUSY; 1855074cc1deSTrond Myklebust 18568c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 18578c209ce7SDavid Howells return -EBUSY; 1858074cc1deSTrond Myklebust 1859a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1860074cc1deSTrond Myklebust } 1861074cc1deSTrond Myklebust #endif 1862074cc1deSTrond Myklebust 1863f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 18641da177e4SLinus Torvalds { 18651da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 1866c0752cdfSAnna Schumaker sizeof(struct nfs_rw_header), 18671da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 186820c2df83SPaul Mundt NULL); 18691da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 18701da177e4SLinus Torvalds return -ENOMEM; 18711da177e4SLinus Torvalds 187293d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 18731da177e4SLinus Torvalds nfs_wdata_cachep); 18741da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 18753dd4765fSJeff Layton goto out_destroy_write_cache; 18761da177e4SLinus Torvalds 18770b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 18780b7c0153SFred Isaman sizeof(struct nfs_commit_data), 18790b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 18800b7c0153SFred Isaman NULL); 18810b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 18823dd4765fSJeff Layton goto out_destroy_write_mempool; 18830b7c0153SFred Isaman 188493d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 18854c100210SYanchuan Nian nfs_cdata_cachep); 18861da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 18873dd4765fSJeff Layton goto out_destroy_commit_cache; 18881da177e4SLinus Torvalds 188989a09141SPeter Zijlstra /* 189089a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 189189a09141SPeter Zijlstra * 189289a09141SPeter Zijlstra * 64MB: 8192k 189389a09141SPeter Zijlstra * 128MB: 11585k 189489a09141SPeter Zijlstra * 256MB: 16384k 189589a09141SPeter Zijlstra * 512MB: 23170k 189689a09141SPeter Zijlstra * 1GB: 32768k 189789a09141SPeter Zijlstra * 2GB: 46340k 189889a09141SPeter Zijlstra * 4GB: 65536k 189989a09141SPeter Zijlstra * 8GB: 92681k 190089a09141SPeter Zijlstra * 16GB: 131072k 190189a09141SPeter Zijlstra * 190289a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 190389a09141SPeter Zijlstra * Limit the default to 256M 190489a09141SPeter Zijlstra */ 190589a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 190689a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 190789a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 190889a09141SPeter Zijlstra 19091da177e4SLinus Torvalds return 0; 19103dd4765fSJeff Layton 19113dd4765fSJeff Layton out_destroy_commit_cache: 19123dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19133dd4765fSJeff Layton out_destroy_write_mempool: 19143dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 19153dd4765fSJeff Layton out_destroy_write_cache: 19163dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 19173dd4765fSJeff Layton return -ENOMEM; 19181da177e4SLinus Torvalds } 19191da177e4SLinus Torvalds 1920266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 19211da177e4SLinus Torvalds { 19221da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 19233dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19241da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 19251a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 19261da177e4SLinus Torvalds } 19271da177e4SLinus Torvalds 1928