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); 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 46061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 47f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 484a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_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 734a0de55cSAnna Schumaker static 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 774a0de55cSAnna Schumaker if (p) 783feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 793feb2d49STrond Myklebust return p; 803feb2d49STrond Myklebust } 813feb2d49STrond Myklebust 824a0de55cSAnna Schumaker static void nfs_writehdr_free(struct nfs_rw_header *whdr) 836c75dc0dSFred Isaman { 84cd841605SFred Isaman mempool_free(whdr, nfs_wdata_mempool); 853feb2d49STrond Myklebust } 861da177e4SLinus Torvalds 877b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 887b159fc1STrond Myklebust { 897b159fc1STrond Myklebust ctx->error = error; 907b159fc1STrond Myklebust smp_wmb(); 917b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 927b159fc1STrond Myklebust } 937b159fc1STrond Myklebust 9429418aa4SMel Gorman static struct nfs_page * 9529418aa4SMel Gorman nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page) 96277459d2STrond Myklebust { 97277459d2STrond Myklebust struct nfs_page *req = NULL; 98277459d2STrond Myklebust 9929418aa4SMel Gorman if (PagePrivate(page)) 100277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 10129418aa4SMel Gorman else if (unlikely(PageSwapCache(page))) { 10229418aa4SMel Gorman struct nfs_page *freq, *t; 10329418aa4SMel Gorman 10429418aa4SMel Gorman /* Linearly search the commit list for the correct req */ 10529418aa4SMel Gorman list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { 10629418aa4SMel Gorman if (freq->wb_page == page) { 10729418aa4SMel Gorman req = freq; 10829418aa4SMel Gorman break; 109277459d2STrond Myklebust } 11029418aa4SMel Gorman } 11129418aa4SMel Gorman } 11229418aa4SMel Gorman 11329418aa4SMel Gorman if (req) 11429418aa4SMel Gorman kref_get(&req->wb_kref); 11529418aa4SMel Gorman 116277459d2STrond Myklebust return req; 117277459d2STrond Myklebust } 118277459d2STrond Myklebust 119277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 120277459d2STrond Myklebust { 121d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 122277459d2STrond Myklebust struct nfs_page *req = NULL; 123277459d2STrond Myklebust 124587142f8STrond Myklebust spin_lock(&inode->i_lock); 12529418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 126587142f8STrond Myklebust spin_unlock(&inode->i_lock); 127277459d2STrond Myklebust return req; 128277459d2STrond Myklebust } 129277459d2STrond Myklebust 1301da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1311da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1321da177e4SLinus Torvalds { 133d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 134a3d01454STrond Myklebust loff_t end, i_size; 135a3d01454STrond Myklebust pgoff_t end_index; 1361da177e4SLinus Torvalds 137a3d01454STrond Myklebust spin_lock(&inode->i_lock); 138a3d01454STrond Myklebust i_size = i_size_read(inode); 139a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 140d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 141a3d01454STrond Myklebust goto out; 142d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1431da177e4SLinus Torvalds if (i_size >= end) 144a3d01454STrond Myklebust goto out; 1451da177e4SLinus Torvalds i_size_write(inode, end); 146a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 147a3d01454STrond Myklebust out: 148a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 152a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 153a301b777STrond Myklebust { 154d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 155a301b777STrond Myklebust } 156a301b777STrond Myklebust 157d72ddcbaSWeston Andros Adamson /* 158d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 159d72ddcbaSWeston Andros Adamson * @head - head request of page group 160d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 161d72ddcbaSWeston Andros Adamson * 162d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 163d72ddcbaSWeston Andros Adamson * page offset @page_offset. 164d72ddcbaSWeston Andros Adamson * 165d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 166d72ddcbaSWeston Andros Adamson * match is found. 167d72ddcbaSWeston Andros Adamson * 168d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 169d72ddcbaSWeston Andros Adamson */ 170d72ddcbaSWeston Andros Adamson static struct nfs_page * 171d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 172d72ddcbaSWeston Andros Adamson { 173d72ddcbaSWeston Andros Adamson struct nfs_page *req; 174d72ddcbaSWeston Andros Adamson 175d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 176d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 177d72ddcbaSWeston Andros Adamson 178d72ddcbaSWeston Andros Adamson req = head; 179d72ddcbaSWeston Andros Adamson do { 180d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 181d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 182d72ddcbaSWeston Andros Adamson return req; 183d72ddcbaSWeston Andros Adamson 184d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 185d72ddcbaSWeston Andros Adamson } while (req != head); 186d72ddcbaSWeston Andros Adamson 187d72ddcbaSWeston Andros Adamson return NULL; 188d72ddcbaSWeston Andros Adamson } 189d72ddcbaSWeston Andros Adamson 190d72ddcbaSWeston Andros Adamson /* 191d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 192d72ddcbaSWeston Andros Adamson * @head - head request of page group 193d72ddcbaSWeston Andros Adamson * 194d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 195d72ddcbaSWeston Andros Adamson * returns false otherwise 196d72ddcbaSWeston Andros Adamson */ 197d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 198d72ddcbaSWeston Andros Adamson { 199d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 200d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 201d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 202d72ddcbaSWeston Andros Adamson 203d72ddcbaSWeston Andros Adamson nfs_page_group_lock(req); 204d72ddcbaSWeston Andros Adamson 205d72ddcbaSWeston Andros Adamson do { 206d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 207d72ddcbaSWeston Andros Adamson if (tmp) { 208d72ddcbaSWeston Andros Adamson /* no way this should happen */ 209d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 210d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 211d72ddcbaSWeston Andros Adamson } 212d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 213d72ddcbaSWeston Andros Adamson 214d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 215d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 216d72ddcbaSWeston Andros Adamson return pos == len; 217d72ddcbaSWeston Andros Adamson } 218d72ddcbaSWeston Andros Adamson 2191da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2201da177e4SLinus Torvalds * covers the full page. 2211da177e4SLinus Torvalds */ 222d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2231da177e4SLinus Torvalds { 224d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2251da177e4SLinus Torvalds return; 226d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2271da177e4SLinus Torvalds return; 228d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2291da177e4SLinus Torvalds } 2301da177e4SLinus Torvalds 2311da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2321da177e4SLinus Torvalds { 2331da177e4SLinus Torvalds if (wbc->for_reclaim) 234c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 235b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 236b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 237b31268acSTrond Myklebust return FLUSH_COND_STABLE; 2381da177e4SLinus Torvalds } 2391da177e4SLinus Torvalds 2401da177e4SLinus Torvalds /* 24189a09141SPeter Zijlstra * NFS congestion control 24289a09141SPeter Zijlstra */ 24389a09141SPeter Zijlstra 24489a09141SPeter Zijlstra int nfs_congestion_kb; 24589a09141SPeter Zijlstra 24689a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 24789a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 24889a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 24989a09141SPeter Zijlstra 250deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 25189a09141SPeter Zijlstra { 252deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2535a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2545a6d41b3STrond Myklebust 255deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 25689a09141SPeter Zijlstra 257277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2588aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2598aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2608aa7e847SJens Axboe BLK_RW_ASYNC); 2618aa7e847SJens Axboe } 26289a09141SPeter Zijlstra } 26389a09141SPeter Zijlstra 26420633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 26589a09141SPeter Zijlstra { 26620633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 26789a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 26889a09141SPeter Zijlstra 26920633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 27020633f04SWeston Andros Adamson return; 27120633f04SWeston Andros Adamson 27220633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 273c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2748aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 27589a09141SPeter Zijlstra } 27689a09141SPeter Zijlstra 277cfb506e1STrond Myklebust static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) 278e261f51fSTrond Myklebust { 279d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 280e261f51fSTrond Myklebust struct nfs_page *req; 281e261f51fSTrond Myklebust int ret; 282e261f51fSTrond Myklebust 283587142f8STrond Myklebust spin_lock(&inode->i_lock); 284e261f51fSTrond Myklebust for (;;) { 28529418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 286074cc1deSTrond Myklebust if (req == NULL) 287074cc1deSTrond Myklebust break; 2887ad84aa9STrond Myklebust if (nfs_lock_request(req)) 289e261f51fSTrond Myklebust break; 290e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 2917ad84aa9STrond Myklebust * then the call to nfs_lock_request() will always 292e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 293e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 294e261f51fSTrond Myklebust */ 295587142f8STrond Myklebust spin_unlock(&inode->i_lock); 296cfb506e1STrond Myklebust if (!nonblock) 297e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 298cfb506e1STrond Myklebust else 299cfb506e1STrond Myklebust ret = -EAGAIN; 300e261f51fSTrond Myklebust nfs_release_request(req); 301e261f51fSTrond Myklebust if (ret != 0) 302074cc1deSTrond Myklebust return ERR_PTR(ret); 303587142f8STrond Myklebust spin_lock(&inode->i_lock); 304e261f51fSTrond Myklebust } 305587142f8STrond Myklebust spin_unlock(&inode->i_lock); 306074cc1deSTrond Myklebust return req; 307612c9384STrond Myklebust } 308074cc1deSTrond Myklebust 309074cc1deSTrond Myklebust /* 310074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 311074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 312074cc1deSTrond Myklebust */ 313074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 314cfb506e1STrond Myklebust struct page *page, bool nonblock) 315074cc1deSTrond Myklebust { 316074cc1deSTrond Myklebust struct nfs_page *req; 317074cc1deSTrond Myklebust int ret = 0; 318074cc1deSTrond Myklebust 319cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, nonblock); 320074cc1deSTrond Myklebust if (!req) 321074cc1deSTrond Myklebust goto out; 322074cc1deSTrond Myklebust ret = PTR_ERR(req); 323074cc1deSTrond Myklebust if (IS_ERR(req)) 324074cc1deSTrond Myklebust goto out; 325074cc1deSTrond Myklebust 326deed85e7STrond Myklebust nfs_set_page_writeback(page); 327deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 328074cc1deSTrond Myklebust 329deed85e7STrond Myklebust ret = 0; 330f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 331f8512ad0SFred Isaman nfs_redirty_request(req); 332074cc1deSTrond Myklebust ret = pgio->pg_error; 333f8512ad0SFred Isaman } 334074cc1deSTrond Myklebust out: 335074cc1deSTrond Myklebust return ret; 336e261f51fSTrond Myklebust } 337e261f51fSTrond Myklebust 338f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 339f758c885STrond Myklebust { 340d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 341cfb506e1STrond Myklebust int ret; 342f758c885STrond Myklebust 343f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 344f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 345f758c885STrond Myklebust 346d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 3471b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 348cfb506e1STrond Myklebust if (ret == -EAGAIN) { 349cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 350cfb506e1STrond Myklebust ret = 0; 351cfb506e1STrond Myklebust } 352cfb506e1STrond Myklebust return ret; 353f758c885STrond Myklebust } 354f758c885STrond Myklebust 355e261f51fSTrond Myklebust /* 3561da177e4SLinus Torvalds * Write an mmapped page to the server. 3571da177e4SLinus Torvalds */ 3584d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3591da177e4SLinus Torvalds { 360f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 361e261f51fSTrond Myklebust int err; 3621da177e4SLinus Torvalds 363a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 364a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 365f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 366f758c885STrond Myklebust nfs_pageio_complete(&pgio); 367f758c885STrond Myklebust if (err < 0) 3684d770ccfSTrond Myklebust return err; 369f758c885STrond Myklebust if (pgio.pg_error < 0) 370f758c885STrond Myklebust return pgio.pg_error; 371f758c885STrond Myklebust return 0; 3724d770ccfSTrond Myklebust } 3734d770ccfSTrond Myklebust 3744d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3754d770ccfSTrond Myklebust { 376f758c885STrond Myklebust int ret; 3774d770ccfSTrond Myklebust 378f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3791da177e4SLinus Torvalds unlock_page(page); 380f758c885STrond Myklebust return ret; 381f758c885STrond Myklebust } 382f758c885STrond Myklebust 383f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 384f758c885STrond Myklebust { 385f758c885STrond Myklebust int ret; 386f758c885STrond Myklebust 387f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 388f758c885STrond Myklebust unlock_page(page); 389f758c885STrond Myklebust return ret; 3901da177e4SLinus Torvalds } 3911da177e4SLinus Torvalds 3921da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3931da177e4SLinus Torvalds { 3941da177e4SLinus Torvalds struct inode *inode = mapping->host; 39572cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 396c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3971da177e4SLinus Torvalds int err; 3981da177e4SLinus Torvalds 39972cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 400*74316201SNeilBrown err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING, 40172cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 40272cb77f4STrond Myklebust if (err) 40372cb77f4STrond Myklebust goto out_err; 40472cb77f4STrond Myklebust 40591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 40691d5b470SChuck Lever 407a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 408a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 409f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 410c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 41172cb77f4STrond Myklebust 41272cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 4134e857c58SPeter Zijlstra smp_mb__after_atomic(); 41472cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 41572cb77f4STrond Myklebust 416f758c885STrond Myklebust if (err < 0) 41772cb77f4STrond Myklebust goto out_err; 41872cb77f4STrond Myklebust err = pgio.pg_error; 41972cb77f4STrond Myklebust if (err < 0) 42072cb77f4STrond Myklebust goto out_err; 421c63c7b05STrond Myklebust return 0; 42272cb77f4STrond Myklebust out_err: 42372cb77f4STrond Myklebust return err; 4241da177e4SLinus Torvalds } 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds /* 4271da177e4SLinus Torvalds * Insert a write request into an inode 4281da177e4SLinus Torvalds */ 429d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 4301da177e4SLinus Torvalds { 4311da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 432e7d39069STrond Myklebust 4332bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 4342bfc6e56SWeston Andros Adamson 435e7d39069STrond Myklebust /* Lock the request! */ 4367ad84aa9STrond Myklebust nfs_lock_request(req); 437e7d39069STrond Myklebust 438e7d39069STrond Myklebust spin_lock(&inode->i_lock); 439011e2a7fSBryan Schumaker if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 440a9a4a87aSTrond Myklebust inode->i_version++; 44129418aa4SMel Gorman /* 44229418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 44329418aa4SMel Gorman * with invalidate/truncate. 44429418aa4SMel Gorman */ 44529418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 4462df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 447deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 448277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 44929418aa4SMel Gorman } 4501da177e4SLinus Torvalds nfsi->npages++; 4512bfc6e56SWeston Andros Adamson set_bit(PG_INODE_REF, &req->wb_flags); 452c03b4024STrond Myklebust kref_get(&req->wb_kref); 453e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 4541da177e4SLinus Torvalds } 4551da177e4SLinus Torvalds 4561da177e4SLinus Torvalds /* 45789a09141SPeter Zijlstra * Remove a write request from an inode 4581da177e4SLinus Torvalds */ 4591da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4601da177e4SLinus Torvalds { 4613d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 4621da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 46320633f04SWeston Andros Adamson struct nfs_page *head; 46420633f04SWeston Andros Adamson 46520633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 46620633f04SWeston Andros Adamson head = req->wb_head; 4671da177e4SLinus Torvalds 468587142f8STrond Myklebust spin_lock(&inode->i_lock); 46920633f04SWeston Andros Adamson if (likely(!PageSwapCache(head->wb_page))) { 47020633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 47120633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 47220633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 47329418aa4SMel Gorman } 4741da177e4SLinus Torvalds nfsi->npages--; 475587142f8STrond Myklebust spin_unlock(&inode->i_lock); 47620633f04SWeston Andros Adamson } 4771da177e4SLinus Torvalds nfs_release_request(req); 4781da177e4SLinus Torvalds } 4791da177e4SLinus Torvalds 48061822ab5STrond Myklebust static void 4816d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 48261822ab5STrond Myklebust { 48361822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 48461822ab5STrond Myklebust } 48561822ab5STrond Myklebust 48689d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 4878dd37758STrond Myklebust /** 4888dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 4898dd37758STrond Myklebust * @req: pointer to a struct nfs_page 490ea2cf228SFred Isaman * @dst: commit list head 491ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 4928dd37758STrond Myklebust * 493ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 4948dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 4958dd37758STrond Myklebust * the MM page stats. 4968dd37758STrond Myklebust * 497ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 4988dd37758STrond Myklebust * holding the nfs_page lock. 4998dd37758STrond Myklebust */ 5008dd37758STrond Myklebust void 501ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 502ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5038dd37758STrond Myklebust { 5048dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 505ea2cf228SFred Isaman spin_lock(cinfo->lock); 506ea2cf228SFred Isaman nfs_list_add_request(req, dst); 507ea2cf228SFred Isaman cinfo->mds->ncommit++; 508ea2cf228SFred Isaman spin_unlock(cinfo->lock); 50956f9cd68SFred Isaman if (!cinfo->dreq) { 5108dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 511d56b4ddfSMel Gorman inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 51256f9cd68SFred Isaman BDI_RECLAIMABLE); 51356f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 51456f9cd68SFred Isaman I_DIRTY_DATASYNC); 51556f9cd68SFred Isaman } 5168dd37758STrond Myklebust } 5178dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 5188dd37758STrond Myklebust 5198dd37758STrond Myklebust /** 5208dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 5218dd37758STrond Myklebust * @req: pointer to a nfs_page 522ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 5238dd37758STrond Myklebust * 524ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 5258dd37758STrond Myklebust * number of outstanding requests requiring a commit 5268dd37758STrond Myklebust * It does not update the MM page stats. 5278dd37758STrond Myklebust * 528ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 5298dd37758STrond Myklebust */ 5308dd37758STrond Myklebust void 531ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 532ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5338dd37758STrond Myklebust { 5348dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 5358dd37758STrond Myklebust return; 5368dd37758STrond Myklebust nfs_list_remove_request(req); 537ea2cf228SFred Isaman cinfo->mds->ncommit--; 5388dd37758STrond Myklebust } 5398dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 5408dd37758STrond Myklebust 541ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 542ea2cf228SFred Isaman struct inode *inode) 543ea2cf228SFred Isaman { 544ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 545ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 546ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 547b359f9d0SFred Isaman cinfo->dreq = NULL; 548f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 549ea2cf228SFred Isaman } 550ea2cf228SFred Isaman 551ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 552ea2cf228SFred Isaman struct inode *inode, 553ea2cf228SFred Isaman struct nfs_direct_req *dreq) 554ea2cf228SFred Isaman { 5551763da12SFred Isaman if (dreq) 5561763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 5571763da12SFred Isaman else 558ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 559ea2cf228SFred Isaman } 560ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 5618dd37758STrond Myklebust 5621da177e4SLinus Torvalds /* 5631da177e4SLinus Torvalds * Add a request to the inode's commit list. 5641da177e4SLinus Torvalds */ 5651763da12SFred Isaman void 566ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 567ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5681da177e4SLinus Torvalds { 569ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 5708dd37758STrond Myklebust return; 571ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 5721da177e4SLinus Torvalds } 5738e821cadSTrond Myklebust 574d6d6dc7cSFred Isaman static void 575d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 576e468bae9STrond Myklebust { 577e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 578d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); 579e468bae9STrond Myklebust } 580d6d6dc7cSFred Isaman 5818dd37758STrond Myklebust static void 582d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 583d6d6dc7cSFred Isaman { 5848dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 5858dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 586ea2cf228SFred Isaman struct nfs_commit_info cinfo; 587d6d6dc7cSFred Isaman 588ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 589ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 590ea2cf228SFred Isaman spin_lock(cinfo.lock); 591ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 592ea2cf228SFred Isaman spin_unlock(cinfo.lock); 593d6d6dc7cSFred Isaman } 5948dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 5958dd37758STrond Myklebust } 596e468bae9STrond Myklebust } 597e468bae9STrond Myklebust 5988e821cadSTrond Myklebust static inline 5999c7e1b3dSAnna Schumaker int nfs_write_need_commit(struct nfs_pgio_data *data) 6008e821cadSTrond Myklebust { 601465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 602cd841605SFred Isaman return data->header->lseg == NULL; 6038e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 6048e821cadSTrond Myklebust } 6058e821cadSTrond Myklebust 6068e821cadSTrond Myklebust #else 60768cd6fa4SBryan Schumaker static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 60868cd6fa4SBryan Schumaker struct inode *inode) 60968cd6fa4SBryan Schumaker { 61068cd6fa4SBryan Schumaker } 61168cd6fa4SBryan Schumaker 61268cd6fa4SBryan Schumaker void nfs_init_cinfo(struct nfs_commit_info *cinfo, 61368cd6fa4SBryan Schumaker struct inode *inode, 61468cd6fa4SBryan Schumaker struct nfs_direct_req *dreq) 61568cd6fa4SBryan Schumaker { 61668cd6fa4SBryan Schumaker } 61768cd6fa4SBryan Schumaker 6181763da12SFred Isaman void 619ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 620ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 6218e821cadSTrond Myklebust { 6228e821cadSTrond Myklebust } 6238e821cadSTrond Myklebust 6248dd37758STrond Myklebust static void 625e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 626e468bae9STrond Myklebust { 627e468bae9STrond Myklebust } 628e468bae9STrond Myklebust 6298e821cadSTrond Myklebust static inline 6309c7e1b3dSAnna Schumaker int nfs_write_need_commit(struct nfs_pgio_data *data) 6318e821cadSTrond Myklebust { 6328e821cadSTrond Myklebust return 0; 6338e821cadSTrond Myklebust } 6348e821cadSTrond Myklebust 6351da177e4SLinus Torvalds #endif 6361da177e4SLinus Torvalds 637061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 6386c75dc0dSFred Isaman { 639ea2cf228SFred Isaman struct nfs_commit_info cinfo; 6406c75dc0dSFred Isaman unsigned long bytes = 0; 6412bfc6e56SWeston Andros Adamson bool do_destroy; 6426c75dc0dSFred Isaman 6436c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 6446c75dc0dSFred Isaman goto out; 645ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 6466c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 6476c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 6486c75dc0dSFred Isaman 6496c75dc0dSFred Isaman bytes += req->wb_bytes; 6506c75dc0dSFred Isaman nfs_list_remove_request(req); 6516c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 6526c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 653d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 6546c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 6556c75dc0dSFred Isaman goto remove_req; 6566c75dc0dSFred Isaman } 6576c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { 6586c75dc0dSFred Isaman nfs_mark_request_dirty(req); 6596c75dc0dSFred Isaman goto next; 6606c75dc0dSFred Isaman } 6616c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { 662f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 663ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 6646c75dc0dSFred Isaman goto next; 6656c75dc0dSFred Isaman } 6666c75dc0dSFred Isaman remove_req: 6676c75dc0dSFred Isaman nfs_inode_remove_request(req); 6686c75dc0dSFred Isaman next: 6691d1afcbcSTrond Myklebust nfs_unlock_request(req); 67020633f04SWeston Andros Adamson nfs_end_page_writeback(req); 6712bfc6e56SWeston Andros Adamson do_destroy = !test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags); 6723aff4ebbSTrond Myklebust nfs_release_request(req); 6736c75dc0dSFred Isaman } 6746c75dc0dSFred Isaman out: 6756c75dc0dSFred Isaman hdr->release(hdr); 6766c75dc0dSFred Isaman } 6776c75dc0dSFred Isaman 67889d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 679ce59515cSAnna Schumaker unsigned long 680ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 681fb8a1f11STrond Myklebust { 682ea2cf228SFred Isaman return cinfo->mds->ncommit; 683fb8a1f11STrond Myklebust } 684fb8a1f11STrond Myklebust 685ea2cf228SFred Isaman /* cinfo->lock held by caller */ 6861763da12SFred Isaman int 687ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 688ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 689d6d6dc7cSFred Isaman { 690d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 691d6d6dc7cSFred Isaman int ret = 0; 692d6d6dc7cSFred Isaman 693d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 6948dd37758STrond Myklebust if (!nfs_lock_request(req)) 6958dd37758STrond Myklebust continue; 6967ad84aa9STrond Myklebust kref_get(&req->wb_kref); 697ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 6983b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 699ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 7008dd37758STrond Myklebust nfs_list_add_request(req, dst); 701d6d6dc7cSFred Isaman ret++; 7021763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 703d6d6dc7cSFred Isaman break; 704d6d6dc7cSFred Isaman } 705d6d6dc7cSFred Isaman return ret; 706d6d6dc7cSFred Isaman } 707d6d6dc7cSFred Isaman 7081da177e4SLinus Torvalds /* 7091da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 7101da177e4SLinus Torvalds * @inode: NFS inode to scan 711ea2cf228SFred Isaman * @dst: mds destination list 712ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 7131da177e4SLinus Torvalds * 7141da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 7151da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 7161da177e4SLinus Torvalds */ 7171763da12SFred Isaman int 718ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 719ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7201da177e4SLinus Torvalds { 721d6d6dc7cSFred Isaman int ret = 0; 722fb8a1f11STrond Myklebust 723ea2cf228SFred Isaman spin_lock(cinfo->lock); 724ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 7258dd37758STrond Myklebust const int max = INT_MAX; 726d6d6dc7cSFred Isaman 727ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 728ea2cf228SFred Isaman cinfo, max); 729ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 730d6d6dc7cSFred Isaman } 731ea2cf228SFred Isaman spin_unlock(cinfo->lock); 732ff778d02STrond Myklebust return ret; 7331da177e4SLinus Torvalds } 734d6d6dc7cSFred Isaman 735c42de9ddSTrond Myklebust #else 736ce59515cSAnna Schumaker unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 737fb8a1f11STrond Myklebust { 738fb8a1f11STrond Myklebust return 0; 739fb8a1f11STrond Myklebust } 740fb8a1f11STrond Myklebust 7411763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 742ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 743c42de9ddSTrond Myklebust { 744c42de9ddSTrond Myklebust return 0; 745c42de9ddSTrond Myklebust } 7461da177e4SLinus Torvalds #endif 7471da177e4SLinus Torvalds 7481da177e4SLinus Torvalds /* 749e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 750e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 7511da177e4SLinus Torvalds * 752e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 753e7d39069STrond Myklebust * to disk. 7541da177e4SLinus Torvalds */ 755e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 756e7d39069STrond Myklebust struct page *page, 757e7d39069STrond Myklebust unsigned int offset, 758e7d39069STrond Myklebust unsigned int bytes) 7591da177e4SLinus Torvalds { 760e7d39069STrond Myklebust struct nfs_page *req; 761e7d39069STrond Myklebust unsigned int rqend; 762e7d39069STrond Myklebust unsigned int end; 7631da177e4SLinus Torvalds int error; 764277459d2STrond Myklebust 765e7d39069STrond Myklebust if (!PagePrivate(page)) 766e7d39069STrond Myklebust return NULL; 767e7d39069STrond Myklebust 768e7d39069STrond Myklebust end = offset + bytes; 769e7d39069STrond Myklebust spin_lock(&inode->i_lock); 770e7d39069STrond Myklebust 771e7d39069STrond Myklebust for (;;) { 77229418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 773e7d39069STrond Myklebust if (req == NULL) 774e7d39069STrond Myklebust goto out_unlock; 775e7d39069STrond Myklebust 7762bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 7772bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 7782bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7792bfc6e56SWeston Andros Adamson 780e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 781e7d39069STrond Myklebust /* 782e7d39069STrond Myklebust * Tell the caller to flush out the request if 783e7d39069STrond Myklebust * the offsets are non-contiguous. 784e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 785e7d39069STrond Myklebust * have flushed out requests having wrong owners. 786e7d39069STrond Myklebust */ 787e468bae9STrond Myklebust if (offset > rqend 788e7d39069STrond Myklebust || end < req->wb_offset) 789e7d39069STrond Myklebust goto out_flushme; 790e7d39069STrond Myklebust 7917ad84aa9STrond Myklebust if (nfs_lock_request(req)) 792e7d39069STrond Myklebust break; 793e7d39069STrond Myklebust 794e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 795587142f8STrond Myklebust spin_unlock(&inode->i_lock); 7961da177e4SLinus Torvalds error = nfs_wait_on_request(req); 7971da177e4SLinus Torvalds nfs_release_request(req); 798e7d39069STrond Myklebust if (error != 0) 799e7d39069STrond Myklebust goto out_err; 800e7d39069STrond Myklebust spin_lock(&inode->i_lock); 8011da177e4SLinus Torvalds } 8021da177e4SLinus Torvalds 8031da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 8041da177e4SLinus Torvalds if (offset < req->wb_offset) { 8051da177e4SLinus Torvalds req->wb_offset = offset; 8061da177e4SLinus Torvalds req->wb_pgbase = offset; 8071da177e4SLinus Torvalds } 8081da177e4SLinus Torvalds if (end > rqend) 8091da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 810e7d39069STrond Myklebust else 811e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 812e7d39069STrond Myklebust out_unlock: 813e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 814ca138f36SFred Isaman if (req) 8158dd37758STrond Myklebust nfs_clear_request_commit(req); 816e7d39069STrond Myklebust return req; 817e7d39069STrond Myklebust out_flushme: 818e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 819e7d39069STrond Myklebust nfs_release_request(req); 820e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 821e7d39069STrond Myklebust out_err: 822e7d39069STrond Myklebust return ERR_PTR(error); 823e7d39069STrond Myklebust } 8241da177e4SLinus Torvalds 825e7d39069STrond Myklebust /* 826e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 827e7d39069STrond Myklebust * 828e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 829e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 830e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 831e7d39069STrond Myklebust */ 832e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 833e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 834e7d39069STrond Myklebust { 835d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 836e7d39069STrond Myklebust struct nfs_page *req; 837e7d39069STrond Myklebust 838e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 839e7d39069STrond Myklebust if (req != NULL) 840e7d39069STrond Myklebust goto out; 8412bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 842e7d39069STrond Myklebust if (IS_ERR(req)) 843e7d39069STrond Myklebust goto out; 844d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 845efc91ed0STrond Myklebust out: 84661e930a9STrond Myklebust return req; 8471da177e4SLinus Torvalds } 8481da177e4SLinus Torvalds 849e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 850e7d39069STrond Myklebust unsigned int offset, unsigned int count) 851e7d39069STrond Myklebust { 852e7d39069STrond Myklebust struct nfs_page *req; 853e7d39069STrond Myklebust 854e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 855e7d39069STrond Myklebust if (IS_ERR(req)) 856e7d39069STrond Myklebust return PTR_ERR(req); 857e7d39069STrond Myklebust /* Update file length */ 858e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 859d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 860a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 8611d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 862e7d39069STrond Myklebust return 0; 863e7d39069STrond Myklebust } 864e7d39069STrond Myklebust 8651da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 8661da177e4SLinus Torvalds { 867cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 8682a369153STrond Myklebust struct nfs_lock_context *l_ctx; 8691da177e4SLinus Torvalds struct nfs_page *req; 8701a54533eSTrond Myklebust int do_flush, status; 8711da177e4SLinus Torvalds /* 8721da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 8731da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 8741da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 8751da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 8761da177e4SLinus Torvalds * Also do the same if we find a request from an existing 8771da177e4SLinus Torvalds * dropped page. 8781da177e4SLinus Torvalds */ 8791a54533eSTrond Myklebust do { 880277459d2STrond Myklebust req = nfs_page_find_request(page); 8811a54533eSTrond Myklebust if (req == NULL) 8821a54533eSTrond Myklebust return 0; 8832a369153STrond Myklebust l_ctx = req->wb_lock_context; 8842a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 8852bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 8862bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 8870f1d2605STrond Myklebust if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { 8882a369153STrond Myklebust do_flush |= l_ctx->lockowner.l_owner != current->files 8892a369153STrond Myklebust || l_ctx->lockowner.l_pid != current->tgid; 8902a369153STrond Myklebust } 8911da177e4SLinus Torvalds nfs_release_request(req); 8921a54533eSTrond Myklebust if (!do_flush) 8931a54533eSTrond Myklebust return 0; 894d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 8951a54533eSTrond Myklebust } while (status == 0); 8961a54533eSTrond Myklebust return status; 8971da177e4SLinus Torvalds } 8981da177e4SLinus Torvalds 8991da177e4SLinus Torvalds /* 900dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 901dc24826bSAndy Adamson * expire soon. 902dc24826bSAndy Adamson * 903dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 904dc24826bSAndy Adamson * 905dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 906dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 907dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 908dc24826bSAndy Adamson */ 909dc24826bSAndy Adamson int 910dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 911dc24826bSAndy Adamson { 912dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 913dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 914dc24826bSAndy Adamson 915dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 916dc24826bSAndy Adamson } 917dc24826bSAndy Adamson 918dc24826bSAndy Adamson /* 919dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 920dc24826bSAndy Adamson */ 921dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 922dc24826bSAndy Adamson { 923dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 924dc24826bSAndy Adamson } 925dc24826bSAndy Adamson 926dc24826bSAndy Adamson /* 9275d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 9285d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 9295d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 9305d47a356STrond Myklebust */ 9318d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 9325d47a356STrond Myklebust { 933d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 934d529ef83SJeff Layton 9358d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 9368d197a56STrond Myklebust goto out; 93718dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 938d529ef83SJeff Layton return false; 9394db72b40SJeff Layton smp_rmb(); 940d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 9418d197a56STrond Myklebust return false; 9428d197a56STrond Myklebust out: 94318dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 94418dd78c4SScott Mayhew return false; 9458d197a56STrond Myklebust return PageUptodate(page) != 0; 9465d47a356STrond Myklebust } 9475d47a356STrond Myklebust 948c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 949c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 950c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 951c7559663SScott Mayhew * inefficiencies. 952c7559663SScott Mayhew * 953263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 954263b4509SScott Mayhew * of the checks. 955c7559663SScott Mayhew */ 956c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 957c7559663SScott Mayhew { 958c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 959c7559663SScott Mayhew return 0; 960263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 961263b4509SScott Mayhew return 0; 962c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 963c7559663SScott Mayhew return 1; 964263b4509SScott Mayhew if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && 965c7559663SScott Mayhew inode->i_flock->fl_end == OFFSET_MAX && 966263b4509SScott Mayhew inode->i_flock->fl_type != F_RDLCK)) 967c7559663SScott Mayhew return 1; 968c7559663SScott Mayhew return 0; 969c7559663SScott Mayhew } 970c7559663SScott Mayhew 9715d47a356STrond Myklebust /* 9721da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 9731da177e4SLinus Torvalds * 9741da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 9751da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 9761da177e4SLinus Torvalds */ 9771da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 9781da177e4SLinus Torvalds unsigned int offset, unsigned int count) 9791da177e4SLinus Torvalds { 980cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 981d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 9821da177e4SLinus Torvalds int status = 0; 9831da177e4SLinus Torvalds 98491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 98591d5b470SChuck Lever 9866de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 9876de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 9881da177e4SLinus Torvalds 989c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 99049a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 9911da177e4SLinus Torvalds offset = 0; 9921da177e4SLinus Torvalds } 9931da177e4SLinus Torvalds 994e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 99503fa9e84STrond Myklebust if (status < 0) 99603fa9e84STrond Myklebust nfs_set_pageerror(page); 99759b7c05fSTrond Myklebust else 99859b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 9991da177e4SLinus Torvalds 100048186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 10011da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 10021da177e4SLinus Torvalds return status; 10031da177e4SLinus Torvalds } 10041da177e4SLinus Torvalds 10053ff7576dSTrond Myklebust static int flush_task_priority(int how) 10061da177e4SLinus Torvalds { 10071da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 10081da177e4SLinus Torvalds case FLUSH_HIGHPRI: 10091da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 10101da177e4SLinus Torvalds case FLUSH_LOWPRI: 10111da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 10121da177e4SLinus Torvalds } 10131da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds 10161ed26f33SAnna Schumaker static void nfs_initiate_write(struct nfs_pgio_data *data, struct rpc_message *msg, 10171ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 10181da177e4SLinus Torvalds { 1019cd841605SFred Isaman struct inode *inode = data->header->inode; 10203ff7576dSTrond Myklebust int priority = flush_task_priority(how); 10211da177e4SLinus Torvalds 10221ed26f33SAnna Schumaker task_setup_data->priority = priority; 10231ed26f33SAnna Schumaker NFS_PROTO(inode)->write_setup(data, msg); 1024d138d5d1SAndy Adamson 10258c21c62cSWeston Andros Adamson nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 10261ed26f33SAnna Schumaker &task_setup_data->rpc_client, msg, data); 1027275acaafSTrond Myklebust } 1028275acaafSTrond Myklebust 10296d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 10306d884e8fSFred * call this on each, which will prepare them to be retried on next 10316d884e8fSFred * writeback using standard nfs. 10326d884e8fSFred */ 10336d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 10346d884e8fSFred { 10356d884e8fSFred nfs_mark_request_dirty(req); 10361d1afcbcSTrond Myklebust nfs_unlock_request(req); 103720633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10383aff4ebbSTrond Myklebust nfs_release_request(req); 10396d884e8fSFred } 10406d884e8fSFred 1041061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 10426c75dc0dSFred Isaman { 10436c75dc0dSFred Isaman struct nfs_page *req; 10446c75dc0dSFred Isaman 10456c75dc0dSFred Isaman while (!list_empty(head)) { 10466c75dc0dSFred Isaman req = nfs_list_entry(head->next); 10476c75dc0dSFred Isaman nfs_list_remove_request(req); 10486c75dc0dSFred Isaman nfs_redirty_request(req); 10496c75dc0dSFred Isaman } 10506c75dc0dSFred Isaman } 10516c75dc0dSFred Isaman 1052061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1053061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1054061ae2edSFred Isaman .completion = nfs_write_completion, 1055061ae2edSFred Isaman }; 1056061ae2edSFred Isaman 105757208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1058a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1059061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 10601751c363STrond Myklebust { 1061a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 106241d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1063a20c93e3SChristoph Hellwig 1064a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1065a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1066a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1067a20c93e3SChristoph Hellwig #endif 10684a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 10694a0de55cSAnna Schumaker server->wsize, ioflags); 10701751c363STrond Myklebust } 1071ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 10721751c363STrond Myklebust 1073dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1074dce81290STrond Myklebust { 107541d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1076dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1077dce81290STrond Myklebust } 10781f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1079dce81290STrond Myklebust 10801da177e4SLinus Torvalds 10810b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 10820b7c0153SFred Isaman { 10830b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 10840b7c0153SFred Isaman 10850b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 10860b7c0153SFred Isaman } 10870b7c0153SFred Isaman 1088a4cdda59SAnna Schumaker static void nfs_writeback_release_common(struct nfs_pgio_data *data) 10891da177e4SLinus Torvalds { 1090cd841605SFred Isaman struct nfs_pgio_header *hdr = data->header; 1091e2fecb21STrond Myklebust int status = data->task.tk_status; 1092788e7a89STrond Myklebust 10936c75dc0dSFred Isaman if ((status >= 0) && nfs_write_need_commit(data)) { 10946c75dc0dSFred Isaman spin_lock(&hdr->lock); 10956c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) 10966c75dc0dSFred Isaman ; /* Do nothing */ 10976c75dc0dSFred Isaman else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) 1098f79d06f5SAnna Schumaker memcpy(&hdr->verf, &data->verf, sizeof(hdr->verf)); 1099f79d06f5SAnna Schumaker else if (memcmp(&hdr->verf, &data->verf, sizeof(hdr->verf))) 11006c75dc0dSFred Isaman set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); 11016c75dc0dSFred Isaman spin_unlock(&hdr->lock); 11021da177e4SLinus Torvalds } 11031da177e4SLinus Torvalds } 11041da177e4SLinus Torvalds 11051f2edbe3STrond Myklebust /* 11061f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 11071f2edbe3STrond Myklebust */ 11081f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 11091f2edbe3STrond Myklebust { 11101f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 11111f2edbe3STrond Myklebust int kill = 0; 1112788e7a89STrond Myklebust 11131f2edbe3STrond Myklebust /* suid always must be killed */ 11141f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 11151f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 11161f2edbe3STrond Myklebust 11171f2edbe3STrond Myklebust /* 11181f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 11191f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 11201f2edbe3STrond Myklebust */ 11211f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 11221f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 11231f2edbe3STrond Myklebust 11241f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 11251f2edbe3STrond Myklebust return kill; 11261f2edbe3STrond Myklebust 11271f2edbe3STrond Myklebust return 0; 11281f2edbe3STrond Myklebust } 1129788e7a89STrond Myklebust 11301da177e4SLinus Torvalds /* 11311da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 11321da177e4SLinus Torvalds */ 11330eecb214SAnna Schumaker static int nfs_writeback_done(struct rpc_task *task, struct nfs_pgio_data *data, 11340eecb214SAnna Schumaker struct inode *inode) 11351da177e4SLinus Torvalds { 1136788e7a89STrond Myklebust int status; 11371da177e4SLinus Torvalds 1138f551e44fSChuck Lever /* 1139f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1140f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1141f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1142f551e44fSChuck Lever * another writer had changed the file, but some applications 1143f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1144f551e44fSChuck Lever */ 1145cd841605SFred Isaman status = NFS_PROTO(inode)->write_done(task, data); 1146788e7a89STrond Myklebust if (status != 0) 11470eecb214SAnna Schumaker return status; 11480eecb214SAnna Schumaker nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, data->res.count); 114991d5b470SChuck Lever 115089d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 11510eecb214SAnna Schumaker if (data->res.verf->committed < data->args.stable && task->tk_status >= 0) { 11521da177e4SLinus Torvalds /* We tried a write call, but the server did not 11531da177e4SLinus Torvalds * commit data to stable storage even though we 11541da177e4SLinus Torvalds * requested it. 11551da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11561da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11571da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11581da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11591da177e4SLinus Torvalds */ 11601da177e4SLinus Torvalds static unsigned long complain; 11611da177e4SLinus Torvalds 1162a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 11631da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11641da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11651da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1166cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 11670eecb214SAnna Schumaker data->res.verf->committed, data->args.stable); 11681da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11691da177e4SLinus Torvalds } 11701da177e4SLinus Torvalds } 11711da177e4SLinus Torvalds #endif 11721f2edbe3STrond Myklebust 11731f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 11741f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 11751f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 11760eecb214SAnna Schumaker return 0; 11770eecb214SAnna Schumaker } 11780eecb214SAnna Schumaker 11790eecb214SAnna Schumaker /* 11800eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 11810eecb214SAnna Schumaker */ 11820eecb214SAnna Schumaker static void nfs_writeback_result(struct rpc_task *task, struct nfs_pgio_data *data) 11830eecb214SAnna Schumaker { 11840eecb214SAnna Schumaker struct nfs_pgio_args *argp = &data->args; 11850eecb214SAnna Schumaker struct nfs_pgio_res *resp = &data->res; 11861f2edbe3STrond Myklebust 11871f2edbe3STrond Myklebust if (resp->count < argp->count) { 11881da177e4SLinus Torvalds static unsigned long complain; 11891da177e4SLinus Torvalds 11906c75dc0dSFred Isaman /* This a short write! */ 11910eecb214SAnna Schumaker nfs_inc_stats(data->header->inode, NFSIOS_SHORTWRITE); 119291d5b470SChuck Lever 11931da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11946c75dc0dSFred Isaman if (resp->count == 0) { 11956c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 11966c75dc0dSFred Isaman printk(KERN_WARNING 11976c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 11986c75dc0dSFred Isaman argp->count); 11996c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 12006c75dc0dSFred Isaman } 12016c75dc0dSFred Isaman nfs_set_pgio_error(data->header, -EIO, argp->offset); 12026c75dc0dSFred Isaman task->tk_status = -EIO; 12036c75dc0dSFred Isaman return; 12046c75dc0dSFred Isaman } 12051da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12061da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12071da177e4SLinus Torvalds /* Resend from where the server left off */ 1208a69aef14SFred Isaman data->mds_offset += resp->count; 12091da177e4SLinus Torvalds argp->offset += resp->count; 12101da177e4SLinus Torvalds argp->pgbase += resp->count; 12111da177e4SLinus Torvalds argp->count -= resp->count; 12121da177e4SLinus Torvalds } else { 12131da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12141da177e4SLinus Torvalds * headaches in the case of a server crash. 12151da177e4SLinus Torvalds */ 12161da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 12171da177e4SLinus Torvalds } 1218d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 12191da177e4SLinus Torvalds } 12201da177e4SLinus Torvalds } 12211da177e4SLinus Torvalds 12221da177e4SLinus Torvalds 122389d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 122471d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 122571d0a611STrond Myklebust { 1226b8413f98STrond Myklebust int ret; 1227b8413f98STrond Myklebust 122871d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 122971d0a611STrond Myklebust return 1; 1230b8413f98STrond Myklebust if (!may_wait) 123171d0a611STrond Myklebust return 0; 1232b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1233b8413f98STrond Myklebust NFS_INO_COMMIT, 1234b8413f98STrond Myklebust nfs_wait_bit_killable, 1235b8413f98STrond Myklebust TASK_KILLABLE); 1236b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 123771d0a611STrond Myklebust } 123871d0a611STrond Myklebust 1239f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 124071d0a611STrond Myklebust { 124171d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 12424e857c58SPeter Zijlstra smp_mb__after_atomic(); 124371d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 124471d0a611STrond Myklebust } 124571d0a611STrond Myklebust 12460b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 12471da177e4SLinus Torvalds { 12480b7c0153SFred Isaman put_nfs_open_context(data->context); 12490b7c0153SFred Isaman nfs_commit_free(data); 12501da177e4SLinus Torvalds } 1251e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 12521da177e4SLinus Torvalds 12530b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 12549ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 12559f0ec176SAndy Adamson int how, int flags) 12561da177e4SLinus Torvalds { 125707737691STrond Myklebust struct rpc_task *task; 12589ace33cdSFred Isaman int priority = flush_task_priority(how); 1259bdc7f021STrond Myklebust struct rpc_message msg = { 1260bdc7f021STrond Myklebust .rpc_argp = &data->args, 1261bdc7f021STrond Myklebust .rpc_resp = &data->res, 12629ace33cdSFred Isaman .rpc_cred = data->cred, 1263bdc7f021STrond Myklebust }; 126484115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 126507737691STrond Myklebust .task = &data->task, 12669ace33cdSFred Isaman .rpc_client = clnt, 1267bdc7f021STrond Myklebust .rpc_message = &msg, 12689ace33cdSFred Isaman .callback_ops = call_ops, 126984115e1cSTrond Myklebust .callback_data = data, 1270101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 12719f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 12723ff7576dSTrond Myklebust .priority = priority, 127384115e1cSTrond Myklebust }; 1274788e7a89STrond Myklebust /* Set up the initial task struct. */ 12759ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 12761da177e4SLinus Torvalds 1277a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1278bdc7f021STrond Myklebust 12798c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 12808c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 12818c21c62cSWeston Andros Adamson 128207737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1283dbae4c73STrond Myklebust if (IS_ERR(task)) 1284dbae4c73STrond Myklebust return PTR_ERR(task); 1285d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1286d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 128707737691STrond Myklebust rpc_put_task(task); 1288dbae4c73STrond Myklebust return 0; 12891da177e4SLinus Torvalds } 1290e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 12911da177e4SLinus Torvalds 12921da177e4SLinus Torvalds /* 12939ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 12949ace33cdSFred Isaman */ 12950b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1296988b6dceSFred Isaman struct list_head *head, 1297f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1298f453a54aSFred Isaman struct nfs_commit_info *cinfo) 12999ace33cdSFred Isaman { 13009ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 13013d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 13029ace33cdSFred Isaman 13039ace33cdSFred Isaman /* Set up the RPC argument and reply structs 13049ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 13059ace33cdSFred Isaman 13069ace33cdSFred Isaman list_splice_init(head, &data->pages); 13079ace33cdSFred Isaman 13089ace33cdSFred Isaman data->inode = inode; 13099ace33cdSFred Isaman data->cred = first->wb_context->cred; 1310988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 13119ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1312f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1313b359f9d0SFred Isaman data->dreq = cinfo->dreq; 13149ace33cdSFred Isaman 13159ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 13169ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 13179ace33cdSFred Isaman data->args.offset = 0; 13189ace33cdSFred Isaman data->args.count = 0; 13190b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 13209ace33cdSFred Isaman data->res.fattr = &data->fattr; 13219ace33cdSFred Isaman data->res.verf = &data->verf; 13229ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 13239ace33cdSFred Isaman } 1324e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 13259ace33cdSFred Isaman 1326e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1327ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1328ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 132964bfeb49SFred Isaman { 133064bfeb49SFred Isaman struct nfs_page *req; 133164bfeb49SFred Isaman 133264bfeb49SFred Isaman while (!list_empty(page_list)) { 133364bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 133464bfeb49SFred Isaman nfs_list_remove_request(req); 1335ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 133656f9cd68SFred Isaman if (!cinfo->dreq) { 133764bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1338d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 133964bfeb49SFred Isaman BDI_RECLAIMABLE); 134056f9cd68SFred Isaman } 13411d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 134264bfeb49SFred Isaman } 134364bfeb49SFred Isaman } 1344e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 134564bfeb49SFred Isaman 13469ace33cdSFred Isaman /* 13471da177e4SLinus Torvalds * Commit dirty pages 13481da177e4SLinus Torvalds */ 13491da177e4SLinus Torvalds static int 1350ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1351ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 13521da177e4SLinus Torvalds { 13530b7c0153SFred Isaman struct nfs_commit_data *data; 13541da177e4SLinus Torvalds 1355c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 13561da177e4SLinus Torvalds 13571da177e4SLinus Torvalds if (!data) 13581da177e4SLinus Torvalds goto out_bad; 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds /* Set up the argument struct */ 1361f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1362f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 13639f0ec176SAndy Adamson return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 13649f0ec176SAndy Adamson how, 0); 13651da177e4SLinus Torvalds out_bad: 1366ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1367f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 13681da177e4SLinus Torvalds return -ENOMEM; 13691da177e4SLinus Torvalds } 13701da177e4SLinus Torvalds 13711da177e4SLinus Torvalds /* 13721da177e4SLinus Torvalds * COMMIT call returned 13731da177e4SLinus Torvalds */ 1374788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 13751da177e4SLinus Torvalds { 13760b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13771da177e4SLinus Torvalds 1378a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 13791da177e4SLinus Torvalds task->tk_pid, task->tk_status); 13801da177e4SLinus Torvalds 1381788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1382c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1383c9d8f89dSTrond Myklebust } 1384c9d8f89dSTrond Myklebust 1385f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1386c9d8f89dSTrond Myklebust { 1387c9d8f89dSTrond Myklebust struct nfs_page *req; 1388c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1389f453a54aSFred Isaman struct nfs_commit_info cinfo; 1390788e7a89STrond Myklebust 13911da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 13921da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 13931da177e4SLinus Torvalds nfs_list_remove_request(req); 1394d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 13951da177e4SLinus Torvalds 13961e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 13973d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 13981e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), 13991da177e4SLinus Torvalds req->wb_bytes, 14001da177e4SLinus Torvalds (long long)req_offset(req)); 1401c9d8f89dSTrond Myklebust if (status < 0) { 1402c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 14031da177e4SLinus Torvalds nfs_inode_remove_request(req); 1404c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 14051da177e4SLinus Torvalds goto next; 14061da177e4SLinus Torvalds } 14071da177e4SLinus Torvalds 14081da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 14091da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 14102f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 14111da177e4SLinus Torvalds /* We have a match */ 14121da177e4SLinus Torvalds nfs_inode_remove_request(req); 14131da177e4SLinus Torvalds dprintk(" OK\n"); 14141da177e4SLinus Torvalds goto next; 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 14171da177e4SLinus Torvalds dprintk(" mismatch\n"); 14186d884e8fSFred nfs_mark_request_dirty(req); 141905990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 14201da177e4SLinus Torvalds next: 14211d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 14221da177e4SLinus Torvalds } 1423f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1424f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1425f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 14265917ce84SFred Isaman } 14275917ce84SFred Isaman 14285917ce84SFred Isaman static void nfs_commit_release(void *calldata) 14295917ce84SFred Isaman { 14300b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14315917ce84SFred Isaman 1432f453a54aSFred Isaman data->completion_ops->completion(data); 1433c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 14341da177e4SLinus Torvalds } 1435788e7a89STrond Myklebust 1436788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 14370b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1438788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1439788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1440788e7a89STrond Myklebust }; 14411da177e4SLinus Torvalds 1442f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1443f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1444f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1445f453a54aSFred Isaman }; 1446f453a54aSFred Isaman 14471763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1448ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 144984c53ab5SFred Isaman { 145084c53ab5SFred Isaman int status; 145184c53ab5SFred Isaman 1452ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 145384c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1454ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 145584c53ab5SFred Isaman return status; 145684c53ab5SFred Isaman } 145784c53ab5SFred Isaman 1458b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 14591da177e4SLinus Torvalds { 14601da177e4SLinus Torvalds LIST_HEAD(head); 1461ea2cf228SFred Isaman struct nfs_commit_info cinfo; 146271d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1463b8413f98STrond Myklebust int res; 14641da177e4SLinus Torvalds 1465b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1466b8413f98STrond Myklebust if (res <= 0) 1467c5efa5fcSTrond Myklebust goto out_mark_dirty; 1468ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1469ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 14701da177e4SLinus Torvalds if (res) { 1471a861a1e1SFred Isaman int error; 1472a861a1e1SFred Isaman 1473ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 14741da177e4SLinus Torvalds if (error < 0) 14751da177e4SLinus Torvalds return error; 1476b8413f98STrond Myklebust if (!may_wait) 1477b8413f98STrond Myklebust goto out_mark_dirty; 1478*74316201SNeilBrown error = wait_on_bit_action(&NFS_I(inode)->flags, 1479b8413f98STrond Myklebust NFS_INO_COMMIT, 148071d0a611STrond Myklebust nfs_wait_bit_killable, 148171d0a611STrond Myklebust TASK_KILLABLE); 1482b8413f98STrond Myklebust if (error < 0) 1483b8413f98STrond Myklebust return error; 148471d0a611STrond Myklebust } else 148571d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1486c5efa5fcSTrond Myklebust return res; 1487c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1488c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1489c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1490c5efa5fcSTrond Myklebust * that the data is on the disk. 1491c5efa5fcSTrond Myklebust */ 1492c5efa5fcSTrond Myklebust out_mark_dirty: 1493c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 14941da177e4SLinus Torvalds return res; 14951da177e4SLinus Torvalds } 14968fc795f7STrond Myklebust 14978fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 14988fc795f7STrond Myklebust { 1499420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1500420e3646STrond Myklebust int flags = FLUSH_SYNC; 1501420e3646STrond Myklebust int ret = 0; 15028fc795f7STrond Myklebust 15033236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1504ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 15053236c3e1SJeff Layton return ret; 15063236c3e1SJeff Layton 1507a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1508a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1509a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1510420e3646STrond Myklebust */ 1511ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1512420e3646STrond Myklebust goto out_mark_dirty; 1513420e3646STrond Myklebust 1514a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1515420e3646STrond Myklebust flags = 0; 1516a00dd6c0SJeff Layton } 1517a00dd6c0SJeff Layton 1518420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1519420e3646STrond Myklebust if (ret >= 0) { 1520420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1521420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1522420e3646STrond Myklebust wbc->nr_to_write -= ret; 1523420e3646STrond Myklebust else 1524420e3646STrond Myklebust wbc->nr_to_write = 0; 1525420e3646STrond Myklebust } 15268fc795f7STrond Myklebust return 0; 1527420e3646STrond Myklebust } 1528420e3646STrond Myklebust out_mark_dirty: 15298fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 15308fc795f7STrond Myklebust return ret; 15318fc795f7STrond Myklebust } 1532c63c7b05STrond Myklebust #else 15338fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 15348fc795f7STrond Myklebust { 15358fc795f7STrond Myklebust return 0; 15368fc795f7STrond Myklebust } 15371da177e4SLinus Torvalds #endif 15381da177e4SLinus Torvalds 15398fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 15408fc795f7STrond Myklebust { 1541a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1542a8d8f02cSBryan Schumaker } 154389d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1544863a3c6cSAndy Adamson 1545acdc53b2STrond Myklebust /* 1546acdc53b2STrond Myklebust * flush the inode to disk. 1547acdc53b2STrond Myklebust */ 1548acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 154934901f70STrond Myklebust { 155034901f70STrond Myklebust struct writeback_control wbc = { 155172cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 155234901f70STrond Myklebust .nr_to_write = LONG_MAX, 1553d7fb1207STrond Myklebust .range_start = 0, 1554d7fb1207STrond Myklebust .range_end = LLONG_MAX, 155534901f70STrond Myklebust }; 1556f4ce1299STrond Myklebust int ret; 155734901f70STrond Myklebust 1558f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1559f4ce1299STrond Myklebust 1560f4ce1299STrond Myklebust ret = sync_inode(inode, &wbc); 1561f4ce1299STrond Myklebust 1562f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1563f4ce1299STrond Myklebust return ret; 15641c75950bSTrond Myklebust } 1565ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 15661c75950bSTrond Myklebust 15671b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 15681b3b4a1aSTrond Myklebust { 15691b3b4a1aSTrond Myklebust struct nfs_page *req; 15701b3b4a1aSTrond Myklebust int ret = 0; 15711b3b4a1aSTrond Myklebust 15721b3b4a1aSTrond Myklebust for (;;) { 1573ba8b06e6STrond Myklebust wait_on_page_writeback(page); 15741b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 15751b3b4a1aSTrond Myklebust if (req == NULL) 15761b3b4a1aSTrond Myklebust break; 15777ad84aa9STrond Myklebust if (nfs_lock_request(req)) { 15788dd37758STrond Myklebust nfs_clear_request_commit(req); 15791b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 15801b3b4a1aSTrond Myklebust /* 15811b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 15821b3b4a1aSTrond Myklebust * page as being dirty 15831b3b4a1aSTrond Myklebust */ 15841b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 15851d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 15861b3b4a1aSTrond Myklebust break; 15871b3b4a1aSTrond Myklebust } 15881b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1589c9edda71STrond Myklebust nfs_release_request(req); 15901b3b4a1aSTrond Myklebust if (ret < 0) 1591c988950eSTrond Myklebust break; 15921b3b4a1aSTrond Myklebust } 15931b3b4a1aSTrond Myklebust return ret; 15941b3b4a1aSTrond Myklebust } 15951b3b4a1aSTrond Myklebust 15961c75950bSTrond Myklebust /* 15971c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15981c75950bSTrond Myklebust */ 15991c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 16001c75950bSTrond Myklebust { 160129418aa4SMel Gorman loff_t range_start = page_file_offset(page); 16027f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 16037f2f12d9STrond Myklebust struct writeback_control wbc = { 16047f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 16057f2f12d9STrond Myklebust .nr_to_write = 0, 16067f2f12d9STrond Myklebust .range_start = range_start, 16077f2f12d9STrond Myklebust .range_end = range_end, 16087f2f12d9STrond Myklebust }; 16097f2f12d9STrond Myklebust int ret; 16107f2f12d9STrond Myklebust 1611f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1612f4ce1299STrond Myklebust 16130522f6adSTrond Myklebust for (;;) { 1614ba8b06e6STrond Myklebust wait_on_page_writeback(page); 16157f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 16167f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 16177f2f12d9STrond Myklebust if (ret < 0) 16187f2f12d9STrond Myklebust goto out_error; 16190522f6adSTrond Myklebust continue; 16207f2f12d9STrond Myklebust } 1621f4ce1299STrond Myklebust ret = 0; 16220522f6adSTrond Myklebust if (!PagePrivate(page)) 16230522f6adSTrond Myklebust break; 16240522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 16257f2f12d9STrond Myklebust if (ret < 0) 16267f2f12d9STrond Myklebust goto out_error; 16277f2f12d9STrond Myklebust } 16287f2f12d9STrond Myklebust out_error: 1629f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 16307f2f12d9STrond Myklebust return ret; 16311c75950bSTrond Myklebust } 16321c75950bSTrond Myklebust 1633074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1634074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1635a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1636074cc1deSTrond Myklebust { 16372da95652SJeff Layton /* 16382da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 16392da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 16402da95652SJeff Layton * 16412da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 16422da95652SJeff Layton * that we can safely release the inode reference while holding 16432da95652SJeff Layton * the page lock. 16442da95652SJeff Layton */ 16452da95652SJeff Layton if (PagePrivate(page)) 16462da95652SJeff Layton return -EBUSY; 1647074cc1deSTrond Myklebust 16488c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 16498c209ce7SDavid Howells return -EBUSY; 1650074cc1deSTrond Myklebust 1651a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1652074cc1deSTrond Myklebust } 1653074cc1deSTrond Myklebust #endif 1654074cc1deSTrond Myklebust 1655f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 16561da177e4SLinus Torvalds { 16571da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 1658c0752cdfSAnna Schumaker sizeof(struct nfs_rw_header), 16591da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 166020c2df83SPaul Mundt NULL); 16611da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 16621da177e4SLinus Torvalds return -ENOMEM; 16631da177e4SLinus Torvalds 166493d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 16651da177e4SLinus Torvalds nfs_wdata_cachep); 16661da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 16673dd4765fSJeff Layton goto out_destroy_write_cache; 16681da177e4SLinus Torvalds 16690b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 16700b7c0153SFred Isaman sizeof(struct nfs_commit_data), 16710b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 16720b7c0153SFred Isaman NULL); 16730b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 16743dd4765fSJeff Layton goto out_destroy_write_mempool; 16750b7c0153SFred Isaman 167693d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 16774c100210SYanchuan Nian nfs_cdata_cachep); 16781da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 16793dd4765fSJeff Layton goto out_destroy_commit_cache; 16801da177e4SLinus Torvalds 168189a09141SPeter Zijlstra /* 168289a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 168389a09141SPeter Zijlstra * 168489a09141SPeter Zijlstra * 64MB: 8192k 168589a09141SPeter Zijlstra * 128MB: 11585k 168689a09141SPeter Zijlstra * 256MB: 16384k 168789a09141SPeter Zijlstra * 512MB: 23170k 168889a09141SPeter Zijlstra * 1GB: 32768k 168989a09141SPeter Zijlstra * 2GB: 46340k 169089a09141SPeter Zijlstra * 4GB: 65536k 169189a09141SPeter Zijlstra * 8GB: 92681k 169289a09141SPeter Zijlstra * 16GB: 131072k 169389a09141SPeter Zijlstra * 169489a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 169589a09141SPeter Zijlstra * Limit the default to 256M 169689a09141SPeter Zijlstra */ 169789a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 169889a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 169989a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 170089a09141SPeter Zijlstra 17011da177e4SLinus Torvalds return 0; 17023dd4765fSJeff Layton 17033dd4765fSJeff Layton out_destroy_commit_cache: 17043dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 17053dd4765fSJeff Layton out_destroy_write_mempool: 17063dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 17073dd4765fSJeff Layton out_destroy_write_cache: 17083dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 17093dd4765fSJeff Layton return -ENOMEM; 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds 1712266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 17131da177e4SLinus Torvalds { 17141da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 17153dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 17161da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 17171a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 17181da177e4SLinus Torvalds } 17191da177e4SLinus Torvalds 17204a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 1721a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 17224a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 17234a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 1724a4cdda59SAnna Schumaker .rw_release = nfs_writeback_release_common, 17250eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 17260eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 17271ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 17284a0de55cSAnna Schumaker }; 1729