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; 49d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 5002d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 5102d1426cSWeston Andros Adamson struct inode *inode); 521da177e4SLinus Torvalds 53e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 543feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 550b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 561da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 571da177e4SLinus Torvalds 580b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 591da177e4SLinus Torvalds { 60192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 6140859d7eSChuck Lever 621da177e4SLinus Torvalds if (p) { 631da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 641da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 651da177e4SLinus Torvalds } 661da177e4SLinus Torvalds return p; 671da177e4SLinus Torvalds } 68e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 691da177e4SLinus Torvalds 700b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 711da177e4SLinus Torvalds { 721da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 731da177e4SLinus Torvalds } 74e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 751da177e4SLinus Torvalds 761e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 773feb2d49STrond Myklebust { 781e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 793feb2d49STrond Myklebust 804a0de55cSAnna Schumaker if (p) 813feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 823feb2d49STrond Myklebust return p; 833feb2d49STrond Myklebust } 843feb2d49STrond Myklebust 851e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 866c75dc0dSFred Isaman { 871e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 883feb2d49STrond Myklebust } 891da177e4SLinus Torvalds 907b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 917b159fc1STrond Myklebust { 927b159fc1STrond Myklebust ctx->error = error; 937b159fc1STrond Myklebust smp_wmb(); 947b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 957b159fc1STrond Myklebust } 967b159fc1STrond Myklebust 9784d3a9a9SWeston Andros Adamson /* 9802d1426cSWeston Andros Adamson * nfs_page_search_commits_for_head_request_locked 9902d1426cSWeston Andros Adamson * 10002d1426cSWeston Andros Adamson * Search through commit lists on @inode for the head request for @page. 10102d1426cSWeston Andros Adamson * Must be called while holding the inode (which is cinfo) lock. 10202d1426cSWeston Andros Adamson * 10302d1426cSWeston Andros Adamson * Returns the head request if found, or NULL if not found. 10402d1426cSWeston Andros Adamson */ 10502d1426cSWeston Andros Adamson static struct nfs_page * 10602d1426cSWeston Andros Adamson nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 10702d1426cSWeston Andros Adamson struct page *page) 10802d1426cSWeston Andros Adamson { 10902d1426cSWeston Andros Adamson struct nfs_page *freq, *t; 11002d1426cSWeston Andros Adamson struct nfs_commit_info cinfo; 11102d1426cSWeston Andros Adamson struct inode *inode = &nfsi->vfs_inode; 11202d1426cSWeston Andros Adamson 11302d1426cSWeston Andros Adamson nfs_init_cinfo_from_inode(&cinfo, inode); 11402d1426cSWeston Andros Adamson 11502d1426cSWeston Andros Adamson /* search through pnfs commit lists */ 11602d1426cSWeston Andros Adamson freq = pnfs_search_commit_reqs(inode, &cinfo, page); 11702d1426cSWeston Andros Adamson if (freq) 11802d1426cSWeston Andros Adamson return freq->wb_head; 11902d1426cSWeston Andros Adamson 12002d1426cSWeston Andros Adamson /* Linearly search the commit list for the correct request */ 12102d1426cSWeston Andros Adamson list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 12202d1426cSWeston Andros Adamson if (freq->wb_page == page) 12302d1426cSWeston Andros Adamson return freq->wb_head; 12402d1426cSWeston Andros Adamson } 12502d1426cSWeston Andros Adamson 12602d1426cSWeston Andros Adamson return NULL; 12702d1426cSWeston Andros Adamson } 12802d1426cSWeston Andros Adamson 12902d1426cSWeston Andros Adamson /* 13084d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 13184d3a9a9SWeston Andros Adamson * 13284d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 13384d3a9a9SWeston Andros Adamson * 13484d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 13584d3a9a9SWeston Andros Adamson */ 13629418aa4SMel Gorman static struct nfs_page * 13784d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 138277459d2STrond Myklebust { 139277459d2STrond Myklebust struct nfs_page *req = NULL; 140277459d2STrond Myklebust 14129418aa4SMel Gorman if (PagePrivate(page)) 142277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 14302d1426cSWeston Andros Adamson else if (unlikely(PageSwapCache(page))) 14402d1426cSWeston Andros Adamson req = nfs_page_search_commits_for_head_request_locked(nfsi, 14502d1426cSWeston Andros Adamson page); 14629418aa4SMel Gorman 14784d3a9a9SWeston Andros Adamson if (req) { 14884d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 14929418aa4SMel Gorman kref_get(&req->wb_kref); 15084d3a9a9SWeston Andros Adamson } 15129418aa4SMel Gorman 152277459d2STrond Myklebust return req; 153277459d2STrond Myklebust } 154277459d2STrond Myklebust 15584d3a9a9SWeston Andros Adamson /* 15684d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 15784d3a9a9SWeston Andros Adamson * 15884d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 15984d3a9a9SWeston Andros Adamson */ 16084d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 161277459d2STrond Myklebust { 162d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 163277459d2STrond Myklebust struct nfs_page *req = NULL; 164277459d2STrond Myklebust 165587142f8STrond Myklebust spin_lock(&inode->i_lock); 16684d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 167587142f8STrond Myklebust spin_unlock(&inode->i_lock); 168277459d2STrond Myklebust return req; 169277459d2STrond Myklebust } 170277459d2STrond Myklebust 1711da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1721da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1731da177e4SLinus Torvalds { 174d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 175a3d01454STrond Myklebust loff_t end, i_size; 176a3d01454STrond Myklebust pgoff_t end_index; 1771da177e4SLinus Torvalds 178a3d01454STrond Myklebust spin_lock(&inode->i_lock); 179a3d01454STrond Myklebust i_size = i_size_read(inode); 180a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 181d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 182a3d01454STrond Myklebust goto out; 183d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1841da177e4SLinus Torvalds if (i_size >= end) 185a3d01454STrond Myklebust goto out; 1861da177e4SLinus Torvalds i_size_write(inode, end); 187a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 188a3d01454STrond Myklebust out: 189a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1901da177e4SLinus Torvalds } 1911da177e4SLinus Torvalds 192a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 193a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 194a301b777STrond Myklebust { 195d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 196a301b777STrond Myklebust } 197a301b777STrond Myklebust 198d72ddcbaSWeston Andros Adamson /* 199d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 200d72ddcbaSWeston Andros Adamson * @head - head request of page group 201d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 202d72ddcbaSWeston Andros Adamson * 203d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 204d72ddcbaSWeston Andros Adamson * page offset @page_offset. 205d72ddcbaSWeston Andros Adamson * 206d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 207d72ddcbaSWeston Andros Adamson * match is found. 208d72ddcbaSWeston Andros Adamson * 209d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 210d72ddcbaSWeston Andros Adamson */ 211d72ddcbaSWeston Andros Adamson static struct nfs_page * 212d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 213d72ddcbaSWeston Andros Adamson { 214d72ddcbaSWeston Andros Adamson struct nfs_page *req; 215d72ddcbaSWeston Andros Adamson 216d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 217d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 218d72ddcbaSWeston Andros Adamson 219d72ddcbaSWeston Andros Adamson req = head; 220d72ddcbaSWeston Andros Adamson do { 221d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 222d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 223d72ddcbaSWeston Andros Adamson return req; 224d72ddcbaSWeston Andros Adamson 225d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 226d72ddcbaSWeston Andros Adamson } while (req != head); 227d72ddcbaSWeston Andros Adamson 228d72ddcbaSWeston Andros Adamson return NULL; 229d72ddcbaSWeston Andros Adamson } 230d72ddcbaSWeston Andros Adamson 231d72ddcbaSWeston Andros Adamson /* 232d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 233d72ddcbaSWeston Andros Adamson * @head - head request of page group 234d72ddcbaSWeston Andros Adamson * 235d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 236d72ddcbaSWeston Andros Adamson * returns false otherwise 237d72ddcbaSWeston Andros Adamson */ 238d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 239d72ddcbaSWeston Andros Adamson { 240d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 241d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 242d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 243d72ddcbaSWeston Andros Adamson 244fd2f3a06SWeston Andros Adamson nfs_page_group_lock(req, false); 245d72ddcbaSWeston Andros Adamson 246d72ddcbaSWeston Andros Adamson do { 247d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 248d72ddcbaSWeston Andros Adamson if (tmp) { 249d72ddcbaSWeston Andros Adamson /* no way this should happen */ 250d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 251d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 252d72ddcbaSWeston Andros Adamson } 253d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 254d72ddcbaSWeston Andros Adamson 255d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 256d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 257d72ddcbaSWeston Andros Adamson return pos == len; 258d72ddcbaSWeston Andros Adamson } 259d72ddcbaSWeston Andros Adamson 2601da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2611da177e4SLinus Torvalds * covers the full page. 2621da177e4SLinus Torvalds */ 263d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2641da177e4SLinus Torvalds { 265d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2661da177e4SLinus Torvalds return; 267d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2681da177e4SLinus Torvalds return; 269d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2701da177e4SLinus Torvalds } 2711da177e4SLinus Torvalds 2721da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2731da177e4SLinus Torvalds { 2741da177e4SLinus Torvalds if (wbc->for_reclaim) 275c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 276b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 277b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 278b31268acSTrond Myklebust return FLUSH_COND_STABLE; 2791da177e4SLinus Torvalds } 2801da177e4SLinus Torvalds 2811da177e4SLinus Torvalds /* 28289a09141SPeter Zijlstra * NFS congestion control 28389a09141SPeter Zijlstra */ 28489a09141SPeter Zijlstra 28589a09141SPeter Zijlstra int nfs_congestion_kb; 28689a09141SPeter Zijlstra 28789a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 28889a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 28989a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 29089a09141SPeter Zijlstra 291deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 29289a09141SPeter Zijlstra { 293deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2945a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2955a6d41b3STrond Myklebust 296deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 29789a09141SPeter Zijlstra 298277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2998aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 3008aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 3018aa7e847SJens Axboe BLK_RW_ASYNC); 3028aa7e847SJens Axboe } 30389a09141SPeter Zijlstra } 30489a09141SPeter Zijlstra 30520633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 30689a09141SPeter Zijlstra { 30720633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 30889a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 30989a09141SPeter Zijlstra 31020633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 31120633f04SWeston Andros Adamson return; 31220633f04SWeston Andros Adamson 31320633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 314c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3158aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 31689a09141SPeter Zijlstra } 31789a09141SPeter Zijlstra 318d4581383SWeston Andros Adamson 319d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 320d4581383SWeston Andros Adamson * @req - an nfs request 321d4581383SWeston Andros Adamson * clears all page group related bits from @req 322d4581383SWeston Andros Adamson */ 323d4581383SWeston Andros Adamson static void 324d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 325e261f51fSTrond Myklebust { 326d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 327d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 328d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 329d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 330d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 331d4581383SWeston Andros Adamson } 332d4581383SWeston Andros Adamson 333d4581383SWeston Andros Adamson 334d4581383SWeston Andros Adamson /* 335d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 336d4581383SWeston Andros Adamson * 337d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 338d4581383SWeston Andros Adamson * 339d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 340d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 341d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 342d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 343d4581383SWeston Andros Adamson * 344d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 345d4581383SWeston Andros Adamson * and BOTH will be released before returning. 346d4581383SWeston Andros Adamson * 347d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 348d4581383SWeston Andros Adamson */ 349d4581383SWeston Andros Adamson static int 350d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 351d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 352d4581383SWeston Andros Adamson __releases(&inode->i_lock) 353d4581383SWeston Andros Adamson { 354d4581383SWeston Andros Adamson struct nfs_page *tmp; 355e261f51fSTrond Myklebust int ret; 356e261f51fSTrond Myklebust 357d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 358d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 359d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 360d4581383SWeston Andros Adamson 361d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 362d4581383SWeston Andros Adamson 363d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 364d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 365d4581383SWeston Andros Adamson 366d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 367587142f8STrond Myklebust spin_unlock(&inode->i_lock); 368d4581383SWeston Andros Adamson 369d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 370d4581383SWeston Andros Adamson nfs_release_request(head); 371d4581383SWeston Andros Adamson 372cfb506e1STrond Myklebust if (!nonblock) 373e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 374cfb506e1STrond Myklebust else 375cfb506e1STrond Myklebust ret = -EAGAIN; 376e261f51fSTrond Myklebust nfs_release_request(req); 377d4581383SWeston Andros Adamson 378d4581383SWeston Andros Adamson return ret; 379e261f51fSTrond Myklebust } 380d4581383SWeston Andros Adamson 381d4581383SWeston Andros Adamson /* 382d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 383d4581383SWeston Andros Adamson * 384d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 385d4581383SWeston Andros Adamson * @old_head - the old head of the list 386d4581383SWeston Andros Adamson * 387d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 388d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 389d4581383SWeston Andros Adamson * with a reference held by some other context. 390d4581383SWeston Andros Adamson */ 391d4581383SWeston Andros Adamson static void 392d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 393d4581383SWeston Andros Adamson struct nfs_page *old_head) 394d4581383SWeston Andros Adamson { 395d4581383SWeston Andros Adamson while (destroy_list) { 396d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 397d4581383SWeston Andros Adamson 398d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 399d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 400d4581383SWeston Andros Adamson 401d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 402d4581383SWeston Andros Adamson 403d4581383SWeston Andros Adamson /* make sure old group is not used */ 404d4581383SWeston Andros Adamson subreq->wb_head = subreq; 405d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 406d4581383SWeston Andros Adamson 407d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 408d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 409d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 410d4581383SWeston Andros Adamson 411d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 412d4581383SWeston Andros Adamson /* release ref on old head request */ 413d4581383SWeston Andros Adamson nfs_release_request(old_head); 414d4581383SWeston Andros Adamson 415d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 416d4581383SWeston Andros Adamson 417d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 418d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 419d4581383SWeston Andros Adamson nfs_release_request(subreq); 420d4581383SWeston Andros Adamson else 421d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 422d4581383SWeston Andros Adamson } else { 423d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 424d4581383SWeston Andros Adamson /* zombie requests have already released the last 425d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 426d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 427d4581383SWeston Andros Adamson * group, simply free the request */ 428d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 429d4581383SWeston Andros Adamson nfs_free_request(subreq); 430d4581383SWeston Andros Adamson } 431d4581383SWeston Andros Adamson } 432d4581383SWeston Andros Adamson } 433d4581383SWeston Andros Adamson 434d4581383SWeston Andros Adamson /* 435d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 436d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 437d4581383SWeston Andros Adamson * operations for this page. 438d4581383SWeston Andros Adamson * 439d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 440d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 441d4581383SWeston Andros Adamson * 442d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 443d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 444d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 445d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 446d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 447d4581383SWeston Andros Adamson * 448d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 449d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 450d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 451d4581383SWeston Andros Adamson * error was encountered. 452d4581383SWeston Andros Adamson */ 453d4581383SWeston Andros Adamson static struct nfs_page * 454d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 455d4581383SWeston Andros Adamson { 456d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 457d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 458d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 459d4581383SWeston Andros Adamson unsigned int total_bytes; 460d4581383SWeston Andros Adamson int ret; 461d4581383SWeston Andros Adamson 462d4581383SWeston Andros Adamson try_again: 463d4581383SWeston Andros Adamson total_bytes = 0; 464d4581383SWeston Andros Adamson 465d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 466d4581383SWeston Andros Adamson 467d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 468d4581383SWeston Andros Adamson 469d4581383SWeston Andros Adamson /* 470d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 471d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 472d4581383SWeston Andros Adamson * until the head reference is released. 473d4581383SWeston Andros Adamson */ 474d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 475d4581383SWeston Andros Adamson 476d4581383SWeston Andros Adamson if (!head) { 477587142f8STrond Myklebust spin_unlock(&inode->i_lock); 478d4581383SWeston Andros Adamson return NULL; 479d4581383SWeston Andros Adamson } 480d4581383SWeston Andros Adamson 481*7c3af975SWeston Andros Adamson /* holding inode lock, so always make a non-blocking call to try the 482*7c3af975SWeston Andros Adamson * page group lock */ 483fd2f3a06SWeston Andros Adamson ret = nfs_page_group_lock(head, true); 48494970014SWeston Andros Adamson if (ret < 0) { 48594970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 486*7c3af975SWeston Andros Adamson 487*7c3af975SWeston Andros Adamson if (!nonblock && ret == -EAGAIN) { 488*7c3af975SWeston Andros Adamson nfs_page_group_lock_wait(head); 489*7c3af975SWeston Andros Adamson nfs_release_request(head); 490*7c3af975SWeston Andros Adamson goto try_again; 491*7c3af975SWeston Andros Adamson } 492*7c3af975SWeston Andros Adamson 49394970014SWeston Andros Adamson nfs_release_request(head); 494e7029206SWeston Andros Adamson return ERR_PTR(ret); 49594970014SWeston Andros Adamson } 496*7c3af975SWeston Andros Adamson 497*7c3af975SWeston Andros Adamson /* lock each request in the page group */ 498d4581383SWeston Andros Adamson subreq = head; 499d4581383SWeston Andros Adamson do { 500d4581383SWeston Andros Adamson /* 501d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 502d4581383SWeston Andros Adamson * and in order. If not, it's a programming error. 503d4581383SWeston Andros Adamson */ 504d4581383SWeston Andros Adamson WARN_ON_ONCE(subreq->wb_offset != 505d4581383SWeston Andros Adamson (head->wb_offset + total_bytes)); 506d4581383SWeston Andros Adamson 507d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 508d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 509d4581383SWeston Andros Adamson 510d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 511d4581383SWeston Andros Adamson /* releases page group bit lock and 512d4581383SWeston Andros Adamson * inode spin lock and all references */ 513d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 514d4581383SWeston Andros Adamson subreq, nonblock); 515d4581383SWeston Andros Adamson 516d4581383SWeston Andros Adamson if (ret == 0) 517d4581383SWeston Andros Adamson goto try_again; 518d4581383SWeston Andros Adamson 519d4581383SWeston Andros Adamson return ERR_PTR(ret); 520d4581383SWeston Andros Adamson } 521d4581383SWeston Andros Adamson 522d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 523d4581383SWeston Andros Adamson } while (subreq != head); 524d4581383SWeston Andros Adamson 525d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 526d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 527d4581383SWeston Andros Adamson subreq = head; 528d4581383SWeston Andros Adamson do { 529411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 530d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 531d4581383SWeston Andros Adamson } while (subreq != head); 532d4581383SWeston Andros Adamson 533d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 534d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 535d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 536d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 537d4581383SWeston Andros Adamson head->wb_this_page = head; 538d4581383SWeston Andros Adamson 539d4581383SWeston Andros Adamson /* change head request to cover whole range that 540d4581383SWeston Andros Adamson * the former page group covered */ 541d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 542d4581383SWeston Andros Adamson } 543d4581383SWeston Andros Adamson 544d4581383SWeston Andros Adamson /* 545d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 546d4581383SWeston Andros Adamson */ 547d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 548d4581383SWeston Andros Adamson 549d4581383SWeston Andros Adamson /* 550d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 551d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 552d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 553d4581383SWeston Andros Adamson */ 554d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 555d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 556d4581383SWeston Andros Adamson 557d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 558d4581383SWeston Andros Adamson 559411a99adSWeston Andros Adamson /* drop lock to clean uprequests on destroy list */ 560d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 561d4581383SWeston Andros Adamson 562d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 563d4581383SWeston Andros Adamson 564d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 565d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 566d4581383SWeston Andros Adamson return head; 567612c9384STrond Myklebust } 568074cc1deSTrond Myklebust 569074cc1deSTrond Myklebust /* 570074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 571074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 572074cc1deSTrond Myklebust */ 573074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 574cfb506e1STrond Myklebust struct page *page, bool nonblock) 575074cc1deSTrond Myklebust { 576074cc1deSTrond Myklebust struct nfs_page *req; 577074cc1deSTrond Myklebust int ret = 0; 578074cc1deSTrond Myklebust 579d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 580074cc1deSTrond Myklebust if (!req) 581074cc1deSTrond Myklebust goto out; 582074cc1deSTrond Myklebust ret = PTR_ERR(req); 583074cc1deSTrond Myklebust if (IS_ERR(req)) 584074cc1deSTrond Myklebust goto out; 585074cc1deSTrond Myklebust 586deed85e7STrond Myklebust nfs_set_page_writeback(page); 587deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 588074cc1deSTrond Myklebust 589deed85e7STrond Myklebust ret = 0; 590f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 591f8512ad0SFred Isaman nfs_redirty_request(req); 592074cc1deSTrond Myklebust ret = pgio->pg_error; 593f8512ad0SFred Isaman } 594074cc1deSTrond Myklebust out: 595074cc1deSTrond Myklebust return ret; 596e261f51fSTrond Myklebust } 597e261f51fSTrond Myklebust 598f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 599f758c885STrond Myklebust { 600d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 601cfb506e1STrond Myklebust int ret; 602f758c885STrond Myklebust 603f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 604f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 605f758c885STrond Myklebust 606d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 6071b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 608cfb506e1STrond Myklebust if (ret == -EAGAIN) { 609cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 610cfb506e1STrond Myklebust ret = 0; 611cfb506e1STrond Myklebust } 612cfb506e1STrond Myklebust return ret; 613f758c885STrond Myklebust } 614f758c885STrond Myklebust 615e261f51fSTrond Myklebust /* 6161da177e4SLinus Torvalds * Write an mmapped page to the server. 6171da177e4SLinus Torvalds */ 6184d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 6191da177e4SLinus Torvalds { 620f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 621e261f51fSTrond Myklebust int err; 6221da177e4SLinus Torvalds 623a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 624a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 625f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 626f758c885STrond Myklebust nfs_pageio_complete(&pgio); 627f758c885STrond Myklebust if (err < 0) 6284d770ccfSTrond Myklebust return err; 629f758c885STrond Myklebust if (pgio.pg_error < 0) 630f758c885STrond Myklebust return pgio.pg_error; 631f758c885STrond Myklebust return 0; 6324d770ccfSTrond Myklebust } 6334d770ccfSTrond Myklebust 6344d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6354d770ccfSTrond Myklebust { 636f758c885STrond Myklebust int ret; 6374d770ccfSTrond Myklebust 638f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6391da177e4SLinus Torvalds unlock_page(page); 640f758c885STrond Myklebust return ret; 641f758c885STrond Myklebust } 642f758c885STrond Myklebust 643f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 644f758c885STrond Myklebust { 645f758c885STrond Myklebust int ret; 646f758c885STrond Myklebust 647f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 648f758c885STrond Myklebust unlock_page(page); 649f758c885STrond Myklebust return ret; 6501da177e4SLinus Torvalds } 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6531da177e4SLinus Torvalds { 6541da177e4SLinus Torvalds struct inode *inode = mapping->host; 65572cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 656c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6571da177e4SLinus Torvalds int err; 6581da177e4SLinus Torvalds 65972cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 66074316201SNeilBrown err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING, 66172cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 66272cb77f4STrond Myklebust if (err) 66372cb77f4STrond Myklebust goto out_err; 66472cb77f4STrond Myklebust 66591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 66691d5b470SChuck Lever 667a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 668a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 669f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 670c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 67172cb77f4STrond Myklebust 67272cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 6734e857c58SPeter Zijlstra smp_mb__after_atomic(); 67472cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 67572cb77f4STrond Myklebust 676f758c885STrond Myklebust if (err < 0) 67772cb77f4STrond Myklebust goto out_err; 67872cb77f4STrond Myklebust err = pgio.pg_error; 67972cb77f4STrond Myklebust if (err < 0) 68072cb77f4STrond Myklebust goto out_err; 681c63c7b05STrond Myklebust return 0; 68272cb77f4STrond Myklebust out_err: 68372cb77f4STrond Myklebust return err; 6841da177e4SLinus Torvalds } 6851da177e4SLinus Torvalds 6861da177e4SLinus Torvalds /* 6871da177e4SLinus Torvalds * Insert a write request into an inode 6881da177e4SLinus Torvalds */ 689d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6901da177e4SLinus Torvalds { 6911da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 692e7d39069STrond Myklebust 6932bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 6942bfc6e56SWeston Andros Adamson 695e7d39069STrond Myklebust /* Lock the request! */ 6967ad84aa9STrond Myklebust nfs_lock_request(req); 697e7d39069STrond Myklebust 698e7d39069STrond Myklebust spin_lock(&inode->i_lock); 699011e2a7fSBryan Schumaker if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 700a9a4a87aSTrond Myklebust inode->i_version++; 70129418aa4SMel Gorman /* 70229418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 70329418aa4SMel Gorman * with invalidate/truncate. 70429418aa4SMel Gorman */ 70529418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7062df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 707deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 708277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 70929418aa4SMel Gorman } 7101da177e4SLinus Torvalds nfsi->npages++; 71117089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 71217089a29SWeston Andros Adamson * extra reference so sub groups can follow suit */ 71317089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 714c03b4024STrond Myklebust kref_get(&req->wb_kref); 715e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds /* 71989a09141SPeter Zijlstra * Remove a write request from an inode 7201da177e4SLinus Torvalds */ 7211da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7221da177e4SLinus Torvalds { 7233d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 7241da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 72520633f04SWeston Andros Adamson struct nfs_page *head; 72620633f04SWeston Andros Adamson 72720633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 72820633f04SWeston Andros Adamson head = req->wb_head; 7291da177e4SLinus Torvalds 730587142f8STrond Myklebust spin_lock(&inode->i_lock); 73120633f04SWeston Andros Adamson if (likely(!PageSwapCache(head->wb_page))) { 73220633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 73320633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 73420633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 73529418aa4SMel Gorman } 7361da177e4SLinus Torvalds nfsi->npages--; 737587142f8STrond Myklebust spin_unlock(&inode->i_lock); 73820633f04SWeston Andros Adamson } 73917089a29SWeston Andros Adamson 74017089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7411da177e4SLinus Torvalds nfs_release_request(req); 742b412ddf0SWeston Andros Adamson else 743b412ddf0SWeston Andros Adamson WARN_ON_ONCE(1); 7441da177e4SLinus Torvalds } 7451da177e4SLinus Torvalds 74661822ab5STrond Myklebust static void 7476d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 74861822ab5STrond Myklebust { 74961822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 75061822ab5STrond Myklebust } 75161822ab5STrond Myklebust 75289d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 7538dd37758STrond Myklebust /** 7548dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 7558dd37758STrond Myklebust * @req: pointer to a struct nfs_page 756ea2cf228SFred Isaman * @dst: commit list head 757ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7588dd37758STrond Myklebust * 759ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 7608dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 7618dd37758STrond Myklebust * the MM page stats. 7628dd37758STrond Myklebust * 763ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 7648dd37758STrond Myklebust * holding the nfs_page lock. 7658dd37758STrond Myklebust */ 7668dd37758STrond Myklebust void 767ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 768ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7698dd37758STrond Myklebust { 7708dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 771ea2cf228SFred Isaman spin_lock(cinfo->lock); 772ea2cf228SFred Isaman nfs_list_add_request(req, dst); 773ea2cf228SFred Isaman cinfo->mds->ncommit++; 774ea2cf228SFred Isaman spin_unlock(cinfo->lock); 77556f9cd68SFred Isaman if (!cinfo->dreq) { 7768dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 777d56b4ddfSMel Gorman inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 77856f9cd68SFred Isaman BDI_RECLAIMABLE); 77956f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 78056f9cd68SFred Isaman I_DIRTY_DATASYNC); 78156f9cd68SFred Isaman } 7828dd37758STrond Myklebust } 7838dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 7848dd37758STrond Myklebust 7858dd37758STrond Myklebust /** 7868dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 7878dd37758STrond Myklebust * @req: pointer to a nfs_page 788ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7898dd37758STrond Myklebust * 790ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 7918dd37758STrond Myklebust * number of outstanding requests requiring a commit 7928dd37758STrond Myklebust * It does not update the MM page stats. 7938dd37758STrond Myklebust * 794ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 7958dd37758STrond Myklebust */ 7968dd37758STrond Myklebust void 797ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 798ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7998dd37758STrond Myklebust { 8008dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 8018dd37758STrond Myklebust return; 8028dd37758STrond Myklebust nfs_list_remove_request(req); 803ea2cf228SFred Isaman cinfo->mds->ncommit--; 8048dd37758STrond Myklebust } 8058dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 8068dd37758STrond Myklebust 807ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 808ea2cf228SFred Isaman struct inode *inode) 809ea2cf228SFred Isaman { 810ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 811ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 812ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 813b359f9d0SFred Isaman cinfo->dreq = NULL; 814f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 815ea2cf228SFred Isaman } 816ea2cf228SFred Isaman 817ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 818ea2cf228SFred Isaman struct inode *inode, 819ea2cf228SFred Isaman struct nfs_direct_req *dreq) 820ea2cf228SFred Isaman { 8211763da12SFred Isaman if (dreq) 8221763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8231763da12SFred Isaman else 824ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 825ea2cf228SFred Isaman } 826ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8278dd37758STrond Myklebust 8281da177e4SLinus Torvalds /* 8291da177e4SLinus Torvalds * Add a request to the inode's commit list. 8301da177e4SLinus Torvalds */ 8311763da12SFred Isaman void 832ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 833ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8341da177e4SLinus Torvalds { 835ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 8368dd37758STrond Myklebust return; 837ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 8381da177e4SLinus Torvalds } 8398e821cadSTrond Myklebust 840d6d6dc7cSFred Isaman static void 841d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 842e468bae9STrond Myklebust { 843e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 844d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); 845e468bae9STrond Myklebust } 846d6d6dc7cSFred Isaman 847411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 8488dd37758STrond Myklebust static void 849d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 850d6d6dc7cSFred Isaman { 8518dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 8528dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 853ea2cf228SFred Isaman struct nfs_commit_info cinfo; 854d6d6dc7cSFred Isaman 855ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 856ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 857ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 858d6d6dc7cSFred Isaman } 8598dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 8608dd37758STrond Myklebust } 861e468bae9STrond Myklebust } 862e468bae9STrond Myklebust 863d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 8648e821cadSTrond Myklebust { 865c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 866d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 867c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 8688e821cadSTrond Myklebust } 8698e821cadSTrond Myklebust 8708e821cadSTrond Myklebust #else 87168cd6fa4SBryan Schumaker static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 87268cd6fa4SBryan Schumaker struct inode *inode) 87368cd6fa4SBryan Schumaker { 87468cd6fa4SBryan Schumaker } 87568cd6fa4SBryan Schumaker 87668cd6fa4SBryan Schumaker void nfs_init_cinfo(struct nfs_commit_info *cinfo, 87768cd6fa4SBryan Schumaker struct inode *inode, 87868cd6fa4SBryan Schumaker struct nfs_direct_req *dreq) 87968cd6fa4SBryan Schumaker { 88068cd6fa4SBryan Schumaker } 88168cd6fa4SBryan Schumaker 8821763da12SFred Isaman void 883ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 884ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8858e821cadSTrond Myklebust { 8868e821cadSTrond Myklebust } 8878e821cadSTrond Myklebust 8888dd37758STrond Myklebust static void 889e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 890e468bae9STrond Myklebust { 891e468bae9STrond Myklebust } 892e468bae9STrond Myklebust 893d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 8948e821cadSTrond Myklebust { 8958e821cadSTrond Myklebust return 0; 8968e821cadSTrond Myklebust } 8978e821cadSTrond Myklebust 8981da177e4SLinus Torvalds #endif 8991da177e4SLinus Torvalds 900061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9016c75dc0dSFred Isaman { 902ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9036c75dc0dSFred Isaman unsigned long bytes = 0; 9046c75dc0dSFred Isaman 9056c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9066c75dc0dSFred Isaman goto out; 907ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9086c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9096c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9106c75dc0dSFred Isaman 9116c75dc0dSFred Isaman bytes += req->wb_bytes; 9126c75dc0dSFred Isaman nfs_list_remove_request(req); 9136c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9146c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 915d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9166c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9176c75dc0dSFred Isaman goto remove_req; 9186c75dc0dSFred Isaman } 919c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 920f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 921ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 9226c75dc0dSFred Isaman goto next; 9236c75dc0dSFred Isaman } 9246c75dc0dSFred Isaman remove_req: 9256c75dc0dSFred Isaman nfs_inode_remove_request(req); 9266c75dc0dSFred Isaman next: 9271d1afcbcSTrond Myklebust nfs_unlock_request(req); 92820633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9293aff4ebbSTrond Myklebust nfs_release_request(req); 9306c75dc0dSFred Isaman } 9316c75dc0dSFred Isaman out: 9326c75dc0dSFred Isaman hdr->release(hdr); 9336c75dc0dSFred Isaman } 9346c75dc0dSFred Isaman 93589d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 936ce59515cSAnna Schumaker unsigned long 937ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 938fb8a1f11STrond Myklebust { 939ea2cf228SFred Isaman return cinfo->mds->ncommit; 940fb8a1f11STrond Myklebust } 941fb8a1f11STrond Myklebust 942ea2cf228SFred Isaman /* cinfo->lock held by caller */ 9431763da12SFred Isaman int 944ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 945ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 946d6d6dc7cSFred Isaman { 947d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 948d6d6dc7cSFred Isaman int ret = 0; 949d6d6dc7cSFred Isaman 950d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9518dd37758STrond Myklebust if (!nfs_lock_request(req)) 9528dd37758STrond Myklebust continue; 9537ad84aa9STrond Myklebust kref_get(&req->wb_kref); 954ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 9553b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 956ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9578dd37758STrond Myklebust nfs_list_add_request(req, dst); 958d6d6dc7cSFred Isaman ret++; 9591763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 960d6d6dc7cSFred Isaman break; 961d6d6dc7cSFred Isaman } 962d6d6dc7cSFred Isaman return ret; 963d6d6dc7cSFred Isaman } 964d6d6dc7cSFred Isaman 9651da177e4SLinus Torvalds /* 9661da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9671da177e4SLinus Torvalds * @inode: NFS inode to scan 968ea2cf228SFred Isaman * @dst: mds destination list 969ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9701da177e4SLinus Torvalds * 9711da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 9721da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 9731da177e4SLinus Torvalds */ 9741763da12SFred Isaman int 975ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 976ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9771da177e4SLinus Torvalds { 978d6d6dc7cSFred Isaman int ret = 0; 979fb8a1f11STrond Myklebust 980ea2cf228SFred Isaman spin_lock(cinfo->lock); 981ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 9828dd37758STrond Myklebust const int max = INT_MAX; 983d6d6dc7cSFred Isaman 984ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 985ea2cf228SFred Isaman cinfo, max); 986ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 987d6d6dc7cSFred Isaman } 988ea2cf228SFred Isaman spin_unlock(cinfo->lock); 989ff778d02STrond Myklebust return ret; 9901da177e4SLinus Torvalds } 991d6d6dc7cSFred Isaman 992c42de9ddSTrond Myklebust #else 993ce59515cSAnna Schumaker unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 994fb8a1f11STrond Myklebust { 995fb8a1f11STrond Myklebust return 0; 996fb8a1f11STrond Myklebust } 997fb8a1f11STrond Myklebust 9981763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 999ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 1000c42de9ddSTrond Myklebust { 1001c42de9ddSTrond Myklebust return 0; 1002c42de9ddSTrond Myklebust } 10031da177e4SLinus Torvalds #endif 10041da177e4SLinus Torvalds 10051da177e4SLinus Torvalds /* 1006e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1007e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10081da177e4SLinus Torvalds * 1009e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1010e7d39069STrond Myklebust * to disk. 10111da177e4SLinus Torvalds */ 1012e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1013e7d39069STrond Myklebust struct page *page, 1014e7d39069STrond Myklebust unsigned int offset, 1015e7d39069STrond Myklebust unsigned int bytes) 10161da177e4SLinus Torvalds { 1017e7d39069STrond Myklebust struct nfs_page *req; 1018e7d39069STrond Myklebust unsigned int rqend; 1019e7d39069STrond Myklebust unsigned int end; 10201da177e4SLinus Torvalds int error; 1021277459d2STrond Myklebust 1022e7d39069STrond Myklebust if (!PagePrivate(page)) 1023e7d39069STrond Myklebust return NULL; 1024e7d39069STrond Myklebust 1025e7d39069STrond Myklebust end = offset + bytes; 1026e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1027e7d39069STrond Myklebust 1028e7d39069STrond Myklebust for (;;) { 102984d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1030e7d39069STrond Myklebust if (req == NULL) 1031e7d39069STrond Myklebust goto out_unlock; 1032e7d39069STrond Myklebust 10332bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10342bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10352bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10362bfc6e56SWeston Andros Adamson 1037e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1038e7d39069STrond Myklebust /* 1039e7d39069STrond Myklebust * Tell the caller to flush out the request if 1040e7d39069STrond Myklebust * the offsets are non-contiguous. 1041e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1042e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1043e7d39069STrond Myklebust */ 1044e468bae9STrond Myklebust if (offset > rqend 1045e7d39069STrond Myklebust || end < req->wb_offset) 1046e7d39069STrond Myklebust goto out_flushme; 1047e7d39069STrond Myklebust 10487ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1049e7d39069STrond Myklebust break; 1050e7d39069STrond Myklebust 1051e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1052587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10531da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10541da177e4SLinus Torvalds nfs_release_request(req); 1055e7d39069STrond Myklebust if (error != 0) 1056e7d39069STrond Myklebust goto out_err; 1057e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10581da177e4SLinus Torvalds } 10591da177e4SLinus Torvalds 10601da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10611da177e4SLinus Torvalds if (offset < req->wb_offset) { 10621da177e4SLinus Torvalds req->wb_offset = offset; 10631da177e4SLinus Torvalds req->wb_pgbase = offset; 10641da177e4SLinus Torvalds } 10651da177e4SLinus Torvalds if (end > rqend) 10661da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1067e7d39069STrond Myklebust else 1068e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1069e7d39069STrond Myklebust out_unlock: 1070ca138f36SFred Isaman if (req) 10718dd37758STrond Myklebust nfs_clear_request_commit(req); 1072411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1073e7d39069STrond Myklebust return req; 1074e7d39069STrond Myklebust out_flushme: 1075e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1076e7d39069STrond Myklebust nfs_release_request(req); 1077e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1078e7d39069STrond Myklebust out_err: 1079e7d39069STrond Myklebust return ERR_PTR(error); 1080e7d39069STrond Myklebust } 10811da177e4SLinus Torvalds 1082e7d39069STrond Myklebust /* 1083e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1084e7d39069STrond Myklebust * 1085e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1086e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1087e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1088e7d39069STrond Myklebust */ 1089e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1090e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1091e7d39069STrond Myklebust { 1092d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1093e7d39069STrond Myklebust struct nfs_page *req; 1094e7d39069STrond Myklebust 1095e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1096e7d39069STrond Myklebust if (req != NULL) 1097e7d39069STrond Myklebust goto out; 10982bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1099e7d39069STrond Myklebust if (IS_ERR(req)) 1100e7d39069STrond Myklebust goto out; 1101d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1102efc91ed0STrond Myklebust out: 110361e930a9STrond Myklebust return req; 11041da177e4SLinus Torvalds } 11051da177e4SLinus Torvalds 1106e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1107e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1108e7d39069STrond Myklebust { 1109e7d39069STrond Myklebust struct nfs_page *req; 1110e7d39069STrond Myklebust 1111e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1112e7d39069STrond Myklebust if (IS_ERR(req)) 1113e7d39069STrond Myklebust return PTR_ERR(req); 1114e7d39069STrond Myklebust /* Update file length */ 1115e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1116d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1117a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11181d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1119e7d39069STrond Myklebust return 0; 1120e7d39069STrond Myklebust } 1121e7d39069STrond Myklebust 11221da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11231da177e4SLinus Torvalds { 1124cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11252a369153STrond Myklebust struct nfs_lock_context *l_ctx; 11261da177e4SLinus Torvalds struct nfs_page *req; 11271a54533eSTrond Myklebust int do_flush, status; 11281da177e4SLinus Torvalds /* 11291da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11301da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11311da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11321da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11331da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11341da177e4SLinus Torvalds * dropped page. 11351da177e4SLinus Torvalds */ 11361a54533eSTrond Myklebust do { 113784d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11381a54533eSTrond Myklebust if (req == NULL) 11391a54533eSTrond Myklebust return 0; 11402a369153STrond Myklebust l_ctx = req->wb_lock_context; 11412a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 11422bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11432bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 11440f1d2605STrond Myklebust if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { 11452a369153STrond Myklebust do_flush |= l_ctx->lockowner.l_owner != current->files 11462a369153STrond Myklebust || l_ctx->lockowner.l_pid != current->tgid; 11472a369153STrond Myklebust } 11481da177e4SLinus Torvalds nfs_release_request(req); 11491a54533eSTrond Myklebust if (!do_flush) 11501a54533eSTrond Myklebust return 0; 1151d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11521a54533eSTrond Myklebust } while (status == 0); 11531a54533eSTrond Myklebust return status; 11541da177e4SLinus Torvalds } 11551da177e4SLinus Torvalds 11561da177e4SLinus Torvalds /* 1157dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1158dc24826bSAndy Adamson * expire soon. 1159dc24826bSAndy Adamson * 1160dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1161dc24826bSAndy Adamson * 1162dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1163dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1164dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1165dc24826bSAndy Adamson */ 1166dc24826bSAndy Adamson int 1167dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1168dc24826bSAndy Adamson { 1169dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1170dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1171dc24826bSAndy Adamson 1172dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1173dc24826bSAndy Adamson } 1174dc24826bSAndy Adamson 1175dc24826bSAndy Adamson /* 1176dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1177dc24826bSAndy Adamson */ 1178dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 1179dc24826bSAndy Adamson { 1180dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 1181dc24826bSAndy Adamson } 1182dc24826bSAndy Adamson 1183dc24826bSAndy Adamson /* 11845d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 11855d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 11865d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 11875d47a356STrond Myklebust */ 11888d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 11895d47a356STrond Myklebust { 1190d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1191d529ef83SJeff Layton 11928d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 11938d197a56STrond Myklebust goto out; 119418dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1195d529ef83SJeff Layton return false; 11964db72b40SJeff Layton smp_rmb(); 1197d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 11988d197a56STrond Myklebust return false; 11998d197a56STrond Myklebust out: 120018dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 120118dd78c4SScott Mayhew return false; 12028d197a56STrond Myklebust return PageUptodate(page) != 0; 12035d47a356STrond Myklebust } 12045d47a356STrond Myklebust 1205c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1206c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1207c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1208c7559663SScott Mayhew * inefficiencies. 1209c7559663SScott Mayhew * 1210263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1211263b4509SScott Mayhew * of the checks. 1212c7559663SScott Mayhew */ 1213c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1214c7559663SScott Mayhew { 1215c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1216c7559663SScott Mayhew return 0; 1217263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1218263b4509SScott Mayhew return 0; 1219c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1220c7559663SScott Mayhew return 1; 1221263b4509SScott Mayhew if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && 1222c7559663SScott Mayhew inode->i_flock->fl_end == OFFSET_MAX && 1223263b4509SScott Mayhew inode->i_flock->fl_type != F_RDLCK)) 1224c7559663SScott Mayhew return 1; 1225c7559663SScott Mayhew return 0; 1226c7559663SScott Mayhew } 1227c7559663SScott Mayhew 12285d47a356STrond Myklebust /* 12291da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12301da177e4SLinus Torvalds * 12311da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12321da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12331da177e4SLinus Torvalds */ 12341da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12351da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12361da177e4SLinus Torvalds { 1237cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1238d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12391da177e4SLinus Torvalds int status = 0; 12401da177e4SLinus Torvalds 124191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 124291d5b470SChuck Lever 12436de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12446de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12451da177e4SLinus Torvalds 1246c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 124749a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12481da177e4SLinus Torvalds offset = 0; 12491da177e4SLinus Torvalds } 12501da177e4SLinus Torvalds 1251e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 125203fa9e84STrond Myklebust if (status < 0) 125303fa9e84STrond Myklebust nfs_set_pageerror(page); 125459b7c05fSTrond Myklebust else 125559b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 12561da177e4SLinus Torvalds 125748186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 12581da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 12591da177e4SLinus Torvalds return status; 12601da177e4SLinus Torvalds } 12611da177e4SLinus Torvalds 12623ff7576dSTrond Myklebust static int flush_task_priority(int how) 12631da177e4SLinus Torvalds { 12641da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 12651da177e4SLinus Torvalds case FLUSH_HIGHPRI: 12661da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 12671da177e4SLinus Torvalds case FLUSH_LOWPRI: 12681da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 12711da177e4SLinus Torvalds } 12721da177e4SLinus Torvalds 1273d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1274d45f60c6SWeston Andros Adamson struct rpc_message *msg, 12751ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 12761da177e4SLinus Torvalds { 1277d45f60c6SWeston Andros Adamson struct inode *inode = hdr->inode; 12783ff7576dSTrond Myklebust int priority = flush_task_priority(how); 12791da177e4SLinus Torvalds 12801ed26f33SAnna Schumaker task_setup_data->priority = priority; 1281d45f60c6SWeston Andros Adamson NFS_PROTO(inode)->write_setup(hdr, msg); 1282d138d5d1SAndy Adamson 12838c21c62cSWeston Andros Adamson nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 1284d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1285275acaafSTrond Myklebust } 1286275acaafSTrond Myklebust 12876d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 12886d884e8fSFred * call this on each, which will prepare them to be retried on next 12896d884e8fSFred * writeback using standard nfs. 12906d884e8fSFred */ 12916d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 12926d884e8fSFred { 12936d884e8fSFred nfs_mark_request_dirty(req); 12941d1afcbcSTrond Myklebust nfs_unlock_request(req); 129520633f04SWeston Andros Adamson nfs_end_page_writeback(req); 12963aff4ebbSTrond Myklebust nfs_release_request(req); 12976d884e8fSFred } 12986d884e8fSFred 1299061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13006c75dc0dSFred Isaman { 13016c75dc0dSFred Isaman struct nfs_page *req; 13026c75dc0dSFred Isaman 13036c75dc0dSFred Isaman while (!list_empty(head)) { 13046c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13056c75dc0dSFred Isaman nfs_list_remove_request(req); 13066c75dc0dSFred Isaman nfs_redirty_request(req); 13076c75dc0dSFred Isaman } 13086c75dc0dSFred Isaman } 13096c75dc0dSFred Isaman 1310061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1311061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1312061ae2edSFred Isaman .completion = nfs_write_completion, 1313061ae2edSFred Isaman }; 1314061ae2edSFred Isaman 131557208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1316a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1317061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 13181751c363STrond Myklebust { 1319a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 132041d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1321a20c93e3SChristoph Hellwig 1322a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1323a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1324a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1325a20c93e3SChristoph Hellwig #endif 13264a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 13274a0de55cSAnna Schumaker server->wsize, ioflags); 13281751c363STrond Myklebust } 1329ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13301751c363STrond Myklebust 1331dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1332dce81290STrond Myklebust { 133341d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1334dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1335dce81290STrond Myklebust } 13361f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1337dce81290STrond Myklebust 13381da177e4SLinus Torvalds 13390b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13400b7c0153SFred Isaman { 13410b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13420b7c0153SFred Isaman 13430b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13440b7c0153SFred Isaman } 13450b7c0153SFred Isaman 1346d45f60c6SWeston Andros Adamson static void nfs_writeback_release_common(struct nfs_pgio_header *hdr) 13471da177e4SLinus Torvalds { 1348c65e6254SWeston Andros Adamson /* do nothing! */ 13491da177e4SLinus Torvalds } 13501da177e4SLinus Torvalds 13511f2edbe3STrond Myklebust /* 13521f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 13531f2edbe3STrond Myklebust */ 13541f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 13551f2edbe3STrond Myklebust { 13561f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 13571f2edbe3STrond Myklebust int kill = 0; 1358788e7a89STrond Myklebust 13591f2edbe3STrond Myklebust /* suid always must be killed */ 13601f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 13611f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 13621f2edbe3STrond Myklebust 13631f2edbe3STrond Myklebust /* 13641f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 13651f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 13661f2edbe3STrond Myklebust */ 13671f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 13681f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 13691f2edbe3STrond Myklebust 13701f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 13711f2edbe3STrond Myklebust return kill; 13721f2edbe3STrond Myklebust 13731f2edbe3STrond Myklebust return 0; 13741f2edbe3STrond Myklebust } 1375788e7a89STrond Myklebust 13761da177e4SLinus Torvalds /* 13771da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 13781da177e4SLinus Torvalds */ 1379d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1380d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 13810eecb214SAnna Schumaker struct inode *inode) 13821da177e4SLinus Torvalds { 1383788e7a89STrond Myklebust int status; 13841da177e4SLinus Torvalds 1385f551e44fSChuck Lever /* 1386f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1387f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1388f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1389f551e44fSChuck Lever * another writer had changed the file, but some applications 1390f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1391f551e44fSChuck Lever */ 1392d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1393788e7a89STrond Myklebust if (status != 0) 13940eecb214SAnna Schumaker return status; 1395d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 139691d5b470SChuck Lever 139789d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 1398d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1399d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 14001da177e4SLinus Torvalds /* We tried a write call, but the server did not 14011da177e4SLinus Torvalds * commit data to stable storage even though we 14021da177e4SLinus Torvalds * requested it. 14031da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 14041da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 14051da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 14061da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 14071da177e4SLinus Torvalds */ 14081da177e4SLinus Torvalds static unsigned long complain; 14091da177e4SLinus Torvalds 1410a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 14111da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 14121da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 14131da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1414cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1415d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 14161da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 14171da177e4SLinus Torvalds } 14181da177e4SLinus Torvalds } 14191da177e4SLinus Torvalds #endif 14201f2edbe3STrond Myklebust 14211f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 14221f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 14231f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 14240eecb214SAnna Schumaker return 0; 14250eecb214SAnna Schumaker } 14260eecb214SAnna Schumaker 14270eecb214SAnna Schumaker /* 14280eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 14290eecb214SAnna Schumaker */ 1430d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1431d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 14320eecb214SAnna Schumaker { 1433d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1434d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 14351f2edbe3STrond Myklebust 14361f2edbe3STrond Myklebust if (resp->count < argp->count) { 14371da177e4SLinus Torvalds static unsigned long complain; 14381da177e4SLinus Torvalds 14396c75dc0dSFred Isaman /* This a short write! */ 1440d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 144191d5b470SChuck Lever 14421da177e4SLinus Torvalds /* Has the server at least made some progress? */ 14436c75dc0dSFred Isaman if (resp->count == 0) { 14446c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 14456c75dc0dSFred Isaman printk(KERN_WARNING 14466c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 14476c75dc0dSFred Isaman argp->count); 14486c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 14496c75dc0dSFred Isaman } 1450d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 14516c75dc0dSFred Isaman task->tk_status = -EIO; 14526c75dc0dSFred Isaman return; 14536c75dc0dSFred Isaman } 14541da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 14551da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 14561da177e4SLinus Torvalds /* Resend from where the server left off */ 1457d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 14581da177e4SLinus Torvalds argp->offset += resp->count; 14591da177e4SLinus Torvalds argp->pgbase += resp->count; 14601da177e4SLinus Torvalds argp->count -= resp->count; 14611da177e4SLinus Torvalds } else { 14621da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 14631da177e4SLinus Torvalds * headaches in the case of a server crash. 14641da177e4SLinus Torvalds */ 14651da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 14661da177e4SLinus Torvalds } 1467d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 14681da177e4SLinus Torvalds } 14691da177e4SLinus Torvalds } 14701da177e4SLinus Torvalds 14711da177e4SLinus Torvalds 147289d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 147371d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 147471d0a611STrond Myklebust { 1475b8413f98STrond Myklebust int ret; 1476b8413f98STrond Myklebust 147771d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 147871d0a611STrond Myklebust return 1; 1479b8413f98STrond Myklebust if (!may_wait) 148071d0a611STrond Myklebust return 0; 1481b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1482b8413f98STrond Myklebust NFS_INO_COMMIT, 1483b8413f98STrond Myklebust nfs_wait_bit_killable, 1484b8413f98STrond Myklebust TASK_KILLABLE); 1485b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 148671d0a611STrond Myklebust } 148771d0a611STrond Myklebust 1488f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 148971d0a611STrond Myklebust { 149071d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 14914e857c58SPeter Zijlstra smp_mb__after_atomic(); 149271d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 149371d0a611STrond Myklebust } 149471d0a611STrond Myklebust 14950b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 14961da177e4SLinus Torvalds { 14970b7c0153SFred Isaman put_nfs_open_context(data->context); 14980b7c0153SFred Isaman nfs_commit_free(data); 14991da177e4SLinus Torvalds } 1500e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15011da177e4SLinus Torvalds 15020b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 15039ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15049f0ec176SAndy Adamson int how, int flags) 15051da177e4SLinus Torvalds { 150607737691STrond Myklebust struct rpc_task *task; 15079ace33cdSFred Isaman int priority = flush_task_priority(how); 1508bdc7f021STrond Myklebust struct rpc_message msg = { 1509bdc7f021STrond Myklebust .rpc_argp = &data->args, 1510bdc7f021STrond Myklebust .rpc_resp = &data->res, 15119ace33cdSFred Isaman .rpc_cred = data->cred, 1512bdc7f021STrond Myklebust }; 151384115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 151407737691STrond Myklebust .task = &data->task, 15159ace33cdSFred Isaman .rpc_client = clnt, 1516bdc7f021STrond Myklebust .rpc_message = &msg, 15179ace33cdSFred Isaman .callback_ops = call_ops, 151884115e1cSTrond Myklebust .callback_data = data, 1519101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 15209f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 15213ff7576dSTrond Myklebust .priority = priority, 152284115e1cSTrond Myklebust }; 1523788e7a89STrond Myklebust /* Set up the initial task struct. */ 15249ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 15251da177e4SLinus Torvalds 1526a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1527bdc7f021STrond Myklebust 15288c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 15298c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 15308c21c62cSWeston Andros Adamson 153107737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1532dbae4c73STrond Myklebust if (IS_ERR(task)) 1533dbae4c73STrond Myklebust return PTR_ERR(task); 1534d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1535d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 153607737691STrond Myklebust rpc_put_task(task); 1537dbae4c73STrond Myklebust return 0; 15381da177e4SLinus Torvalds } 1539e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 15401da177e4SLinus Torvalds 15411da177e4SLinus Torvalds /* 15429ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 15439ace33cdSFred Isaman */ 15440b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1545988b6dceSFred Isaman struct list_head *head, 1546f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1547f453a54aSFred Isaman struct nfs_commit_info *cinfo) 15489ace33cdSFred Isaman { 15499ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 15503d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 15519ace33cdSFred Isaman 15529ace33cdSFred Isaman /* Set up the RPC argument and reply structs 15539ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 15549ace33cdSFred Isaman 15559ace33cdSFred Isaman list_splice_init(head, &data->pages); 15569ace33cdSFred Isaman 15579ace33cdSFred Isaman data->inode = inode; 15589ace33cdSFred Isaman data->cred = first->wb_context->cred; 1559988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 15609ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1561f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1562b359f9d0SFred Isaman data->dreq = cinfo->dreq; 15639ace33cdSFred Isaman 15649ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 15659ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 15669ace33cdSFred Isaman data->args.offset = 0; 15679ace33cdSFred Isaman data->args.count = 0; 15680b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 15699ace33cdSFred Isaman data->res.fattr = &data->fattr; 15709ace33cdSFred Isaman data->res.verf = &data->verf; 15719ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 15729ace33cdSFred Isaman } 1573e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 15749ace33cdSFred Isaman 1575e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1576ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1577ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 157864bfeb49SFred Isaman { 157964bfeb49SFred Isaman struct nfs_page *req; 158064bfeb49SFred Isaman 158164bfeb49SFred Isaman while (!list_empty(page_list)) { 158264bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 158364bfeb49SFred Isaman nfs_list_remove_request(req); 1584ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 158556f9cd68SFred Isaman if (!cinfo->dreq) { 158664bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1587d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 158864bfeb49SFred Isaman BDI_RECLAIMABLE); 158956f9cd68SFred Isaman } 15901d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 159164bfeb49SFred Isaman } 159264bfeb49SFred Isaman } 1593e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 159464bfeb49SFred Isaman 15959ace33cdSFred Isaman /* 15961da177e4SLinus Torvalds * Commit dirty pages 15971da177e4SLinus Torvalds */ 15981da177e4SLinus Torvalds static int 1599ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1600ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 16011da177e4SLinus Torvalds { 16020b7c0153SFred Isaman struct nfs_commit_data *data; 16031da177e4SLinus Torvalds 1604c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 16051da177e4SLinus Torvalds 16061da177e4SLinus Torvalds if (!data) 16071da177e4SLinus Torvalds goto out_bad; 16081da177e4SLinus Torvalds 16091da177e4SLinus Torvalds /* Set up the argument struct */ 1610f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1611f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 16129f0ec176SAndy Adamson return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 16139f0ec176SAndy Adamson how, 0); 16141da177e4SLinus Torvalds out_bad: 1615ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1616f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 16171da177e4SLinus Torvalds return -ENOMEM; 16181da177e4SLinus Torvalds } 16191da177e4SLinus Torvalds 16201da177e4SLinus Torvalds /* 16211da177e4SLinus Torvalds * COMMIT call returned 16221da177e4SLinus Torvalds */ 1623788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 16241da177e4SLinus Torvalds { 16250b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16261da177e4SLinus Torvalds 1627a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 16281da177e4SLinus Torvalds task->tk_pid, task->tk_status); 16291da177e4SLinus Torvalds 1630788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1631c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1632c9d8f89dSTrond Myklebust } 1633c9d8f89dSTrond Myklebust 1634f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1635c9d8f89dSTrond Myklebust { 1636c9d8f89dSTrond Myklebust struct nfs_page *req; 1637c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1638f453a54aSFred Isaman struct nfs_commit_info cinfo; 1639788e7a89STrond Myklebust 16401da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 16411da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 16421da177e4SLinus Torvalds nfs_list_remove_request(req); 1643d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 16441da177e4SLinus Torvalds 16451e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 16463d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 16471e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), 16481da177e4SLinus Torvalds req->wb_bytes, 16491da177e4SLinus Torvalds (long long)req_offset(req)); 1650c9d8f89dSTrond Myklebust if (status < 0) { 1651c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 16521da177e4SLinus Torvalds nfs_inode_remove_request(req); 1653c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 16541da177e4SLinus Torvalds goto next; 16551da177e4SLinus Torvalds } 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 16581da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 16592f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 16601da177e4SLinus Torvalds /* We have a match */ 16611da177e4SLinus Torvalds nfs_inode_remove_request(req); 16621da177e4SLinus Torvalds dprintk(" OK\n"); 16631da177e4SLinus Torvalds goto next; 16641da177e4SLinus Torvalds } 16651da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 16661da177e4SLinus Torvalds dprintk(" mismatch\n"); 16676d884e8fSFred nfs_mark_request_dirty(req); 166805990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 16691da177e4SLinus Torvalds next: 16701d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 16711da177e4SLinus Torvalds } 1672f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1673f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1674f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 16755917ce84SFred Isaman } 16765917ce84SFred Isaman 16775917ce84SFred Isaman static void nfs_commit_release(void *calldata) 16785917ce84SFred Isaman { 16790b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16805917ce84SFred Isaman 1681f453a54aSFred Isaman data->completion_ops->completion(data); 1682c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 16831da177e4SLinus Torvalds } 1684788e7a89STrond Myklebust 1685788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 16860b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1687788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1688788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1689788e7a89STrond Myklebust }; 16901da177e4SLinus Torvalds 1691f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1692f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1693f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1694f453a54aSFred Isaman }; 1695f453a54aSFred Isaman 16961763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1697ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 169884c53ab5SFred Isaman { 169984c53ab5SFred Isaman int status; 170084c53ab5SFred Isaman 1701ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 170284c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1703ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 170484c53ab5SFred Isaman return status; 170584c53ab5SFred Isaman } 170684c53ab5SFred Isaman 1707b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 17081da177e4SLinus Torvalds { 17091da177e4SLinus Torvalds LIST_HEAD(head); 1710ea2cf228SFred Isaman struct nfs_commit_info cinfo; 171171d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1712b8413f98STrond Myklebust int res; 17131da177e4SLinus Torvalds 1714b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1715b8413f98STrond Myklebust if (res <= 0) 1716c5efa5fcSTrond Myklebust goto out_mark_dirty; 1717ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1718ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 17191da177e4SLinus Torvalds if (res) { 1720a861a1e1SFred Isaman int error; 1721a861a1e1SFred Isaman 1722ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 17231da177e4SLinus Torvalds if (error < 0) 17241da177e4SLinus Torvalds return error; 1725b8413f98STrond Myklebust if (!may_wait) 1726b8413f98STrond Myklebust goto out_mark_dirty; 172774316201SNeilBrown error = wait_on_bit_action(&NFS_I(inode)->flags, 1728b8413f98STrond Myklebust NFS_INO_COMMIT, 172971d0a611STrond Myklebust nfs_wait_bit_killable, 173071d0a611STrond Myklebust TASK_KILLABLE); 1731b8413f98STrond Myklebust if (error < 0) 1732b8413f98STrond Myklebust return error; 173371d0a611STrond Myklebust } else 173471d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1735c5efa5fcSTrond Myklebust return res; 1736c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1737c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1738c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1739c5efa5fcSTrond Myklebust * that the data is on the disk. 1740c5efa5fcSTrond Myklebust */ 1741c5efa5fcSTrond Myklebust out_mark_dirty: 1742c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17431da177e4SLinus Torvalds return res; 17441da177e4SLinus Torvalds } 17458fc795f7STrond Myklebust 17468fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17478fc795f7STrond Myklebust { 1748420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1749420e3646STrond Myklebust int flags = FLUSH_SYNC; 1750420e3646STrond Myklebust int ret = 0; 17518fc795f7STrond Myklebust 17523236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1753ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 17543236c3e1SJeff Layton return ret; 17553236c3e1SJeff Layton 1756a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1757a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1758a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1759420e3646STrond Myklebust */ 1760ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1761420e3646STrond Myklebust goto out_mark_dirty; 1762420e3646STrond Myklebust 1763a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1764420e3646STrond Myklebust flags = 0; 1765a00dd6c0SJeff Layton } 1766a00dd6c0SJeff Layton 1767420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1768420e3646STrond Myklebust if (ret >= 0) { 1769420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1770420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1771420e3646STrond Myklebust wbc->nr_to_write -= ret; 1772420e3646STrond Myklebust else 1773420e3646STrond Myklebust wbc->nr_to_write = 0; 1774420e3646STrond Myklebust } 17758fc795f7STrond Myklebust return 0; 1776420e3646STrond Myklebust } 1777420e3646STrond Myklebust out_mark_dirty: 17788fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17798fc795f7STrond Myklebust return ret; 17808fc795f7STrond Myklebust } 1781c63c7b05STrond Myklebust #else 17828fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17838fc795f7STrond Myklebust { 17848fc795f7STrond Myklebust return 0; 17858fc795f7STrond Myklebust } 17861da177e4SLinus Torvalds #endif 17871da177e4SLinus Torvalds 17888fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 17898fc795f7STrond Myklebust { 1790a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1791a8d8f02cSBryan Schumaker } 179289d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1793863a3c6cSAndy Adamson 1794acdc53b2STrond Myklebust /* 1795acdc53b2STrond Myklebust * flush the inode to disk. 1796acdc53b2STrond Myklebust */ 1797acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 179834901f70STrond Myklebust { 179934901f70STrond Myklebust struct writeback_control wbc = { 180072cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 180134901f70STrond Myklebust .nr_to_write = LONG_MAX, 1802d7fb1207STrond Myklebust .range_start = 0, 1803d7fb1207STrond Myklebust .range_end = LLONG_MAX, 180434901f70STrond Myklebust }; 1805f4ce1299STrond Myklebust int ret; 180634901f70STrond Myklebust 1807f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1808f4ce1299STrond Myklebust 1809f4ce1299STrond Myklebust ret = sync_inode(inode, &wbc); 1810f4ce1299STrond Myklebust 1811f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1812f4ce1299STrond Myklebust return ret; 18131c75950bSTrond Myklebust } 1814ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 18151c75950bSTrond Myklebust 18161b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 18171b3b4a1aSTrond Myklebust { 18181b3b4a1aSTrond Myklebust struct nfs_page *req; 18191b3b4a1aSTrond Myklebust int ret = 0; 18201b3b4a1aSTrond Myklebust 1821ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18223e217045SWeston Andros Adamson 18233e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 18243e217045SWeston Andros Adamson * request */ 18253e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 18263e217045SWeston Andros Adamson 18273e217045SWeston Andros Adamson if (IS_ERR(req)) { 18283e217045SWeston Andros Adamson ret = PTR_ERR(req); 18293e217045SWeston Andros Adamson } else if (req) { 18303e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 18313e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 18323e217045SWeston Andros Adamson * request from the inode / page_private pointer and 18333e217045SWeston Andros Adamson * release it */ 18341b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 18351b3b4a1aSTrond Myklebust /* 18361b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 18371b3b4a1aSTrond Myklebust * page as being dirty 18381b3b4a1aSTrond Myklebust */ 18391b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 18401d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18411b3b4a1aSTrond Myklebust } 18423e217045SWeston Andros Adamson 18431b3b4a1aSTrond Myklebust return ret; 18441b3b4a1aSTrond Myklebust } 18451b3b4a1aSTrond Myklebust 18461c75950bSTrond Myklebust /* 18471c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 18481c75950bSTrond Myklebust */ 18491c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 18501c75950bSTrond Myklebust { 185129418aa4SMel Gorman loff_t range_start = page_file_offset(page); 18527f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 18537f2f12d9STrond Myklebust struct writeback_control wbc = { 18547f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 18557f2f12d9STrond Myklebust .nr_to_write = 0, 18567f2f12d9STrond Myklebust .range_start = range_start, 18577f2f12d9STrond Myklebust .range_end = range_end, 18587f2f12d9STrond Myklebust }; 18597f2f12d9STrond Myklebust int ret; 18607f2f12d9STrond Myklebust 1861f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1862f4ce1299STrond Myklebust 18630522f6adSTrond Myklebust for (;;) { 1864ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18657f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 18667f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 18677f2f12d9STrond Myklebust if (ret < 0) 18687f2f12d9STrond Myklebust goto out_error; 18690522f6adSTrond Myklebust continue; 18707f2f12d9STrond Myklebust } 1871f4ce1299STrond Myklebust ret = 0; 18720522f6adSTrond Myklebust if (!PagePrivate(page)) 18730522f6adSTrond Myklebust break; 18740522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 18757f2f12d9STrond Myklebust if (ret < 0) 18767f2f12d9STrond Myklebust goto out_error; 18777f2f12d9STrond Myklebust } 18787f2f12d9STrond Myklebust out_error: 1879f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 18807f2f12d9STrond Myklebust return ret; 18811c75950bSTrond Myklebust } 18821c75950bSTrond Myklebust 1883074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1884074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1885a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1886074cc1deSTrond Myklebust { 18872da95652SJeff Layton /* 18882da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 18892da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 18902da95652SJeff Layton * 18912da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 18922da95652SJeff Layton * that we can safely release the inode reference while holding 18932da95652SJeff Layton * the page lock. 18942da95652SJeff Layton */ 18952da95652SJeff Layton if (PagePrivate(page)) 18962da95652SJeff Layton return -EBUSY; 1897074cc1deSTrond Myklebust 18988c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 18998c209ce7SDavid Howells return -EBUSY; 1900074cc1deSTrond Myklebust 1901a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1902074cc1deSTrond Myklebust } 1903074cc1deSTrond Myklebust #endif 1904074cc1deSTrond Myklebust 1905f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 19061da177e4SLinus Torvalds { 19071da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 19081e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 19091da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 191020c2df83SPaul Mundt NULL); 19111da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 19121da177e4SLinus Torvalds return -ENOMEM; 19131da177e4SLinus Torvalds 191493d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 19151da177e4SLinus Torvalds nfs_wdata_cachep); 19161da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 19173dd4765fSJeff Layton goto out_destroy_write_cache; 19181da177e4SLinus Torvalds 19190b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 19200b7c0153SFred Isaman sizeof(struct nfs_commit_data), 19210b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 19220b7c0153SFred Isaman NULL); 19230b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 19243dd4765fSJeff Layton goto out_destroy_write_mempool; 19250b7c0153SFred Isaman 192693d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 19274c100210SYanchuan Nian nfs_cdata_cachep); 19281da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 19293dd4765fSJeff Layton goto out_destroy_commit_cache; 19301da177e4SLinus Torvalds 193189a09141SPeter Zijlstra /* 193289a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 193389a09141SPeter Zijlstra * 193489a09141SPeter Zijlstra * 64MB: 8192k 193589a09141SPeter Zijlstra * 128MB: 11585k 193689a09141SPeter Zijlstra * 256MB: 16384k 193789a09141SPeter Zijlstra * 512MB: 23170k 193889a09141SPeter Zijlstra * 1GB: 32768k 193989a09141SPeter Zijlstra * 2GB: 46340k 194089a09141SPeter Zijlstra * 4GB: 65536k 194189a09141SPeter Zijlstra * 8GB: 92681k 194289a09141SPeter Zijlstra * 16GB: 131072k 194389a09141SPeter Zijlstra * 194489a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 194589a09141SPeter Zijlstra * Limit the default to 256M 194689a09141SPeter Zijlstra */ 194789a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 194889a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 194989a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 195089a09141SPeter Zijlstra 19511da177e4SLinus Torvalds return 0; 19523dd4765fSJeff Layton 19533dd4765fSJeff Layton out_destroy_commit_cache: 19543dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19553dd4765fSJeff Layton out_destroy_write_mempool: 19563dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 19573dd4765fSJeff Layton out_destroy_write_cache: 19583dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 19593dd4765fSJeff Layton return -ENOMEM; 19601da177e4SLinus Torvalds } 19611da177e4SLinus Torvalds 1962266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 19631da177e4SLinus Torvalds { 19641da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 19653dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19661da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 19671a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 19681da177e4SLinus Torvalds } 19691da177e4SLinus Torvalds 19704a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 1971a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 19724a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 19734a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 1974a4cdda59SAnna Schumaker .rw_release = nfs_writeback_release_common, 19750eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 19760eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 19771ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 19784a0de55cSAnna Schumaker }; 1979