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); 523a3908c8STrond Myklebust static struct nfs_page * 533a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 543a3908c8STrond Myklebust struct page *page); 551da177e4SLinus Torvalds 56e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 573feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 580b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 591da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 601da177e4SLinus Torvalds 610b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 621da177e4SLinus Torvalds { 63192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 6440859d7eSChuck Lever 651da177e4SLinus Torvalds if (p) { 661da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 671da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 681da177e4SLinus Torvalds } 691da177e4SLinus Torvalds return p; 701da177e4SLinus Torvalds } 71e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 721da177e4SLinus Torvalds 730b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 741da177e4SLinus Torvalds { 751da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 761da177e4SLinus Torvalds } 77e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 781da177e4SLinus Torvalds 791e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 803feb2d49STrond Myklebust { 811e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 823feb2d49STrond Myklebust 834a0de55cSAnna Schumaker if (p) 843feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 853feb2d49STrond Myklebust return p; 863feb2d49STrond Myklebust } 873feb2d49STrond Myklebust 881e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 896c75dc0dSFred Isaman { 901e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 913feb2d49STrond Myklebust } 921da177e4SLinus Torvalds 937b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 947b159fc1STrond Myklebust { 957b159fc1STrond Myklebust ctx->error = error; 967b159fc1STrond Myklebust smp_wmb(); 977b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 987b159fc1STrond Myklebust } 997b159fc1STrond Myklebust 10084d3a9a9SWeston Andros Adamson /* 10184d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 10284d3a9a9SWeston Andros Adamson * 10384d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 10484d3a9a9SWeston Andros Adamson * 10584d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 10684d3a9a9SWeston Andros Adamson */ 10729418aa4SMel Gorman static struct nfs_page * 10884d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 109277459d2STrond Myklebust { 110277459d2STrond Myklebust struct nfs_page *req = NULL; 111277459d2STrond Myklebust 11229418aa4SMel Gorman if (PagePrivate(page)) 113277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 11402d1426cSWeston Andros Adamson else if (unlikely(PageSwapCache(page))) 11502d1426cSWeston Andros Adamson req = nfs_page_search_commits_for_head_request_locked(nfsi, 11602d1426cSWeston Andros Adamson page); 11729418aa4SMel Gorman 11884d3a9a9SWeston Andros Adamson if (req) { 11984d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 12029418aa4SMel Gorman kref_get(&req->wb_kref); 12184d3a9a9SWeston Andros Adamson } 12229418aa4SMel Gorman 123277459d2STrond Myklebust return req; 124277459d2STrond Myklebust } 125277459d2STrond Myklebust 12684d3a9a9SWeston Andros Adamson /* 12784d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 12884d3a9a9SWeston Andros Adamson * 12984d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 13084d3a9a9SWeston Andros Adamson */ 13184d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 132277459d2STrond Myklebust { 133d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 134277459d2STrond Myklebust struct nfs_page *req = NULL; 135277459d2STrond Myklebust 136587142f8STrond Myklebust spin_lock(&inode->i_lock); 13784d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 138587142f8STrond Myklebust spin_unlock(&inode->i_lock); 139277459d2STrond Myklebust return req; 140277459d2STrond Myklebust } 141277459d2STrond Myklebust 1421da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1431da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1441da177e4SLinus Torvalds { 145d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 146a3d01454STrond Myklebust loff_t end, i_size; 147a3d01454STrond Myklebust pgoff_t end_index; 1481da177e4SLinus Torvalds 149a3d01454STrond Myklebust spin_lock(&inode->i_lock); 150a3d01454STrond Myklebust i_size = i_size_read(inode); 151a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 152d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 153a3d01454STrond Myklebust goto out; 154d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1551da177e4SLinus Torvalds if (i_size >= end) 156a3d01454STrond Myklebust goto out; 1571da177e4SLinus Torvalds i_size_write(inode, end); 158a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 159a3d01454STrond Myklebust out: 160a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1611da177e4SLinus Torvalds } 1621da177e4SLinus Torvalds 163a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 164a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 165a301b777STrond Myklebust { 166d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 167a301b777STrond Myklebust } 168a301b777STrond Myklebust 169d72ddcbaSWeston Andros Adamson /* 170d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 171d72ddcbaSWeston Andros Adamson * @head - head request of page group 172d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 173d72ddcbaSWeston Andros Adamson * 174d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 175d72ddcbaSWeston Andros Adamson * page offset @page_offset. 176d72ddcbaSWeston Andros Adamson * 177d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 178d72ddcbaSWeston Andros Adamson * match is found. 179d72ddcbaSWeston Andros Adamson * 180d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 181d72ddcbaSWeston Andros Adamson */ 182d72ddcbaSWeston Andros Adamson static struct nfs_page * 183d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 184d72ddcbaSWeston Andros Adamson { 185d72ddcbaSWeston Andros Adamson struct nfs_page *req; 186d72ddcbaSWeston Andros Adamson 187d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 188d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 189d72ddcbaSWeston Andros Adamson 190d72ddcbaSWeston Andros Adamson req = head; 191d72ddcbaSWeston Andros Adamson do { 192d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 193d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 194d72ddcbaSWeston Andros Adamson return req; 195d72ddcbaSWeston Andros Adamson 196d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 197d72ddcbaSWeston Andros Adamson } while (req != head); 198d72ddcbaSWeston Andros Adamson 199d72ddcbaSWeston Andros Adamson return NULL; 200d72ddcbaSWeston Andros Adamson } 201d72ddcbaSWeston Andros Adamson 202d72ddcbaSWeston Andros Adamson /* 203d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 204d72ddcbaSWeston Andros Adamson * @head - head request of page group 205d72ddcbaSWeston Andros Adamson * 206d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 207d72ddcbaSWeston Andros Adamson * returns false otherwise 208d72ddcbaSWeston Andros Adamson */ 209d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 210d72ddcbaSWeston Andros Adamson { 211d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 212d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 213d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 214d72ddcbaSWeston Andros Adamson 215fd2f3a06SWeston Andros Adamson nfs_page_group_lock(req, false); 216d72ddcbaSWeston Andros Adamson 217d72ddcbaSWeston Andros Adamson do { 218d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 219d72ddcbaSWeston Andros Adamson if (tmp) { 220d72ddcbaSWeston Andros Adamson /* no way this should happen */ 221d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 222d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 223d72ddcbaSWeston Andros Adamson } 224d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 225d72ddcbaSWeston Andros Adamson 226d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 227d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 228d72ddcbaSWeston Andros Adamson return pos == len; 229d72ddcbaSWeston Andros Adamson } 230d72ddcbaSWeston Andros Adamson 2311da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2321da177e4SLinus Torvalds * covers the full page. 2331da177e4SLinus Torvalds */ 234d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2351da177e4SLinus Torvalds { 236d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2371da177e4SLinus Torvalds return; 238d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2391da177e4SLinus Torvalds return; 240d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2411da177e4SLinus Torvalds } 2421da177e4SLinus Torvalds 2431da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2441da177e4SLinus Torvalds { 245e87b4c7aSNeilBrown int ret = 0; 2461da177e4SLinus Torvalds if (wbc->for_reclaim) 247c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 248e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 249e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 250b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 251e87b4c7aSNeilBrown ret |= FLUSH_LOWPRI; 252e87b4c7aSNeilBrown return ret; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2551da177e4SLinus Torvalds /* 25689a09141SPeter Zijlstra * NFS congestion control 25789a09141SPeter Zijlstra */ 25889a09141SPeter Zijlstra 25989a09141SPeter Zijlstra int nfs_congestion_kb; 26089a09141SPeter Zijlstra 26189a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 26289a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 26389a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 26489a09141SPeter Zijlstra 265deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 26689a09141SPeter Zijlstra { 267deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2685a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2695a6d41b3STrond Myklebust 270deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 27189a09141SPeter Zijlstra 272277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2738aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2748aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2758aa7e847SJens Axboe BLK_RW_ASYNC); 2768aa7e847SJens Axboe } 27789a09141SPeter Zijlstra } 27889a09141SPeter Zijlstra 27920633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 28089a09141SPeter Zijlstra { 28120633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 28289a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 28389a09141SPeter Zijlstra 28420633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 28520633f04SWeston Andros Adamson return; 28620633f04SWeston Andros Adamson 28720633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 288c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2898aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 29089a09141SPeter Zijlstra } 29189a09141SPeter Zijlstra 292d4581383SWeston Andros Adamson 293d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 294d4581383SWeston Andros Adamson * @req - an nfs request 295d4581383SWeston Andros Adamson * clears all page group related bits from @req 296d4581383SWeston Andros Adamson */ 297d4581383SWeston Andros Adamson static void 298d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 299e261f51fSTrond Myklebust { 300d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 301d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 302d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 303d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 304d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 305d4581383SWeston Andros Adamson } 306d4581383SWeston Andros Adamson 307d4581383SWeston Andros Adamson 308d4581383SWeston Andros Adamson /* 309d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 310d4581383SWeston Andros Adamson * 311d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 312d4581383SWeston Andros Adamson * 313d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 314d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 315d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 316d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 317d4581383SWeston Andros Adamson * 318d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 319d4581383SWeston Andros Adamson * and BOTH will be released before returning. 320d4581383SWeston Andros Adamson * 321d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 322d4581383SWeston Andros Adamson */ 323d4581383SWeston Andros Adamson static int 324d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 325d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 326d4581383SWeston Andros Adamson __releases(&inode->i_lock) 327d4581383SWeston Andros Adamson { 328d4581383SWeston Andros Adamson struct nfs_page *tmp; 329e261f51fSTrond Myklebust int ret; 330e261f51fSTrond Myklebust 331d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 332d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 333d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 334d4581383SWeston Andros Adamson 335d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 336d4581383SWeston Andros Adamson 337d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 338d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 339d4581383SWeston Andros Adamson 340d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 341587142f8STrond Myklebust spin_unlock(&inode->i_lock); 342d4581383SWeston Andros Adamson 343d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 344d4581383SWeston Andros Adamson nfs_release_request(head); 345d4581383SWeston Andros Adamson 346cfb506e1STrond Myklebust if (!nonblock) 347e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 348cfb506e1STrond Myklebust else 349cfb506e1STrond Myklebust ret = -EAGAIN; 350e261f51fSTrond Myklebust nfs_release_request(req); 351d4581383SWeston Andros Adamson 352d4581383SWeston Andros Adamson return ret; 353e261f51fSTrond Myklebust } 354d4581383SWeston Andros Adamson 355d4581383SWeston Andros Adamson /* 356d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 357d4581383SWeston Andros Adamson * 358d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 359d4581383SWeston Andros Adamson * @old_head - the old head of the list 360d4581383SWeston Andros Adamson * 361d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 362d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 363d4581383SWeston Andros Adamson * with a reference held by some other context. 364d4581383SWeston Andros Adamson */ 365d4581383SWeston Andros Adamson static void 366d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 367d4581383SWeston Andros Adamson struct nfs_page *old_head) 368d4581383SWeston Andros Adamson { 369d4581383SWeston Andros Adamson while (destroy_list) { 370d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 371d4581383SWeston Andros Adamson 372d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 373d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 374d4581383SWeston Andros Adamson 375d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 376d4581383SWeston Andros Adamson 377d4581383SWeston Andros Adamson /* make sure old group is not used */ 378d4581383SWeston Andros Adamson subreq->wb_head = subreq; 379d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 380d4581383SWeston Andros Adamson 381d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 382d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 383d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 384d4581383SWeston Andros Adamson 385d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 386d4581383SWeston Andros Adamson /* release ref on old head request */ 387d4581383SWeston Andros Adamson nfs_release_request(old_head); 388d4581383SWeston Andros Adamson 389d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 390d4581383SWeston Andros Adamson 391d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 392d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 393d4581383SWeston Andros Adamson nfs_release_request(subreq); 394d4581383SWeston Andros Adamson else 395d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 396d4581383SWeston Andros Adamson } else { 397d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 398d4581383SWeston Andros Adamson /* zombie requests have already released the last 399d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 400d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 401d4581383SWeston Andros Adamson * group, simply free the request */ 402d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 403d4581383SWeston Andros Adamson nfs_free_request(subreq); 404d4581383SWeston Andros Adamson } 405d4581383SWeston Andros Adamson } 406d4581383SWeston Andros Adamson } 407d4581383SWeston Andros Adamson 408d4581383SWeston Andros Adamson /* 409d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 410d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 411d4581383SWeston Andros Adamson * operations for this page. 412d4581383SWeston Andros Adamson * 413d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 414d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 415d4581383SWeston Andros Adamson * 416d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 417d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 418d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 419d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 420d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 421d4581383SWeston Andros Adamson * 422d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 423d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 424d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 425d4581383SWeston Andros Adamson * error was encountered. 426d4581383SWeston Andros Adamson */ 427d4581383SWeston Andros Adamson static struct nfs_page * 428d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 429d4581383SWeston Andros Adamson { 430d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 431d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 432d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 433d4581383SWeston Andros Adamson unsigned int total_bytes; 434d4581383SWeston Andros Adamson int ret; 435d4581383SWeston Andros Adamson 436d4581383SWeston Andros Adamson try_again: 437d4581383SWeston Andros Adamson total_bytes = 0; 438d4581383SWeston Andros Adamson 439d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 440d4581383SWeston Andros Adamson 441d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 442d4581383SWeston Andros Adamson 443d4581383SWeston Andros Adamson /* 444d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 445d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 446d4581383SWeston Andros Adamson * until the head reference is released. 447d4581383SWeston Andros Adamson */ 448d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 449d4581383SWeston Andros Adamson 450d4581383SWeston Andros Adamson if (!head) { 451587142f8STrond Myklebust spin_unlock(&inode->i_lock); 452d4581383SWeston Andros Adamson return NULL; 453d4581383SWeston Andros Adamson } 454d4581383SWeston Andros Adamson 4557c3af975SWeston Andros Adamson /* holding inode lock, so always make a non-blocking call to try the 4567c3af975SWeston Andros Adamson * page group lock */ 457fd2f3a06SWeston Andros Adamson ret = nfs_page_group_lock(head, true); 45894970014SWeston Andros Adamson if (ret < 0) { 45994970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 4607c3af975SWeston Andros Adamson 4617c3af975SWeston Andros Adamson if (!nonblock && ret == -EAGAIN) { 4627c3af975SWeston Andros Adamson nfs_page_group_lock_wait(head); 4637c3af975SWeston Andros Adamson nfs_release_request(head); 4647c3af975SWeston Andros Adamson goto try_again; 4657c3af975SWeston Andros Adamson } 4667c3af975SWeston Andros Adamson 46794970014SWeston Andros Adamson nfs_release_request(head); 468e7029206SWeston Andros Adamson return ERR_PTR(ret); 46994970014SWeston Andros Adamson } 4707c3af975SWeston Andros Adamson 4717c3af975SWeston Andros Adamson /* lock each request in the page group */ 472d4581383SWeston Andros Adamson subreq = head; 473d4581383SWeston Andros Adamson do { 474d4581383SWeston Andros Adamson /* 475d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 476309a1d65SWeston Andros Adamson * and in order - but may be repeated (mirrored writes). 477d4581383SWeston Andros Adamson */ 478309a1d65SWeston Andros Adamson if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 479d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 480d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 481309a1d65SWeston Andros Adamson } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 482309a1d65SWeston Andros Adamson ((subreq->wb_offset + subreq->wb_bytes) > 483309a1d65SWeston Andros Adamson (head->wb_offset + total_bytes)))) { 484309a1d65SWeston Andros Adamson nfs_page_group_unlock(head); 485309a1d65SWeston Andros Adamson spin_unlock(&inode->i_lock); 486309a1d65SWeston Andros Adamson return ERR_PTR(-EIO); 487309a1d65SWeston Andros Adamson } 488d4581383SWeston Andros Adamson 489d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 490d4581383SWeston Andros Adamson /* releases page group bit lock and 491d4581383SWeston Andros Adamson * inode spin lock and all references */ 492d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 493d4581383SWeston Andros Adamson subreq, nonblock); 494d4581383SWeston Andros Adamson 495d4581383SWeston Andros Adamson if (ret == 0) 496d4581383SWeston Andros Adamson goto try_again; 497d4581383SWeston Andros Adamson 498d4581383SWeston Andros Adamson return ERR_PTR(ret); 499d4581383SWeston Andros Adamson } 500d4581383SWeston Andros Adamson 501d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 502d4581383SWeston Andros Adamson } while (subreq != head); 503d4581383SWeston Andros Adamson 504d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 505d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 506d4581383SWeston Andros Adamson subreq = head; 507d4581383SWeston Andros Adamson do { 508411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 509d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 510d4581383SWeston Andros Adamson } while (subreq != head); 511d4581383SWeston Andros Adamson 512d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 513d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 514d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 515d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 516d4581383SWeston Andros Adamson head->wb_this_page = head; 517d4581383SWeston Andros Adamson 518d4581383SWeston Andros Adamson /* change head request to cover whole range that 519d4581383SWeston Andros Adamson * the former page group covered */ 520d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 521d4581383SWeston Andros Adamson } 522d4581383SWeston Andros Adamson 523d4581383SWeston Andros Adamson /* 524d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 525d4581383SWeston Andros Adamson */ 526d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 527d4581383SWeston Andros Adamson 528d4581383SWeston Andros Adamson /* 529d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 530d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 531d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 532d4581383SWeston Andros Adamson */ 533d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 534d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 535d4581383SWeston Andros Adamson 536d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 537d4581383SWeston Andros Adamson 538411a99adSWeston Andros Adamson /* drop lock to clean uprequests on destroy list */ 539d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 540d4581383SWeston Andros Adamson 541d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 542d4581383SWeston Andros Adamson 543d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 544d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 545d4581383SWeston Andros Adamson return head; 546612c9384STrond Myklebust } 547074cc1deSTrond Myklebust 548074cc1deSTrond Myklebust /* 549074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 550074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 551074cc1deSTrond Myklebust */ 552074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 553cfb506e1STrond Myklebust struct page *page, bool nonblock) 554074cc1deSTrond Myklebust { 555074cc1deSTrond Myklebust struct nfs_page *req; 556074cc1deSTrond Myklebust int ret = 0; 557074cc1deSTrond Myklebust 558d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 559074cc1deSTrond Myklebust if (!req) 560074cc1deSTrond Myklebust goto out; 561074cc1deSTrond Myklebust ret = PTR_ERR(req); 562074cc1deSTrond Myklebust if (IS_ERR(req)) 563074cc1deSTrond Myklebust goto out; 564074cc1deSTrond Myklebust 565deed85e7STrond Myklebust nfs_set_page_writeback(page); 566deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 567074cc1deSTrond Myklebust 568deed85e7STrond Myklebust ret = 0; 569f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 570f8512ad0SFred Isaman nfs_redirty_request(req); 571074cc1deSTrond Myklebust ret = pgio->pg_error; 572f8512ad0SFred Isaman } 573074cc1deSTrond Myklebust out: 574074cc1deSTrond Myklebust return ret; 575e261f51fSTrond Myklebust } 576e261f51fSTrond Myklebust 577f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 578f758c885STrond Myklebust { 579d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 580cfb506e1STrond Myklebust int ret; 581f758c885STrond Myklebust 582f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 5833708f842SNicolas Iooss nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 584f758c885STrond Myklebust 585d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 5861b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 587cfb506e1STrond Myklebust if (ret == -EAGAIN) { 588cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 589cfb506e1STrond Myklebust ret = 0; 590cfb506e1STrond Myklebust } 591cfb506e1STrond Myklebust return ret; 592f758c885STrond Myklebust } 593f758c885STrond Myklebust 594e261f51fSTrond Myklebust /* 5951da177e4SLinus Torvalds * Write an mmapped page to the server. 5961da177e4SLinus Torvalds */ 5974d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 5981da177e4SLinus Torvalds { 599f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 600e261f51fSTrond Myklebust int err; 6011da177e4SLinus Torvalds 602a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 603a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 604f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 605f758c885STrond Myklebust nfs_pageio_complete(&pgio); 606f758c885STrond Myklebust if (err < 0) 6074d770ccfSTrond Myklebust return err; 608f758c885STrond Myklebust if (pgio.pg_error < 0) 609f758c885STrond Myklebust return pgio.pg_error; 610f758c885STrond Myklebust return 0; 6114d770ccfSTrond Myklebust } 6124d770ccfSTrond Myklebust 6134d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6144d770ccfSTrond Myklebust { 615f758c885STrond Myklebust int ret; 6164d770ccfSTrond Myklebust 617f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6181da177e4SLinus Torvalds unlock_page(page); 619f758c885STrond Myklebust return ret; 620f758c885STrond Myklebust } 621f758c885STrond Myklebust 622f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 623f758c885STrond Myklebust { 624f758c885STrond Myklebust int ret; 625f758c885STrond Myklebust 626f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 627f758c885STrond Myklebust unlock_page(page); 628f758c885STrond Myklebust return ret; 6291da177e4SLinus Torvalds } 6301da177e4SLinus Torvalds 6311da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6321da177e4SLinus Torvalds { 6331da177e4SLinus Torvalds struct inode *inode = mapping->host; 63472cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 635c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6361da177e4SLinus Torvalds int err; 6371da177e4SLinus Torvalds 63872cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 63974316201SNeilBrown err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING, 64072cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 64172cb77f4STrond Myklebust if (err) 64272cb77f4STrond Myklebust goto out_err; 64372cb77f4STrond Myklebust 64491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 64591d5b470SChuck Lever 646a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 647a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 648f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 649c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 65072cb77f4STrond Myklebust 65172cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 6524e857c58SPeter Zijlstra smp_mb__after_atomic(); 65372cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 65472cb77f4STrond Myklebust 655f758c885STrond Myklebust if (err < 0) 65672cb77f4STrond Myklebust goto out_err; 65772cb77f4STrond Myklebust err = pgio.pg_error; 65872cb77f4STrond Myklebust if (err < 0) 65972cb77f4STrond Myklebust goto out_err; 660c63c7b05STrond Myklebust return 0; 66172cb77f4STrond Myklebust out_err: 66272cb77f4STrond Myklebust return err; 6631da177e4SLinus Torvalds } 6641da177e4SLinus Torvalds 6651da177e4SLinus Torvalds /* 6661da177e4SLinus Torvalds * Insert a write request into an inode 6671da177e4SLinus Torvalds */ 668d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6691da177e4SLinus Torvalds { 6701da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 671e7d39069STrond Myklebust 6722bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 6732bfc6e56SWeston Andros Adamson 674e7d39069STrond Myklebust /* Lock the request! */ 6757ad84aa9STrond Myklebust nfs_lock_request(req); 676e7d39069STrond Myklebust 677e7d39069STrond Myklebust spin_lock(&inode->i_lock); 678cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 679cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 680a9a4a87aSTrond Myklebust inode->i_version++; 68129418aa4SMel Gorman /* 68229418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 68329418aa4SMel Gorman * with invalidate/truncate. 68429418aa4SMel Gorman */ 68529418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 6862df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 687deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 688277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 68929418aa4SMel Gorman } 690cb1410c7SWeston Andros Adamson nfsi->nrequests++; 69117089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 692cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 693cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 694cb1410c7SWeston Andros Adamson * adding subrequests. */ 69517089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 696c03b4024STrond Myklebust kref_get(&req->wb_kref); 697e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 6981da177e4SLinus Torvalds } 6991da177e4SLinus Torvalds 7001da177e4SLinus Torvalds /* 70189a09141SPeter Zijlstra * Remove a write request from an inode 7021da177e4SLinus Torvalds */ 7031da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7041da177e4SLinus Torvalds { 7052b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7061da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 70720633f04SWeston Andros Adamson struct nfs_page *head; 70820633f04SWeston Andros Adamson 70920633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 71020633f04SWeston Andros Adamson head = req->wb_head; 7111da177e4SLinus Torvalds 712587142f8STrond Myklebust spin_lock(&inode->i_lock); 71320633f04SWeston Andros Adamson if (likely(!PageSwapCache(head->wb_page))) { 71420633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 71520633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 71695905446SNeilBrown smp_mb__after_atomic(); 71795905446SNeilBrown wake_up_page(head->wb_page, PG_private); 71820633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 71929418aa4SMel Gorman } 720cb1410c7SWeston Andros Adamson nfsi->nrequests--; 721cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 722cb1410c7SWeston Andros Adamson } else { 723cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 724cb1410c7SWeston Andros Adamson nfsi->nrequests--; 725587142f8STrond Myklebust spin_unlock(&inode->i_lock); 72620633f04SWeston Andros Adamson } 72717089a29SWeston Andros Adamson 72817089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7291da177e4SLinus Torvalds nfs_release_request(req); 7301da177e4SLinus Torvalds } 7311da177e4SLinus Torvalds 73261822ab5STrond Myklebust static void 7336d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 73461822ab5STrond Myklebust { 73561822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 73661822ab5STrond Myklebust } 73761822ab5STrond Myklebust 7383a3908c8STrond Myklebust /* 7393a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 7403a3908c8STrond Myklebust * 7413a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 7423a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 7433a3908c8STrond Myklebust * 7443a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 7453a3908c8STrond Myklebust */ 7463a3908c8STrond Myklebust static struct nfs_page * 7473a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 7483a3908c8STrond Myklebust struct page *page) 7493a3908c8STrond Myklebust { 7503a3908c8STrond Myklebust struct nfs_page *freq, *t; 7513a3908c8STrond Myklebust struct nfs_commit_info cinfo; 7523a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 7533a3908c8STrond Myklebust 7543a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 7553a3908c8STrond Myklebust 7563a3908c8STrond Myklebust /* search through pnfs commit lists */ 7573a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 7583a3908c8STrond Myklebust if (freq) 7593a3908c8STrond Myklebust return freq->wb_head; 7603a3908c8STrond Myklebust 7613a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 7623a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 7633a3908c8STrond Myklebust if (freq->wb_page == page) 7643a3908c8STrond Myklebust return freq->wb_head; 7653a3908c8STrond Myklebust } 7663a3908c8STrond Myklebust 7673a3908c8STrond Myklebust return NULL; 7683a3908c8STrond Myklebust } 7693a3908c8STrond Myklebust 7708dd37758STrond Myklebust /** 7718dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 7728dd37758STrond Myklebust * @req: pointer to a struct nfs_page 773ea2cf228SFred Isaman * @dst: commit list head 774ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7758dd37758STrond Myklebust * 776ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 7778dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 7788dd37758STrond Myklebust * the MM page stats. 7798dd37758STrond Myklebust * 780ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 7818dd37758STrond Myklebust * holding the nfs_page lock. 7828dd37758STrond Myklebust */ 7838dd37758STrond Myklebust void 784ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 785ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7868dd37758STrond Myklebust { 7878dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 788ea2cf228SFred Isaman spin_lock(cinfo->lock); 789ea2cf228SFred Isaman nfs_list_add_request(req, dst); 790ea2cf228SFred Isaman cinfo->mds->ncommit++; 791ea2cf228SFred Isaman spin_unlock(cinfo->lock); 792d15bc38dSTom Haynes if (!cinfo->dreq) 793d15bc38dSTom Haynes nfs_mark_page_unstable(req->wb_page); 7948dd37758STrond Myklebust } 7958dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 7968dd37758STrond Myklebust 7978dd37758STrond Myklebust /** 7988dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 7998dd37758STrond Myklebust * @req: pointer to a nfs_page 800ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8018dd37758STrond Myklebust * 802ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8038dd37758STrond Myklebust * number of outstanding requests requiring a commit 8048dd37758STrond Myklebust * It does not update the MM page stats. 8058dd37758STrond Myklebust * 806ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8078dd37758STrond Myklebust */ 8088dd37758STrond Myklebust void 809ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 810ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8118dd37758STrond Myklebust { 8128dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 8138dd37758STrond Myklebust return; 8148dd37758STrond Myklebust nfs_list_remove_request(req); 815ea2cf228SFred Isaman cinfo->mds->ncommit--; 8168dd37758STrond Myklebust } 8178dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 8188dd37758STrond Myklebust 819ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 820ea2cf228SFred Isaman struct inode *inode) 821ea2cf228SFred Isaman { 822ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 823ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 824ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 825b359f9d0SFred Isaman cinfo->dreq = NULL; 826f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 827ea2cf228SFred Isaman } 828ea2cf228SFred Isaman 829ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 830ea2cf228SFred Isaman struct inode *inode, 831ea2cf228SFred Isaman struct nfs_direct_req *dreq) 832ea2cf228SFred Isaman { 8331763da12SFred Isaman if (dreq) 8341763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8351763da12SFred Isaman else 836ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 837ea2cf228SFred Isaman } 838ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8398dd37758STrond Myklebust 8401da177e4SLinus Torvalds /* 8411da177e4SLinus Torvalds * Add a request to the inode's commit list. 8421da177e4SLinus Torvalds */ 8431763da12SFred Isaman void 844ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 845b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 8461da177e4SLinus Torvalds { 847b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 8488dd37758STrond Myklebust return; 849ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 8501da177e4SLinus Torvalds } 8518e821cadSTrond Myklebust 852d6d6dc7cSFred Isaman static void 853d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 854e468bae9STrond Myklebust { 855e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 856de1414a6SChristoph Hellwig dec_bdi_stat(inode_to_bdi(page_file_mapping(page)->host), BDI_RECLAIMABLE); 857e468bae9STrond Myklebust } 858d6d6dc7cSFred Isaman 859411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 8608dd37758STrond Myklebust static void 861d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 862d6d6dc7cSFred Isaman { 8638dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 8642b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 865ea2cf228SFred Isaman struct nfs_commit_info cinfo; 866d6d6dc7cSFred Isaman 867ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 868ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 869ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 870d6d6dc7cSFred Isaman } 8718dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 8728dd37758STrond Myklebust } 873e468bae9STrond Myklebust } 874e468bae9STrond Myklebust 875d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 8768e821cadSTrond Myklebust { 877c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 878d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 879c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 8808e821cadSTrond Myklebust } 8818e821cadSTrond Myklebust 882061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 8836c75dc0dSFred Isaman { 884ea2cf228SFred Isaman struct nfs_commit_info cinfo; 8856c75dc0dSFred Isaman unsigned long bytes = 0; 8866c75dc0dSFred Isaman 8876c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 8886c75dc0dSFred Isaman goto out; 889ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 8906c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 8916c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 8926c75dc0dSFred Isaman 8936c75dc0dSFred Isaman bytes += req->wb_bytes; 8946c75dc0dSFred Isaman nfs_list_remove_request(req); 8956c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 8966c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 897d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 8986c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 8996c75dc0dSFred Isaman goto remove_req; 9006c75dc0dSFred Isaman } 901c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 902f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 903b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 904a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 9056c75dc0dSFred Isaman goto next; 9066c75dc0dSFred Isaman } 9076c75dc0dSFred Isaman remove_req: 9086c75dc0dSFred Isaman nfs_inode_remove_request(req); 9096c75dc0dSFred Isaman next: 9101d1afcbcSTrond Myklebust nfs_unlock_request(req); 91120633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9123aff4ebbSTrond Myklebust nfs_release_request(req); 9136c75dc0dSFred Isaman } 9146c75dc0dSFred Isaman out: 9156c75dc0dSFred Isaman hdr->release(hdr); 9166c75dc0dSFred Isaman } 9176c75dc0dSFred Isaman 918ce59515cSAnna Schumaker unsigned long 919ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 920fb8a1f11STrond Myklebust { 921ea2cf228SFred Isaman return cinfo->mds->ncommit; 922fb8a1f11STrond Myklebust } 923fb8a1f11STrond Myklebust 924ea2cf228SFred Isaman /* cinfo->lock held by caller */ 9251763da12SFred Isaman int 926ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 927ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 928d6d6dc7cSFred Isaman { 929d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 930d6d6dc7cSFred Isaman int ret = 0; 931d6d6dc7cSFred Isaman 932d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9338dd37758STrond Myklebust if (!nfs_lock_request(req)) 9348dd37758STrond Myklebust continue; 9357ad84aa9STrond Myklebust kref_get(&req->wb_kref); 936ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 9373b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 938ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9398dd37758STrond Myklebust nfs_list_add_request(req, dst); 940d6d6dc7cSFred Isaman ret++; 9411763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 942d6d6dc7cSFred Isaman break; 943d6d6dc7cSFred Isaman } 944d6d6dc7cSFred Isaman return ret; 945d6d6dc7cSFred Isaman } 946d6d6dc7cSFred Isaman 9471da177e4SLinus Torvalds /* 9481da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9491da177e4SLinus Torvalds * @inode: NFS inode to scan 950ea2cf228SFred Isaman * @dst: mds destination list 951ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9521da177e4SLinus Torvalds * 9531da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 9541da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 9551da177e4SLinus Torvalds */ 9561763da12SFred Isaman int 957ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 958ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9591da177e4SLinus Torvalds { 960d6d6dc7cSFred Isaman int ret = 0; 961fb8a1f11STrond Myklebust 962ea2cf228SFred Isaman spin_lock(cinfo->lock); 963ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 9648dd37758STrond Myklebust const int max = INT_MAX; 965d6d6dc7cSFred Isaman 966ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 967ea2cf228SFred Isaman cinfo, max); 968ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 969d6d6dc7cSFred Isaman } 970ea2cf228SFred Isaman spin_unlock(cinfo->lock); 971ff778d02STrond Myklebust return ret; 9721da177e4SLinus Torvalds } 973d6d6dc7cSFred Isaman 9741da177e4SLinus Torvalds /* 975e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 976e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 9771da177e4SLinus Torvalds * 978e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 979e7d39069STrond Myklebust * to disk. 9801da177e4SLinus Torvalds */ 981e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 982e7d39069STrond Myklebust struct page *page, 983e7d39069STrond Myklebust unsigned int offset, 984e7d39069STrond Myklebust unsigned int bytes) 9851da177e4SLinus Torvalds { 986e7d39069STrond Myklebust struct nfs_page *req; 987e7d39069STrond Myklebust unsigned int rqend; 988e7d39069STrond Myklebust unsigned int end; 9891da177e4SLinus Torvalds int error; 990277459d2STrond Myklebust 991e7d39069STrond Myklebust if (!PagePrivate(page)) 992e7d39069STrond Myklebust return NULL; 993e7d39069STrond Myklebust 994e7d39069STrond Myklebust end = offset + bytes; 995e7d39069STrond Myklebust spin_lock(&inode->i_lock); 996e7d39069STrond Myklebust 997e7d39069STrond Myklebust for (;;) { 99884d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 999e7d39069STrond Myklebust if (req == NULL) 1000e7d39069STrond Myklebust goto out_unlock; 1001e7d39069STrond Myklebust 10022bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10032bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10042bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10052bfc6e56SWeston Andros Adamson 1006e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1007e7d39069STrond Myklebust /* 1008e7d39069STrond Myklebust * Tell the caller to flush out the request if 1009e7d39069STrond Myklebust * the offsets are non-contiguous. 1010e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1011e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1012e7d39069STrond Myklebust */ 1013e468bae9STrond Myklebust if (offset > rqend 1014e7d39069STrond Myklebust || end < req->wb_offset) 1015e7d39069STrond Myklebust goto out_flushme; 1016e7d39069STrond Myklebust 10177ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1018e7d39069STrond Myklebust break; 1019e7d39069STrond Myklebust 1020e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1021587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10221da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10231da177e4SLinus Torvalds nfs_release_request(req); 1024e7d39069STrond Myklebust if (error != 0) 1025e7d39069STrond Myklebust goto out_err; 1026e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10301da177e4SLinus Torvalds if (offset < req->wb_offset) { 10311da177e4SLinus Torvalds req->wb_offset = offset; 10321da177e4SLinus Torvalds req->wb_pgbase = offset; 10331da177e4SLinus Torvalds } 10341da177e4SLinus Torvalds if (end > rqend) 10351da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1036e7d39069STrond Myklebust else 1037e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1038e7d39069STrond Myklebust out_unlock: 1039ca138f36SFred Isaman if (req) 10408dd37758STrond Myklebust nfs_clear_request_commit(req); 1041411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1042e7d39069STrond Myklebust return req; 1043e7d39069STrond Myklebust out_flushme: 1044e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1045e7d39069STrond Myklebust nfs_release_request(req); 1046e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1047e7d39069STrond Myklebust out_err: 1048e7d39069STrond Myklebust return ERR_PTR(error); 1049e7d39069STrond Myklebust } 10501da177e4SLinus Torvalds 1051e7d39069STrond Myklebust /* 1052e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1053e7d39069STrond Myklebust * 1054e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1055e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1056e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1057e7d39069STrond Myklebust */ 1058e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1059e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1060e7d39069STrond Myklebust { 1061d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1062e7d39069STrond Myklebust struct nfs_page *req; 1063e7d39069STrond Myklebust 1064e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1065e7d39069STrond Myklebust if (req != NULL) 1066e7d39069STrond Myklebust goto out; 10672bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1068e7d39069STrond Myklebust if (IS_ERR(req)) 1069e7d39069STrond Myklebust goto out; 1070d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1071efc91ed0STrond Myklebust out: 107261e930a9STrond Myklebust return req; 10731da177e4SLinus Torvalds } 10741da177e4SLinus Torvalds 1075e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1076e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1077e7d39069STrond Myklebust { 1078e7d39069STrond Myklebust struct nfs_page *req; 1079e7d39069STrond Myklebust 1080e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1081e7d39069STrond Myklebust if (IS_ERR(req)) 1082e7d39069STrond Myklebust return PTR_ERR(req); 1083e7d39069STrond Myklebust /* Update file length */ 1084e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1085d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1086a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 10871d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1088e7d39069STrond Myklebust return 0; 1089e7d39069STrond Myklebust } 1090e7d39069STrond Myklebust 10911da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 10921da177e4SLinus Torvalds { 1093cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 10942a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1095bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 10961da177e4SLinus Torvalds struct nfs_page *req; 10971a54533eSTrond Myklebust int do_flush, status; 10981da177e4SLinus Torvalds /* 10991da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11001da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11011da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11021da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11031da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11041da177e4SLinus Torvalds * dropped page. 11051da177e4SLinus Torvalds */ 11061a54533eSTrond Myklebust do { 110784d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11081a54533eSTrond Myklebust if (req == NULL) 11091a54533eSTrond Myklebust return 0; 11102a369153STrond Myklebust l_ctx = req->wb_lock_context; 11112a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 11122bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11132bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1114bd61e0a9SJeff Layton if (l_ctx && flctx && 1115bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1116bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 11175263e31eSJeff Layton do_flush |= l_ctx->lockowner.l_owner != current->files 11185263e31eSJeff Layton || l_ctx->lockowner.l_pid != current->tgid; 11195263e31eSJeff Layton } 11201da177e4SLinus Torvalds nfs_release_request(req); 11211a54533eSTrond Myklebust if (!do_flush) 11221a54533eSTrond Myklebust return 0; 1123d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11241a54533eSTrond Myklebust } while (status == 0); 11251a54533eSTrond Myklebust return status; 11261da177e4SLinus Torvalds } 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds /* 1129dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1130dc24826bSAndy Adamson * expire soon. 1131dc24826bSAndy Adamson * 1132dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1133dc24826bSAndy Adamson * 1134dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1135dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1136dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1137dc24826bSAndy Adamson */ 1138dc24826bSAndy Adamson int 1139dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1140dc24826bSAndy Adamson { 1141dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1142dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1143dc24826bSAndy Adamson 1144dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1145dc24826bSAndy Adamson } 1146dc24826bSAndy Adamson 1147dc24826bSAndy Adamson /* 1148dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1149dc24826bSAndy Adamson */ 1150dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 1151dc24826bSAndy Adamson { 1152dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 1153dc24826bSAndy Adamson } 1154dc24826bSAndy Adamson 1155dc24826bSAndy Adamson /* 11565d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 11575d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 11585d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 11595d47a356STrond Myklebust */ 11608d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 11615d47a356STrond Myklebust { 1162d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1163d529ef83SJeff Layton 11648d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 11658d197a56STrond Myklebust goto out; 116618dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1167d529ef83SJeff Layton return false; 11684db72b40SJeff Layton smp_rmb(); 1169d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 11708d197a56STrond Myklebust return false; 11718d197a56STrond Myklebust out: 117218dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 117318dd78c4SScott Mayhew return false; 11748d197a56STrond Myklebust return PageUptodate(page) != 0; 11755d47a356STrond Myklebust } 11765d47a356STrond Myklebust 11775263e31eSJeff Layton static bool 11785263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 11795263e31eSJeff Layton { 11805263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 11815263e31eSJeff Layton fl->fl_type == F_WRLCK; 11825263e31eSJeff Layton } 11835263e31eSJeff Layton 1184c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1185c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1186c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1187c7559663SScott Mayhew * inefficiencies. 1188c7559663SScott Mayhew * 1189263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1190263b4509SScott Mayhew * of the checks. 1191c7559663SScott Mayhew */ 1192c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1193c7559663SScott Mayhew { 11945263e31eSJeff Layton int ret; 11955263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 11965263e31eSJeff Layton struct file_lock *fl; 11975263e31eSJeff Layton 1198c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1199c7559663SScott Mayhew return 0; 1200263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1201263b4509SScott Mayhew return 0; 1202c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1203c7559663SScott Mayhew return 1; 1204bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1205bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 1206c7559663SScott Mayhew return 0; 12075263e31eSJeff Layton 12085263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12095263e31eSJeff Layton ret = 0; 12106109c850SJeff Layton spin_lock(&flctx->flc_lock); 1211bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1212bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1213bd61e0a9SJeff Layton fl_list); 1214bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12155263e31eSJeff Layton ret = 1; 1216bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 12175263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 12185263e31eSJeff Layton fl_list); 12195263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 12205263e31eSJeff Layton ret = 1; 12215263e31eSJeff Layton } 12226109c850SJeff Layton spin_unlock(&flctx->flc_lock); 12235263e31eSJeff Layton return ret; 1224c7559663SScott Mayhew } 1225c7559663SScott Mayhew 12265d47a356STrond Myklebust /* 12271da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12281da177e4SLinus Torvalds * 12291da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12301da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12311da177e4SLinus Torvalds */ 12321da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12331da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12341da177e4SLinus Torvalds { 1235cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1236d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12371da177e4SLinus Torvalds int status = 0; 12381da177e4SLinus Torvalds 123991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 124091d5b470SChuck Lever 12416de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12426de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12431da177e4SLinus Torvalds 1244c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 124549a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12461da177e4SLinus Torvalds offset = 0; 12471da177e4SLinus Torvalds } 12481da177e4SLinus Torvalds 1249e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 125003fa9e84STrond Myklebust if (status < 0) 125103fa9e84STrond Myklebust nfs_set_pageerror(page); 125259b7c05fSTrond Myklebust else 125359b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 12541da177e4SLinus Torvalds 125548186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 12561da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 12571da177e4SLinus Torvalds return status; 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds 12603ff7576dSTrond Myklebust static int flush_task_priority(int how) 12611da177e4SLinus Torvalds { 12621da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 12631da177e4SLinus Torvalds case FLUSH_HIGHPRI: 12641da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 12651da177e4SLinus Torvalds case FLUSH_LOWPRI: 12661da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 12671da177e4SLinus Torvalds } 12681da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 1271d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1272d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1273abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 12741ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 12751da177e4SLinus Torvalds { 12763ff7576dSTrond Myklebust int priority = flush_task_priority(how); 12771da177e4SLinus Torvalds 12781ed26f33SAnna Schumaker task_setup_data->priority = priority; 1279abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1280d138d5d1SAndy Adamson 1281abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1282d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1283275acaafSTrond Myklebust } 1284275acaafSTrond Myklebust 12856d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 12866d884e8fSFred * call this on each, which will prepare them to be retried on next 12876d884e8fSFred * writeback using standard nfs. 12886d884e8fSFred */ 12896d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 12906d884e8fSFred { 12916d884e8fSFred nfs_mark_request_dirty(req); 12921d1afcbcSTrond Myklebust nfs_unlock_request(req); 129320633f04SWeston Andros Adamson nfs_end_page_writeback(req); 12943aff4ebbSTrond Myklebust nfs_release_request(req); 12956d884e8fSFred } 12966d884e8fSFred 1297061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 12986c75dc0dSFred Isaman { 12996c75dc0dSFred Isaman struct nfs_page *req; 13006c75dc0dSFred Isaman 13016c75dc0dSFred Isaman while (!list_empty(head)) { 13026c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13036c75dc0dSFred Isaman nfs_list_remove_request(req); 13046c75dc0dSFred Isaman nfs_redirty_request(req); 13056c75dc0dSFred Isaman } 13066c75dc0dSFred Isaman } 13076c75dc0dSFred Isaman 1308061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1309061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1310061ae2edSFred Isaman .completion = nfs_write_completion, 1311061ae2edSFred Isaman }; 1312061ae2edSFred Isaman 131357208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1314a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1315061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 13161751c363STrond Myklebust { 1317a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 131841d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1319a20c93e3SChristoph Hellwig 1320a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1321a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1322a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1323a20c93e3SChristoph Hellwig #endif 13244a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 13254a0de55cSAnna Schumaker server->wsize, ioflags); 13261751c363STrond Myklebust } 1327ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13281751c363STrond Myklebust 1329dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1330dce81290STrond Myklebust { 1331a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1332a7d42ddbSWeston Andros Adamson 133341d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1334a7d42ddbSWeston Andros Adamson 1335a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1336a7d42ddbSWeston Andros Adamson 1337a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1338a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1339dce81290STrond Myklebust } 13401f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1341dce81290STrond Myklebust 13421da177e4SLinus Torvalds 13430b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13440b7c0153SFred Isaman { 13450b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13460b7c0153SFred Isaman 13470b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13480b7c0153SFred Isaman } 13490b7c0153SFred Isaman 1350d45f60c6SWeston Andros Adamson static void nfs_writeback_release_common(struct nfs_pgio_header *hdr) 13511da177e4SLinus Torvalds { 1352c65e6254SWeston Andros Adamson /* do nothing! */ 13531da177e4SLinus Torvalds } 13541da177e4SLinus Torvalds 13551f2edbe3STrond Myklebust /* 13561f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 13571f2edbe3STrond Myklebust */ 13581f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 13591f2edbe3STrond Myklebust { 13601f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 13611f2edbe3STrond Myklebust int kill = 0; 1362788e7a89STrond Myklebust 13631f2edbe3STrond Myklebust /* suid always must be killed */ 13641f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 13651f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 13661f2edbe3STrond Myklebust 13671f2edbe3STrond Myklebust /* 13681f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 13691f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 13701f2edbe3STrond Myklebust */ 13711f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 13721f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 13731f2edbe3STrond Myklebust 13741f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 13751f2edbe3STrond Myklebust return kill; 13761f2edbe3STrond Myklebust 13771f2edbe3STrond Myklebust return 0; 13781f2edbe3STrond Myklebust } 1379788e7a89STrond Myklebust 1380a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1381a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1382a08a8cd3STrond Myklebust { 1383a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1384a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 1385a08a8cd3STrond Myklebust 1386a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 1387a08a8cd3STrond Myklebust return; 1388a08a8cd3STrond Myklebust if (argp->offset + resp->count != fattr->size) 1389a08a8cd3STrond Myklebust return; 1390a08a8cd3STrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) 1391a08a8cd3STrond Myklebust return; 1392a08a8cd3STrond Myklebust /* Set attribute barrier */ 1393a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 1394a08a8cd3STrond Myklebust } 1395a08a8cd3STrond Myklebust 1396a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1397a08a8cd3STrond Myklebust { 1398a08a8cd3STrond Myklebust struct nfs_fattr *fattr = hdr->res.fattr; 1399a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1400a08a8cd3STrond Myklebust 1401a08a8cd3STrond Myklebust if (fattr == NULL) 1402a08a8cd3STrond Myklebust return; 1403a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1404a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1405a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1406a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1407a08a8cd3STrond Myklebust } 1408a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1409a08a8cd3STrond Myklebust 14101da177e4SLinus Torvalds /* 14111da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 14121da177e4SLinus Torvalds */ 1413d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1414d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 14150eecb214SAnna Schumaker struct inode *inode) 14161da177e4SLinus Torvalds { 1417788e7a89STrond Myklebust int status; 14181da177e4SLinus Torvalds 1419f551e44fSChuck Lever /* 1420f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1421f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1422f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1423f551e44fSChuck Lever * another writer had changed the file, but some applications 1424f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1425f551e44fSChuck Lever */ 1426d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1427788e7a89STrond Myklebust if (status != 0) 14280eecb214SAnna Schumaker return status; 1429d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 143091d5b470SChuck Lever 1431d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1432d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 14331da177e4SLinus Torvalds /* We tried a write call, but the server did not 14341da177e4SLinus Torvalds * commit data to stable storage even though we 14351da177e4SLinus Torvalds * requested it. 14361da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 14371da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 14381da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 14391da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 14401da177e4SLinus Torvalds */ 14411da177e4SLinus Torvalds static unsigned long complain; 14421da177e4SLinus Torvalds 1443a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 14441da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 14451da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 14461da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1447cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1448d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 14491da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 14501da177e4SLinus Torvalds } 14511da177e4SLinus Torvalds } 14521f2edbe3STrond Myklebust 14531f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 14541f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 14551f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 14560eecb214SAnna Schumaker return 0; 14570eecb214SAnna Schumaker } 14580eecb214SAnna Schumaker 14590eecb214SAnna Schumaker /* 14600eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 14610eecb214SAnna Schumaker */ 1462d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1463d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 14640eecb214SAnna Schumaker { 1465d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1466d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 14671f2edbe3STrond Myklebust 14681f2edbe3STrond Myklebust if (resp->count < argp->count) { 14691da177e4SLinus Torvalds static unsigned long complain; 14701da177e4SLinus Torvalds 14716c75dc0dSFred Isaman /* This a short write! */ 1472d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 147391d5b470SChuck Lever 14741da177e4SLinus Torvalds /* Has the server at least made some progress? */ 14756c75dc0dSFred Isaman if (resp->count == 0) { 14766c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 14776c75dc0dSFred Isaman printk(KERN_WARNING 14786c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 14796c75dc0dSFred Isaman argp->count); 14806c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 14816c75dc0dSFred Isaman } 1482d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 14836c75dc0dSFred Isaman task->tk_status = -EIO; 14846c75dc0dSFred Isaman return; 14856c75dc0dSFred Isaman } 14861da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 14871da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 14881da177e4SLinus Torvalds /* Resend from where the server left off */ 1489d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 14901da177e4SLinus Torvalds argp->offset += resp->count; 14911da177e4SLinus Torvalds argp->pgbase += resp->count; 14921da177e4SLinus Torvalds argp->count -= resp->count; 14931da177e4SLinus Torvalds } else { 14941da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 14951da177e4SLinus Torvalds * headaches in the case of a server crash. 14961da177e4SLinus Torvalds */ 14971da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 14981da177e4SLinus Torvalds } 1499d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 15001da177e4SLinus Torvalds } 15011da177e4SLinus Torvalds } 15021da177e4SLinus Torvalds 15031da177e4SLinus Torvalds 150471d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 150571d0a611STrond Myklebust { 1506b8413f98STrond Myklebust int ret; 1507b8413f98STrond Myklebust 150871d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 150971d0a611STrond Myklebust return 1; 1510b8413f98STrond Myklebust if (!may_wait) 151171d0a611STrond Myklebust return 0; 1512b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1513b8413f98STrond Myklebust NFS_INO_COMMIT, 1514b8413f98STrond Myklebust nfs_wait_bit_killable, 1515b8413f98STrond Myklebust TASK_KILLABLE); 1516b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 151771d0a611STrond Myklebust } 151871d0a611STrond Myklebust 1519f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 152071d0a611STrond Myklebust { 152171d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 15224e857c58SPeter Zijlstra smp_mb__after_atomic(); 152371d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 152471d0a611STrond Myklebust } 152571d0a611STrond Myklebust 15260b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 15271da177e4SLinus Torvalds { 15280b7c0153SFred Isaman put_nfs_open_context(data->context); 15290b7c0153SFred Isaman nfs_commit_free(data); 15301da177e4SLinus Torvalds } 1531e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15321da177e4SLinus Torvalds 15330b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1534c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 15359ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15369f0ec176SAndy Adamson int how, int flags) 15371da177e4SLinus Torvalds { 153807737691STrond Myklebust struct rpc_task *task; 15399ace33cdSFred Isaman int priority = flush_task_priority(how); 1540bdc7f021STrond Myklebust struct rpc_message msg = { 1541bdc7f021STrond Myklebust .rpc_argp = &data->args, 1542bdc7f021STrond Myklebust .rpc_resp = &data->res, 15439ace33cdSFred Isaman .rpc_cred = data->cred, 1544bdc7f021STrond Myklebust }; 154584115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 154607737691STrond Myklebust .task = &data->task, 15479ace33cdSFred Isaman .rpc_client = clnt, 1548bdc7f021STrond Myklebust .rpc_message = &msg, 15499ace33cdSFred Isaman .callback_ops = call_ops, 155084115e1cSTrond Myklebust .callback_data = data, 1551101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 15529f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 15533ff7576dSTrond Myklebust .priority = priority, 155484115e1cSTrond Myklebust }; 1555788e7a89STrond Myklebust /* Set up the initial task struct. */ 1556c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 15571da177e4SLinus Torvalds 1558a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1559bdc7f021STrond Myklebust 15608c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 15618c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 15628c21c62cSWeston Andros Adamson 156307737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1564dbae4c73STrond Myklebust if (IS_ERR(task)) 1565dbae4c73STrond Myklebust return PTR_ERR(task); 1566d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1567d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 156807737691STrond Myklebust rpc_put_task(task); 1569dbae4c73STrond Myklebust return 0; 15701da177e4SLinus Torvalds } 1571e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 15721da177e4SLinus Torvalds 1573378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1574378520b8SPeng Tao { 1575378520b8SPeng Tao loff_t lwb = 0; 1576378520b8SPeng Tao struct nfs_page *req; 1577378520b8SPeng Tao 1578378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1579378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1580378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1581378520b8SPeng Tao 1582378520b8SPeng Tao return lwb; 1583378520b8SPeng Tao } 1584378520b8SPeng Tao 15851da177e4SLinus Torvalds /* 15869ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 15879ace33cdSFred Isaman */ 15880b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1589988b6dceSFred Isaman struct list_head *head, 1590f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1591f453a54aSFred Isaman struct nfs_commit_info *cinfo) 15929ace33cdSFred Isaman { 15939ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 15942b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 15959ace33cdSFred Isaman 15969ace33cdSFred Isaman /* Set up the RPC argument and reply structs 15979ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 15989ace33cdSFred Isaman 15999ace33cdSFred Isaman list_splice_init(head, &data->pages); 16009ace33cdSFred Isaman 16019ace33cdSFred Isaman data->inode = inode; 16029ace33cdSFred Isaman data->cred = first->wb_context->cred; 1603988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1604378520b8SPeng Tao /* only set lwb for pnfs commit */ 1605378520b8SPeng Tao if (lseg) 1606378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 16079ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1608f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1609b359f9d0SFred Isaman data->dreq = cinfo->dreq; 16109ace33cdSFred Isaman 16119ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 16129ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 16139ace33cdSFred Isaman data->args.offset = 0; 16149ace33cdSFred Isaman data->args.count = 0; 16150b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 16169ace33cdSFred Isaman data->res.fattr = &data->fattr; 16179ace33cdSFred Isaman data->res.verf = &data->verf; 16189ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 16199ace33cdSFred Isaman } 1620e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 16219ace33cdSFred Isaman 1622e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1623ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1624b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1625b57ff130SWeston Andros Adamson u32 ds_commit_idx) 162664bfeb49SFred Isaman { 162764bfeb49SFred Isaman struct nfs_page *req; 162864bfeb49SFred Isaman 162964bfeb49SFred Isaman while (!list_empty(page_list)) { 163064bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 163164bfeb49SFred Isaman nfs_list_remove_request(req); 1632b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1633487b9b8aSTom Haynes if (!cinfo->dreq) 1634487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 16351d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 163664bfeb49SFred Isaman } 163764bfeb49SFred Isaman } 1638e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 163964bfeb49SFred Isaman 16409ace33cdSFred Isaman /* 16411da177e4SLinus Torvalds * Commit dirty pages 16421da177e4SLinus Torvalds */ 16431da177e4SLinus Torvalds static int 1644ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1645ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 16461da177e4SLinus Torvalds { 16470b7c0153SFred Isaman struct nfs_commit_data *data; 16481da177e4SLinus Torvalds 1649c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 16501da177e4SLinus Torvalds 16511da177e4SLinus Torvalds if (!data) 16521da177e4SLinus Torvalds goto out_bad; 16531da177e4SLinus Torvalds 16541da177e4SLinus Torvalds /* Set up the argument struct */ 1655f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1656f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1657c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1658c36aae9aSPeng Tao data->mds_ops, how, 0); 16591da177e4SLinus Torvalds out_bad: 1660b57ff130SWeston Andros Adamson nfs_retry_commit(head, NULL, cinfo, 0); 1661f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 16621da177e4SLinus Torvalds return -ENOMEM; 16631da177e4SLinus Torvalds } 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds /* 16661da177e4SLinus Torvalds * COMMIT call returned 16671da177e4SLinus Torvalds */ 1668788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 16691da177e4SLinus Torvalds { 16700b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16711da177e4SLinus Torvalds 1672a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 16731da177e4SLinus Torvalds task->tk_pid, task->tk_status); 16741da177e4SLinus Torvalds 1675788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1676c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1677c9d8f89dSTrond Myklebust } 1678c9d8f89dSTrond Myklebust 1679f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1680c9d8f89dSTrond Myklebust { 1681c9d8f89dSTrond Myklebust struct nfs_page *req; 1682c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1683f453a54aSFred Isaman struct nfs_commit_info cinfo; 1684353db796SNeilBrown struct nfs_server *nfss; 1685788e7a89STrond Myklebust 16861da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 16871da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 16881da177e4SLinus Torvalds nfs_list_remove_request(req); 1689d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 16901da177e4SLinus Torvalds 16911e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 16923d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 16932b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 16941da177e4SLinus Torvalds req->wb_bytes, 16951da177e4SLinus Torvalds (long long)req_offset(req)); 1696c9d8f89dSTrond Myklebust if (status < 0) { 1697c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 16981da177e4SLinus Torvalds nfs_inode_remove_request(req); 1699c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 17001da177e4SLinus Torvalds goto next; 17011da177e4SLinus Torvalds } 17021da177e4SLinus Torvalds 17031da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 17041da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 17052f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 17061da177e4SLinus Torvalds /* We have a match */ 17071da177e4SLinus Torvalds nfs_inode_remove_request(req); 17081da177e4SLinus Torvalds dprintk(" OK\n"); 17091da177e4SLinus Torvalds goto next; 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 17121da177e4SLinus Torvalds dprintk(" mismatch\n"); 17136d884e8fSFred nfs_mark_request_dirty(req); 171405990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 17151da177e4SLinus Torvalds next: 17161d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 17171da177e4SLinus Torvalds } 1718353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1719353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 1720353db796SNeilBrown clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 1721353db796SNeilBrown 1722f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1723f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1724f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 17255917ce84SFred Isaman } 17265917ce84SFred Isaman 17275917ce84SFred Isaman static void nfs_commit_release(void *calldata) 17285917ce84SFred Isaman { 17290b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17305917ce84SFred Isaman 1731f453a54aSFred Isaman data->completion_ops->completion(data); 1732c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 17331da177e4SLinus Torvalds } 1734788e7a89STrond Myklebust 1735788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 17360b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1737788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1738788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1739788e7a89STrond Myklebust }; 17401da177e4SLinus Torvalds 1741f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1742f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1743f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1744f453a54aSFred Isaman }; 1745f453a54aSFred Isaman 17461763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1747ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 174884c53ab5SFred Isaman { 174984c53ab5SFred Isaman int status; 175084c53ab5SFred Isaman 1751ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 175284c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1753ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 175484c53ab5SFred Isaman return status; 175584c53ab5SFred Isaman } 175684c53ab5SFred Isaman 1757b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 17581da177e4SLinus Torvalds { 17591da177e4SLinus Torvalds LIST_HEAD(head); 1760ea2cf228SFred Isaman struct nfs_commit_info cinfo; 176171d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1762b8413f98STrond Myklebust int res; 17631da177e4SLinus Torvalds 1764b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1765b8413f98STrond Myklebust if (res <= 0) 1766c5efa5fcSTrond Myklebust goto out_mark_dirty; 1767ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1768ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 17691da177e4SLinus Torvalds if (res) { 1770a861a1e1SFred Isaman int error; 1771a861a1e1SFred Isaman 1772ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 17731da177e4SLinus Torvalds if (error < 0) 17741da177e4SLinus Torvalds return error; 1775b8413f98STrond Myklebust if (!may_wait) 1776b8413f98STrond Myklebust goto out_mark_dirty; 177774316201SNeilBrown error = wait_on_bit_action(&NFS_I(inode)->flags, 1778b8413f98STrond Myklebust NFS_INO_COMMIT, 177971d0a611STrond Myklebust nfs_wait_bit_killable, 178071d0a611STrond Myklebust TASK_KILLABLE); 1781b8413f98STrond Myklebust if (error < 0) 1782b8413f98STrond Myklebust return error; 178371d0a611STrond Myklebust } else 178471d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1785c5efa5fcSTrond Myklebust return res; 1786c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1787c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1788c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1789c5efa5fcSTrond Myklebust * that the data is on the disk. 1790c5efa5fcSTrond Myklebust */ 1791c5efa5fcSTrond Myklebust out_mark_dirty: 1792c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17931da177e4SLinus Torvalds return res; 17941da177e4SLinus Torvalds } 17958fc795f7STrond Myklebust 17968fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17978fc795f7STrond Myklebust { 1798420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1799420e3646STrond Myklebust int flags = FLUSH_SYNC; 1800420e3646STrond Myklebust int ret = 0; 18018fc795f7STrond Myklebust 18023236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1803ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 18043236c3e1SJeff Layton return ret; 18053236c3e1SJeff Layton 1806a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1807a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1808a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1809420e3646STrond Myklebust */ 1810cb1410c7SWeston Andros Adamson if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) 1811420e3646STrond Myklebust goto out_mark_dirty; 1812420e3646STrond Myklebust 1813a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1814420e3646STrond Myklebust flags = 0; 1815a00dd6c0SJeff Layton } 1816a00dd6c0SJeff Layton 1817420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1818420e3646STrond Myklebust if (ret >= 0) { 1819420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1820420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1821420e3646STrond Myklebust wbc->nr_to_write -= ret; 1822420e3646STrond Myklebust else 1823420e3646STrond Myklebust wbc->nr_to_write = 0; 1824420e3646STrond Myklebust } 18258fc795f7STrond Myklebust return 0; 1826420e3646STrond Myklebust } 1827420e3646STrond Myklebust out_mark_dirty: 18288fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18298fc795f7STrond Myklebust return ret; 18308fc795f7STrond Myklebust } 18311da177e4SLinus Torvalds 18328fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18338fc795f7STrond Myklebust { 1834a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1835a8d8f02cSBryan Schumaker } 183689d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1837863a3c6cSAndy Adamson 1838acdc53b2STrond Myklebust /* 1839acdc53b2STrond Myklebust * flush the inode to disk. 1840acdc53b2STrond Myklebust */ 1841acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 184234901f70STrond Myklebust { 1843f4ce1299STrond Myklebust int ret; 184434901f70STrond Myklebust 1845f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1846f4ce1299STrond Myklebust 18475bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 1848*6b196875SChuck Lever if (ret) 1849*6b196875SChuck Lever goto out; 18505bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 1851*6b196875SChuck Lever if (ret < 0) 1852*6b196875SChuck Lever goto out; 18535bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 1854*6b196875SChuck Lever ret = 0; 1855f4ce1299STrond Myklebust 1856*6b196875SChuck Lever out: 1857f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1858f4ce1299STrond Myklebust return ret; 18591c75950bSTrond Myklebust } 1860ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 18611c75950bSTrond Myklebust 18621b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 18631b3b4a1aSTrond Myklebust { 18641b3b4a1aSTrond Myklebust struct nfs_page *req; 18651b3b4a1aSTrond Myklebust int ret = 0; 18661b3b4a1aSTrond Myklebust 1867ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18683e217045SWeston Andros Adamson 18693e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 18703e217045SWeston Andros Adamson * request */ 18713e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 18723e217045SWeston Andros Adamson 18733e217045SWeston Andros Adamson if (IS_ERR(req)) { 18743e217045SWeston Andros Adamson ret = PTR_ERR(req); 18753e217045SWeston Andros Adamson } else if (req) { 18763e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 18773e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 18783e217045SWeston Andros Adamson * request from the inode / page_private pointer and 18793e217045SWeston Andros Adamson * release it */ 18801b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 18811d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18821b3b4a1aSTrond Myklebust } 18833e217045SWeston Andros Adamson 18841b3b4a1aSTrond Myklebust return ret; 18851b3b4a1aSTrond Myklebust } 18861b3b4a1aSTrond Myklebust 18871c75950bSTrond Myklebust /* 18881c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 18891c75950bSTrond Myklebust */ 18901c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 18911c75950bSTrond Myklebust { 189229418aa4SMel Gorman loff_t range_start = page_file_offset(page); 18937f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 18947f2f12d9STrond Myklebust struct writeback_control wbc = { 18957f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 18967f2f12d9STrond Myklebust .nr_to_write = 0, 18977f2f12d9STrond Myklebust .range_start = range_start, 18987f2f12d9STrond Myklebust .range_end = range_end, 18997f2f12d9STrond Myklebust }; 19007f2f12d9STrond Myklebust int ret; 19017f2f12d9STrond Myklebust 1902f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1903f4ce1299STrond Myklebust 19040522f6adSTrond Myklebust for (;;) { 1905ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19067f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 19077f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 19087f2f12d9STrond Myklebust if (ret < 0) 19097f2f12d9STrond Myklebust goto out_error; 19100522f6adSTrond Myklebust continue; 19117f2f12d9STrond Myklebust } 1912f4ce1299STrond Myklebust ret = 0; 19130522f6adSTrond Myklebust if (!PagePrivate(page)) 19140522f6adSTrond Myklebust break; 19150522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19167f2f12d9STrond Myklebust if (ret < 0) 19177f2f12d9STrond Myklebust goto out_error; 19187f2f12d9STrond Myklebust } 19197f2f12d9STrond Myklebust out_error: 1920f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 19217f2f12d9STrond Myklebust return ret; 19221c75950bSTrond Myklebust } 19231c75950bSTrond Myklebust 1924074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1925074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1926a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1927074cc1deSTrond Myklebust { 19282da95652SJeff Layton /* 19292da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 19302da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 19312da95652SJeff Layton * 19322da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 19332da95652SJeff Layton * that we can safely release the inode reference while holding 19342da95652SJeff Layton * the page lock. 19352da95652SJeff Layton */ 19362da95652SJeff Layton if (PagePrivate(page)) 19372da95652SJeff Layton return -EBUSY; 1938074cc1deSTrond Myklebust 19398c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 19408c209ce7SDavid Howells return -EBUSY; 1941074cc1deSTrond Myklebust 1942a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1943074cc1deSTrond Myklebust } 1944074cc1deSTrond Myklebust #endif 1945074cc1deSTrond Myklebust 1946f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 19471da177e4SLinus Torvalds { 19481da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 19491e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 19501da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 195120c2df83SPaul Mundt NULL); 19521da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 19531da177e4SLinus Torvalds return -ENOMEM; 19541da177e4SLinus Torvalds 195593d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 19561da177e4SLinus Torvalds nfs_wdata_cachep); 19571da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 19583dd4765fSJeff Layton goto out_destroy_write_cache; 19591da177e4SLinus Torvalds 19600b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 19610b7c0153SFred Isaman sizeof(struct nfs_commit_data), 19620b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 19630b7c0153SFred Isaman NULL); 19640b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 19653dd4765fSJeff Layton goto out_destroy_write_mempool; 19660b7c0153SFred Isaman 196793d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 19684c100210SYanchuan Nian nfs_cdata_cachep); 19691da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 19703dd4765fSJeff Layton goto out_destroy_commit_cache; 19711da177e4SLinus Torvalds 197289a09141SPeter Zijlstra /* 197389a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 197489a09141SPeter Zijlstra * 197589a09141SPeter Zijlstra * 64MB: 8192k 197689a09141SPeter Zijlstra * 128MB: 11585k 197789a09141SPeter Zijlstra * 256MB: 16384k 197889a09141SPeter Zijlstra * 512MB: 23170k 197989a09141SPeter Zijlstra * 1GB: 32768k 198089a09141SPeter Zijlstra * 2GB: 46340k 198189a09141SPeter Zijlstra * 4GB: 65536k 198289a09141SPeter Zijlstra * 8GB: 92681k 198389a09141SPeter Zijlstra * 16GB: 131072k 198489a09141SPeter Zijlstra * 198589a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 198689a09141SPeter Zijlstra * Limit the default to 256M 198789a09141SPeter Zijlstra */ 198889a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 198989a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 199089a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 199189a09141SPeter Zijlstra 19921da177e4SLinus Torvalds return 0; 19933dd4765fSJeff Layton 19943dd4765fSJeff Layton out_destroy_commit_cache: 19953dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19963dd4765fSJeff Layton out_destroy_write_mempool: 19973dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 19983dd4765fSJeff Layton out_destroy_write_cache: 19993dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 20003dd4765fSJeff Layton return -ENOMEM; 20011da177e4SLinus Torvalds } 20021da177e4SLinus Torvalds 2003266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 20041da177e4SLinus Torvalds { 20051da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 20063dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20071da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 20081a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 20091da177e4SLinus Torvalds } 20101da177e4SLinus Torvalds 20114a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 2012a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 20134a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 20144a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 2015a4cdda59SAnna Schumaker .rw_release = nfs_writeback_release_common, 20160eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 20170eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 20181ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 20194a0de55cSAnna Schumaker }; 2020