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 /** 77186d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 77286d80f97STrond Myklebust * @req: pointer to a struct nfs_page 77386d80f97STrond Myklebust * @dst: commit list head 77486d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 77586d80f97STrond Myklebust * 77686d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 77786d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 77886d80f97STrond Myklebust * the MM page stats. 77986d80f97STrond Myklebust * 78086d80f97STrond Myklebust * The caller must hold the cinfo->lock, and the nfs_page lock. 78186d80f97STrond Myklebust */ 78286d80f97STrond Myklebust void 78386d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 78486d80f97STrond Myklebust struct nfs_commit_info *cinfo) 78586d80f97STrond Myklebust { 78686d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 78786d80f97STrond Myklebust nfs_list_add_request(req, dst); 78886d80f97STrond Myklebust cinfo->mds->ncommit++; 78986d80f97STrond Myklebust } 79086d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 79186d80f97STrond Myklebust 79286d80f97STrond Myklebust /** 7938dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 7948dd37758STrond Myklebust * @req: pointer to a struct nfs_page 795ea2cf228SFred Isaman * @dst: commit list head 796ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7978dd37758STrond Myklebust * 798ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 7998dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8008dd37758STrond Myklebust * the MM page stats. 8018dd37758STrond Myklebust * 802ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8038dd37758STrond Myklebust * holding the nfs_page lock. 8048dd37758STrond Myklebust */ 8058dd37758STrond Myklebust void 806ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 807ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8088dd37758STrond Myklebust { 809ea2cf228SFred Isaman spin_lock(cinfo->lock); 81086d80f97STrond Myklebust nfs_request_add_commit_list_locked(req, dst, cinfo); 811ea2cf228SFred Isaman spin_unlock(cinfo->lock); 81286d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8138dd37758STrond Myklebust } 8148dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8158dd37758STrond Myklebust 8168dd37758STrond Myklebust /** 8178dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8188dd37758STrond Myklebust * @req: pointer to a nfs_page 819ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8208dd37758STrond Myklebust * 821ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8228dd37758STrond Myklebust * number of outstanding requests requiring a commit 8238dd37758STrond Myklebust * It does not update the MM page stats. 8248dd37758STrond Myklebust * 825ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8268dd37758STrond Myklebust */ 8278dd37758STrond Myklebust void 828ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 829ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8308dd37758STrond Myklebust { 8318dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 8328dd37758STrond Myklebust return; 8338dd37758STrond Myklebust nfs_list_remove_request(req); 834ea2cf228SFred Isaman cinfo->mds->ncommit--; 8358dd37758STrond Myklebust } 8368dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 8378dd37758STrond Myklebust 838ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 839ea2cf228SFred Isaman struct inode *inode) 840ea2cf228SFred Isaman { 841ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 842ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 843ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 844b359f9d0SFred Isaman cinfo->dreq = NULL; 845f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 846ea2cf228SFred Isaman } 847ea2cf228SFred Isaman 848ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 849ea2cf228SFred Isaman struct inode *inode, 850ea2cf228SFred Isaman struct nfs_direct_req *dreq) 851ea2cf228SFred Isaman { 8521763da12SFred Isaman if (dreq) 8531763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8541763da12SFred Isaman else 855ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 856ea2cf228SFred Isaman } 857ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8588dd37758STrond Myklebust 8591da177e4SLinus Torvalds /* 8601da177e4SLinus Torvalds * Add a request to the inode's commit list. 8611da177e4SLinus Torvalds */ 8621763da12SFred Isaman void 863ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 864b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 8651da177e4SLinus Torvalds { 866b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 8678dd37758STrond Myklebust return; 868ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 8691da177e4SLinus Torvalds } 8708e821cadSTrond Myklebust 871d6d6dc7cSFred Isaman static void 872d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 873e468bae9STrond Myklebust { 874e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 87593f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 87693f78d88STejun Heo WB_RECLAIMABLE); 877e468bae9STrond Myklebust } 878d6d6dc7cSFred Isaman 879411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 8808dd37758STrond Myklebust static void 881d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 882d6d6dc7cSFred Isaman { 8838dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 8842b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 885ea2cf228SFred Isaman struct nfs_commit_info cinfo; 886d6d6dc7cSFred Isaman 887ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 888ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 889ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 890d6d6dc7cSFred Isaman } 8918dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 8928dd37758STrond Myklebust } 893e468bae9STrond Myklebust } 894e468bae9STrond Myklebust 895d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 8968e821cadSTrond Myklebust { 897c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 898d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 899c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9008e821cadSTrond Myklebust } 9018e821cadSTrond Myklebust 902061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9036c75dc0dSFred Isaman { 904ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9056c75dc0dSFred Isaman unsigned long bytes = 0; 9066c75dc0dSFred Isaman 9076c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9086c75dc0dSFred Isaman goto out; 909ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9106c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9116c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9126c75dc0dSFred Isaman 9136c75dc0dSFred Isaman bytes += req->wb_bytes; 9146c75dc0dSFred Isaman nfs_list_remove_request(req); 9156c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9166c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 917d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9186c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9196c75dc0dSFred Isaman goto remove_req; 9206c75dc0dSFred Isaman } 921c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 922f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 923b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 924a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 9256c75dc0dSFred Isaman goto next; 9266c75dc0dSFred Isaman } 9276c75dc0dSFred Isaman remove_req: 9286c75dc0dSFred Isaman nfs_inode_remove_request(req); 9296c75dc0dSFred Isaman next: 9301d1afcbcSTrond Myklebust nfs_unlock_request(req); 93120633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9323aff4ebbSTrond Myklebust nfs_release_request(req); 9336c75dc0dSFred Isaman } 9346c75dc0dSFred Isaman out: 9356c75dc0dSFred Isaman hdr->release(hdr); 9366c75dc0dSFred Isaman } 9376c75dc0dSFred Isaman 938ce59515cSAnna Schumaker unsigned long 939ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 940fb8a1f11STrond Myklebust { 941ea2cf228SFred Isaman return cinfo->mds->ncommit; 942fb8a1f11STrond Myklebust } 943fb8a1f11STrond Myklebust 944ea2cf228SFred Isaman /* cinfo->lock held by caller */ 9451763da12SFred Isaman int 946ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 947ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 948d6d6dc7cSFred Isaman { 949d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 950d6d6dc7cSFred Isaman int ret = 0; 951d6d6dc7cSFred Isaman 952d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9538dd37758STrond Myklebust if (!nfs_lock_request(req)) 9548dd37758STrond Myklebust continue; 9557ad84aa9STrond Myklebust kref_get(&req->wb_kref); 956ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 9573b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 958ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9598dd37758STrond Myklebust nfs_list_add_request(req, dst); 960d6d6dc7cSFred Isaman ret++; 9611763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 962d6d6dc7cSFred Isaman break; 963d6d6dc7cSFred Isaman } 964d6d6dc7cSFred Isaman return ret; 965d6d6dc7cSFred Isaman } 966d6d6dc7cSFred Isaman 9671da177e4SLinus Torvalds /* 9681da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9691da177e4SLinus Torvalds * @inode: NFS inode to scan 970ea2cf228SFred Isaman * @dst: mds destination list 971ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9721da177e4SLinus Torvalds * 9731da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 9741da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 9751da177e4SLinus Torvalds */ 9761763da12SFred Isaman int 977ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 978ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9791da177e4SLinus Torvalds { 980d6d6dc7cSFred Isaman int ret = 0; 981fb8a1f11STrond Myklebust 982ea2cf228SFred Isaman spin_lock(cinfo->lock); 983ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 9848dd37758STrond Myklebust const int max = INT_MAX; 985d6d6dc7cSFred Isaman 986ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 987ea2cf228SFred Isaman cinfo, max); 988ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 989d6d6dc7cSFred Isaman } 990ea2cf228SFred Isaman spin_unlock(cinfo->lock); 991ff778d02STrond Myklebust return ret; 9921da177e4SLinus Torvalds } 993d6d6dc7cSFred Isaman 9941da177e4SLinus Torvalds /* 995e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 996e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 9971da177e4SLinus Torvalds * 998e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 999e7d39069STrond Myklebust * to disk. 10001da177e4SLinus Torvalds */ 1001e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1002e7d39069STrond Myklebust struct page *page, 1003e7d39069STrond Myklebust unsigned int offset, 1004e7d39069STrond Myklebust unsigned int bytes) 10051da177e4SLinus Torvalds { 1006e7d39069STrond Myklebust struct nfs_page *req; 1007e7d39069STrond Myklebust unsigned int rqend; 1008e7d39069STrond Myklebust unsigned int end; 10091da177e4SLinus Torvalds int error; 1010277459d2STrond Myklebust 1011e7d39069STrond Myklebust if (!PagePrivate(page)) 1012e7d39069STrond Myklebust return NULL; 1013e7d39069STrond Myklebust 1014e7d39069STrond Myklebust end = offset + bytes; 1015e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1016e7d39069STrond Myklebust 1017e7d39069STrond Myklebust for (;;) { 101884d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1019e7d39069STrond Myklebust if (req == NULL) 1020e7d39069STrond Myklebust goto out_unlock; 1021e7d39069STrond Myklebust 10222bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10232bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10242bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10252bfc6e56SWeston Andros Adamson 1026e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1027e7d39069STrond Myklebust /* 1028e7d39069STrond Myklebust * Tell the caller to flush out the request if 1029e7d39069STrond Myklebust * the offsets are non-contiguous. 1030e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1031e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1032e7d39069STrond Myklebust */ 1033e468bae9STrond Myklebust if (offset > rqend 1034e7d39069STrond Myklebust || end < req->wb_offset) 1035e7d39069STrond Myklebust goto out_flushme; 1036e7d39069STrond Myklebust 10377ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1038e7d39069STrond Myklebust break; 1039e7d39069STrond Myklebust 1040e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1041587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10421da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10431da177e4SLinus Torvalds nfs_release_request(req); 1044e7d39069STrond Myklebust if (error != 0) 1045e7d39069STrond Myklebust goto out_err; 1046e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10471da177e4SLinus Torvalds } 10481da177e4SLinus Torvalds 10491da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10501da177e4SLinus Torvalds if (offset < req->wb_offset) { 10511da177e4SLinus Torvalds req->wb_offset = offset; 10521da177e4SLinus Torvalds req->wb_pgbase = offset; 10531da177e4SLinus Torvalds } 10541da177e4SLinus Torvalds if (end > rqend) 10551da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1056e7d39069STrond Myklebust else 1057e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1058e7d39069STrond Myklebust out_unlock: 1059ca138f36SFred Isaman if (req) 10608dd37758STrond Myklebust nfs_clear_request_commit(req); 1061411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1062e7d39069STrond Myklebust return req; 1063e7d39069STrond Myklebust out_flushme: 1064e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1065e7d39069STrond Myklebust nfs_release_request(req); 1066e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1067e7d39069STrond Myklebust out_err: 1068e7d39069STrond Myklebust return ERR_PTR(error); 1069e7d39069STrond Myklebust } 10701da177e4SLinus Torvalds 1071e7d39069STrond Myklebust /* 1072e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1073e7d39069STrond Myklebust * 1074e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1075e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1076e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1077e7d39069STrond Myklebust */ 1078e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1079e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1080e7d39069STrond Myklebust { 1081d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1082e7d39069STrond Myklebust struct nfs_page *req; 1083e7d39069STrond Myklebust 1084e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1085e7d39069STrond Myklebust if (req != NULL) 1086e7d39069STrond Myklebust goto out; 10872bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1088e7d39069STrond Myklebust if (IS_ERR(req)) 1089e7d39069STrond Myklebust goto out; 1090d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1091efc91ed0STrond Myklebust out: 109261e930a9STrond Myklebust return req; 10931da177e4SLinus Torvalds } 10941da177e4SLinus Torvalds 1095e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1096e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1097e7d39069STrond Myklebust { 1098e7d39069STrond Myklebust struct nfs_page *req; 1099e7d39069STrond Myklebust 1100e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1101e7d39069STrond Myklebust if (IS_ERR(req)) 1102e7d39069STrond Myklebust return PTR_ERR(req); 1103e7d39069STrond Myklebust /* Update file length */ 1104e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1105d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1106a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11071d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1108e7d39069STrond Myklebust return 0; 1109e7d39069STrond Myklebust } 1110e7d39069STrond Myklebust 11111da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11121da177e4SLinus Torvalds { 1113cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11142a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1115bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11161da177e4SLinus Torvalds struct nfs_page *req; 11171a54533eSTrond Myklebust int do_flush, status; 11181da177e4SLinus Torvalds /* 11191da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11201da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11211da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11221da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11231da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11241da177e4SLinus Torvalds * dropped page. 11251da177e4SLinus Torvalds */ 11261a54533eSTrond Myklebust do { 112784d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11281a54533eSTrond Myklebust if (req == NULL) 11291a54533eSTrond Myklebust return 0; 11302a369153STrond Myklebust l_ctx = req->wb_lock_context; 11312a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 11322bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11332bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1134bd61e0a9SJeff Layton if (l_ctx && flctx && 1135bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1136bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 11375263e31eSJeff Layton do_flush |= l_ctx->lockowner.l_owner != current->files 11385263e31eSJeff Layton || l_ctx->lockowner.l_pid != current->tgid; 11395263e31eSJeff Layton } 11401da177e4SLinus Torvalds nfs_release_request(req); 11411a54533eSTrond Myklebust if (!do_flush) 11421a54533eSTrond Myklebust return 0; 1143d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11441a54533eSTrond Myklebust } while (status == 0); 11451a54533eSTrond Myklebust return status; 11461da177e4SLinus Torvalds } 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds /* 1149dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1150dc24826bSAndy Adamson * expire soon. 1151dc24826bSAndy Adamson * 1152dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1153dc24826bSAndy Adamson * 1154dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1155dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1156dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1157dc24826bSAndy Adamson */ 1158dc24826bSAndy Adamson int 1159dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1160dc24826bSAndy Adamson { 1161dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1162dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1163dc24826bSAndy Adamson 1164dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1165dc24826bSAndy Adamson } 1166dc24826bSAndy Adamson 1167dc24826bSAndy Adamson /* 1168dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1169dc24826bSAndy Adamson */ 1170dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 1171dc24826bSAndy Adamson { 1172dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 1173dc24826bSAndy Adamson } 1174dc24826bSAndy Adamson 1175dc24826bSAndy Adamson /* 11765d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 11775d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 11785d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 11795d47a356STrond Myklebust */ 11808d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 11815d47a356STrond Myklebust { 1182d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1183d529ef83SJeff Layton 11848d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 11858d197a56STrond Myklebust goto out; 118618dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1187d529ef83SJeff Layton return false; 11884db72b40SJeff Layton smp_rmb(); 1189d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 11908d197a56STrond Myklebust return false; 11918d197a56STrond Myklebust out: 119218dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 119318dd78c4SScott Mayhew return false; 11948d197a56STrond Myklebust return PageUptodate(page) != 0; 11955d47a356STrond Myklebust } 11965d47a356STrond Myklebust 11975263e31eSJeff Layton static bool 11985263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 11995263e31eSJeff Layton { 12005263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12015263e31eSJeff Layton fl->fl_type == F_WRLCK; 12025263e31eSJeff Layton } 12035263e31eSJeff Layton 1204c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1205c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1206c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1207c7559663SScott Mayhew * inefficiencies. 1208c7559663SScott Mayhew * 1209263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1210263b4509SScott Mayhew * of the checks. 1211c7559663SScott Mayhew */ 1212c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1213c7559663SScott Mayhew { 12145263e31eSJeff Layton int ret; 12155263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12165263e31eSJeff Layton struct file_lock *fl; 12175263e31eSJeff Layton 1218c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1219c7559663SScott Mayhew return 0; 1220263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1221263b4509SScott Mayhew return 0; 1222c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1223c7559663SScott Mayhew return 1; 1224bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1225bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 1226c7559663SScott Mayhew return 0; 12275263e31eSJeff Layton 12285263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12295263e31eSJeff Layton ret = 0; 12306109c850SJeff Layton spin_lock(&flctx->flc_lock); 1231bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1232bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1233bd61e0a9SJeff Layton fl_list); 1234bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12355263e31eSJeff Layton ret = 1; 1236bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 12375263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 12385263e31eSJeff Layton fl_list); 12395263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 12405263e31eSJeff Layton ret = 1; 12415263e31eSJeff Layton } 12426109c850SJeff Layton spin_unlock(&flctx->flc_lock); 12435263e31eSJeff Layton return ret; 1244c7559663SScott Mayhew } 1245c7559663SScott Mayhew 12465d47a356STrond Myklebust /* 12471da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12481da177e4SLinus Torvalds * 12491da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12501da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12511da177e4SLinus Torvalds */ 12521da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12531da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12541da177e4SLinus Torvalds { 1255cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1256d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12571da177e4SLinus Torvalds int status = 0; 12581da177e4SLinus Torvalds 125991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 126091d5b470SChuck Lever 12616de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12626de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12631da177e4SLinus Torvalds 1264c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 126549a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12661da177e4SLinus Torvalds offset = 0; 12671da177e4SLinus Torvalds } 12681da177e4SLinus Torvalds 1269e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 127003fa9e84STrond Myklebust if (status < 0) 127103fa9e84STrond Myklebust nfs_set_pageerror(page); 127259b7c05fSTrond Myklebust else 127359b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 12741da177e4SLinus Torvalds 127548186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 12761da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 12771da177e4SLinus Torvalds return status; 12781da177e4SLinus Torvalds } 12791da177e4SLinus Torvalds 12803ff7576dSTrond Myklebust static int flush_task_priority(int how) 12811da177e4SLinus Torvalds { 12821da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 12831da177e4SLinus Torvalds case FLUSH_HIGHPRI: 12841da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 12851da177e4SLinus Torvalds case FLUSH_LOWPRI: 12861da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 12871da177e4SLinus Torvalds } 12881da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 12891da177e4SLinus Torvalds } 12901da177e4SLinus Torvalds 1291d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1292d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1293abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 12941ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 12951da177e4SLinus Torvalds { 12963ff7576dSTrond Myklebust int priority = flush_task_priority(how); 12971da177e4SLinus Torvalds 12981ed26f33SAnna Schumaker task_setup_data->priority = priority; 1299abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1300d138d5d1SAndy Adamson 1301abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1302d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1303275acaafSTrond Myklebust } 1304275acaafSTrond Myklebust 13056d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13066d884e8fSFred * call this on each, which will prepare them to be retried on next 13076d884e8fSFred * writeback using standard nfs. 13086d884e8fSFred */ 13096d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13106d884e8fSFred { 13116d884e8fSFred nfs_mark_request_dirty(req); 1312c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 13131d1afcbcSTrond Myklebust nfs_unlock_request(req); 131420633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13153aff4ebbSTrond Myklebust nfs_release_request(req); 13166d884e8fSFred } 13176d884e8fSFred 1318061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13196c75dc0dSFred Isaman { 13206c75dc0dSFred Isaman struct nfs_page *req; 13216c75dc0dSFred Isaman 13226c75dc0dSFred Isaman while (!list_empty(head)) { 13236c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13246c75dc0dSFred Isaman nfs_list_remove_request(req); 13256c75dc0dSFred Isaman nfs_redirty_request(req); 13266c75dc0dSFred Isaman } 13276c75dc0dSFred Isaman } 13286c75dc0dSFred Isaman 1329061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1330061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1331061ae2edSFred Isaman .completion = nfs_write_completion, 1332061ae2edSFred Isaman }; 1333061ae2edSFred Isaman 133457208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1335a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1336061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 13371751c363STrond Myklebust { 1338a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 133941d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1340a20c93e3SChristoph Hellwig 1341a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1342a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1343a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1344a20c93e3SChristoph Hellwig #endif 13454a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 13464a0de55cSAnna Schumaker server->wsize, ioflags); 13471751c363STrond Myklebust } 1348ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13491751c363STrond Myklebust 1350dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1351dce81290STrond Myklebust { 1352a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1353a7d42ddbSWeston Andros Adamson 135441d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1355a7d42ddbSWeston Andros Adamson 1356a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1357a7d42ddbSWeston Andros Adamson 1358a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1359a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1360dce81290STrond Myklebust } 13611f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1362dce81290STrond Myklebust 13631da177e4SLinus Torvalds 13640b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13650b7c0153SFred Isaman { 13660b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13670b7c0153SFred Isaman 13680b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13690b7c0153SFred Isaman } 13700b7c0153SFred Isaman 13711f2edbe3STrond Myklebust /* 13721f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 13731f2edbe3STrond Myklebust */ 13741f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 13751f2edbe3STrond Myklebust { 13761f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 13771f2edbe3STrond Myklebust int kill = 0; 1378788e7a89STrond Myklebust 13791f2edbe3STrond Myklebust /* suid always must be killed */ 13801f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 13811f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 13821f2edbe3STrond Myklebust 13831f2edbe3STrond Myklebust /* 13841f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 13851f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 13861f2edbe3STrond Myklebust */ 13871f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 13881f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 13891f2edbe3STrond Myklebust 13901f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 13911f2edbe3STrond Myklebust return kill; 13921f2edbe3STrond Myklebust 13931f2edbe3STrond Myklebust return 0; 13941f2edbe3STrond Myklebust } 1395788e7a89STrond Myklebust 1396a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1397a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1398a08a8cd3STrond Myklebust { 1399a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1400a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14012b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1402a08a8cd3STrond Myklebust 1403a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14042b83d3deSTrond Myklebust fattr->size = size; 14052b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 14062b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1407a08a8cd3STrond Myklebust return; 14082b83d3deSTrond Myklebust } 14092b83d3deSTrond Myklebust if (size != fattr->size) 1410a08a8cd3STrond Myklebust return; 1411a08a8cd3STrond Myklebust /* Set attribute barrier */ 1412a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 14132b83d3deSTrond Myklebust /* ...and update size */ 14142b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1415a08a8cd3STrond Myklebust } 1416a08a8cd3STrond Myklebust 1417a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1418a08a8cd3STrond Myklebust { 14192b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1420a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1421a08a8cd3STrond Myklebust 1422a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1423a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1424a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1425a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1426a08a8cd3STrond Myklebust } 1427a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1428a08a8cd3STrond Myklebust 14291da177e4SLinus Torvalds /* 14301da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 14311da177e4SLinus Torvalds */ 1432d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1433d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 14340eecb214SAnna Schumaker struct inode *inode) 14351da177e4SLinus Torvalds { 1436788e7a89STrond Myklebust int status; 14371da177e4SLinus Torvalds 1438f551e44fSChuck Lever /* 1439f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1440f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1441f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1442f551e44fSChuck Lever * another writer had changed the file, but some applications 1443f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1444f551e44fSChuck Lever */ 1445d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1446788e7a89STrond Myklebust if (status != 0) 14470eecb214SAnna Schumaker return status; 1448d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 144991d5b470SChuck Lever 1450d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1451d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 14521da177e4SLinus Torvalds /* We tried a write call, but the server did not 14531da177e4SLinus Torvalds * commit data to stable storage even though we 14541da177e4SLinus Torvalds * requested it. 14551da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 14561da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 14571da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 14581da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 14591da177e4SLinus Torvalds */ 14601da177e4SLinus Torvalds static unsigned long complain; 14611da177e4SLinus Torvalds 1462a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 14631da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 14641da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 14651da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1466cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1467d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 14681da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 14691da177e4SLinus Torvalds } 14701da177e4SLinus Torvalds } 14711f2edbe3STrond Myklebust 14721f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 14731f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 14741f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 14750eecb214SAnna Schumaker return 0; 14760eecb214SAnna Schumaker } 14770eecb214SAnna Schumaker 14780eecb214SAnna Schumaker /* 14790eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 14800eecb214SAnna Schumaker */ 1481d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1482d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 14830eecb214SAnna Schumaker { 1484d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1485d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 14861f2edbe3STrond Myklebust 14871f2edbe3STrond Myklebust if (resp->count < argp->count) { 14881da177e4SLinus Torvalds static unsigned long complain; 14891da177e4SLinus Torvalds 14906c75dc0dSFred Isaman /* This a short write! */ 1491d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 149291d5b470SChuck Lever 14931da177e4SLinus Torvalds /* Has the server at least made some progress? */ 14946c75dc0dSFred Isaman if (resp->count == 0) { 14956c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 14966c75dc0dSFred Isaman printk(KERN_WARNING 14976c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 14986c75dc0dSFred Isaman argp->count); 14996c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15006c75dc0dSFred Isaman } 1501d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15026c75dc0dSFred Isaman task->tk_status = -EIO; 15036c75dc0dSFred Isaman return; 15046c75dc0dSFred Isaman } 15051da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 15061da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 15071da177e4SLinus Torvalds /* Resend from where the server left off */ 1508d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 15091da177e4SLinus Torvalds argp->offset += resp->count; 15101da177e4SLinus Torvalds argp->pgbase += resp->count; 15111da177e4SLinus Torvalds argp->count -= resp->count; 15121da177e4SLinus Torvalds } else { 15131da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 15141da177e4SLinus Torvalds * headaches in the case of a server crash. 15151da177e4SLinus Torvalds */ 15161da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 15171da177e4SLinus Torvalds } 1518d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 15191da177e4SLinus Torvalds } 15201da177e4SLinus Torvalds } 15211da177e4SLinus Torvalds 15221da177e4SLinus Torvalds 152371d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 152471d0a611STrond Myklebust { 1525b8413f98STrond Myklebust int ret; 1526b8413f98STrond Myklebust 152771d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 152871d0a611STrond Myklebust return 1; 1529b8413f98STrond Myklebust if (!may_wait) 153071d0a611STrond Myklebust return 0; 1531b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1532b8413f98STrond Myklebust NFS_INO_COMMIT, 1533b8413f98STrond Myklebust nfs_wait_bit_killable, 1534b8413f98STrond Myklebust TASK_KILLABLE); 1535b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 153671d0a611STrond Myklebust } 153771d0a611STrond Myklebust 1538f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 153971d0a611STrond Myklebust { 154071d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 15414e857c58SPeter Zijlstra smp_mb__after_atomic(); 154271d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 154371d0a611STrond Myklebust } 154471d0a611STrond Myklebust 15450b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 15461da177e4SLinus Torvalds { 15470b7c0153SFred Isaman put_nfs_open_context(data->context); 15480b7c0153SFred Isaman nfs_commit_free(data); 15491da177e4SLinus Torvalds } 1550e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15511da177e4SLinus Torvalds 15520b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1553c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 15549ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15559f0ec176SAndy Adamson int how, int flags) 15561da177e4SLinus Torvalds { 155707737691STrond Myklebust struct rpc_task *task; 15589ace33cdSFred Isaman int priority = flush_task_priority(how); 1559bdc7f021STrond Myklebust struct rpc_message msg = { 1560bdc7f021STrond Myklebust .rpc_argp = &data->args, 1561bdc7f021STrond Myklebust .rpc_resp = &data->res, 15629ace33cdSFred Isaman .rpc_cred = data->cred, 1563bdc7f021STrond Myklebust }; 156484115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 156507737691STrond Myklebust .task = &data->task, 15669ace33cdSFred Isaman .rpc_client = clnt, 1567bdc7f021STrond Myklebust .rpc_message = &msg, 15689ace33cdSFred Isaman .callback_ops = call_ops, 156984115e1cSTrond Myklebust .callback_data = data, 1570101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 15719f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 15723ff7576dSTrond Myklebust .priority = priority, 157384115e1cSTrond Myklebust }; 1574788e7a89STrond Myklebust /* Set up the initial task struct. */ 1575c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 15761da177e4SLinus Torvalds 1577b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1578bdc7f021STrond Myklebust 15798c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 15808c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 15818c21c62cSWeston Andros Adamson 158207737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1583dbae4c73STrond Myklebust if (IS_ERR(task)) 1584dbae4c73STrond Myklebust return PTR_ERR(task); 1585d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1586d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 158707737691STrond Myklebust rpc_put_task(task); 1588dbae4c73STrond Myklebust return 0; 15891da177e4SLinus Torvalds } 1590e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 15911da177e4SLinus Torvalds 1592378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1593378520b8SPeng Tao { 1594378520b8SPeng Tao loff_t lwb = 0; 1595378520b8SPeng Tao struct nfs_page *req; 1596378520b8SPeng Tao 1597378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1598378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1599378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1600378520b8SPeng Tao 1601378520b8SPeng Tao return lwb; 1602378520b8SPeng Tao } 1603378520b8SPeng Tao 16041da177e4SLinus Torvalds /* 16059ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16069ace33cdSFred Isaman */ 16070b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1608988b6dceSFred Isaman struct list_head *head, 1609f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1610f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16119ace33cdSFred Isaman { 16129ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16132b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16149ace33cdSFred Isaman 16159ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16169ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 16179ace33cdSFred Isaman 16189ace33cdSFred Isaman list_splice_init(head, &data->pages); 16199ace33cdSFred Isaman 16209ace33cdSFred Isaman data->inode = inode; 16219ace33cdSFred Isaman data->cred = first->wb_context->cred; 1622988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1623378520b8SPeng Tao /* only set lwb for pnfs commit */ 1624378520b8SPeng Tao if (lseg) 1625378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 16269ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1627f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1628b359f9d0SFred Isaman data->dreq = cinfo->dreq; 16299ace33cdSFred Isaman 16309ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 16319ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 16329ace33cdSFred Isaman data->args.offset = 0; 16339ace33cdSFred Isaman data->args.count = 0; 16340b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 16359ace33cdSFred Isaman data->res.fattr = &data->fattr; 16369ace33cdSFred Isaman data->res.verf = &data->verf; 16379ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 16389ace33cdSFred Isaman } 1639e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 16409ace33cdSFred Isaman 1641e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1642ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1643b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1644b57ff130SWeston Andros Adamson u32 ds_commit_idx) 164564bfeb49SFred Isaman { 164664bfeb49SFred Isaman struct nfs_page *req; 164764bfeb49SFred Isaman 164864bfeb49SFred Isaman while (!list_empty(page_list)) { 164964bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 165064bfeb49SFred Isaman nfs_list_remove_request(req); 1651b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1652487b9b8aSTom Haynes if (!cinfo->dreq) 1653487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 16541d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 165564bfeb49SFred Isaman } 165664bfeb49SFred Isaman } 1657e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 165864bfeb49SFred Isaman 16599ace33cdSFred Isaman /* 16601da177e4SLinus Torvalds * Commit dirty pages 16611da177e4SLinus Torvalds */ 16621da177e4SLinus Torvalds static int 1663ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1664ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 16651da177e4SLinus Torvalds { 16660b7c0153SFred Isaman struct nfs_commit_data *data; 16671da177e4SLinus Torvalds 1668c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 16691da177e4SLinus Torvalds 16701da177e4SLinus Torvalds if (!data) 16711da177e4SLinus Torvalds goto out_bad; 16721da177e4SLinus Torvalds 16731da177e4SLinus Torvalds /* Set up the argument struct */ 1674f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1675f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1676c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1677c36aae9aSPeng Tao data->mds_ops, how, 0); 16781da177e4SLinus Torvalds out_bad: 1679b57ff130SWeston Andros Adamson nfs_retry_commit(head, NULL, cinfo, 0); 1680f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 16811da177e4SLinus Torvalds return -ENOMEM; 16821da177e4SLinus Torvalds } 16831da177e4SLinus Torvalds 16841da177e4SLinus Torvalds /* 16851da177e4SLinus Torvalds * COMMIT call returned 16861da177e4SLinus Torvalds */ 1687788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 16881da177e4SLinus Torvalds { 16890b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16901da177e4SLinus Torvalds 1691a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 16921da177e4SLinus Torvalds task->tk_pid, task->tk_status); 16931da177e4SLinus Torvalds 1694788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1695c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1696c9d8f89dSTrond Myklebust } 1697c9d8f89dSTrond Myklebust 1698f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1699c9d8f89dSTrond Myklebust { 1700c9d8f89dSTrond Myklebust struct nfs_page *req; 1701c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1702f453a54aSFred Isaman struct nfs_commit_info cinfo; 1703353db796SNeilBrown struct nfs_server *nfss; 1704788e7a89STrond Myklebust 17051da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17061da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17071da177e4SLinus Torvalds nfs_list_remove_request(req); 1708d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17091da177e4SLinus Torvalds 17101e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17113d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 17122b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 17131da177e4SLinus Torvalds req->wb_bytes, 17141da177e4SLinus Torvalds (long long)req_offset(req)); 1715c9d8f89dSTrond Myklebust if (status < 0) { 1716c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 17171da177e4SLinus Torvalds nfs_inode_remove_request(req); 1718c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 17191da177e4SLinus Torvalds goto next; 17201da177e4SLinus Torvalds } 17211da177e4SLinus Torvalds 17221da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 17231da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 17242f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 17251da177e4SLinus Torvalds /* We have a match */ 17261da177e4SLinus Torvalds nfs_inode_remove_request(req); 17271da177e4SLinus Torvalds dprintk(" OK\n"); 17281da177e4SLinus Torvalds goto next; 17291da177e4SLinus Torvalds } 17301da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 17311da177e4SLinus Torvalds dprintk(" mismatch\n"); 17326d884e8fSFred nfs_mark_request_dirty(req); 173305990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 17341da177e4SLinus Torvalds next: 17351d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 17361da177e4SLinus Torvalds } 1737353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1738353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 1739353db796SNeilBrown clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 1740353db796SNeilBrown 1741f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1742f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1743f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 17445917ce84SFred Isaman } 17455917ce84SFred Isaman 17465917ce84SFred Isaman static void nfs_commit_release(void *calldata) 17475917ce84SFred Isaman { 17480b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17495917ce84SFred Isaman 1750f453a54aSFred Isaman data->completion_ops->completion(data); 1751c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 17521da177e4SLinus Torvalds } 1753788e7a89STrond Myklebust 1754788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 17550b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1756788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1757788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1758788e7a89STrond Myklebust }; 17591da177e4SLinus Torvalds 1760f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1761f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1762f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1763f453a54aSFred Isaman }; 1764f453a54aSFred Isaman 17651763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1766ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 176784c53ab5SFred Isaman { 176884c53ab5SFred Isaman int status; 176984c53ab5SFred Isaman 1770ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 177184c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1772ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 177384c53ab5SFred Isaman return status; 177484c53ab5SFred Isaman } 177584c53ab5SFred Isaman 1776b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 17771da177e4SLinus Torvalds { 17781da177e4SLinus Torvalds LIST_HEAD(head); 1779ea2cf228SFred Isaman struct nfs_commit_info cinfo; 178071d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1781b8413f98STrond Myklebust int res; 17821da177e4SLinus Torvalds 1783b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1784b8413f98STrond Myklebust if (res <= 0) 1785c5efa5fcSTrond Myklebust goto out_mark_dirty; 1786ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1787ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 17881da177e4SLinus Torvalds if (res) { 1789a861a1e1SFred Isaman int error; 1790a861a1e1SFred Isaman 1791ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 17921da177e4SLinus Torvalds if (error < 0) 17931da177e4SLinus Torvalds return error; 1794b8413f98STrond Myklebust if (!may_wait) 1795b8413f98STrond Myklebust goto out_mark_dirty; 179674316201SNeilBrown error = wait_on_bit_action(&NFS_I(inode)->flags, 1797b8413f98STrond Myklebust NFS_INO_COMMIT, 179871d0a611STrond Myklebust nfs_wait_bit_killable, 179971d0a611STrond Myklebust TASK_KILLABLE); 1800b8413f98STrond Myklebust if (error < 0) 1801b8413f98STrond Myklebust return error; 180271d0a611STrond Myklebust } else 180371d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1804c5efa5fcSTrond Myklebust return res; 1805c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1806c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1807c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1808c5efa5fcSTrond Myklebust * that the data is on the disk. 1809c5efa5fcSTrond Myklebust */ 1810c5efa5fcSTrond Myklebust out_mark_dirty: 1811c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18121da177e4SLinus Torvalds return res; 18131da177e4SLinus Torvalds } 18148fc795f7STrond Myklebust 1815ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18168fc795f7STrond Myklebust { 1817420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1818420e3646STrond Myklebust int flags = FLUSH_SYNC; 1819420e3646STrond Myklebust int ret = 0; 18208fc795f7STrond Myklebust 18213236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1822ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 18233236c3e1SJeff Layton return ret; 18243236c3e1SJeff Layton 1825a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1826a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1827a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1828420e3646STrond Myklebust */ 1829cb1410c7SWeston Andros Adamson if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) 1830420e3646STrond Myklebust goto out_mark_dirty; 1831420e3646STrond Myklebust 1832a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1833420e3646STrond Myklebust flags = 0; 1834a00dd6c0SJeff Layton } 1835a00dd6c0SJeff Layton 1836420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1837420e3646STrond Myklebust if (ret >= 0) { 1838420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1839420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1840420e3646STrond Myklebust wbc->nr_to_write -= ret; 1841420e3646STrond Myklebust else 1842420e3646STrond Myklebust wbc->nr_to_write = 0; 1843420e3646STrond Myklebust } 18448fc795f7STrond Myklebust return 0; 1845420e3646STrond Myklebust } 1846420e3646STrond Myklebust out_mark_dirty: 18478fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18488fc795f7STrond Myklebust return ret; 18498fc795f7STrond Myklebust } 185089d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1851863a3c6cSAndy Adamson 1852acdc53b2STrond Myklebust /* 1853acdc53b2STrond Myklebust * flush the inode to disk. 1854acdc53b2STrond Myklebust */ 1855acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 185634901f70STrond Myklebust { 1857f4ce1299STrond Myklebust int ret; 185834901f70STrond Myklebust 1859f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1860f4ce1299STrond Myklebust 18615bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 18626b196875SChuck Lever if (ret) 18636b196875SChuck Lever goto out; 18645bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 18656b196875SChuck Lever if (ret < 0) 18666b196875SChuck Lever goto out; 18675bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 18686b196875SChuck Lever ret = 0; 1869f4ce1299STrond Myklebust 18706b196875SChuck Lever out: 1871f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1872f4ce1299STrond Myklebust return ret; 18731c75950bSTrond Myklebust } 1874ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 18751c75950bSTrond Myklebust 18761b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 18771b3b4a1aSTrond Myklebust { 18781b3b4a1aSTrond Myklebust struct nfs_page *req; 18791b3b4a1aSTrond Myklebust int ret = 0; 18801b3b4a1aSTrond Myklebust 1881ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18823e217045SWeston Andros Adamson 18833e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 18843e217045SWeston Andros Adamson * request */ 18853e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 18863e217045SWeston Andros Adamson 18873e217045SWeston Andros Adamson if (IS_ERR(req)) { 18883e217045SWeston Andros Adamson ret = PTR_ERR(req); 18893e217045SWeston Andros Adamson } else if (req) { 18903e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 18913e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 18923e217045SWeston Andros Adamson * request from the inode / page_private pointer and 18933e217045SWeston Andros Adamson * release it */ 18941b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 18951d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18961b3b4a1aSTrond Myklebust } 18973e217045SWeston Andros Adamson 18981b3b4a1aSTrond Myklebust return ret; 18991b3b4a1aSTrond Myklebust } 19001b3b4a1aSTrond Myklebust 19011c75950bSTrond Myklebust /* 19021c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 19031c75950bSTrond Myklebust */ 19041c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 19051c75950bSTrond Myklebust { 190629418aa4SMel Gorman loff_t range_start = page_file_offset(page); 19077f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 19087f2f12d9STrond Myklebust struct writeback_control wbc = { 19097f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 19107f2f12d9STrond Myklebust .nr_to_write = 0, 19117f2f12d9STrond Myklebust .range_start = range_start, 19127f2f12d9STrond Myklebust .range_end = range_end, 19137f2f12d9STrond Myklebust }; 19147f2f12d9STrond Myklebust int ret; 19157f2f12d9STrond Myklebust 1916f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1917f4ce1299STrond Myklebust 19180522f6adSTrond Myklebust for (;;) { 1919ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19207f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 19217f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 19227f2f12d9STrond Myklebust if (ret < 0) 19237f2f12d9STrond Myklebust goto out_error; 19240522f6adSTrond Myklebust continue; 19257f2f12d9STrond Myklebust } 1926f4ce1299STrond Myklebust ret = 0; 19270522f6adSTrond Myklebust if (!PagePrivate(page)) 19280522f6adSTrond Myklebust break; 19290522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19307f2f12d9STrond Myklebust if (ret < 0) 19317f2f12d9STrond Myklebust goto out_error; 19327f2f12d9STrond Myklebust } 19337f2f12d9STrond Myklebust out_error: 1934f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 19357f2f12d9STrond Myklebust return ret; 19361c75950bSTrond Myklebust } 19371c75950bSTrond Myklebust 1938074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1939074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1940a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1941074cc1deSTrond Myklebust { 19422da95652SJeff Layton /* 19432da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 19442da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 19452da95652SJeff Layton * 19462da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 19472da95652SJeff Layton * that we can safely release the inode reference while holding 19482da95652SJeff Layton * the page lock. 19492da95652SJeff Layton */ 19502da95652SJeff Layton if (PagePrivate(page)) 19512da95652SJeff Layton return -EBUSY; 1952074cc1deSTrond Myklebust 19538c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 19548c209ce7SDavid Howells return -EBUSY; 1955074cc1deSTrond Myklebust 1956a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1957074cc1deSTrond Myklebust } 1958074cc1deSTrond Myklebust #endif 1959074cc1deSTrond Myklebust 1960f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 19611da177e4SLinus Torvalds { 19621da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 19631e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 19641da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 196520c2df83SPaul Mundt NULL); 19661da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 19671da177e4SLinus Torvalds return -ENOMEM; 19681da177e4SLinus Torvalds 196993d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 19701da177e4SLinus Torvalds nfs_wdata_cachep); 19711da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 19723dd4765fSJeff Layton goto out_destroy_write_cache; 19731da177e4SLinus Torvalds 19740b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 19750b7c0153SFred Isaman sizeof(struct nfs_commit_data), 19760b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 19770b7c0153SFred Isaman NULL); 19780b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 19793dd4765fSJeff Layton goto out_destroy_write_mempool; 19800b7c0153SFred Isaman 198193d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 19824c100210SYanchuan Nian nfs_cdata_cachep); 19831da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 19843dd4765fSJeff Layton goto out_destroy_commit_cache; 19851da177e4SLinus Torvalds 198689a09141SPeter Zijlstra /* 198789a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 198889a09141SPeter Zijlstra * 198989a09141SPeter Zijlstra * 64MB: 8192k 199089a09141SPeter Zijlstra * 128MB: 11585k 199189a09141SPeter Zijlstra * 256MB: 16384k 199289a09141SPeter Zijlstra * 512MB: 23170k 199389a09141SPeter Zijlstra * 1GB: 32768k 199489a09141SPeter Zijlstra * 2GB: 46340k 199589a09141SPeter Zijlstra * 4GB: 65536k 199689a09141SPeter Zijlstra * 8GB: 92681k 199789a09141SPeter Zijlstra * 16GB: 131072k 199889a09141SPeter Zijlstra * 199989a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 200089a09141SPeter Zijlstra * Limit the default to 256M 200189a09141SPeter Zijlstra */ 200289a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 200389a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 200489a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 200589a09141SPeter Zijlstra 20061da177e4SLinus Torvalds return 0; 20073dd4765fSJeff Layton 20083dd4765fSJeff Layton out_destroy_commit_cache: 20093dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20103dd4765fSJeff Layton out_destroy_write_mempool: 20113dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 20123dd4765fSJeff Layton out_destroy_write_cache: 20133dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 20143dd4765fSJeff Layton return -ENOMEM; 20151da177e4SLinus Torvalds } 20161da177e4SLinus Torvalds 2017266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 20181da177e4SLinus Torvalds { 20191da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 20203dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20211da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 20221a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 20231da177e4SLinus Torvalds } 20241da177e4SLinus Torvalds 20254a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 2026a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 20274a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 20284a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 20290eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 20300eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 20311ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 20324a0de55cSAnna Schumaker }; 2033