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> 24af7cf057STrond Myklebust #include <linux/freezer.h> 25af7cf057STrond Myklebust #include <linux/wait.h> 263fcfab16SAndrew Morton 277c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 281da177e4SLinus Torvalds 291da177e4SLinus Torvalds #include "delegation.h" 3049a70f27STrond Myklebust #include "internal.h" 3191d5b470SChuck Lever #include "iostat.h" 32def6ed7eSAndy Adamson #include "nfs4_fs.h" 33074cc1deSTrond Myklebust #include "fscache.h" 3494ad1c80SFred Isaman #include "pnfs.h" 351da177e4SLinus Torvalds 36f4ce1299STrond Myklebust #include "nfstrace.h" 37f4ce1299STrond Myklebust 381da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 391da177e4SLinus Torvalds 401da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 411da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 421da177e4SLinus Torvalds 431da177e4SLinus Torvalds /* 441da177e4SLinus Torvalds * Local function declarations 451da177e4SLinus Torvalds */ 46f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 47788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 48061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 49f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 504a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 51d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 5202d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 5302d1426cSWeston Andros Adamson struct inode *inode); 543a3908c8STrond Myklebust static struct nfs_page * 553a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 563a3908c8STrond Myklebust struct page *page); 571da177e4SLinus Torvalds 58e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 593feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 600b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 611da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 621da177e4SLinus Torvalds 630b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 641da177e4SLinus Torvalds { 65192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 6640859d7eSChuck Lever 671da177e4SLinus Torvalds if (p) { 681da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 691da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 701da177e4SLinus Torvalds } 711da177e4SLinus Torvalds return p; 721da177e4SLinus Torvalds } 73e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 741da177e4SLinus Torvalds 750b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 761da177e4SLinus Torvalds { 771da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 781da177e4SLinus Torvalds } 79e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 801da177e4SLinus Torvalds 811e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 823feb2d49STrond Myklebust { 831e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 843feb2d49STrond Myklebust 854a0de55cSAnna Schumaker if (p) 863feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 873feb2d49STrond Myklebust return p; 883feb2d49STrond Myklebust } 893feb2d49STrond Myklebust 901e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 916c75dc0dSFred Isaman { 921e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 933feb2d49STrond Myklebust } 941da177e4SLinus Torvalds 957b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 967b159fc1STrond Myklebust { 977b159fc1STrond Myklebust ctx->error = error; 987b159fc1STrond Myklebust smp_wmb(); 997b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1007b159fc1STrond Myklebust } 1017b159fc1STrond Myklebust 10284d3a9a9SWeston Andros Adamson /* 10384d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 10484d3a9a9SWeston Andros Adamson * 10584d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 10684d3a9a9SWeston Andros Adamson * 10784d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 10884d3a9a9SWeston Andros Adamson */ 10929418aa4SMel Gorman static struct nfs_page * 11084d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 111277459d2STrond Myklebust { 112277459d2STrond Myklebust struct nfs_page *req = NULL; 113277459d2STrond Myklebust 11429418aa4SMel Gorman if (PagePrivate(page)) 115277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 11602d1426cSWeston Andros Adamson else if (unlikely(PageSwapCache(page))) 11702d1426cSWeston Andros Adamson req = nfs_page_search_commits_for_head_request_locked(nfsi, 11802d1426cSWeston Andros Adamson page); 11929418aa4SMel Gorman 12084d3a9a9SWeston Andros Adamson if (req) { 12184d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 12229418aa4SMel Gorman kref_get(&req->wb_kref); 12384d3a9a9SWeston Andros Adamson } 12429418aa4SMel Gorman 125277459d2STrond Myklebust return req; 126277459d2STrond Myklebust } 127277459d2STrond Myklebust 12884d3a9a9SWeston Andros Adamson /* 12984d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 13084d3a9a9SWeston Andros Adamson * 13184d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 13284d3a9a9SWeston Andros Adamson */ 13384d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 134277459d2STrond Myklebust { 135d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 136277459d2STrond Myklebust struct nfs_page *req = NULL; 137277459d2STrond Myklebust 138587142f8STrond Myklebust spin_lock(&inode->i_lock); 13984d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 140587142f8STrond Myklebust spin_unlock(&inode->i_lock); 141277459d2STrond Myklebust return req; 142277459d2STrond Myklebust } 143277459d2STrond Myklebust 1441da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1451da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1461da177e4SLinus Torvalds { 147d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 148a3d01454STrond Myklebust loff_t end, i_size; 149a3d01454STrond Myklebust pgoff_t end_index; 1501da177e4SLinus Torvalds 151a3d01454STrond Myklebust spin_lock(&inode->i_lock); 152a3d01454STrond Myklebust i_size = i_size_read(inode); 15309cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 1548cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 155a3d01454STrond Myklebust goto out; 156d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1571da177e4SLinus Torvalds if (i_size >= end) 158a3d01454STrond Myklebust goto out; 1591da177e4SLinus Torvalds i_size_write(inode, end); 160a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 161a3d01454STrond Myklebust out: 162a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds 165a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 166a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 167a301b777STrond Myklebust { 168d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 169a301b777STrond Myklebust } 170a301b777STrond Myklebust 171d72ddcbaSWeston Andros Adamson /* 172d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 173d72ddcbaSWeston Andros Adamson * @head - head request of page group 174d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 175d72ddcbaSWeston Andros Adamson * 176d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 177d72ddcbaSWeston Andros Adamson * page offset @page_offset. 178d72ddcbaSWeston Andros Adamson * 179d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 180d72ddcbaSWeston Andros Adamson * match is found. 181d72ddcbaSWeston Andros Adamson * 182d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 183d72ddcbaSWeston Andros Adamson */ 184d72ddcbaSWeston Andros Adamson static struct nfs_page * 185d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 186d72ddcbaSWeston Andros Adamson { 187d72ddcbaSWeston Andros Adamson struct nfs_page *req; 188d72ddcbaSWeston Andros Adamson 189d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 190d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 191d72ddcbaSWeston Andros Adamson 192d72ddcbaSWeston Andros Adamson req = head; 193d72ddcbaSWeston Andros Adamson do { 194d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 195d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 196d72ddcbaSWeston Andros Adamson return req; 197d72ddcbaSWeston Andros Adamson 198d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 199d72ddcbaSWeston Andros Adamson } while (req != head); 200d72ddcbaSWeston Andros Adamson 201d72ddcbaSWeston Andros Adamson return NULL; 202d72ddcbaSWeston Andros Adamson } 203d72ddcbaSWeston Andros Adamson 204d72ddcbaSWeston Andros Adamson /* 205d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 206d72ddcbaSWeston Andros Adamson * @head - head request of page group 207d72ddcbaSWeston Andros Adamson * 208d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 209d72ddcbaSWeston Andros Adamson * returns false otherwise 210d72ddcbaSWeston Andros Adamson */ 211d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 212d72ddcbaSWeston Andros Adamson { 213d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 214d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 215d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 216d72ddcbaSWeston Andros Adamson 217fd2f3a06SWeston Andros Adamson nfs_page_group_lock(req, false); 218d72ddcbaSWeston Andros Adamson 219d72ddcbaSWeston Andros Adamson do { 220d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 221d72ddcbaSWeston Andros Adamson if (tmp) { 222d72ddcbaSWeston Andros Adamson /* no way this should happen */ 223d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 224d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 225d72ddcbaSWeston Andros Adamson } 226d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 227d72ddcbaSWeston Andros Adamson 228d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 229d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 230d72ddcbaSWeston Andros Adamson return pos == len; 231d72ddcbaSWeston Andros Adamson } 232d72ddcbaSWeston Andros Adamson 2331da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2341da177e4SLinus Torvalds * covers the full page. 2351da177e4SLinus Torvalds */ 236d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2371da177e4SLinus Torvalds { 238d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2391da177e4SLinus Torvalds return; 240d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2411da177e4SLinus Torvalds return; 242d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2431da177e4SLinus Torvalds } 2441da177e4SLinus Torvalds 2451da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2461da177e4SLinus Torvalds { 247e87b4c7aSNeilBrown int ret = 0; 248cca588d6STrond Myklebust 249e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 250e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 251e87b4c7aSNeilBrown return ret; 2521da177e4SLinus Torvalds } 2531da177e4SLinus Torvalds 2541da177e4SLinus Torvalds /* 25589a09141SPeter Zijlstra * NFS congestion control 25689a09141SPeter Zijlstra */ 25789a09141SPeter Zijlstra 25889a09141SPeter Zijlstra int nfs_congestion_kb; 25989a09141SPeter Zijlstra 26089a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 26189a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 26289a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 26389a09141SPeter Zijlstra 264deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 26589a09141SPeter Zijlstra { 266*0db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 267*0db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 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) > 273*0db10944SJan Kara NFS_CONGESTION_ON_THRESH) 274*0db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 27589a09141SPeter Zijlstra } 27689a09141SPeter Zijlstra 27720633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 27889a09141SPeter Zijlstra { 27920633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 28089a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 28189a09141SPeter Zijlstra 28220633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 28320633f04SWeston Andros Adamson return; 28420633f04SWeston Andros Adamson 28520633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 286c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 287*0db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 28889a09141SPeter Zijlstra } 28989a09141SPeter Zijlstra 290d4581383SWeston Andros Adamson 291d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 292d4581383SWeston Andros Adamson * @req - an nfs request 293d4581383SWeston Andros Adamson * clears all page group related bits from @req 294d4581383SWeston Andros Adamson */ 295d4581383SWeston Andros Adamson static void 296d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 297e261f51fSTrond Myklebust { 298d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 299d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 300d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 301d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 302d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 303d4581383SWeston Andros Adamson } 304d4581383SWeston Andros Adamson 305d4581383SWeston Andros Adamson 306d4581383SWeston Andros Adamson /* 307d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 308d4581383SWeston Andros Adamson * 309d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 310d4581383SWeston Andros Adamson * 311d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 312d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 313d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 314d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 315d4581383SWeston Andros Adamson * 316d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 317d4581383SWeston Andros Adamson * and BOTH will be released before returning. 318d4581383SWeston Andros Adamson * 319d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 320d4581383SWeston Andros Adamson */ 321d4581383SWeston Andros Adamson static int 322d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 323d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 324d4581383SWeston Andros Adamson __releases(&inode->i_lock) 325d4581383SWeston Andros Adamson { 326d4581383SWeston Andros Adamson struct nfs_page *tmp; 327e261f51fSTrond Myklebust int ret; 328e261f51fSTrond Myklebust 329d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 330d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 331d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 332d4581383SWeston Andros Adamson 333d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 334d4581383SWeston Andros Adamson 335d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 336d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 337d4581383SWeston Andros Adamson 338d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 339587142f8STrond Myklebust spin_unlock(&inode->i_lock); 340d4581383SWeston Andros Adamson 341d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 342d4581383SWeston Andros Adamson nfs_release_request(head); 343d4581383SWeston Andros Adamson 344cfb506e1STrond Myklebust if (!nonblock) 345e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 346cfb506e1STrond Myklebust else 347cfb506e1STrond Myklebust ret = -EAGAIN; 348e261f51fSTrond Myklebust nfs_release_request(req); 349d4581383SWeston Andros Adamson 350d4581383SWeston Andros Adamson return ret; 351e261f51fSTrond Myklebust } 352d4581383SWeston Andros Adamson 353d4581383SWeston Andros Adamson /* 354d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 355d4581383SWeston Andros Adamson * 356d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 357d4581383SWeston Andros Adamson * @old_head - the old head of the list 358d4581383SWeston Andros Adamson * 359d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 360d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 361d4581383SWeston Andros Adamson * with a reference held by some other context. 362d4581383SWeston Andros Adamson */ 363d4581383SWeston Andros Adamson static void 364d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 365d4581383SWeston Andros Adamson struct nfs_page *old_head) 366d4581383SWeston Andros Adamson { 367d4581383SWeston Andros Adamson while (destroy_list) { 368d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 369d4581383SWeston Andros Adamson 370d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 371d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 372d4581383SWeston Andros Adamson 373d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 374d4581383SWeston Andros Adamson 375d4581383SWeston Andros Adamson /* make sure old group is not used */ 376d4581383SWeston Andros Adamson subreq->wb_head = subreq; 377d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 378d4581383SWeston Andros Adamson 379d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 380d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 381d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 382d4581383SWeston Andros Adamson 383d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 384d4581383SWeston Andros Adamson /* release ref on old head request */ 385d4581383SWeston Andros Adamson nfs_release_request(old_head); 386d4581383SWeston Andros Adamson 387d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 388d4581383SWeston Andros Adamson 389d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 390d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 391d4581383SWeston Andros Adamson nfs_release_request(subreq); 392d4581383SWeston Andros Adamson else 393d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 394d4581383SWeston Andros Adamson } else { 395d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 396d4581383SWeston Andros Adamson /* zombie requests have already released the last 397d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 398d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 399d4581383SWeston Andros Adamson * group, simply free the request */ 400d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 401d4581383SWeston Andros Adamson nfs_free_request(subreq); 402d4581383SWeston Andros Adamson } 403d4581383SWeston Andros Adamson } 404d4581383SWeston Andros Adamson } 405d4581383SWeston Andros Adamson 406d4581383SWeston Andros Adamson /* 407d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 408d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 409d4581383SWeston Andros Adamson * operations for this page. 410d4581383SWeston Andros Adamson * 411d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 412d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 413d4581383SWeston Andros Adamson * 414d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 415d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 416d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 417d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 418d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 419d4581383SWeston Andros Adamson * 420d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 421d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 422d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 423d4581383SWeston Andros Adamson * error was encountered. 424d4581383SWeston Andros Adamson */ 425d4581383SWeston Andros Adamson static struct nfs_page * 426d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 427d4581383SWeston Andros Adamson { 428d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 429d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 430d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 431d4581383SWeston Andros Adamson unsigned int total_bytes; 432d4581383SWeston Andros Adamson int ret; 433d4581383SWeston Andros Adamson 434d4581383SWeston Andros Adamson try_again: 435d4581383SWeston Andros Adamson total_bytes = 0; 436d4581383SWeston Andros Adamson 437d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 438d4581383SWeston Andros Adamson 439d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 440d4581383SWeston Andros Adamson 441d4581383SWeston Andros Adamson /* 442d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 443d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 444d4581383SWeston Andros Adamson * until the head reference is released. 445d4581383SWeston Andros Adamson */ 446d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 447d4581383SWeston Andros Adamson 448d4581383SWeston Andros Adamson if (!head) { 449587142f8STrond Myklebust spin_unlock(&inode->i_lock); 450d4581383SWeston Andros Adamson return NULL; 451d4581383SWeston Andros Adamson } 452d4581383SWeston Andros Adamson 4537c3af975SWeston Andros Adamson /* holding inode lock, so always make a non-blocking call to try the 4547c3af975SWeston Andros Adamson * page group lock */ 455fd2f3a06SWeston Andros Adamson ret = nfs_page_group_lock(head, true); 45694970014SWeston Andros Adamson if (ret < 0) { 45794970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 4587c3af975SWeston Andros Adamson 4597c3af975SWeston Andros Adamson if (!nonblock && ret == -EAGAIN) { 4607c3af975SWeston Andros Adamson nfs_page_group_lock_wait(head); 4617c3af975SWeston Andros Adamson nfs_release_request(head); 4627c3af975SWeston Andros Adamson goto try_again; 4637c3af975SWeston Andros Adamson } 4647c3af975SWeston Andros Adamson 46594970014SWeston Andros Adamson nfs_release_request(head); 466e7029206SWeston Andros Adamson return ERR_PTR(ret); 46794970014SWeston Andros Adamson } 4687c3af975SWeston Andros Adamson 4697c3af975SWeston Andros Adamson /* lock each request in the page group */ 470d4581383SWeston Andros Adamson subreq = head; 471d4581383SWeston Andros Adamson do { 472d4581383SWeston Andros Adamson /* 473d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 474309a1d65SWeston Andros Adamson * and in order - but may be repeated (mirrored writes). 475d4581383SWeston Andros Adamson */ 476309a1d65SWeston Andros Adamson if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 477d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 478d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 479309a1d65SWeston Andros Adamson } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 480309a1d65SWeston Andros Adamson ((subreq->wb_offset + subreq->wb_bytes) > 481309a1d65SWeston Andros Adamson (head->wb_offset + total_bytes)))) { 482309a1d65SWeston Andros Adamson nfs_page_group_unlock(head); 483309a1d65SWeston Andros Adamson spin_unlock(&inode->i_lock); 484309a1d65SWeston Andros Adamson return ERR_PTR(-EIO); 485309a1d65SWeston Andros Adamson } 486d4581383SWeston Andros Adamson 487d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 488d4581383SWeston Andros Adamson /* releases page group bit lock and 489d4581383SWeston Andros Adamson * inode spin lock and all references */ 490d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 491d4581383SWeston Andros Adamson subreq, nonblock); 492d4581383SWeston Andros Adamson 493d4581383SWeston Andros Adamson if (ret == 0) 494d4581383SWeston Andros Adamson goto try_again; 495d4581383SWeston Andros Adamson 496d4581383SWeston Andros Adamson return ERR_PTR(ret); 497d4581383SWeston Andros Adamson } 498d4581383SWeston Andros Adamson 499d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 500d4581383SWeston Andros Adamson } while (subreq != head); 501d4581383SWeston Andros Adamson 502d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 503d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 504d4581383SWeston Andros Adamson subreq = head; 505d4581383SWeston Andros Adamson do { 506411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 507d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 508d4581383SWeston Andros Adamson } while (subreq != head); 509d4581383SWeston Andros Adamson 510d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 511d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 512d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 513d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 514d4581383SWeston Andros Adamson head->wb_this_page = head; 515d4581383SWeston Andros Adamson 516d4581383SWeston Andros Adamson /* change head request to cover whole range that 517d4581383SWeston Andros Adamson * the former page group covered */ 518d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 519d4581383SWeston Andros Adamson } 520d4581383SWeston Andros Adamson 521d4581383SWeston Andros Adamson /* 522d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 523d4581383SWeston Andros Adamson */ 524d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 525d4581383SWeston Andros Adamson 526d4581383SWeston Andros Adamson /* 527d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 528d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 529d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 530d4581383SWeston Andros Adamson */ 531d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 532d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 533d4581383SWeston Andros Adamson 534d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 535d4581383SWeston Andros Adamson 536411a99adSWeston Andros Adamson /* drop lock to clean uprequests on destroy list */ 537d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 538d4581383SWeston Andros Adamson 539d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 540d4581383SWeston Andros Adamson 541d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 542d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 543d4581383SWeston Andros Adamson return head; 544612c9384STrond Myklebust } 545074cc1deSTrond Myklebust 5460bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5470bcbf039SPeng Tao { 5480bcbf039SPeng Tao nfs_unlock_request(req); 5490bcbf039SPeng Tao nfs_end_page_writeback(req); 5500bcbf039SPeng Tao nfs_release_request(req); 5510bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5520bcbf039SPeng Tao req->wb_page); 5530bcbf039SPeng Tao } 5540bcbf039SPeng Tao 555074cc1deSTrond Myklebust /* 556074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 557074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 558074cc1deSTrond Myklebust */ 559074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 560d6c843b9SPeng Tao struct page *page, bool nonblock, 561d6c843b9SPeng Tao bool launder) 562074cc1deSTrond Myklebust { 563074cc1deSTrond Myklebust struct nfs_page *req; 564074cc1deSTrond Myklebust int ret = 0; 565074cc1deSTrond Myklebust 566d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 567074cc1deSTrond Myklebust if (!req) 568074cc1deSTrond Myklebust goto out; 569074cc1deSTrond Myklebust ret = PTR_ERR(req); 570074cc1deSTrond Myklebust if (IS_ERR(req)) 571074cc1deSTrond Myklebust goto out; 572074cc1deSTrond Myklebust 573deed85e7STrond Myklebust nfs_set_page_writeback(page); 574deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 575074cc1deSTrond Myklebust 576deed85e7STrond Myklebust ret = 0; 577f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 578074cc1deSTrond Myklebust ret = pgio->pg_error; 5790bcbf039SPeng Tao /* 580d6c843b9SPeng Tao * Remove the problematic req upon fatal errors 581d6c843b9SPeng Tao * in launder case, while other dirty pages can 582d6c843b9SPeng Tao * still be around until they get flushed. 5830bcbf039SPeng Tao */ 5840bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 5850bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 586d6c843b9SPeng Tao if (launder) { 5870bcbf039SPeng Tao nfs_write_error_remove_page(req); 588d6c843b9SPeng Tao goto out; 589d6c843b9SPeng Tao } 590d6c843b9SPeng Tao } 5910bcbf039SPeng Tao nfs_redirty_request(req); 5920bcbf039SPeng Tao ret = -EAGAIN; 59340f90271STrond Myklebust } else 59440f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 59540f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 596074cc1deSTrond Myklebust out: 597074cc1deSTrond Myklebust return ret; 598e261f51fSTrond Myklebust } 599e261f51fSTrond Myklebust 600d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 601d6c843b9SPeng Tao struct nfs_pageio_descriptor *pgio, bool launder) 602f758c885STrond Myklebust { 603cfb506e1STrond Myklebust int ret; 604f758c885STrond Myklebust 6058cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 606d6c843b9SPeng Tao ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE, 607d6c843b9SPeng Tao launder); 608cfb506e1STrond Myklebust if (ret == -EAGAIN) { 609cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 610cfb506e1STrond Myklebust ret = 0; 611cfb506e1STrond Myklebust } 612cfb506e1STrond Myklebust return ret; 613f758c885STrond Myklebust } 614f758c885STrond Myklebust 615e261f51fSTrond Myklebust /* 6161da177e4SLinus Torvalds * Write an mmapped page to the server. 6171da177e4SLinus Torvalds */ 618d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 619d6c843b9SPeng Tao struct writeback_control *wbc, 620d6c843b9SPeng Tao bool launder) 6211da177e4SLinus Torvalds { 622f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 62340f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 624e261f51fSTrond Myklebust int err; 6251da177e4SLinus Torvalds 62640f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 627811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 628a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 629d6c843b9SPeng Tao err = nfs_do_writepage(page, wbc, &pgio, launder); 630f758c885STrond Myklebust nfs_pageio_complete(&pgio); 631f758c885STrond Myklebust if (err < 0) 6324d770ccfSTrond Myklebust return err; 633f758c885STrond Myklebust if (pgio.pg_error < 0) 634f758c885STrond Myklebust return pgio.pg_error; 635f758c885STrond Myklebust return 0; 6364d770ccfSTrond Myklebust } 6374d770ccfSTrond Myklebust 6384d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6394d770ccfSTrond Myklebust { 640f758c885STrond Myklebust int ret; 6414d770ccfSTrond Myklebust 642d6c843b9SPeng Tao ret = nfs_writepage_locked(page, wbc, false); 6431da177e4SLinus Torvalds unlock_page(page); 644f758c885STrond Myklebust return ret; 645f758c885STrond Myklebust } 646f758c885STrond Myklebust 647f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 648f758c885STrond Myklebust { 649f758c885STrond Myklebust int ret; 650f758c885STrond Myklebust 651d6c843b9SPeng Tao ret = nfs_do_writepage(page, wbc, data, false); 652f758c885STrond Myklebust unlock_page(page); 653f758c885STrond Myklebust return ret; 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds 6561da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6571da177e4SLinus Torvalds { 6581da177e4SLinus Torvalds struct inode *inode = mapping->host; 659c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6601da177e4SLinus Torvalds int err; 6611da177e4SLinus Torvalds 66291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 66391d5b470SChuck Lever 664a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 665a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 666f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 667c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 66872cb77f4STrond Myklebust 669f758c885STrond Myklebust if (err < 0) 67072cb77f4STrond Myklebust goto out_err; 67172cb77f4STrond Myklebust err = pgio.pg_error; 67272cb77f4STrond Myklebust if (err < 0) 67372cb77f4STrond Myklebust goto out_err; 674c63c7b05STrond Myklebust return 0; 67572cb77f4STrond Myklebust out_err: 67672cb77f4STrond Myklebust return err; 6771da177e4SLinus Torvalds } 6781da177e4SLinus Torvalds 6791da177e4SLinus Torvalds /* 6801da177e4SLinus Torvalds * Insert a write request into an inode 6811da177e4SLinus Torvalds */ 682d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6831da177e4SLinus Torvalds { 6841da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 685e7d39069STrond Myklebust 6862bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 6872bfc6e56SWeston Andros Adamson 688e7d39069STrond Myklebust /* Lock the request! */ 6897ad84aa9STrond Myklebust nfs_lock_request(req); 690e7d39069STrond Myklebust 691e7d39069STrond Myklebust spin_lock(&inode->i_lock); 692cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 693cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 694a9a4a87aSTrond Myklebust inode->i_version++; 69529418aa4SMel Gorman /* 69629418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 69729418aa4SMel Gorman * with invalidate/truncate. 69829418aa4SMel Gorman */ 69929418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7002df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 701deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 702277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 70329418aa4SMel Gorman } 704cb1410c7SWeston Andros Adamson nfsi->nrequests++; 70517089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 706cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 707cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 708cb1410c7SWeston Andros Adamson * adding subrequests. */ 70917089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 710c03b4024STrond Myklebust kref_get(&req->wb_kref); 711e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7121da177e4SLinus Torvalds } 7131da177e4SLinus Torvalds 7141da177e4SLinus Torvalds /* 71589a09141SPeter Zijlstra * Remove a write request from an inode 7161da177e4SLinus Torvalds */ 7171da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7181da177e4SLinus Torvalds { 7192b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7201da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 72120633f04SWeston Andros Adamson struct nfs_page *head; 72220633f04SWeston Andros Adamson 72320633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 72420633f04SWeston Andros Adamson head = req->wb_head; 7251da177e4SLinus Torvalds 726587142f8STrond Myklebust spin_lock(&inode->i_lock); 72767911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 72820633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 72920633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 73020633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 73129418aa4SMel Gorman } 732cb1410c7SWeston Andros Adamson nfsi->nrequests--; 733cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 734cb1410c7SWeston Andros Adamson } else { 735cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 736cb1410c7SWeston Andros Adamson nfsi->nrequests--; 737587142f8STrond Myklebust spin_unlock(&inode->i_lock); 73820633f04SWeston Andros Adamson } 73917089a29SWeston Andros Adamson 74017089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7411da177e4SLinus Torvalds nfs_release_request(req); 7421da177e4SLinus Torvalds } 7431da177e4SLinus Torvalds 74461822ab5STrond Myklebust static void 7456d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 74661822ab5STrond Myklebust { 74767911c8fSAnna Schumaker if (req->wb_page) 74861822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 74961822ab5STrond Myklebust } 75061822ab5STrond Myklebust 7513a3908c8STrond Myklebust /* 7523a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 7533a3908c8STrond Myklebust * 7543a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 7553a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 7563a3908c8STrond Myklebust * 7573a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 7583a3908c8STrond Myklebust */ 7593a3908c8STrond Myklebust static struct nfs_page * 7603a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 7613a3908c8STrond Myklebust struct page *page) 7623a3908c8STrond Myklebust { 7633a3908c8STrond Myklebust struct nfs_page *freq, *t; 7643a3908c8STrond Myklebust struct nfs_commit_info cinfo; 7653a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 7663a3908c8STrond Myklebust 7673a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 7683a3908c8STrond Myklebust 7693a3908c8STrond Myklebust /* search through pnfs commit lists */ 7703a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 7713a3908c8STrond Myklebust if (freq) 7723a3908c8STrond Myklebust return freq->wb_head; 7733a3908c8STrond Myklebust 7743a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 7753a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 7763a3908c8STrond Myklebust if (freq->wb_page == page) 7773a3908c8STrond Myklebust return freq->wb_head; 7783a3908c8STrond Myklebust } 7793a3908c8STrond Myklebust 7803a3908c8STrond Myklebust return NULL; 7813a3908c8STrond Myklebust } 7823a3908c8STrond Myklebust 7838dd37758STrond Myklebust /** 78486d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 78586d80f97STrond Myklebust * @req: pointer to a struct nfs_page 78686d80f97STrond Myklebust * @dst: commit list head 78786d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 78886d80f97STrond Myklebust * 78986d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 79086d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 79186d80f97STrond Myklebust * the MM page stats. 79286d80f97STrond Myklebust * 793fe238e60SDave Wysochanski * The caller must hold cinfo->inode->i_lock, and the nfs_page lock. 79486d80f97STrond Myklebust */ 79586d80f97STrond Myklebust void 79686d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 79786d80f97STrond Myklebust struct nfs_commit_info *cinfo) 79886d80f97STrond Myklebust { 79986d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 80086d80f97STrond Myklebust nfs_list_add_request(req, dst); 80186d80f97STrond Myklebust cinfo->mds->ncommit++; 80286d80f97STrond Myklebust } 80386d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 80486d80f97STrond Myklebust 80586d80f97STrond Myklebust /** 8068dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8078dd37758STrond Myklebust * @req: pointer to a struct nfs_page 808ea2cf228SFred Isaman * @dst: commit list head 809ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8108dd37758STrond Myklebust * 811ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8128dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8138dd37758STrond Myklebust * the MM page stats. 8148dd37758STrond Myklebust * 815ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8168dd37758STrond Myklebust * holding the nfs_page lock. 8178dd37758STrond Myklebust */ 8188dd37758STrond Myklebust void 8196272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8208dd37758STrond Myklebust { 821fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 8226272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 823fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 82467911c8fSAnna Schumaker if (req->wb_page) 82586d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8268dd37758STrond Myklebust } 8278dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8288dd37758STrond Myklebust 8298dd37758STrond Myklebust /** 8308dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8318dd37758STrond Myklebust * @req: pointer to a nfs_page 832ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8338dd37758STrond Myklebust * 834ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8358dd37758STrond Myklebust * number of outstanding requests requiring a commit 8368dd37758STrond Myklebust * It does not update the MM page stats. 8378dd37758STrond Myklebust * 838ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8398dd37758STrond Myklebust */ 8408dd37758STrond Myklebust void 841ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 842ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8438dd37758STrond Myklebust { 8448dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 8458dd37758STrond Myklebust return; 8468dd37758STrond Myklebust nfs_list_remove_request(req); 847ea2cf228SFred Isaman cinfo->mds->ncommit--; 8488dd37758STrond Myklebust } 8498dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 8508dd37758STrond Myklebust 851ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 852ea2cf228SFred Isaman struct inode *inode) 853ea2cf228SFred Isaman { 854fe238e60SDave Wysochanski cinfo->inode = inode; 855ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 856ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 857b359f9d0SFred Isaman cinfo->dreq = NULL; 858f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 859ea2cf228SFred Isaman } 860ea2cf228SFred Isaman 861ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 862ea2cf228SFred Isaman struct inode *inode, 863ea2cf228SFred Isaman struct nfs_direct_req *dreq) 864ea2cf228SFred Isaman { 8651763da12SFred Isaman if (dreq) 8661763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8671763da12SFred Isaman else 868ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 869ea2cf228SFred Isaman } 870ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8718dd37758STrond Myklebust 8721da177e4SLinus Torvalds /* 8731da177e4SLinus Torvalds * Add a request to the inode's commit list. 8741da177e4SLinus Torvalds */ 8751763da12SFred Isaman void 876ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 877b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 8781da177e4SLinus Torvalds { 879b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 8808dd37758STrond Myklebust return; 8816272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 8821da177e4SLinus Torvalds } 8838e821cadSTrond Myklebust 884d6d6dc7cSFred Isaman static void 885d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 886e468bae9STrond Myklebust { 88711fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 88893f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 88993f78d88STejun Heo WB_RECLAIMABLE); 890e468bae9STrond Myklebust } 891d6d6dc7cSFred Isaman 892411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 8938dd37758STrond Myklebust static void 894d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 895d6d6dc7cSFred Isaman { 8968dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 8972b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 898ea2cf228SFred Isaman struct nfs_commit_info cinfo; 899d6d6dc7cSFred Isaman 900ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 901ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 902ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 903d6d6dc7cSFred Isaman } 9048dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9058dd37758STrond Myklebust } 906e468bae9STrond Myklebust } 907e468bae9STrond Myklebust 908d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9098e821cadSTrond Myklebust { 910c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 911d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 912c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9138e821cadSTrond Myklebust } 9148e821cadSTrond Myklebust 915061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9166c75dc0dSFred Isaman { 917ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9186c75dc0dSFred Isaman unsigned long bytes = 0; 9196c75dc0dSFred Isaman 9206c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9216c75dc0dSFred Isaman goto out; 922ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9236c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9246c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9256c75dc0dSFred Isaman 9266c75dc0dSFred Isaman bytes += req->wb_bytes; 9276c75dc0dSFred Isaman nfs_list_remove_request(req); 9286c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9296c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 930d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9316c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9326c75dc0dSFred Isaman goto remove_req; 9336c75dc0dSFred Isaman } 934c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 935f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 936b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 937a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 9386c75dc0dSFred Isaman goto next; 9396c75dc0dSFred Isaman } 9406c75dc0dSFred Isaman remove_req: 9416c75dc0dSFred Isaman nfs_inode_remove_request(req); 9426c75dc0dSFred Isaman next: 9431d1afcbcSTrond Myklebust nfs_unlock_request(req); 94420633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9453aff4ebbSTrond Myklebust nfs_release_request(req); 9466c75dc0dSFred Isaman } 9476c75dc0dSFred Isaman out: 9486c75dc0dSFred Isaman hdr->release(hdr); 9496c75dc0dSFred Isaman } 9506c75dc0dSFred Isaman 951ce59515cSAnna Schumaker unsigned long 952ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 953fb8a1f11STrond Myklebust { 954ea2cf228SFred Isaman return cinfo->mds->ncommit; 955fb8a1f11STrond Myklebust } 956fb8a1f11STrond Myklebust 957fe238e60SDave Wysochanski /* cinfo->inode->i_lock held by caller */ 9581763da12SFred Isaman int 959ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 960ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 961d6d6dc7cSFred Isaman { 962d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 963d6d6dc7cSFred Isaman int ret = 0; 964d6d6dc7cSFred Isaman 965d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9668dd37758STrond Myklebust if (!nfs_lock_request(req)) 9678dd37758STrond Myklebust continue; 9687ad84aa9STrond Myklebust kref_get(&req->wb_kref); 969fe238e60SDave Wysochanski if (cond_resched_lock(&cinfo->inode->i_lock)) 9703b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 971ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9728dd37758STrond Myklebust nfs_list_add_request(req, dst); 973d6d6dc7cSFred Isaman ret++; 9741763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 975d6d6dc7cSFred Isaman break; 976d6d6dc7cSFred Isaman } 977d6d6dc7cSFred Isaman return ret; 978d6d6dc7cSFred Isaman } 979d6d6dc7cSFred Isaman 9801da177e4SLinus Torvalds /* 9811da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9821da177e4SLinus Torvalds * @inode: NFS inode to scan 983ea2cf228SFred Isaman * @dst: mds destination list 984ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9851da177e4SLinus Torvalds * 9861da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 9871da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 9881da177e4SLinus Torvalds */ 9891763da12SFred Isaman int 990ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 991ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9921da177e4SLinus Torvalds { 993d6d6dc7cSFred Isaman int ret = 0; 994fb8a1f11STrond Myklebust 995fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 996ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 9978dd37758STrond Myklebust const int max = INT_MAX; 998d6d6dc7cSFred Isaman 999ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1000ea2cf228SFred Isaman cinfo, max); 1001ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1002d6d6dc7cSFred Isaman } 1003fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 1004ff778d02STrond Myklebust return ret; 10051da177e4SLinus Torvalds } 1006d6d6dc7cSFred Isaman 10071da177e4SLinus Torvalds /* 1008e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1009e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10101da177e4SLinus Torvalds * 1011e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1012e7d39069STrond Myklebust * to disk. 10131da177e4SLinus Torvalds */ 1014e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1015e7d39069STrond Myklebust struct page *page, 1016e7d39069STrond Myklebust unsigned int offset, 1017e7d39069STrond Myklebust unsigned int bytes) 10181da177e4SLinus Torvalds { 1019e7d39069STrond Myklebust struct nfs_page *req; 1020e7d39069STrond Myklebust unsigned int rqend; 1021e7d39069STrond Myklebust unsigned int end; 10221da177e4SLinus Torvalds int error; 1023277459d2STrond Myklebust 1024e7d39069STrond Myklebust if (!PagePrivate(page)) 1025e7d39069STrond Myklebust return NULL; 1026e7d39069STrond Myklebust 1027e7d39069STrond Myklebust end = offset + bytes; 1028e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1029e7d39069STrond Myklebust 1030e7d39069STrond Myklebust for (;;) { 103184d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1032e7d39069STrond Myklebust if (req == NULL) 1033e7d39069STrond Myklebust goto out_unlock; 1034e7d39069STrond Myklebust 10352bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10362bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10372bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10382bfc6e56SWeston Andros Adamson 1039e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1040e7d39069STrond Myklebust /* 1041e7d39069STrond Myklebust * Tell the caller to flush out the request if 1042e7d39069STrond Myklebust * the offsets are non-contiguous. 1043e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1044e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1045e7d39069STrond Myklebust */ 1046e468bae9STrond Myklebust if (offset > rqend 1047e7d39069STrond Myklebust || end < req->wb_offset) 1048e7d39069STrond Myklebust goto out_flushme; 1049e7d39069STrond Myklebust 10507ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1051e7d39069STrond Myklebust break; 1052e7d39069STrond Myklebust 1053e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1054587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10551da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10561da177e4SLinus Torvalds nfs_release_request(req); 1057e7d39069STrond Myklebust if (error != 0) 1058e7d39069STrond Myklebust goto out_err; 1059e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10601da177e4SLinus Torvalds } 10611da177e4SLinus Torvalds 10621da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10631da177e4SLinus Torvalds if (offset < req->wb_offset) { 10641da177e4SLinus Torvalds req->wb_offset = offset; 10651da177e4SLinus Torvalds req->wb_pgbase = offset; 10661da177e4SLinus Torvalds } 10671da177e4SLinus Torvalds if (end > rqend) 10681da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1069e7d39069STrond Myklebust else 1070e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1071e7d39069STrond Myklebust out_unlock: 1072ca138f36SFred Isaman if (req) 10738dd37758STrond Myklebust nfs_clear_request_commit(req); 1074411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1075e7d39069STrond Myklebust return req; 1076e7d39069STrond Myklebust out_flushme: 1077e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1078e7d39069STrond Myklebust nfs_release_request(req); 1079e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1080e7d39069STrond Myklebust out_err: 1081e7d39069STrond Myklebust return ERR_PTR(error); 1082e7d39069STrond Myklebust } 10831da177e4SLinus Torvalds 1084e7d39069STrond Myklebust /* 1085e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1086e7d39069STrond Myklebust * 1087e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1088e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1089e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1090e7d39069STrond Myklebust */ 1091e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1092e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1093e7d39069STrond Myklebust { 1094d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1095e7d39069STrond Myklebust struct nfs_page *req; 1096e7d39069STrond Myklebust 1097e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1098e7d39069STrond Myklebust if (req != NULL) 1099e7d39069STrond Myklebust goto out; 11002bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1101e7d39069STrond Myklebust if (IS_ERR(req)) 1102e7d39069STrond Myklebust goto out; 1103d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1104efc91ed0STrond Myklebust out: 110561e930a9STrond Myklebust return req; 11061da177e4SLinus Torvalds } 11071da177e4SLinus Torvalds 1108e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1109e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1110e7d39069STrond Myklebust { 1111e7d39069STrond Myklebust struct nfs_page *req; 1112e7d39069STrond Myklebust 1113e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1114e7d39069STrond Myklebust if (IS_ERR(req)) 1115e7d39069STrond Myklebust return PTR_ERR(req); 1116e7d39069STrond Myklebust /* Update file length */ 1117e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1118d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1119a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11201d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1121e7d39069STrond Myklebust return 0; 1122e7d39069STrond Myklebust } 1123e7d39069STrond Myklebust 11241da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11251da177e4SLinus Torvalds { 1126cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11272a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1128bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11291da177e4SLinus Torvalds struct nfs_page *req; 11301a54533eSTrond Myklebust int do_flush, status; 11311da177e4SLinus Torvalds /* 11321da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11331da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11341da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11351da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11361da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11371da177e4SLinus Torvalds * dropped page. 11381da177e4SLinus Torvalds */ 11391a54533eSTrond Myklebust do { 114084d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11411a54533eSTrond Myklebust if (req == NULL) 11421a54533eSTrond Myklebust return 0; 11432a369153STrond Myklebust l_ctx = req->wb_lock_context; 1144138a2935STrond Myklebust do_flush = req->wb_page != page || 1145138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 11462bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11472bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1148bd61e0a9SJeff Layton if (l_ctx && flctx && 1149bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1150bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1151d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 11525263e31eSJeff Layton } 11531da177e4SLinus Torvalds nfs_release_request(req); 11541a54533eSTrond Myklebust if (!do_flush) 11551a54533eSTrond Myklebust return 0; 1156d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11571a54533eSTrond Myklebust } while (status == 0); 11581a54533eSTrond Myklebust return status; 11591da177e4SLinus Torvalds } 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds /* 1162dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1163dc24826bSAndy Adamson * expire soon. 1164dc24826bSAndy Adamson * 1165dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1166dc24826bSAndy Adamson * 1167dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1168dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1169dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1170dc24826bSAndy Adamson */ 1171dc24826bSAndy Adamson int 1172dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1173dc24826bSAndy Adamson { 1174dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1175dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1176dc24826bSAndy Adamson 1177dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1178dc24826bSAndy Adamson } 1179dc24826bSAndy Adamson 1180dc24826bSAndy Adamson /* 1181dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1182dc24826bSAndy Adamson */ 1183ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1184dc24826bSAndy Adamson { 1185ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1186ce52914eSScott Mayhew 1187ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1188dc24826bSAndy Adamson } 1189dc24826bSAndy Adamson 1190dc24826bSAndy Adamson /* 11915d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 11925d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 11935d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 11945d47a356STrond Myklebust */ 11958d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 11965d47a356STrond Myklebust { 1197d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1198d529ef83SJeff Layton 11998d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12008d197a56STrond Myklebust goto out; 120118dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1202d529ef83SJeff Layton return false; 12034db72b40SJeff Layton smp_rmb(); 1204d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12058d197a56STrond Myklebust return false; 12068d197a56STrond Myklebust out: 120718dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 120818dd78c4SScott Mayhew return false; 12098d197a56STrond Myklebust return PageUptodate(page) != 0; 12105d47a356STrond Myklebust } 12115d47a356STrond Myklebust 12125263e31eSJeff Layton static bool 12135263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12145263e31eSJeff Layton { 12155263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12165263e31eSJeff Layton fl->fl_type == F_WRLCK; 12175263e31eSJeff Layton } 12185263e31eSJeff Layton 1219c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1220c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1221c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1222c7559663SScott Mayhew * inefficiencies. 1223c7559663SScott Mayhew * 1224263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1225263b4509SScott Mayhew * of the checks. 1226c7559663SScott Mayhew */ 1227c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1228c7559663SScott Mayhew { 12295263e31eSJeff Layton int ret; 12305263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12315263e31eSJeff Layton struct file_lock *fl; 12325263e31eSJeff Layton 1233c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1234c7559663SScott Mayhew return 0; 1235263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1236263b4509SScott Mayhew return 0; 1237c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1238c7559663SScott Mayhew return 1; 1239bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1240bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 12418fa4592aSTrond Myklebust return 1; 12425263e31eSJeff Layton 12435263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12445263e31eSJeff Layton ret = 0; 12456109c850SJeff Layton spin_lock(&flctx->flc_lock); 1246bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1247bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1248bd61e0a9SJeff Layton fl_list); 1249bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12505263e31eSJeff Layton ret = 1; 1251bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 12525263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 12535263e31eSJeff Layton fl_list); 12545263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 12555263e31eSJeff Layton ret = 1; 12565263e31eSJeff Layton } 12576109c850SJeff Layton spin_unlock(&flctx->flc_lock); 12585263e31eSJeff Layton return ret; 1259c7559663SScott Mayhew } 1260c7559663SScott Mayhew 12615d47a356STrond Myklebust /* 12621da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12631da177e4SLinus Torvalds * 12641da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12651da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12661da177e4SLinus Torvalds */ 12671da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12681da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12691da177e4SLinus Torvalds { 1270cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1271d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12721da177e4SLinus Torvalds int status = 0; 12731da177e4SLinus Torvalds 127491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 127591d5b470SChuck Lever 12766de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12776de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12781da177e4SLinus Torvalds 1279149a4fddSBenjamin Coddington if (!count) 1280149a4fddSBenjamin Coddington goto out; 1281149a4fddSBenjamin Coddington 1282c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 128349a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12841da177e4SLinus Torvalds offset = 0; 12851da177e4SLinus Torvalds } 12861da177e4SLinus Torvalds 1287e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 128803fa9e84STrond Myklebust if (status < 0) 128903fa9e84STrond Myklebust nfs_set_pageerror(page); 129059b7c05fSTrond Myklebust else 129159b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1292149a4fddSBenjamin Coddington out: 129348186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 12941da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 12951da177e4SLinus Torvalds return status; 12961da177e4SLinus Torvalds } 12971da177e4SLinus Torvalds 12983ff7576dSTrond Myklebust static int flush_task_priority(int how) 12991da177e4SLinus Torvalds { 13001da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13011da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13021da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13031da177e4SLinus Torvalds case FLUSH_LOWPRI: 13041da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13051da177e4SLinus Torvalds } 13061da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13071da177e4SLinus Torvalds } 13081da177e4SLinus Torvalds 1309d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1310d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1311abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13121ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13131da177e4SLinus Torvalds { 13143ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13151da177e4SLinus Torvalds 13161ed26f33SAnna Schumaker task_setup_data->priority = priority; 1317abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1318d138d5d1SAndy Adamson 1319abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1320d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1321275acaafSTrond Myklebust } 1322275acaafSTrond Myklebust 13236d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13246d884e8fSFred * call this on each, which will prepare them to be retried on next 13256d884e8fSFred * writeback using standard nfs. 13266d884e8fSFred */ 13276d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13286d884e8fSFred { 13296d884e8fSFred nfs_mark_request_dirty(req); 1330c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 13311d1afcbcSTrond Myklebust nfs_unlock_request(req); 133220633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13333aff4ebbSTrond Myklebust nfs_release_request(req); 13346d884e8fSFred } 13356d884e8fSFred 1336061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13376c75dc0dSFred Isaman { 13386c75dc0dSFred Isaman struct nfs_page *req; 13396c75dc0dSFred Isaman 13406c75dc0dSFred Isaman while (!list_empty(head)) { 13416c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13426c75dc0dSFred Isaman nfs_list_remove_request(req); 13436c75dc0dSFred Isaman nfs_redirty_request(req); 13446c75dc0dSFred Isaman } 13456c75dc0dSFred Isaman } 13466c75dc0dSFred Isaman 1347dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1348dc602dd7STrond Myklebust { 1349dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1350dc602dd7STrond Myklebust } 1351dc602dd7STrond Myklebust 1352061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1353061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1354061ae2edSFred Isaman .completion = nfs_write_completion, 1355dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1356061ae2edSFred Isaman }; 1357061ae2edSFred Isaman 135857208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1359a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1360061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 13611751c363STrond Myklebust { 1362a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 136341d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1364a20c93e3SChristoph Hellwig 1365a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1366a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1367a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1368a20c93e3SChristoph Hellwig #endif 13694a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 13704a0de55cSAnna Schumaker server->wsize, ioflags); 13711751c363STrond Myklebust } 1372ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13731751c363STrond Myklebust 1374dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1375dce81290STrond Myklebust { 1376a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1377a7d42ddbSWeston Andros Adamson 13786f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 13796f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 13806f29b9bbSKinglong Mee 138141d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1382a7d42ddbSWeston Andros Adamson 1383a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1384a7d42ddbSWeston Andros Adamson 1385a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1386a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1387dce81290STrond Myklebust } 13881f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1389dce81290STrond Myklebust 13901da177e4SLinus Torvalds 13910b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13920b7c0153SFred Isaman { 13930b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13940b7c0153SFred Isaman 13950b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13960b7c0153SFred Isaman } 13970b7c0153SFred Isaman 13981f2edbe3STrond Myklebust /* 13991f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14001f2edbe3STrond Myklebust */ 14011f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14021f2edbe3STrond Myklebust { 14031f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14041f2edbe3STrond Myklebust int kill = 0; 1405788e7a89STrond Myklebust 14061f2edbe3STrond Myklebust /* suid always must be killed */ 14071f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14081f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14091f2edbe3STrond Myklebust 14101f2edbe3STrond Myklebust /* 14111f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14121f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14131f2edbe3STrond Myklebust */ 14141f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14151f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14161f2edbe3STrond Myklebust 14171f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14181f2edbe3STrond Myklebust return kill; 14191f2edbe3STrond Myklebust 14201f2edbe3STrond Myklebust return 0; 14211f2edbe3STrond Myklebust } 1422788e7a89STrond Myklebust 1423a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1424a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1425a08a8cd3STrond Myklebust { 1426a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1427a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14282b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1429a08a8cd3STrond Myklebust 1430a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14312b83d3deSTrond Myklebust fattr->size = size; 14322b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 14332b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1434a08a8cd3STrond Myklebust return; 14352b83d3deSTrond Myklebust } 14362b83d3deSTrond Myklebust if (size != fattr->size) 1437a08a8cd3STrond Myklebust return; 1438a08a8cd3STrond Myklebust /* Set attribute barrier */ 1439a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 14402b83d3deSTrond Myklebust /* ...and update size */ 14412b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1442a08a8cd3STrond Myklebust } 1443a08a8cd3STrond Myklebust 1444a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1445a08a8cd3STrond Myklebust { 14462b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1447a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1448a08a8cd3STrond Myklebust 1449a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1450a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1451a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1452a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1453a08a8cd3STrond Myklebust } 1454a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1455a08a8cd3STrond Myklebust 14561da177e4SLinus Torvalds /* 14571da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 14581da177e4SLinus Torvalds */ 1459d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1460d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 14610eecb214SAnna Schumaker struct inode *inode) 14621da177e4SLinus Torvalds { 1463788e7a89STrond Myklebust int status; 14641da177e4SLinus Torvalds 1465f551e44fSChuck Lever /* 1466f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1467f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1468f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1469f551e44fSChuck Lever * another writer had changed the file, but some applications 1470f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1471f551e44fSChuck Lever */ 1472d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1473788e7a89STrond Myklebust if (status != 0) 14740eecb214SAnna Schumaker return status; 1475d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 147691d5b470SChuck Lever 1477d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1478d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 14791da177e4SLinus Torvalds /* We tried a write call, but the server did not 14801da177e4SLinus Torvalds * commit data to stable storage even though we 14811da177e4SLinus Torvalds * requested it. 14821da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 14831da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 14841da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 14851da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 14861da177e4SLinus Torvalds */ 14871da177e4SLinus Torvalds static unsigned long complain; 14881da177e4SLinus Torvalds 1489a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 14901da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 14911da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 14921da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1493cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1494d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 14951da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 14961da177e4SLinus Torvalds } 14971da177e4SLinus Torvalds } 14981f2edbe3STrond Myklebust 14991f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15001f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15011f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15020eecb214SAnna Schumaker return 0; 15030eecb214SAnna Schumaker } 15040eecb214SAnna Schumaker 15050eecb214SAnna Schumaker /* 15060eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15070eecb214SAnna Schumaker */ 1508d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1509d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15100eecb214SAnna Schumaker { 1511d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1512d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15131f2edbe3STrond Myklebust 15141f2edbe3STrond Myklebust if (resp->count < argp->count) { 15151da177e4SLinus Torvalds static unsigned long complain; 15161da177e4SLinus Torvalds 15176c75dc0dSFred Isaman /* This a short write! */ 1518d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 151991d5b470SChuck Lever 15201da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15216c75dc0dSFred Isaman if (resp->count == 0) { 15226c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15236c75dc0dSFred Isaman printk(KERN_WARNING 15246c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 15256c75dc0dSFred Isaman argp->count); 15266c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15276c75dc0dSFred Isaman } 1528d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15296c75dc0dSFred Isaman task->tk_status = -EIO; 15306c75dc0dSFred Isaman return; 15316c75dc0dSFred Isaman } 1532f8417b48SKinglong Mee 1533f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1534f8417b48SKinglong Mee if (!task->tk_ops) { 1535f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1536f8417b48SKinglong Mee return; 1537f8417b48SKinglong Mee } 1538f8417b48SKinglong Mee 15391da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 15401da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 15411da177e4SLinus Torvalds /* Resend from where the server left off */ 1542d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 15431da177e4SLinus Torvalds argp->offset += resp->count; 15441da177e4SLinus Torvalds argp->pgbase += resp->count; 15451da177e4SLinus Torvalds argp->count -= resp->count; 15461da177e4SLinus Torvalds } else { 15471da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 15481da177e4SLinus Torvalds * headaches in the case of a server crash. 15491da177e4SLinus Torvalds */ 15501da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 15511da177e4SLinus Torvalds } 1552d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 15531da177e4SLinus Torvalds } 15541da177e4SLinus Torvalds } 15551da177e4SLinus Torvalds 1556af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 155771d0a611STrond Myklebust { 1558af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1559af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1560af7cf057STrond Myklebust } 1561af7cf057STrond Myklebust 1562af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1563af7cf057STrond Myklebust { 1564af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1565af7cf057STrond Myklebust } 1566af7cf057STrond Myklebust 1567af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1568af7cf057STrond Myklebust { 1569af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1570af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 157171d0a611STrond Myklebust } 157271d0a611STrond Myklebust 15730b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 15741da177e4SLinus Torvalds { 15750b7c0153SFred Isaman put_nfs_open_context(data->context); 15760b7c0153SFred Isaman nfs_commit_free(data); 15771da177e4SLinus Torvalds } 1578e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15791da177e4SLinus Torvalds 15800b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1581c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 15829ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15839f0ec176SAndy Adamson int how, int flags) 15841da177e4SLinus Torvalds { 158507737691STrond Myklebust struct rpc_task *task; 15869ace33cdSFred Isaman int priority = flush_task_priority(how); 1587bdc7f021STrond Myklebust struct rpc_message msg = { 1588bdc7f021STrond Myklebust .rpc_argp = &data->args, 1589bdc7f021STrond Myklebust .rpc_resp = &data->res, 15909ace33cdSFred Isaman .rpc_cred = data->cred, 1591bdc7f021STrond Myklebust }; 159284115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 159307737691STrond Myklebust .task = &data->task, 15949ace33cdSFred Isaman .rpc_client = clnt, 1595bdc7f021STrond Myklebust .rpc_message = &msg, 15969ace33cdSFred Isaman .callback_ops = call_ops, 159784115e1cSTrond Myklebust .callback_data = data, 1598101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 15999f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16003ff7576dSTrond Myklebust .priority = priority, 160184115e1cSTrond Myklebust }; 1602788e7a89STrond Myklebust /* Set up the initial task struct. */ 1603c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16041da177e4SLinus Torvalds 1605b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1606bdc7f021STrond Myklebust 16078c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16088c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16098c21c62cSWeston Andros Adamson 161007737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1611dbae4c73STrond Myklebust if (IS_ERR(task)) 1612dbae4c73STrond Myklebust return PTR_ERR(task); 1613d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1614d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 161507737691STrond Myklebust rpc_put_task(task); 1616dbae4c73STrond Myklebust return 0; 16171da177e4SLinus Torvalds } 1618e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16191da177e4SLinus Torvalds 1620378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1621378520b8SPeng Tao { 1622378520b8SPeng Tao loff_t lwb = 0; 1623378520b8SPeng Tao struct nfs_page *req; 1624378520b8SPeng Tao 1625378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1626378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1627378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1628378520b8SPeng Tao 1629378520b8SPeng Tao return lwb; 1630378520b8SPeng Tao } 1631378520b8SPeng Tao 16321da177e4SLinus Torvalds /* 16339ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16349ace33cdSFred Isaman */ 16350b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1636988b6dceSFred Isaman struct list_head *head, 1637f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1638f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16399ace33cdSFred Isaman { 16409ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16412b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16429ace33cdSFred Isaman 16439ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16449ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 16459ace33cdSFred Isaman 16469ace33cdSFred Isaman list_splice_init(head, &data->pages); 16479ace33cdSFred Isaman 16489ace33cdSFred Isaman data->inode = inode; 16499ace33cdSFred Isaman data->cred = first->wb_context->cred; 1650988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1651378520b8SPeng Tao /* only set lwb for pnfs commit */ 1652378520b8SPeng Tao if (lseg) 1653378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 16549ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1655f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1656b359f9d0SFred Isaman data->dreq = cinfo->dreq; 16579ace33cdSFred Isaman 16589ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 16599ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 16609ace33cdSFred Isaman data->args.offset = 0; 16619ace33cdSFred Isaman data->args.count = 0; 16620b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 16639ace33cdSFred Isaman data->res.fattr = &data->fattr; 16649ace33cdSFred Isaman data->res.verf = &data->verf; 16659ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 16669ace33cdSFred Isaman } 1667e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 16689ace33cdSFred Isaman 1669e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1670ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1671b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1672b57ff130SWeston Andros Adamson u32 ds_commit_idx) 167364bfeb49SFred Isaman { 167464bfeb49SFred Isaman struct nfs_page *req; 167564bfeb49SFred Isaman 167664bfeb49SFred Isaman while (!list_empty(page_list)) { 167764bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 167864bfeb49SFred Isaman nfs_list_remove_request(req); 1679b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1680487b9b8aSTom Haynes if (!cinfo->dreq) 1681487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 16821d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 168364bfeb49SFred Isaman } 168464bfeb49SFred Isaman } 1685e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 168664bfeb49SFred Isaman 1687b20135d0STrond Myklebust static void 1688b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1689b20135d0STrond Myklebust struct nfs_page *req) 1690b20135d0STrond Myklebust { 1691b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1692b20135d0STrond Myklebust } 1693b20135d0STrond Myklebust 16949ace33cdSFred Isaman /* 16951da177e4SLinus Torvalds * Commit dirty pages 16961da177e4SLinus Torvalds */ 16971da177e4SLinus Torvalds static int 1698ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1699ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17001da177e4SLinus Torvalds { 17010b7c0153SFred Isaman struct nfs_commit_data *data; 17021da177e4SLinus Torvalds 1703ade8febdSWeston Andros Adamson /* another commit raced with us */ 1704ade8febdSWeston Andros Adamson if (list_empty(head)) 1705ade8febdSWeston Andros Adamson return 0; 1706ade8febdSWeston Andros Adamson 1707c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 17081da177e4SLinus Torvalds 17091da177e4SLinus Torvalds if (!data) 17101da177e4SLinus Torvalds goto out_bad; 17111da177e4SLinus Torvalds 17121da177e4SLinus Torvalds /* Set up the argument struct */ 1713f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1714f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1715c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1716c36aae9aSPeng Tao data->mds_ops, how, 0); 17171da177e4SLinus Torvalds out_bad: 1718b57ff130SWeston Andros Adamson nfs_retry_commit(head, NULL, cinfo, 0); 17191da177e4SLinus Torvalds return -ENOMEM; 17201da177e4SLinus Torvalds } 17211da177e4SLinus Torvalds 172267911c8fSAnna Schumaker int nfs_commit_file(struct file *file, struct nfs_write_verifier *verf) 172367911c8fSAnna Schumaker { 172467911c8fSAnna Schumaker struct inode *inode = file_inode(file); 172567911c8fSAnna Schumaker struct nfs_open_context *open; 172667911c8fSAnna Schumaker struct nfs_commit_info cinfo; 172767911c8fSAnna Schumaker struct nfs_page *req; 172867911c8fSAnna Schumaker int ret; 172967911c8fSAnna Schumaker 173067911c8fSAnna Schumaker open = get_nfs_open_context(nfs_file_open_context(file)); 173167911c8fSAnna Schumaker req = nfs_create_request(open, NULL, NULL, 0, i_size_read(inode)); 17322997bfd0SDan Carpenter if (IS_ERR(req)) { 17332997bfd0SDan Carpenter ret = PTR_ERR(req); 173467911c8fSAnna Schumaker goto out_put; 173567911c8fSAnna Schumaker } 173667911c8fSAnna Schumaker 173767911c8fSAnna Schumaker nfs_init_cinfo_from_inode(&cinfo, inode); 173867911c8fSAnna Schumaker 173967911c8fSAnna Schumaker memcpy(&req->wb_verf, verf, sizeof(struct nfs_write_verifier)); 174067911c8fSAnna Schumaker nfs_request_add_commit_list(req, &cinfo); 174167911c8fSAnna Schumaker ret = nfs_commit_inode(inode, FLUSH_SYNC); 174267911c8fSAnna Schumaker if (ret > 0) 174367911c8fSAnna Schumaker ret = 0; 174467911c8fSAnna Schumaker 174567911c8fSAnna Schumaker nfs_free_request(req); 174667911c8fSAnna Schumaker out_put: 174767911c8fSAnna Schumaker put_nfs_open_context(open); 174867911c8fSAnna Schumaker return ret; 174967911c8fSAnna Schumaker } 175067911c8fSAnna Schumaker EXPORT_SYMBOL_GPL(nfs_commit_file); 175167911c8fSAnna Schumaker 17521da177e4SLinus Torvalds /* 17531da177e4SLinus Torvalds * COMMIT call returned 17541da177e4SLinus Torvalds */ 1755788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17561da177e4SLinus Torvalds { 17570b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17581da177e4SLinus Torvalds 1759a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17601da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17611da177e4SLinus Torvalds 1762788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1763c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1764c9d8f89dSTrond Myklebust } 1765c9d8f89dSTrond Myklebust 1766f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1767c9d8f89dSTrond Myklebust { 1768c9d8f89dSTrond Myklebust struct nfs_page *req; 1769c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1770f453a54aSFred Isaman struct nfs_commit_info cinfo; 1771353db796SNeilBrown struct nfs_server *nfss; 1772788e7a89STrond Myklebust 17731da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17741da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17751da177e4SLinus Torvalds nfs_list_remove_request(req); 177667911c8fSAnna Schumaker if (req->wb_page) 1777d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17781da177e4SLinus Torvalds 17791e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17803d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 17812b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 17821da177e4SLinus Torvalds req->wb_bytes, 17831da177e4SLinus Torvalds (long long)req_offset(req)); 1784c9d8f89dSTrond Myklebust if (status < 0) { 1785c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 178638a33101SKinglong Mee if (req->wb_page) 17871da177e4SLinus Torvalds nfs_inode_remove_request(req); 1788ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 17891da177e4SLinus Torvalds goto next; 17901da177e4SLinus Torvalds } 17911da177e4SLinus Torvalds 17921da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 17931da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 17948fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 17951da177e4SLinus Torvalds /* We have a match */ 179638a33101SKinglong Mee if (req->wb_page) 17971da177e4SLinus Torvalds nfs_inode_remove_request(req); 1798ddeaa637SJoe Perches dprintk_cont(" OK\n"); 17991da177e4SLinus Torvalds goto next; 18001da177e4SLinus Torvalds } 18011da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1802ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18036d884e8fSFred nfs_mark_request_dirty(req); 180405990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18051da177e4SLinus Torvalds next: 18061d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18071da177e4SLinus Torvalds } 1808353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1809353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 1810*0db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1811353db796SNeilBrown 1812f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1813af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18145917ce84SFred Isaman } 18155917ce84SFred Isaman 18165917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18175917ce84SFred Isaman { 18180b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18195917ce84SFred Isaman 1820f453a54aSFred Isaman data->completion_ops->completion(data); 1821c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18221da177e4SLinus Torvalds } 1823788e7a89STrond Myklebust 1824788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18250b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1826788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1827788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1828788e7a89STrond Myklebust }; 18291da177e4SLinus Torvalds 1830f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1831f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1832b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1833f453a54aSFred Isaman }; 1834f453a54aSFred Isaman 18351763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1836ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 183784c53ab5SFred Isaman { 183884c53ab5SFred Isaman int status; 183984c53ab5SFred Isaman 1840ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 184184c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1842ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 184384c53ab5SFred Isaman return status; 184484c53ab5SFred Isaman } 184584c53ab5SFred Isaman 1846b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18471da177e4SLinus Torvalds { 18481da177e4SLinus Torvalds LIST_HEAD(head); 1849ea2cf228SFred Isaman struct nfs_commit_info cinfo; 185071d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1851af7cf057STrond Myklebust int error = 0; 1852b8413f98STrond Myklebust int res; 18531da177e4SLinus Torvalds 1854ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1855af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1856ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1857af7cf057STrond Myklebust if (res) 1858ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1859af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18601da177e4SLinus Torvalds if (error < 0) 1861af7cf057STrond Myklebust goto out_error; 1862b8413f98STrond Myklebust if (!may_wait) 1863b8413f98STrond Myklebust goto out_mark_dirty; 1864af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1865b8413f98STrond Myklebust if (error < 0) 1866b8413f98STrond Myklebust return error; 1867c5efa5fcSTrond Myklebust return res; 1868af7cf057STrond Myklebust out_error: 1869af7cf057STrond Myklebust res = error; 1870c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1871c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1872c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1873c5efa5fcSTrond Myklebust * that the data is on the disk. 1874c5efa5fcSTrond Myklebust */ 1875c5efa5fcSTrond Myklebust out_mark_dirty: 1876c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18771da177e4SLinus Torvalds return res; 18781da177e4SLinus Torvalds } 1879b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 18808fc795f7STrond Myklebust 1881ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18828fc795f7STrond Myklebust { 1883420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1884420e3646STrond Myklebust int flags = FLUSH_SYNC; 1885420e3646STrond Myklebust int ret = 0; 18868fc795f7STrond Myklebust 18873236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1888ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 18893236c3e1SJeff Layton return ret; 18903236c3e1SJeff Layton 1891a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1892a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1893a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1894420e3646STrond Myklebust */ 1895cb1410c7SWeston Andros Adamson if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) 1896420e3646STrond Myklebust goto out_mark_dirty; 1897420e3646STrond Myklebust 1898a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1899420e3646STrond Myklebust flags = 0; 1900a00dd6c0SJeff Layton } 1901a00dd6c0SJeff Layton 1902420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1903420e3646STrond Myklebust if (ret >= 0) { 1904420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1905420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1906420e3646STrond Myklebust wbc->nr_to_write -= ret; 1907420e3646STrond Myklebust else 1908420e3646STrond Myklebust wbc->nr_to_write = 0; 1909420e3646STrond Myklebust } 19108fc795f7STrond Myklebust return 0; 1911420e3646STrond Myklebust } 1912420e3646STrond Myklebust out_mark_dirty: 19138fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19148fc795f7STrond Myklebust return ret; 19158fc795f7STrond Myklebust } 191689d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1917863a3c6cSAndy Adamson 1918acdc53b2STrond Myklebust /* 1919837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1920837bb1d7STrond Myklebust * 1921837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1922837bb1d7STrond Myklebust * client. 1923837bb1d7STrond Myklebust */ 1924837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1925837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1926837bb1d7STrond Myklebust { 1927837bb1d7STrond Myklebust int ret; 1928837bb1d7STrond Myklebust 1929837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1930837bb1d7STrond Myklebust if (ret == 0) 1931837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1932837bb1d7STrond Myklebust return ret; 1933837bb1d7STrond Myklebust } 1934837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1935837bb1d7STrond Myklebust 1936837bb1d7STrond Myklebust /* 1937acdc53b2STrond Myklebust * flush the inode to disk. 1938acdc53b2STrond Myklebust */ 1939acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 194034901f70STrond Myklebust { 1941f4ce1299STrond Myklebust int ret; 194234901f70STrond Myklebust 1943f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1944f4ce1299STrond Myklebust 19455bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19466b196875SChuck Lever if (ret) 19476b196875SChuck Lever goto out; 19485bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19496b196875SChuck Lever if (ret < 0) 19506b196875SChuck Lever goto out; 19515bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19526b196875SChuck Lever ret = 0; 1953f4ce1299STrond Myklebust 19546b196875SChuck Lever out: 1955f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1956f4ce1299STrond Myklebust return ret; 19571c75950bSTrond Myklebust } 1958ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19591c75950bSTrond Myklebust 19601b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19611b3b4a1aSTrond Myklebust { 19621b3b4a1aSTrond Myklebust struct nfs_page *req; 19631b3b4a1aSTrond Myklebust int ret = 0; 19641b3b4a1aSTrond Myklebust 1965ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19663e217045SWeston Andros Adamson 19673e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 19683e217045SWeston Andros Adamson * request */ 19693e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 19703e217045SWeston Andros Adamson 19713e217045SWeston Andros Adamson if (IS_ERR(req)) { 19723e217045SWeston Andros Adamson ret = PTR_ERR(req); 19733e217045SWeston Andros Adamson } else if (req) { 19743e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 19753e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 19763e217045SWeston Andros Adamson * request from the inode / page_private pointer and 19773e217045SWeston Andros Adamson * release it */ 19781b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 19791d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19801b3b4a1aSTrond Myklebust } 19813e217045SWeston Andros Adamson 19821b3b4a1aSTrond Myklebust return ret; 19831b3b4a1aSTrond Myklebust } 19841b3b4a1aSTrond Myklebust 19851c75950bSTrond Myklebust /* 19861c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 19871c75950bSTrond Myklebust */ 1988d6c843b9SPeng Tao int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder) 19891c75950bSTrond Myklebust { 199029418aa4SMel Gorman loff_t range_start = page_file_offset(page); 199109cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 19927f2f12d9STrond Myklebust struct writeback_control wbc = { 19937f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 19947f2f12d9STrond Myklebust .nr_to_write = 0, 19957f2f12d9STrond Myklebust .range_start = range_start, 19967f2f12d9STrond Myklebust .range_end = range_end, 19977f2f12d9STrond Myklebust }; 19987f2f12d9STrond Myklebust int ret; 19997f2f12d9STrond Myklebust 2000f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2001f4ce1299STrond Myklebust 20020522f6adSTrond Myklebust for (;;) { 2003ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20047f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2005d6c843b9SPeng Tao ret = nfs_writepage_locked(page, &wbc, launder); 20067f2f12d9STrond Myklebust if (ret < 0) 20077f2f12d9STrond Myklebust goto out_error; 20080522f6adSTrond Myklebust continue; 20097f2f12d9STrond Myklebust } 2010f4ce1299STrond Myklebust ret = 0; 20110522f6adSTrond Myklebust if (!PagePrivate(page)) 20120522f6adSTrond Myklebust break; 20130522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20147f2f12d9STrond Myklebust if (ret < 0) 20157f2f12d9STrond Myklebust goto out_error; 20167f2f12d9STrond Myklebust } 20177f2f12d9STrond Myklebust out_error: 2018f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20197f2f12d9STrond Myklebust return ret; 20201c75950bSTrond Myklebust } 20211c75950bSTrond Myklebust 2022074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2023074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2024a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2025074cc1deSTrond Myklebust { 20262da95652SJeff Layton /* 20272da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20282da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20292da95652SJeff Layton * 20302da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20312da95652SJeff Layton * that we can safely release the inode reference while holding 20322da95652SJeff Layton * the page lock. 20332da95652SJeff Layton */ 20342da95652SJeff Layton if (PagePrivate(page)) 20352da95652SJeff Layton return -EBUSY; 2036074cc1deSTrond Myklebust 20378c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20388c209ce7SDavid Howells return -EBUSY; 2039074cc1deSTrond Myklebust 2040a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2041074cc1deSTrond Myklebust } 2042074cc1deSTrond Myklebust #endif 2043074cc1deSTrond Myklebust 2044f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20451da177e4SLinus Torvalds { 20461da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20471e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20481da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 204920c2df83SPaul Mundt NULL); 20501da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20511da177e4SLinus Torvalds return -ENOMEM; 20521da177e4SLinus Torvalds 205393d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20541da177e4SLinus Torvalds nfs_wdata_cachep); 20551da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20563dd4765fSJeff Layton goto out_destroy_write_cache; 20571da177e4SLinus Torvalds 20580b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20590b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20600b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20610b7c0153SFred Isaman NULL); 20620b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20633dd4765fSJeff Layton goto out_destroy_write_mempool; 20640b7c0153SFred Isaman 206593d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20664c100210SYanchuan Nian nfs_cdata_cachep); 20671da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 20683dd4765fSJeff Layton goto out_destroy_commit_cache; 20691da177e4SLinus Torvalds 207089a09141SPeter Zijlstra /* 207189a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 207289a09141SPeter Zijlstra * 207389a09141SPeter Zijlstra * 64MB: 8192k 207489a09141SPeter Zijlstra * 128MB: 11585k 207589a09141SPeter Zijlstra * 256MB: 16384k 207689a09141SPeter Zijlstra * 512MB: 23170k 207789a09141SPeter Zijlstra * 1GB: 32768k 207889a09141SPeter Zijlstra * 2GB: 46340k 207989a09141SPeter Zijlstra * 4GB: 65536k 208089a09141SPeter Zijlstra * 8GB: 92681k 208189a09141SPeter Zijlstra * 16GB: 131072k 208289a09141SPeter Zijlstra * 208389a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 208489a09141SPeter Zijlstra * Limit the default to 256M 208589a09141SPeter Zijlstra */ 208689a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 208789a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 208889a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 208989a09141SPeter Zijlstra 20901da177e4SLinus Torvalds return 0; 20913dd4765fSJeff Layton 20923dd4765fSJeff Layton out_destroy_commit_cache: 20933dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20943dd4765fSJeff Layton out_destroy_write_mempool: 20953dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 20963dd4765fSJeff Layton out_destroy_write_cache: 20973dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 20983dd4765fSJeff Layton return -ENOMEM; 20991da177e4SLinus Torvalds } 21001da177e4SLinus Torvalds 2101266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21021da177e4SLinus Torvalds { 21031da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21043dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21051da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21061a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21071da177e4SLinus Torvalds } 21081da177e4SLinus Torvalds 21094a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 2110a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 21114a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21124a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21130eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21140eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21151ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21164a0de55cSAnna Schumaker }; 2117