1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/nfs/write.c 41da177e4SLinus Torvalds * 57c85d900STrond Myklebust * Write file data over NFS. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 101da177e4SLinus Torvalds #include <linux/types.h> 111da177e4SLinus Torvalds #include <linux/slab.h> 121da177e4SLinus Torvalds #include <linux/mm.h> 131da177e4SLinus Torvalds #include <linux/pagemap.h> 141da177e4SLinus Torvalds #include <linux/file.h> 151da177e4SLinus Torvalds #include <linux/writeback.h> 1689a09141SPeter Zijlstra #include <linux/swap.h> 17074cc1deSTrond Myklebust #include <linux/migrate.h> 181da177e4SLinus Torvalds 191da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 201da177e4SLinus Torvalds #include <linux/nfs_fs.h> 211da177e4SLinus Torvalds #include <linux/nfs_mount.h> 221da177e4SLinus Torvalds #include <linux/nfs_page.h> 233fcfab16SAndrew Morton #include <linux/backing-dev.h> 24afeacc8cSPaul Gortmaker #include <linux/export.h> 25af7cf057STrond Myklebust #include <linux/freezer.h> 26af7cf057STrond Myklebust #include <linux/wait.h> 271eb5d98fSJeff Layton #include <linux/iversion.h> 285970e15dSJeff Layton #include <linux/filelock.h> 293fcfab16SAndrew Morton 307c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 31875bc3fbSTrond Myklebust #include <linux/sched/mm.h> 321da177e4SLinus Torvalds 331da177e4SLinus Torvalds #include "delegation.h" 3449a70f27STrond Myklebust #include "internal.h" 3591d5b470SChuck Lever #include "iostat.h" 36def6ed7eSAndy Adamson #include "nfs4_fs.h" 37074cc1deSTrond Myklebust #include "fscache.h" 3894ad1c80SFred Isaman #include "pnfs.h" 391da177e4SLinus Torvalds 40f4ce1299STrond Myklebust #include "nfstrace.h" 41f4ce1299STrond Myklebust 421da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 431da177e4SLinus Torvalds 441da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 451da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 461da177e4SLinus Torvalds 47919e3bd9STrond Myklebust struct nfs_io_completion { 48919e3bd9STrond Myklebust void (*complete)(void *data); 49919e3bd9STrond Myklebust void *data; 50919e3bd9STrond Myklebust struct kref refcount; 51919e3bd9STrond Myklebust }; 52919e3bd9STrond Myklebust 531da177e4SLinus Torvalds /* 541da177e4SLinus Torvalds * Local function declarations 551da177e4SLinus Torvalds */ 56f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 57788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 58061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 59f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 604a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 6106c9fdf3STrond Myklebust static void nfs_inode_remove_request(struct nfs_page *req); 62b193a78dSTrond Myklebust static void nfs_clear_request_commit(struct nfs_commit_info *cinfo, 63b193a78dSTrond Myklebust struct nfs_page *req); 6402d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 6502d1426cSWeston Andros Adamson struct inode *inode); 663a3908c8STrond Myklebust static struct nfs_page * 673a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 680c493b5cSTrond Myklebust struct folio *folio); 691da177e4SLinus Torvalds 70e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 713feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 720b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 731da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 741da177e4SLinus Torvalds 75515dcdcdSTrond Myklebust struct nfs_commit_data *nfs_commitdata_alloc(void) 761da177e4SLinus Torvalds { 77518662e0SNeilBrown struct nfs_commit_data *p; 7840859d7eSChuck Lever 79515dcdcdSTrond Myklebust p = kmem_cache_zalloc(nfs_cdata_cachep, nfs_io_gfp_mask()); 80515dcdcdSTrond Myklebust if (!p) { 81518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); 82518662e0SNeilBrown if (!p) 83518662e0SNeilBrown return NULL; 841da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 85515dcdcdSTrond Myklebust } 861da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 871da177e4SLinus Torvalds return p; 881da177e4SLinus Torvalds } 89e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 901da177e4SLinus Torvalds 910b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 921da177e4SLinus Torvalds { 931da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 941da177e4SLinus Torvalds } 95e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 961da177e4SLinus Torvalds 971e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 983feb2d49STrond Myklebust { 990bae835bSTrond Myklebust struct nfs_pgio_header *p; 1003feb2d49STrond Myklebust 1010bae835bSTrond Myklebust p = kmem_cache_zalloc(nfs_wdata_cachep, nfs_io_gfp_mask()); 1020bae835bSTrond Myklebust if (!p) { 1030bae835bSTrond Myklebust p = mempool_alloc(nfs_wdata_mempool, GFP_NOWAIT); 1040bae835bSTrond Myklebust if (!p) 1050bae835bSTrond Myklebust return NULL; 1063feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 1070bae835bSTrond Myklebust } 108fbe77c30SBenjamin Coddington p->rw_mode = FMODE_WRITE; 1093feb2d49STrond Myklebust return p; 1103feb2d49STrond Myklebust } 1113feb2d49STrond Myklebust 1121e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 1136c75dc0dSFred Isaman { 1141e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 1153feb2d49STrond Myklebust } 1161da177e4SLinus Torvalds 117919e3bd9STrond Myklebust static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) 118919e3bd9STrond Myklebust { 119919e3bd9STrond Myklebust return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); 120919e3bd9STrond Myklebust } 121919e3bd9STrond Myklebust 122919e3bd9STrond Myklebust static void nfs_io_completion_init(struct nfs_io_completion *ioc, 123919e3bd9STrond Myklebust void (*complete)(void *), void *data) 124919e3bd9STrond Myklebust { 125919e3bd9STrond Myklebust ioc->complete = complete; 126919e3bd9STrond Myklebust ioc->data = data; 127919e3bd9STrond Myklebust kref_init(&ioc->refcount); 128919e3bd9STrond Myklebust } 129919e3bd9STrond Myklebust 130919e3bd9STrond Myklebust static void nfs_io_completion_release(struct kref *kref) 131919e3bd9STrond Myklebust { 132919e3bd9STrond Myklebust struct nfs_io_completion *ioc = container_of(kref, 133919e3bd9STrond Myklebust struct nfs_io_completion, refcount); 134919e3bd9STrond Myklebust ioc->complete(ioc->data); 135919e3bd9STrond Myklebust kfree(ioc); 136919e3bd9STrond Myklebust } 137919e3bd9STrond Myklebust 138919e3bd9STrond Myklebust static void nfs_io_completion_get(struct nfs_io_completion *ioc) 139919e3bd9STrond Myklebust { 140919e3bd9STrond Myklebust if (ioc != NULL) 141919e3bd9STrond Myklebust kref_get(&ioc->refcount); 142919e3bd9STrond Myklebust } 143919e3bd9STrond Myklebust 144919e3bd9STrond Myklebust static void nfs_io_completion_put(struct nfs_io_completion *ioc) 145919e3bd9STrond Myklebust { 146919e3bd9STrond Myklebust if (ioc != NULL) 147919e3bd9STrond Myklebust kref_put(&ioc->refcount, nfs_io_completion_release); 148919e3bd9STrond Myklebust } 149919e3bd9STrond Myklebust 150e00ed89dSTrond Myklebust static void 151e00ed89dSTrond Myklebust nfs_page_set_inode_ref(struct nfs_page *req, struct inode *inode) 152e00ed89dSTrond Myklebust { 153e00ed89dSTrond Myklebust if (!test_and_set_bit(PG_INODE_REF, &req->wb_flags)) { 154e00ed89dSTrond Myklebust kref_get(&req->wb_kref); 155e00ed89dSTrond Myklebust atomic_long_inc(&NFS_I(inode)->nrequests); 156e00ed89dSTrond Myklebust } 157e00ed89dSTrond Myklebust } 158e00ed89dSTrond Myklebust 159e00ed89dSTrond Myklebust static int 160e00ed89dSTrond Myklebust nfs_cancel_remove_inode(struct nfs_page *req, struct inode *inode) 161e00ed89dSTrond Myklebust { 162e00ed89dSTrond Myklebust int ret; 163e00ed89dSTrond Myklebust 164e00ed89dSTrond Myklebust if (!test_bit(PG_REMOVE, &req->wb_flags)) 165e00ed89dSTrond Myklebust return 0; 166e00ed89dSTrond Myklebust ret = nfs_page_group_lock(req); 167e00ed89dSTrond Myklebust if (ret) 168e00ed89dSTrond Myklebust return ret; 169e00ed89dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &req->wb_flags)) 170e00ed89dSTrond Myklebust nfs_page_set_inode_ref(req, inode); 171e00ed89dSTrond Myklebust nfs_page_group_unlock(req); 172e00ed89dSTrond Myklebust return 0; 173e00ed89dSTrond Myklebust } 174e00ed89dSTrond Myklebust 1750c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_private_request(struct folio *folio) 176bd37d6fcSTrond Myklebust { 1770c493b5cSTrond Myklebust return folio_get_private(folio); 178bd37d6fcSTrond Myklebust } 179bd37d6fcSTrond Myklebust 1800c493b5cSTrond Myklebust /** 1810c493b5cSTrond Myklebust * nfs_folio_find_private_request - find head request associated with a folio 1820c493b5cSTrond Myklebust * @folio: pointer to folio 18384d3a9a9SWeston Andros Adamson * 18484d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 18584d3a9a9SWeston Andros Adamson * 18684d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 18784d3a9a9SWeston Andros Adamson */ 1880c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_find_private_request(struct folio *folio) 189277459d2STrond Myklebust { 1900c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 191bd37d6fcSTrond Myklebust struct nfs_page *req; 192277459d2STrond Myklebust 1930c493b5cSTrond Myklebust if (!folio_test_private(folio)) 194b30d2f04STrond Myklebust return NULL; 195600f111eSMatthew Wilcox (Oracle) spin_lock(&mapping->i_private_lock); 1960c493b5cSTrond Myklebust req = nfs_folio_private_request(folio); 19784d3a9a9SWeston Andros Adamson if (req) { 19884d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 19929418aa4SMel Gorman kref_get(&req->wb_kref); 20084d3a9a9SWeston Andros Adamson } 201600f111eSMatthew Wilcox (Oracle) spin_unlock(&mapping->i_private_lock); 202b30d2f04STrond Myklebust return req; 203b30d2f04STrond Myklebust } 20429418aa4SMel Gorman 2050c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_find_swap_request(struct folio *folio) 206b30d2f04STrond Myklebust { 2070c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 208b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 209b30d2f04STrond Myklebust struct nfs_page *req = NULL; 2100c493b5cSTrond Myklebust if (!folio_test_swapcache(folio)) 211b30d2f04STrond Myklebust return NULL; 212e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 2130c493b5cSTrond Myklebust if (folio_test_swapcache(folio)) { 214b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 2150c493b5cSTrond Myklebust folio); 216b30d2f04STrond Myklebust if (req) { 217b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 218b30d2f04STrond Myklebust kref_get(&req->wb_kref); 219b30d2f04STrond Myklebust } 220b30d2f04STrond Myklebust } 221e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 222277459d2STrond Myklebust return req; 223277459d2STrond Myklebust } 224277459d2STrond Myklebust 2250c493b5cSTrond Myklebust /** 2260c493b5cSTrond Myklebust * nfs_folio_find_head_request - find head request associated with a folio 2270c493b5cSTrond Myklebust * @folio: pointer to folio 22884d3a9a9SWeston Andros Adamson * 22984d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 23084d3a9a9SWeston Andros Adamson */ 2310c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_find_head_request(struct folio *folio) 232277459d2STrond Myklebust { 233b30d2f04STrond Myklebust struct nfs_page *req; 234277459d2STrond Myklebust 2350c493b5cSTrond Myklebust req = nfs_folio_find_private_request(folio); 236b30d2f04STrond Myklebust if (!req) 2370c493b5cSTrond Myklebust req = nfs_folio_find_swap_request(folio); 238277459d2STrond Myklebust return req; 239277459d2STrond Myklebust } 240277459d2STrond Myklebust 2410c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_find_and_lock_request(struct folio *folio) 242e00ed89dSTrond Myklebust { 2430c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 244e00ed89dSTrond Myklebust struct nfs_page *req, *head; 245e00ed89dSTrond Myklebust int ret; 246e00ed89dSTrond Myklebust 247e00ed89dSTrond Myklebust for (;;) { 2480c493b5cSTrond Myklebust req = nfs_folio_find_head_request(folio); 249e00ed89dSTrond Myklebust if (!req) 250e00ed89dSTrond Myklebust return req; 251e00ed89dSTrond Myklebust head = nfs_page_group_lock_head(req); 252e00ed89dSTrond Myklebust if (head != req) 253e00ed89dSTrond Myklebust nfs_release_request(req); 254e00ed89dSTrond Myklebust if (IS_ERR(head)) 255e00ed89dSTrond Myklebust return head; 256e00ed89dSTrond Myklebust ret = nfs_cancel_remove_inode(head, inode); 257e00ed89dSTrond Myklebust if (ret < 0) { 258e00ed89dSTrond Myklebust nfs_unlock_and_release_request(head); 259e00ed89dSTrond Myklebust return ERR_PTR(ret); 260e00ed89dSTrond Myklebust } 261e00ed89dSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 2620c493b5cSTrond Myklebust if (head == nfs_folio_private_request(folio)) 263e00ed89dSTrond Myklebust break; 2640c493b5cSTrond Myklebust if (folio_test_swapcache(folio)) 265e00ed89dSTrond Myklebust break; 266e00ed89dSTrond Myklebust nfs_unlock_and_release_request(head); 267e00ed89dSTrond Myklebust } 268e00ed89dSTrond Myklebust return head; 269e00ed89dSTrond Myklebust } 270e00ed89dSTrond Myklebust 2711da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2720c493b5cSTrond Myklebust static void nfs_grow_file(struct folio *folio, unsigned int offset, 2730c493b5cSTrond Myklebust unsigned int count) 2741da177e4SLinus Torvalds { 2750c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 276a3d01454STrond Myklebust loff_t end, i_size; 277a3d01454STrond Myklebust pgoff_t end_index; 2781da177e4SLinus Torvalds 279a3d01454STrond Myklebust spin_lock(&inode->i_lock); 280a3d01454STrond Myklebust i_size = i_size_read(inode); 2810c493b5cSTrond Myklebust end_index = ((i_size - 1) >> folio_shift(folio)) << folio_order(folio); 2820c493b5cSTrond Myklebust if (i_size > 0 && folio_index(folio) < end_index) 283a3d01454STrond Myklebust goto out; 2840c493b5cSTrond Myklebust end = folio_file_pos(folio) + (loff_t)offset + (loff_t)count; 2851da177e4SLinus Torvalds if (i_size >= end) 286a3d01454STrond Myklebust goto out; 287110cb2d2SChuck Lever trace_nfs_size_grow(inode, end); 2881da177e4SLinus Torvalds i_size_write(inode, end); 289f6cdfa6dSTrond Myklebust NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 290a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 291a3d01454STrond Myklebust out: 292a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 293a6b5a28eSDave Wysochanski nfs_fscache_invalidate(inode, 0); 2941da177e4SLinus Torvalds } 2951da177e4SLinus Torvalds 296a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 297d2ceb7e5SBenjamin Coddington static void nfs_set_pageerror(struct address_space *mapping) 298a301b777STrond Myklebust { 2990df68cedSTrond Myklebust struct inode *inode = mapping->host; 3000df68cedSTrond Myklebust 301d2ceb7e5SBenjamin Coddington nfs_zap_mapping(mapping->host, mapping); 3020df68cedSTrond Myklebust /* Force file size revalidation */ 3030df68cedSTrond Myklebust spin_lock(&inode->i_lock); 304ac46b3d7STrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_REVAL_FORCED | 30588a6099fSTrond Myklebust NFS_INO_INVALID_CHANGE | 306ac46b3d7STrond Myklebust NFS_INO_INVALID_SIZE); 3070df68cedSTrond Myklebust spin_unlock(&inode->i_lock); 308a301b777STrond Myklebust } 309a301b777STrond Myklebust 3100c493b5cSTrond Myklebust static void nfs_mapping_set_error(struct folio *folio, int error) 3116fbda89bSTrond Myklebust { 3120c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 313b8946d7bSTrond Myklebust 3140c493b5cSTrond Myklebust folio_set_error(folio); 3156c984083STrond Myklebust filemap_set_wb_err(mapping, error); 3166c984083STrond Myklebust if (mapping->host) 3176c984083STrond Myklebust errseq_set(&mapping->host->i_sb->s_wb_err, 3186c984083STrond Myklebust error == -ENOSPC ? -ENOSPC : -EIO); 319b8946d7bSTrond Myklebust nfs_set_pageerror(mapping); 3206fbda89bSTrond Myklebust } 3216fbda89bSTrond Myklebust 322d72ddcbaSWeston Andros Adamson /* 323d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 324d72ddcbaSWeston Andros Adamson * @head - head request of page group 325d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 326d72ddcbaSWeston Andros Adamson * 327d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 328d72ddcbaSWeston Andros Adamson * page offset @page_offset. 329d72ddcbaSWeston Andros Adamson * 330d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 331d72ddcbaSWeston Andros Adamson * match is found. 332d72ddcbaSWeston Andros Adamson * 333d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 334d72ddcbaSWeston Andros Adamson */ 335d72ddcbaSWeston Andros Adamson static struct nfs_page * 336d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 337d72ddcbaSWeston Andros Adamson { 338d72ddcbaSWeston Andros Adamson struct nfs_page *req; 339d72ddcbaSWeston Andros Adamson 340d72ddcbaSWeston Andros Adamson req = head; 341d72ddcbaSWeston Andros Adamson do { 342d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 343d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 344d72ddcbaSWeston Andros Adamson return req; 345d72ddcbaSWeston Andros Adamson 346d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 347d72ddcbaSWeston Andros Adamson } while (req != head); 348d72ddcbaSWeston Andros Adamson 349d72ddcbaSWeston Andros Adamson return NULL; 350d72ddcbaSWeston Andros Adamson } 351d72ddcbaSWeston Andros Adamson 352d72ddcbaSWeston Andros Adamson /* 353d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 354d72ddcbaSWeston Andros Adamson * @head - head request of page group 355d72ddcbaSWeston Andros Adamson * 356d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 357d72ddcbaSWeston Andros Adamson * returns false otherwise 358d72ddcbaSWeston Andros Adamson */ 359d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 360d72ddcbaSWeston Andros Adamson { 3610c493b5cSTrond Myklebust unsigned int len = nfs_folio_length(nfs_page_to_folio(req)); 362d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 363d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 364d72ddcbaSWeston Andros Adamson 3651344b7eaSTrond Myklebust nfs_page_group_lock(req); 366d72ddcbaSWeston Andros Adamson 3677e8a30f8STrond Myklebust for (;;) { 368d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 3697e8a30f8STrond Myklebust if (!tmp) 3707e8a30f8STrond Myklebust break; 3717e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 372d72ddcbaSWeston Andros Adamson } 373d72ddcbaSWeston Andros Adamson 374d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3757e8a30f8STrond Myklebust return pos >= len; 376d72ddcbaSWeston Andros Adamson } 377d72ddcbaSWeston Andros Adamson 3781da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3791da177e4SLinus Torvalds * covers the full page. 3801da177e4SLinus Torvalds */ 381d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3821da177e4SLinus Torvalds { 3830c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 3840c493b5cSTrond Myklebust 3850c493b5cSTrond Myklebust if (folio_test_uptodate(folio)) 3861da177e4SLinus Torvalds return; 387d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3881da177e4SLinus Torvalds return; 3890c493b5cSTrond Myklebust folio_mark_uptodate(folio); 3901da177e4SLinus Torvalds } 3911da177e4SLinus Torvalds 3921da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3931da177e4SLinus Torvalds { 394e87b4c7aSNeilBrown int ret = 0; 395cca588d6STrond Myklebust 396e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 397e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 398e87b4c7aSNeilBrown return ret; 3991da177e4SLinus Torvalds } 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds /* 40289a09141SPeter Zijlstra * NFS congestion control 40389a09141SPeter Zijlstra */ 40489a09141SPeter Zijlstra 40589a09141SPeter Zijlstra int nfs_congestion_kb; 40689a09141SPeter Zijlstra 40789a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 40889a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 40989a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 41089a09141SPeter Zijlstra 4110c493b5cSTrond Myklebust static void nfs_folio_set_writeback(struct folio *folio) 41289a09141SPeter Zijlstra { 4130c493b5cSTrond Myklebust struct nfs_server *nfss = NFS_SERVER(folio_file_mapping(folio)->host); 4145a6d41b3STrond Myklebust 4150c493b5cSTrond Myklebust folio_start_writeback(folio); 4160c493b5cSTrond Myklebust if (atomic_long_inc_return(&nfss->writeback) > NFS_CONGESTION_ON_THRESH) 4176df25e58SNeilBrown nfss->write_congested = 1; 41889a09141SPeter Zijlstra } 41989a09141SPeter Zijlstra 4200c493b5cSTrond Myklebust static void nfs_folio_end_writeback(struct folio *folio) 42189a09141SPeter Zijlstra { 4220c493b5cSTrond Myklebust struct nfs_server *nfss = NFS_SERVER(folio_file_mapping(folio)->host); 42389a09141SPeter Zijlstra 4240c493b5cSTrond Myklebust folio_end_writeback(folio); 4250c493b5cSTrond Myklebust if (atomic_long_dec_return(&nfss->writeback) < 4260c493b5cSTrond Myklebust NFS_CONGESTION_OFF_THRESH) 4276df25e58SNeilBrown nfss->write_congested = 0; 42889a09141SPeter Zijlstra } 42989a09141SPeter Zijlstra 4300c493b5cSTrond Myklebust static void nfs_page_end_writeback(struct nfs_page *req) 4310c493b5cSTrond Myklebust { 4320c493b5cSTrond Myklebust if (nfs_page_group_sync_on_bit(req, PG_WB_END)) { 4330c493b5cSTrond Myklebust nfs_unlock_request(req); 4340c493b5cSTrond Myklebust nfs_folio_end_writeback(nfs_page_to_folio(req)); 4350c493b5cSTrond Myklebust } else 4360c493b5cSTrond Myklebust nfs_unlock_request(req); 4370c493b5cSTrond Myklebust } 4380c493b5cSTrond Myklebust 439d4581383SWeston Andros Adamson /* 440d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 441d4581383SWeston Andros Adamson * 442d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 443d4581383SWeston Andros Adamson * @old_head - the old head of the list 444d4581383SWeston Andros Adamson * 445d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 446d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 447d4581383SWeston Andros Adamson * with a reference held by some other context. 448d4581383SWeston Andros Adamson */ 449d4581383SWeston Andros Adamson static void 450d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 451b66aaa8dSTrond Myklebust struct nfs_page *old_head, 452b66aaa8dSTrond Myklebust struct inode *inode) 453d4581383SWeston Andros Adamson { 454d4581383SWeston Andros Adamson while (destroy_list) { 455d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 456d4581383SWeston Andros Adamson 457d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 458d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 459d4581383SWeston Andros Adamson 46008ca8b21STrond Myklebust /* Note: lock subreq in order to change subreq->wb_head */ 46108ca8b21STrond Myklebust nfs_page_set_headlock(subreq); 462d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 463d4581383SWeston Andros Adamson 464d4581383SWeston Andros Adamson /* make sure old group is not used */ 465d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 46608ca8b21STrond Myklebust subreq->wb_head = subreq; 467d4581383SWeston Andros Adamson 468902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 469902a4c00STrond Myklebust 4705b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4715b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4725b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 47308ca8b21STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) { 47408ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 4755b2b5187STrond Myklebust nfs_free_request(subreq); 47608ca8b21STrond Myklebust } else 47708ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 4785b2b5187STrond Myklebust continue; 4795b2b5187STrond Myklebust } 48008ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 481d4581383SWeston Andros Adamson 482add42de3STrond Myklebust nfs_release_request(old_head); 4835b2b5187STrond Myklebust 484b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 485d4581383SWeston Andros Adamson nfs_release_request(subreq); 486a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 487d4581383SWeston Andros Adamson } 4885b2b5187STrond Myklebust 4895b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4905b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4915b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 492d4581383SWeston Andros Adamson } 493d4581383SWeston Andros Adamson } 494d4581383SWeston Andros Adamson 495d4581383SWeston Andros Adamson /* 496e00ed89dSTrond Myklebust * nfs_join_page_group - destroy subrequests of the head req 497e00ed89dSTrond Myklebust * @head: the page used to lookup the "page group" of nfs_page structures 498e00ed89dSTrond Myklebust * @inode: Inode to which the request belongs. 499d4581383SWeston Andros Adamson * 500d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 501d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 502d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 503d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 504d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 505d4581383SWeston Andros Adamson */ 506b193a78dSTrond Myklebust void nfs_join_page_group(struct nfs_page *head, struct nfs_commit_info *cinfo, 507b193a78dSTrond Myklebust struct inode *inode) 508d4581383SWeston Andros Adamson { 509e00ed89dSTrond Myklebust struct nfs_page *subreq; 510d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 511a62f8e3bSTrond Myklebust unsigned int pgbase, off, bytes; 512a62f8e3bSTrond Myklebust 513a62f8e3bSTrond Myklebust pgbase = head->wb_pgbase; 514a62f8e3bSTrond Myklebust bytes = head->wb_bytes; 515a62f8e3bSTrond Myklebust off = head->wb_offset; 516a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 517a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 518a62f8e3bSTrond Myklebust /* Subrequests should always form a contiguous range */ 519a62f8e3bSTrond Myklebust if (pgbase > subreq->wb_pgbase) { 520a62f8e3bSTrond Myklebust off -= pgbase - subreq->wb_pgbase; 521a62f8e3bSTrond Myklebust bytes += pgbase - subreq->wb_pgbase; 522a62f8e3bSTrond Myklebust pgbase = subreq->wb_pgbase; 523a62f8e3bSTrond Myklebust } 524a62f8e3bSTrond Myklebust bytes = max(subreq->wb_pgbase + subreq->wb_bytes 525a62f8e3bSTrond Myklebust - pgbase, bytes); 5261bd5d6d0STrond Myklebust } 5271bd5d6d0STrond Myklebust 528a62f8e3bSTrond Myklebust /* Set the head request's range to cover the former page group */ 529a62f8e3bSTrond Myklebust head->wb_pgbase = pgbase; 530a62f8e3bSTrond Myklebust head->wb_bytes = bytes; 531a62f8e3bSTrond Myklebust head->wb_offset = off; 532d4581383SWeston Andros Adamson 533d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 534d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 535d4581383SWeston Andros Adamson subreq = head; 536d4581383SWeston Andros Adamson do { 537b193a78dSTrond Myklebust nfs_clear_request_commit(cinfo, subreq); 538d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 539d4581383SWeston Andros Adamson } while (subreq != head); 540d4581383SWeston Andros Adamson 541d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 542d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 543d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 544d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 545d4581383SWeston Andros Adamson head->wb_this_page = head; 546d4581383SWeston Andros Adamson } 547d4581383SWeston Andros Adamson 548b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 549b5bab9bfSTrond Myklebust } 550b5bab9bfSTrond Myklebust 551e00ed89dSTrond Myklebust /* 552e00ed89dSTrond Myklebust * nfs_lock_and_join_requests - join all subreqs to the head req 5530c493b5cSTrond Myklebust * @folio: the folio used to lookup the "page group" of nfs_page structures 554e00ed89dSTrond Myklebust * 555e00ed89dSTrond Myklebust * This function joins all sub requests to the head request by first 556e00ed89dSTrond Myklebust * locking all requests in the group, cancelling any pending operations 557e00ed89dSTrond Myklebust * and finally updating the head request to cover the whole range covered by 558e00ed89dSTrond Myklebust * the (former) group. All subrequests are removed from any write or commit 559e00ed89dSTrond Myklebust * lists, unlinked from the group and destroyed. 560e00ed89dSTrond Myklebust * 561e00ed89dSTrond Myklebust * Returns a locked, referenced pointer to the head request - which after 562e00ed89dSTrond Myklebust * this call is guaranteed to be the only request associated with the page. 5630c493b5cSTrond Myklebust * Returns NULL if no requests are found for @folio, or a ERR_PTR if an 564e00ed89dSTrond Myklebust * error was encountered. 565e00ed89dSTrond Myklebust */ 5660c493b5cSTrond Myklebust static struct nfs_page *nfs_lock_and_join_requests(struct folio *folio) 567e00ed89dSTrond Myklebust { 5680c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 569e00ed89dSTrond Myklebust struct nfs_page *head; 570b193a78dSTrond Myklebust struct nfs_commit_info cinfo; 571e00ed89dSTrond Myklebust int ret; 572e00ed89dSTrond Myklebust 573b193a78dSTrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 574e00ed89dSTrond Myklebust /* 575e00ed89dSTrond Myklebust * A reference is taken only on the head request which acts as a 576e00ed89dSTrond Myklebust * reference to the whole page group - the group will not be destroyed 577e00ed89dSTrond Myklebust * until the head reference is released. 578e00ed89dSTrond Myklebust */ 5790c493b5cSTrond Myklebust head = nfs_folio_find_and_lock_request(folio); 580e00ed89dSTrond Myklebust if (IS_ERR_OR_NULL(head)) 581d4581383SWeston Andros Adamson return head; 5820671d8f1SMarkus Elfring 583e00ed89dSTrond Myklebust /* lock each request in the page group */ 584e00ed89dSTrond Myklebust ret = nfs_page_group_lock_subrequests(head); 585e00ed89dSTrond Myklebust if (ret < 0) { 5860671d8f1SMarkus Elfring nfs_unlock_and_release_request(head); 5870671d8f1SMarkus Elfring return ERR_PTR(ret); 588612c9384STrond Myklebust } 589074cc1deSTrond Myklebust 590b193a78dSTrond Myklebust nfs_join_page_group(head, &cinfo, inode); 591e00ed89dSTrond Myklebust 592e00ed89dSTrond Myklebust return head; 593e00ed89dSTrond Myklebust } 594e00ed89dSTrond Myklebust 5956fbda89bSTrond Myklebust static void nfs_write_error(struct nfs_page *req, int error) 5960bcbf039SPeng Tao { 5976dd85e83STrond Myklebust trace_nfs_write_error(nfs_page_to_inode(req), req, error); 5980c493b5cSTrond Myklebust nfs_mapping_set_error(nfs_page_to_folio(req), error); 59906c9fdf3STrond Myklebust nfs_inode_remove_request(req); 6000c493b5cSTrond Myklebust nfs_page_end_writeback(req); 6011f84ccdfSFred Isaman nfs_release_request(req); 6020bcbf039SPeng Tao } 6030bcbf039SPeng Tao 604074cc1deSTrond Myklebust /* 605074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 606074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 607074cc1deSTrond Myklebust */ 6080c493b5cSTrond Myklebust static int nfs_page_async_flush(struct folio *folio, 609c6fd3511STrond Myklebust struct writeback_control *wbc, 610c6fd3511STrond Myklebust struct nfs_pageio_descriptor *pgio) 611074cc1deSTrond Myklebust { 612074cc1deSTrond Myklebust struct nfs_page *req; 613074cc1deSTrond Myklebust int ret = 0; 614074cc1deSTrond Myklebust 6150c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 616074cc1deSTrond Myklebust if (!req) 617074cc1deSTrond Myklebust goto out; 618074cc1deSTrond Myklebust ret = PTR_ERR(req); 619074cc1deSTrond Myklebust if (IS_ERR(req)) 620074cc1deSTrond Myklebust goto out; 621074cc1deSTrond Myklebust 6220c493b5cSTrond Myklebust nfs_folio_set_writeback(folio); 623deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 624074cc1deSTrond Myklebust 625a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 62696c41455STrond Myklebust ret = pgio->pg_error; 62796c41455STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 628a6598813STrond Myklebust goto out_launder; 629a6598813STrond Myklebust 63096c41455STrond Myklebust ret = 0; 631f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 632074cc1deSTrond Myklebust ret = pgio->pg_error; 6330bcbf039SPeng Tao /* 634c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6350bcbf039SPeng Tao */ 636c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 637a6598813STrond Myklebust goto out_launder; 638c6fd3511STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) 639c6fd3511STrond Myklebust ret = AOP_WRITEPAGE_ACTIVATE; 6400c493b5cSTrond Myklebust folio_redirty_for_writepage(wbc, folio); 6418fc75bedSTrond Myklebust nfs_redirty_request(req); 64296c41455STrond Myklebust pgio->pg_error = 0; 64340f90271STrond Myklebust } else 6440c493b5cSTrond Myklebust nfs_add_stats(folio_file_mapping(folio)->host, 64540f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 646074cc1deSTrond Myklebust out: 647074cc1deSTrond Myklebust return ret; 648a6598813STrond Myklebust out_launder: 6496fbda89bSTrond Myklebust nfs_write_error(req, ret); 65014bebe3cSTrond Myklebust return 0; 651e261f51fSTrond Myklebust } 652e261f51fSTrond Myklebust 6530c493b5cSTrond Myklebust static int nfs_do_writepage(struct folio *folio, struct writeback_control *wbc, 654c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 655f758c885STrond Myklebust { 6560c493b5cSTrond Myklebust nfs_pageio_cond_complete(pgio, folio_index(folio)); 6570c493b5cSTrond Myklebust return nfs_page_async_flush(folio, wbc, pgio); 658f758c885STrond Myklebust } 659f758c885STrond Myklebust 660e261f51fSTrond Myklebust /* 6611da177e4SLinus Torvalds * Write an mmapped page to the server. 6621da177e4SLinus Torvalds */ 6630c493b5cSTrond Myklebust static int nfs_writepage_locked(struct folio *folio, 664c373fff7STrond Myklebust struct writeback_control *wbc) 6651da177e4SLinus Torvalds { 666f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 6670c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 668e261f51fSTrond Myklebust int err; 6691da177e4SLinus Torvalds 6706df25e58SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 6716df25e58SNeilBrown NFS_SERVER(inode)->write_congested) 6726df25e58SNeilBrown return AOP_WRITEPAGE_ACTIVATE; 6736df25e58SNeilBrown 67440f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 6750c493b5cSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, false, 6760c493b5cSTrond Myklebust &nfs_async_write_completion_ops); 6770c493b5cSTrond Myklebust err = nfs_do_writepage(folio, wbc, &pgio); 67896c41455STrond Myklebust pgio.pg_error = 0; 679f758c885STrond Myklebust nfs_pageio_complete(&pgio); 6804d770ccfSTrond Myklebust return err; 6814d770ccfSTrond Myklebust } 6824d770ccfSTrond Myklebust 683d585bdbeSMatthew Wilcox (Oracle) static int nfs_writepages_callback(struct folio *folio, 6840c493b5cSTrond Myklebust struct writeback_control *wbc, void *data) 685f758c885STrond Myklebust { 686f758c885STrond Myklebust int ret; 687f758c885STrond Myklebust 6880c493b5cSTrond Myklebust ret = nfs_do_writepage(folio, wbc, data); 68996c41455STrond Myklebust if (ret != AOP_WRITEPAGE_ACTIVATE) 690d585bdbeSMatthew Wilcox (Oracle) folio_unlock(folio); 691f758c885STrond Myklebust return ret; 6921da177e4SLinus Torvalds } 6931da177e4SLinus Torvalds 694919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 695919e3bd9STrond Myklebust { 696919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 697919e3bd9STrond Myklebust } 698919e3bd9STrond Myklebust 6991da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7001da177e4SLinus Torvalds { 7011da177e4SLinus Torvalds struct inode *inode = mapping->host; 702c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 703ed7bcdb3STrond Myklebust struct nfs_io_completion *ioc = NULL; 704ed7bcdb3STrond Myklebust unsigned int mntflags = NFS_SERVER(inode)->flags; 705ed7bcdb3STrond Myklebust int priority = 0; 7061da177e4SLinus Torvalds int err; 7071da177e4SLinus Torvalds 7086df25e58SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 7096df25e58SNeilBrown NFS_SERVER(inode)->write_congested) 7106df25e58SNeilBrown return 0; 7116df25e58SNeilBrown 71291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 71391d5b470SChuck Lever 714ed7bcdb3STrond Myklebust if (!(mntflags & NFS_MOUNT_WRITE_EAGER) || wbc->for_kupdate || 715ed7bcdb3STrond Myklebust wbc->for_background || wbc->for_sync || wbc->for_reclaim) { 7162b17d725STrond Myklebust ioc = nfs_io_completion_alloc(GFP_KERNEL); 717919e3bd9STrond Myklebust if (ioc) 718ed7bcdb3STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, 719ed7bcdb3STrond Myklebust inode); 720ed7bcdb3STrond Myklebust priority = wb_priority(wbc); 721ed7bcdb3STrond Myklebust } 722919e3bd9STrond Myklebust 723c6fd3511STrond Myklebust do { 724ed7bcdb3STrond Myklebust nfs_pageio_init_write(&pgio, inode, priority, false, 725a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 726919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 727c6fd3511STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, 728c6fd3511STrond Myklebust &pgio); 72996c41455STrond Myklebust pgio.pg_error = 0; 730c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 7316e7434abSTrond Myklebust if (err == -EAGAIN && mntflags & NFS_MOUNT_SOFTERR) 7326e7434abSTrond Myklebust break; 733c6fd3511STrond Myklebust } while (err < 0 && !nfs_error_is_fatal(err)); 734919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 73572cb77f4STrond Myklebust 736f758c885STrond Myklebust if (err < 0) 73772cb77f4STrond Myklebust goto out_err; 738c63c7b05STrond Myklebust return 0; 73972cb77f4STrond Myklebust out_err: 74072cb77f4STrond Myklebust return err; 7411da177e4SLinus Torvalds } 7421da177e4SLinus Torvalds 7431da177e4SLinus Torvalds /* 7441da177e4SLinus Torvalds * Insert a write request into an inode 7451da177e4SLinus Torvalds */ 7460c493b5cSTrond Myklebust static void nfs_inode_add_request(struct nfs_page *req) 7471da177e4SLinus Torvalds { 7480c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 7490c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 7500c493b5cSTrond Myklebust struct nfs_inode *nfsi = NFS_I(mapping->host); 751e7d39069STrond Myklebust 7522bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7532bfc6e56SWeston Andros Adamson 754e7d39069STrond Myklebust /* Lock the request! */ 7557ad84aa9STrond Myklebust nfs_lock_request(req); 756e7d39069STrond Myklebust 75729418aa4SMel Gorman /* 75829418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 75929418aa4SMel Gorman * with invalidate/truncate. 76029418aa4SMel Gorman */ 761600f111eSMatthew Wilcox (Oracle) spin_lock(&mapping->i_private_lock); 7620c493b5cSTrond Myklebust if (likely(!folio_test_swapcache(folio))) { 7632df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 7640c493b5cSTrond Myklebust folio_set_private(folio); 7650c493b5cSTrond Myklebust folio->private = req; 76629418aa4SMel Gorman } 767600f111eSMatthew Wilcox (Oracle) spin_unlock(&mapping->i_private_lock); 768a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 76917089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 770cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 771cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 772cb1410c7SWeston Andros Adamson * adding subrequests. */ 77317089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 774c03b4024STrond Myklebust kref_get(&req->wb_kref); 7751da177e4SLinus Torvalds } 7761da177e4SLinus Torvalds 7771da177e4SLinus Torvalds /* 77889a09141SPeter Zijlstra * Remove a write request from an inode 7791da177e4SLinus Torvalds */ 7801da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7811da177e4SLinus Torvalds { 7826a6d4644SScott Mayhew struct nfs_inode *nfsi = NFS_I(nfs_page_to_inode(req)); 7836a6d4644SScott Mayhew 78420633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 7850c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req->wb_head); 7860c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 7871da177e4SLinus Torvalds 788600f111eSMatthew Wilcox (Oracle) spin_lock(&mapping->i_private_lock); 7890c493b5cSTrond Myklebust if (likely(folio && !folio_test_swapcache(folio))) { 7900c493b5cSTrond Myklebust folio->private = NULL; 7910c493b5cSTrond Myklebust folio_clear_private(folio); 7920c493b5cSTrond Myklebust clear_bit(PG_MAPPED, &req->wb_head->wb_flags); 79329418aa4SMel Gorman } 794600f111eSMatthew Wilcox (Oracle) spin_unlock(&mapping->i_private_lock); 79520633f04SWeston Andros Adamson } 79617089a29SWeston Andros Adamson 79733ea5aaaSZhangXiaoxu if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) { 7986a6d4644SScott Mayhew atomic_long_dec(&nfsi->nrequests); 799dd1b2026SJeff Layton nfs_release_request(req); 80033ea5aaaSZhangXiaoxu } 8011da177e4SLinus Torvalds } 8021da177e4SLinus Torvalds 8030c493b5cSTrond Myklebust static void nfs_mark_request_dirty(struct nfs_page *req) 80461822ab5STrond Myklebust { 8050c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 8060c493b5cSTrond Myklebust if (folio) 8070c493b5cSTrond Myklebust filemap_dirty_folio(folio_mapping(folio), folio); 80861822ab5STrond Myklebust } 80961822ab5STrond Myklebust 8103a3908c8STrond Myklebust /* 8113a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8123a3908c8STrond Myklebust * 8130c493b5cSTrond Myklebust * Search through commit lists on @inode for the head request for @folio. 8143a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8153a3908c8STrond Myklebust * 8163a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8173a3908c8STrond Myklebust */ 8183a3908c8STrond Myklebust static struct nfs_page * 8193a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8200c493b5cSTrond Myklebust struct folio *folio) 8213a3908c8STrond Myklebust { 8223a3908c8STrond Myklebust struct nfs_page *freq, *t; 8233a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8243a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8253a3908c8STrond Myklebust 8263a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8273a3908c8STrond Myklebust 8283a3908c8STrond Myklebust /* search through pnfs commit lists */ 8290c493b5cSTrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, folio); 8303a3908c8STrond Myklebust if (freq) 8313a3908c8STrond Myklebust return freq->wb_head; 8323a3908c8STrond Myklebust 8333a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8343a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8350c493b5cSTrond Myklebust if (nfs_page_to_folio(freq) == folio) 8363a3908c8STrond Myklebust return freq->wb_head; 8373a3908c8STrond Myklebust } 8383a3908c8STrond Myklebust 8393a3908c8STrond Myklebust return NULL; 8403a3908c8STrond Myklebust } 8413a3908c8STrond Myklebust 8428dd37758STrond Myklebust /** 84386d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 84486d80f97STrond Myklebust * @req: pointer to a struct nfs_page 84586d80f97STrond Myklebust * @dst: commit list head 84686d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 84786d80f97STrond Myklebust * 84886d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 84986d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 85086d80f97STrond Myklebust * the MM page stats. 85186d80f97STrond Myklebust * 852e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 853e824f99aSTrond Myklebust * nfs_page lock. 85486d80f97STrond Myklebust */ 85586d80f97STrond Myklebust void 85686d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 85786d80f97STrond Myklebust struct nfs_commit_info *cinfo) 85886d80f97STrond Myklebust { 85986d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 86086d80f97STrond Myklebust nfs_list_add_request(req, dst); 8615cb953d4STrond Myklebust atomic_long_inc(&cinfo->mds->ncommit); 86286d80f97STrond Myklebust } 86386d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 86486d80f97STrond Myklebust 86586d80f97STrond Myklebust /** 8668dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8678dd37758STrond Myklebust * @req: pointer to a struct nfs_page 868ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8698dd37758STrond Myklebust * 870ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8718dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8728dd37758STrond Myklebust * the MM page stats. 8738dd37758STrond Myklebust * 874ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8758dd37758STrond Myklebust * holding the nfs_page lock. 8768dd37758STrond Myklebust */ 8778dd37758STrond Myklebust void 8786272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8798dd37758STrond Myklebust { 880e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8816272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 882e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 8830c493b5cSTrond Myklebust nfs_folio_mark_unstable(nfs_page_to_folio(req), cinfo); 8848dd37758STrond Myklebust } 8858dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8868dd37758STrond Myklebust 8878dd37758STrond Myklebust /** 8888dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8898dd37758STrond Myklebust * @req: pointer to a nfs_page 890ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8918dd37758STrond Myklebust * 892ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8938dd37758STrond Myklebust * number of outstanding requests requiring a commit 8948dd37758STrond Myklebust * It does not update the MM page stats. 8958dd37758STrond Myklebust * 896ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8978dd37758STrond Myklebust */ 8988dd37758STrond Myklebust void 899ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 900ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9018dd37758STrond Myklebust { 9028dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9038dd37758STrond Myklebust return; 9048dd37758STrond Myklebust nfs_list_remove_request(req); 9055cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9068dd37758STrond Myklebust } 9078dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9088dd37758STrond Myklebust 909ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 910ea2cf228SFred Isaman struct inode *inode) 911ea2cf228SFred Isaman { 912fe238e60SDave Wysochanski cinfo->inode = inode; 913ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 914ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 915b359f9d0SFred Isaman cinfo->dreq = NULL; 916f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 917ea2cf228SFred Isaman } 918ea2cf228SFred Isaman 919ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 920ea2cf228SFred Isaman struct inode *inode, 921ea2cf228SFred Isaman struct nfs_direct_req *dreq) 922ea2cf228SFred Isaman { 9231763da12SFred Isaman if (dreq) 9241763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9251763da12SFred Isaman else 926ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 927ea2cf228SFred Isaman } 928ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9298dd37758STrond Myklebust 9301da177e4SLinus Torvalds /* 9311da177e4SLinus Torvalds * Add a request to the inode's commit list. 9321da177e4SLinus Torvalds */ 9331763da12SFred Isaman void 934ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 935b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9361da177e4SLinus Torvalds { 937b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9388dd37758STrond Myklebust return; 9396272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9401da177e4SLinus Torvalds } 9418e821cadSTrond Myklebust 9420c493b5cSTrond Myklebust static void nfs_folio_clear_commit(struct folio *folio) 943e468bae9STrond Myklebust { 9440c493b5cSTrond Myklebust if (folio) { 9450c493b5cSTrond Myklebust long nr = folio_nr_pages(folio); 9460c493b5cSTrond Myklebust 9470c493b5cSTrond Myklebust node_stat_mod_folio(folio, NR_WRITEBACK, -nr); 9480c493b5cSTrond Myklebust wb_stat_mod(&inode_to_bdi(folio_file_mapping(folio)->host)->wb, 9490c493b5cSTrond Myklebust WB_WRITEBACK, -nr); 9500c493b5cSTrond Myklebust } 951e468bae9STrond Myklebust } 952d6d6dc7cSFred Isaman 953b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 954b193a78dSTrond Myklebust static void nfs_clear_request_commit(struct nfs_commit_info *cinfo, 955b193a78dSTrond Myklebust struct nfs_page *req) 956d6d6dc7cSFred Isaman { 9578dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9589fcd5960STrond Myklebust struct nfs_open_context *ctx = nfs_req_openctx(req); 9599fcd5960STrond Myklebust struct inode *inode = d_inode(ctx->dentry); 960d6d6dc7cSFred Isaman 961e824f99aSTrond Myklebust mutex_lock(&NFS_I(inode)->commit_mutex); 962b193a78dSTrond Myklebust if (!pnfs_clear_request_commit(req, cinfo)) { 963b193a78dSTrond Myklebust nfs_request_remove_commit_list(req, cinfo); 964d6d6dc7cSFred Isaman } 965e824f99aSTrond Myklebust mutex_unlock(&NFS_I(inode)->commit_mutex); 9660c493b5cSTrond Myklebust nfs_folio_clear_commit(nfs_page_to_folio(req)); 9678dd37758STrond Myklebust } 968e468bae9STrond Myklebust } 969e468bae9STrond Myklebust 970d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9718e821cadSTrond Myklebust { 972c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 973d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 974c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9758e821cadSTrond Myklebust } 9768e821cadSTrond Myklebust 977919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 978919e3bd9STrond Myklebust { 979919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 980919e3bd9STrond Myklebust } 981919e3bd9STrond Myklebust 982061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9836c75dc0dSFred Isaman { 984ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9856c75dc0dSFred Isaman unsigned long bytes = 0; 9866c75dc0dSFred Isaman 9876c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9886c75dc0dSFred Isaman goto out; 989ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9906c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9916c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9926c75dc0dSFred Isaman 9936c75dc0dSFred Isaman bytes += req->wb_bytes; 9946c75dc0dSFred Isaman nfs_list_remove_request(req); 9956c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9966c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 997af887e43STrond Myklebust trace_nfs_comp_error(hdr->inode, req, hdr->error); 9980c493b5cSTrond Myklebust nfs_mapping_set_error(nfs_page_to_folio(req), 9990c493b5cSTrond Myklebust hdr->error); 10006c75dc0dSFred Isaman goto remove_req; 10016c75dc0dSFred Isaman } 1002c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 100333344e0fSTrond Myklebust /* Reset wb_nio, since the write was successful. */ 100433344e0fSTrond Myklebust req->wb_nio = 0; 1005f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1006b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1007a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10086c75dc0dSFred Isaman goto next; 10096c75dc0dSFred Isaman } 10106c75dc0dSFred Isaman remove_req: 10116c75dc0dSFred Isaman nfs_inode_remove_request(req); 10126c75dc0dSFred Isaman next: 10130c493b5cSTrond Myklebust nfs_page_end_writeback(req); 10143aff4ebbSTrond Myklebust nfs_release_request(req); 10156c75dc0dSFred Isaman } 10166c75dc0dSFred Isaman out: 1017919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10186c75dc0dSFred Isaman hdr->release(hdr); 10196c75dc0dSFred Isaman } 10206c75dc0dSFred Isaman 1021ce59515cSAnna Schumaker unsigned long 1022ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1023fb8a1f11STrond Myklebust { 10245cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1025fb8a1f11STrond Myklebust } 1026fb8a1f11STrond Myklebust 1027e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10281763da12SFred Isaman int 1029ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1030ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1031d6d6dc7cSFred Isaman { 1032137da553STrond Myklebust struct nfs_page *req, *tmp; 1033d6d6dc7cSFred Isaman int ret = 0; 1034d6d6dc7cSFred Isaman 1035137da553STrond Myklebust list_for_each_entry_safe(req, tmp, src, wb_list) { 10367ad84aa9STrond Myklebust kref_get(&req->wb_kref); 10372ce209c4STrond Myklebust if (!nfs_lock_request(req)) { 1038137da553STrond Myklebust nfs_release_request(req); 1039137da553STrond Myklebust continue; 1040137da553STrond Myklebust } 1041ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10425d2a9d9dSTrond Myklebust clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); 10438dd37758STrond Myklebust nfs_list_add_request(req, dst); 1044d6d6dc7cSFred Isaman ret++; 10451763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1046d6d6dc7cSFred Isaman break; 1047e824f99aSTrond Myklebust cond_resched(); 1048d6d6dc7cSFred Isaman } 1049d6d6dc7cSFred Isaman return ret; 1050d6d6dc7cSFred Isaman } 10515d2a9d9dSTrond Myklebust EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 1052d6d6dc7cSFred Isaman 10531da177e4SLinus Torvalds /* 10541da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10551da177e4SLinus Torvalds * @inode: NFS inode to scan 1056ea2cf228SFred Isaman * @dst: mds destination list 1057ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10581da177e4SLinus Torvalds * 10591da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10601da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10611da177e4SLinus Torvalds */ 10621763da12SFred Isaman int 1063ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1064ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10651da177e4SLinus Torvalds { 1066d6d6dc7cSFred Isaman int ret = 0; 1067fb8a1f11STrond Myklebust 10685cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10695cb953d4STrond Myklebust return 0; 1070e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10715cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10728dd37758STrond Myklebust const int max = INT_MAX; 1073d6d6dc7cSFred Isaman 1074ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1075ea2cf228SFred Isaman cinfo, max); 1076ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1077d6d6dc7cSFred Isaman } 1078e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1079ff778d02STrond Myklebust return ret; 10801da177e4SLinus Torvalds } 1081d6d6dc7cSFred Isaman 10821da177e4SLinus Torvalds /* 1083e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1084e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10851da177e4SLinus Torvalds * 1086e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1087e7d39069STrond Myklebust * to disk. 10881da177e4SLinus Torvalds */ 10890c493b5cSTrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct folio *folio, 1090e7d39069STrond Myklebust unsigned int offset, 1091e7d39069STrond Myklebust unsigned int bytes) 10921da177e4SLinus Torvalds { 1093e7d39069STrond Myklebust struct nfs_page *req; 1094e7d39069STrond Myklebust unsigned int rqend; 1095e7d39069STrond Myklebust unsigned int end; 10961da177e4SLinus Torvalds int error; 1097277459d2STrond Myklebust 1098e7d39069STrond Myklebust end = offset + bytes; 1099e7d39069STrond Myklebust 11000c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 1101f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1102f6032f21STrond Myklebust return req; 11032bfc6e56SWeston Andros Adamson 1104e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1105e7d39069STrond Myklebust /* 1106e7d39069STrond Myklebust * Tell the caller to flush out the request if 1107e7d39069STrond Myklebust * the offsets are non-contiguous. 1108e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1109e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1110e7d39069STrond Myklebust */ 1111f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1112e7d39069STrond Myklebust goto out_flushme; 1113e7d39069STrond Myklebust 11141da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11151da177e4SLinus Torvalds if (offset < req->wb_offset) { 11161da177e4SLinus Torvalds req->wb_offset = offset; 11171da177e4SLinus Torvalds req->wb_pgbase = offset; 11181da177e4SLinus Torvalds } 11191da177e4SLinus Torvalds if (end > rqend) 11201da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1121e7d39069STrond Myklebust else 1122e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 112333344e0fSTrond Myklebust req->wb_nio = 0; 1124e7d39069STrond Myklebust return req; 1125e7d39069STrond Myklebust out_flushme: 1126f6032f21STrond Myklebust /* 1127f6032f21STrond Myklebust * Note: we mark the request dirty here because 1128f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1129f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1130f6032f21STrond Myklebust */ 1131f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1132f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 11330c493b5cSTrond Myklebust error = nfs_wb_folio(folio_file_mapping(folio)->host, folio); 1134f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1135e7d39069STrond Myklebust } 11361da177e4SLinus Torvalds 1137e7d39069STrond Myklebust /* 1138e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1139e7d39069STrond Myklebust * 1140e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1141e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1142e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1143e7d39069STrond Myklebust */ 1144e7d39069STrond Myklebust static struct nfs_page *nfs_setup_write_request(struct nfs_open_context *ctx, 11450c493b5cSTrond Myklebust struct folio *folio, 11460c493b5cSTrond Myklebust unsigned int offset, 11470c493b5cSTrond Myklebust unsigned int bytes) 1148e7d39069STrond Myklebust { 1149e7d39069STrond Myklebust struct nfs_page *req; 1150e7d39069STrond Myklebust 11510c493b5cSTrond Myklebust req = nfs_try_to_update_request(folio, offset, bytes); 1152e7d39069STrond Myklebust if (req != NULL) 1153e7d39069STrond Myklebust goto out; 11540c493b5cSTrond Myklebust req = nfs_page_create_from_folio(ctx, folio, offset, bytes); 1155e7d39069STrond Myklebust if (IS_ERR(req)) 1156e7d39069STrond Myklebust goto out; 11570c493b5cSTrond Myklebust nfs_inode_add_request(req); 1158efc91ed0STrond Myklebust out: 115961e930a9STrond Myklebust return req; 11601da177e4SLinus Torvalds } 11611da177e4SLinus Torvalds 11620c493b5cSTrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, 11630c493b5cSTrond Myklebust struct folio *folio, unsigned int offset, 11640c493b5cSTrond Myklebust unsigned int count) 1165e7d39069STrond Myklebust { 1166e7d39069STrond Myklebust struct nfs_page *req; 1167e7d39069STrond Myklebust 11680c493b5cSTrond Myklebust req = nfs_setup_write_request(ctx, folio, offset, count); 1169e7d39069STrond Myklebust if (IS_ERR(req)) 1170e7d39069STrond Myklebust return PTR_ERR(req); 1171e7d39069STrond Myklebust /* Update file length */ 11720c493b5cSTrond Myklebust nfs_grow_file(folio, offset, count); 1173d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1174a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11751d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1176e7d39069STrond Myklebust return 0; 1177e7d39069STrond Myklebust } 1178e7d39069STrond Myklebust 11790c493b5cSTrond Myklebust int nfs_flush_incompatible(struct file *file, struct folio *folio) 11801da177e4SLinus Torvalds { 1181cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11822a369153STrond Myklebust struct nfs_lock_context *l_ctx; 118317b985deSJeff Layton struct file_lock_context *flctx = locks_inode_context(file_inode(file)); 11841da177e4SLinus Torvalds struct nfs_page *req; 11851a54533eSTrond Myklebust int do_flush, status; 11861da177e4SLinus Torvalds /* 11871da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11881da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11891da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11901da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11911da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11921da177e4SLinus Torvalds * dropped page. 11931da177e4SLinus Torvalds */ 11941a54533eSTrond Myklebust do { 11950c493b5cSTrond Myklebust req = nfs_folio_find_head_request(folio); 11961a54533eSTrond Myklebust if (req == NULL) 11971a54533eSTrond Myklebust return 0; 11982a369153STrond Myklebust l_ctx = req->wb_lock_context; 11990c493b5cSTrond Myklebust do_flush = nfs_page_to_folio(req) != folio || 12009fcd5960STrond Myklebust !nfs_match_open_context(nfs_req_openctx(req), ctx); 1201bd61e0a9SJeff Layton if (l_ctx && flctx && 1202bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1203bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1204d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12055263e31eSJeff Layton } 12061da177e4SLinus Torvalds nfs_release_request(req); 12071a54533eSTrond Myklebust if (!do_flush) 12081a54533eSTrond Myklebust return 0; 12090c493b5cSTrond Myklebust status = nfs_wb_folio(folio_file_mapping(folio)->host, folio); 12101a54533eSTrond Myklebust } while (status == 0); 12111a54533eSTrond Myklebust return status; 12121da177e4SLinus Torvalds } 12131da177e4SLinus Torvalds 12141da177e4SLinus Torvalds /* 1215dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1216dc24826bSAndy Adamson * expire soon. 1217dc24826bSAndy Adamson * 1218dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1219dc24826bSAndy Adamson * 1220dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1221dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1222dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1223dc24826bSAndy Adamson */ 1224dc24826bSAndy Adamson int 1225dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1226dc24826bSAndy Adamson { 1227dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1228dc24826bSAndy Adamson 1229ddf529eeSNeilBrown if (nfs_ctx_key_to_expire(ctx, inode) && 1230ca05cbaeSTrond Myklebust !rcu_access_pointer(ctx->ll_cred)) 1231ddf529eeSNeilBrown /* Already expired! */ 1232ddf529eeSNeilBrown return -EACCES; 1233ddf529eeSNeilBrown return 0; 1234dc24826bSAndy Adamson } 1235dc24826bSAndy Adamson 1236dc24826bSAndy Adamson /* 1237dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1238dc24826bSAndy Adamson */ 1239ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1240dc24826bSAndy Adamson { 1241ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1242ca05cbaeSTrond Myklebust struct rpc_cred *cred, *new, *old = NULL; 1243ddf529eeSNeilBrown struct auth_cred acred = { 1244a52458b4SNeilBrown .cred = ctx->cred, 1245ddf529eeSNeilBrown }; 1246ca05cbaeSTrond Myklebust bool ret = false; 1247ce52914eSScott Mayhew 1248ca05cbaeSTrond Myklebust rcu_read_lock(); 1249ca05cbaeSTrond Myklebust cred = rcu_dereference(ctx->ll_cred); 1250ca05cbaeSTrond Myklebust if (cred && !(cred->cr_ops->crkey_timeout && 1251ca05cbaeSTrond Myklebust cred->cr_ops->crkey_timeout(cred))) 1252ca05cbaeSTrond Myklebust goto out; 1253ca05cbaeSTrond Myklebust rcu_read_unlock(); 1254ca05cbaeSTrond Myklebust 1255ca05cbaeSTrond Myklebust new = auth->au_ops->lookup_cred(auth, &acred, 0); 1256ca05cbaeSTrond Myklebust if (new == cred) { 1257ca05cbaeSTrond Myklebust put_rpccred(new); 1258ddf529eeSNeilBrown return true; 1259ca05cbaeSTrond Myklebust } 1260ca05cbaeSTrond Myklebust if (IS_ERR_OR_NULL(new)) { 1261ca05cbaeSTrond Myklebust new = NULL; 1262ca05cbaeSTrond Myklebust ret = true; 1263ca05cbaeSTrond Myklebust } else if (new->cr_ops->crkey_timeout && 1264ca05cbaeSTrond Myklebust new->cr_ops->crkey_timeout(new)) 1265ca05cbaeSTrond Myklebust ret = true; 1266ca05cbaeSTrond Myklebust 1267ca05cbaeSTrond Myklebust rcu_read_lock(); 1268ca05cbaeSTrond Myklebust old = rcu_dereference_protected(xchg(&ctx->ll_cred, 1269ca05cbaeSTrond Myklebust RCU_INITIALIZER(new)), 1); 1270ca05cbaeSTrond Myklebust out: 1271ca05cbaeSTrond Myklebust rcu_read_unlock(); 1272ca05cbaeSTrond Myklebust put_rpccred(old); 1273ca05cbaeSTrond Myklebust return ret; 1274dc24826bSAndy Adamson } 1275dc24826bSAndy Adamson 1276dc24826bSAndy Adamson /* 12775d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12785d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12795d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12805d47a356STrond Myklebust */ 12810c493b5cSTrond Myklebust static bool nfs_folio_write_uptodate(struct folio *folio, unsigned int pagelen) 12825d47a356STrond Myklebust { 12830c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 1284d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1285d529ef83SJeff Layton 12868d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12878d197a56STrond Myklebust goto out; 1288fc9dc401STrond Myklebust if (nfsi->cache_validity & 128913c0b082STrond Myklebust (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE)) 1290d529ef83SJeff Layton return false; 12914db72b40SJeff Layton smp_rmb(); 1292fc9dc401STrond Myklebust if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags) && pagelen != 0) 12938d197a56STrond Myklebust return false; 12948d197a56STrond Myklebust out: 1295fc9dc401STrond Myklebust if (nfsi->cache_validity & NFS_INO_INVALID_DATA && pagelen != 0) 129618dd78c4SScott Mayhew return false; 12970c493b5cSTrond Myklebust return folio_test_uptodate(folio) != 0; 12985d47a356STrond Myklebust } 12995d47a356STrond Myklebust 13005263e31eSJeff Layton static bool 13015263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 13025263e31eSJeff Layton { 13035263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 1304*d7c9616bSJeff Layton lock_is_write(fl); 13055263e31eSJeff Layton } 13065263e31eSJeff Layton 1307c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1308c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1309c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1310c7559663SScott Mayhew * inefficiencies. 1311c7559663SScott Mayhew * 1312263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1313263b4509SScott Mayhew * of the checks. 1314c7559663SScott Mayhew */ 13150c493b5cSTrond Myklebust static int nfs_can_extend_write(struct file *file, struct folio *folio, 13160c493b5cSTrond Myklebust unsigned int pagelen) 1317c7559663SScott Mayhew { 13180c493b5cSTrond Myklebust struct inode *inode = file_inode(file); 131917b985deSJeff Layton struct file_lock_context *flctx = locks_inode_context(inode); 13205263e31eSJeff Layton struct file_lock *fl; 13210c493b5cSTrond Myklebust int ret; 13225263e31eSJeff Layton 1323c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1324c7559663SScott Mayhew return 0; 13250c493b5cSTrond Myklebust if (!nfs_folio_write_uptodate(folio, pagelen)) 1326263b4509SScott Mayhew return 0; 1327c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1328c7559663SScott Mayhew return 1; 1329bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1330bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13318fa4592aSTrond Myklebust return 1; 13325263e31eSJeff Layton 13335263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13345263e31eSJeff Layton ret = 0; 13356109c850SJeff Layton spin_lock(&flctx->flc_lock); 1336bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1337bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1338bd61e0a9SJeff Layton fl_list); 1339bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13405263e31eSJeff Layton ret = 1; 1341bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13425263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13435263e31eSJeff Layton fl_list); 1344*d7c9616bSJeff Layton if (lock_is_write(fl)) 13455263e31eSJeff Layton ret = 1; 13465263e31eSJeff Layton } 13476109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13485263e31eSJeff Layton return ret; 1349c7559663SScott Mayhew } 1350c7559663SScott Mayhew 13515d47a356STrond Myklebust /* 13521da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13531da177e4SLinus Torvalds * 13541da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13551da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13561da177e4SLinus Torvalds */ 13570c493b5cSTrond Myklebust int nfs_update_folio(struct file *file, struct folio *folio, 13581da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13591da177e4SLinus Torvalds { 1360cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 13610c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 1362d2ceb7e5SBenjamin Coddington struct inode *inode = mapping->host; 13630c493b5cSTrond Myklebust unsigned int pagelen = nfs_folio_length(folio); 13641da177e4SLinus Torvalds int status = 0; 13651da177e4SLinus Torvalds 136691d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 136791d5b470SChuck Lever 13680c493b5cSTrond Myklebust dprintk("NFS: nfs_update_folio(%pD2 %d@%lld)\n", file, count, 13690c493b5cSTrond Myklebust (long long)(folio_file_pos(folio) + offset)); 13701da177e4SLinus Torvalds 1371149a4fddSBenjamin Coddington if (!count) 1372149a4fddSBenjamin Coddington goto out; 1373149a4fddSBenjamin Coddington 13740c493b5cSTrond Myklebust if (nfs_can_extend_write(file, folio, pagelen)) { 1375fc9dc401STrond Myklebust count = max(count + offset, pagelen); 13761da177e4SLinus Torvalds offset = 0; 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds 13790c493b5cSTrond Myklebust status = nfs_writepage_setup(ctx, folio, offset, count); 138003fa9e84STrond Myklebust if (status < 0) 1381d2ceb7e5SBenjamin Coddington nfs_set_pageerror(mapping); 1382149a4fddSBenjamin Coddington out: 13830c493b5cSTrond Myklebust dprintk("NFS: nfs_update_folio returns %d (isize %lld)\n", 13841da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13851da177e4SLinus Torvalds return status; 13861da177e4SLinus Torvalds } 13871da177e4SLinus Torvalds 13883ff7576dSTrond Myklebust static int flush_task_priority(int how) 13891da177e4SLinus Torvalds { 13901da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13911da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13921da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13931da177e4SLinus Torvalds case FLUSH_LOWPRI: 13941da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13971da177e4SLinus Torvalds } 13981da177e4SLinus Torvalds 1399d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1400d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1401abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 14021ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 14031da177e4SLinus Torvalds { 14043ff7576dSTrond Myklebust int priority = flush_task_priority(how); 14051da177e4SLinus Torvalds 14068db55a03SNeilBrown if (IS_SWAPFILE(hdr->inode)) 14078db55a03SNeilBrown task_setup_data->flags |= RPC_TASK_SWAPPER; 14081ed26f33SAnna Schumaker task_setup_data->priority = priority; 1409fb91fb0eSAnna Schumaker rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); 14105bb2a7cbSTrond Myklebust trace_nfs_initiate_write(hdr); 1411275acaafSTrond Myklebust } 1412275acaafSTrond Myklebust 14136d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 14146d884e8fSFred * call this on each, which will prepare them to be retried on next 14156d884e8fSFred * writeback using standard nfs. 14166d884e8fSFred */ 14176d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14186d884e8fSFred { 14196dd85e83STrond Myklebust struct nfs_inode *nfsi = NFS_I(nfs_page_to_inode(req)); 142067f4b5dcSTrond Myklebust 142133344e0fSTrond Myklebust /* Bump the transmission count */ 142233344e0fSTrond Myklebust req->wb_nio++; 14236d884e8fSFred nfs_mark_request_dirty(req); 142467f4b5dcSTrond Myklebust atomic_long_inc(&nfsi->redirtied_pages); 14250c493b5cSTrond Myklebust nfs_page_end_writeback(req); 14263aff4ebbSTrond Myklebust nfs_release_request(req); 14276d884e8fSFred } 14286d884e8fSFred 1429df3accb8STrond Myklebust static void nfs_async_write_error(struct list_head *head, int error) 14306c75dc0dSFred Isaman { 14316c75dc0dSFred Isaman struct nfs_page *req; 14326c75dc0dSFred Isaman 14336c75dc0dSFred Isaman while (!list_empty(head)) { 14346c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14356c75dc0dSFred Isaman nfs_list_remove_request(req); 1436cea9ba72STrond Myklebust if (nfs_error_is_fatal_on_server(error)) 14376fbda89bSTrond Myklebust nfs_write_error(req, error); 14386fbda89bSTrond Myklebust else 14396c75dc0dSFred Isaman nfs_redirty_request(req); 14406c75dc0dSFred Isaman } 14416c75dc0dSFred Isaman } 14426c75dc0dSFred Isaman 1443dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1444dc602dd7STrond Myklebust { 1445df3accb8STrond Myklebust nfs_async_write_error(&hdr->pages, 0); 1446dc602dd7STrond Myklebust } 1447dc602dd7STrond Myklebust 1448061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1449919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1450061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1451061ae2edSFred Isaman .completion = nfs_write_completion, 1452dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1453061ae2edSFred Isaman }; 1454061ae2edSFred Isaman 145557208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1456a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1457061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14581751c363STrond Myklebust { 1459a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 146041d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1461a20c93e3SChristoph Hellwig 1462a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1463a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1464a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1465a20c93e3SChristoph Hellwig #endif 14664a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 14673bde7afdSTrond Myklebust server->wsize, ioflags); 14681751c363STrond Myklebust } 1469ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14701751c363STrond Myklebust 1471dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1472dce81290STrond Myklebust { 1473a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1474a7d42ddbSWeston Andros Adamson 14756f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14766f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14776f29b9bbSKinglong Mee 147841d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1479a7d42ddbSWeston Andros Adamson 1480a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1481a7d42ddbSWeston Andros Adamson 1482a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1483a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1484dce81290STrond Myklebust } 14851f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1486dce81290STrond Myklebust 14871da177e4SLinus Torvalds 14880b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14890b7c0153SFred Isaman { 14900b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14910b7c0153SFred Isaman 14920b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14930b7c0153SFred Isaman } 14940b7c0153SFred Isaman 1495a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1496a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1497a08a8cd3STrond Myklebust { 1498a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1499a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 15002b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1501a08a8cd3STrond Myklebust 1502a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15032b83d3deSTrond Myklebust fattr->size = size; 15042b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15052b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1506a08a8cd3STrond Myklebust return; 15072b83d3deSTrond Myklebust } 15082b83d3deSTrond Myklebust if (size != fattr->size) 1509a08a8cd3STrond Myklebust return; 1510a08a8cd3STrond Myklebust /* Set attribute barrier */ 1511a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15122b83d3deSTrond Myklebust /* ...and update size */ 15132b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1514a08a8cd3STrond Myklebust } 1515a08a8cd3STrond Myklebust 1516a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1517a08a8cd3STrond Myklebust { 15182b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1519a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1520a08a8cd3STrond Myklebust 1521a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1522a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1523a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1524a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1525a08a8cd3STrond Myklebust } 1526a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1527a08a8cd3STrond Myklebust 15281da177e4SLinus Torvalds /* 15291da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15301da177e4SLinus Torvalds */ 1531d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1532d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15330eecb214SAnna Schumaker struct inode *inode) 15341da177e4SLinus Torvalds { 1535788e7a89STrond Myklebust int status; 15361da177e4SLinus Torvalds 1537f551e44fSChuck Lever /* 1538f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1539f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1540f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1541f551e44fSChuck Lever * another writer had changed the file, but some applications 1542f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1543f551e44fSChuck Lever */ 1544d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1545788e7a89STrond Myklebust if (status != 0) 15460eecb214SAnna Schumaker return status; 15478224b273SChuck Lever 1548d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 15495bb2a7cbSTrond Myklebust trace_nfs_writeback_done(task, hdr); 155091d5b470SChuck Lever 155169d96651SJeff Layton if (task->tk_status >= 0) { 155269d96651SJeff Layton enum nfs3_stable_how committed = hdr->res.verf->committed; 155369d96651SJeff Layton 155469d96651SJeff Layton if (committed == NFS_UNSTABLE) { 155569d96651SJeff Layton /* 155669d96651SJeff Layton * We have some uncommitted data on the server at 155769d96651SJeff Layton * this point, so ensure that we keep track of that 155869d96651SJeff Layton * fact irrespective of what later writes do. 155969d96651SJeff Layton */ 156069d96651SJeff Layton set_bit(NFS_IOHDR_UNSTABLE_WRITES, &hdr->flags); 156169d96651SJeff Layton } 156269d96651SJeff Layton 156369d96651SJeff Layton if (committed < hdr->args.stable) { 15641da177e4SLinus Torvalds /* We tried a write call, but the server did not 15651da177e4SLinus Torvalds * commit data to stable storage even though we 15661da177e4SLinus Torvalds * requested it. 15671da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15681da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15691da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15701da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15711da177e4SLinus Torvalds */ 15721da177e4SLinus Torvalds static unsigned long complain; 15731da177e4SLinus Torvalds 1574a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15751da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15761da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15771da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1578cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 157969d96651SJeff Layton committed, hdr->args.stable); 15801da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15811da177e4SLinus Torvalds } 15821da177e4SLinus Torvalds } 158369d96651SJeff Layton } 15841f2edbe3STrond Myklebust 15851f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 158616e14375STrond Myklebust if (nfs_should_remove_suid(inode)) { 158716e14375STrond Myklebust spin_lock(&inode->i_lock); 1588720869ebSTrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); 158916e14375STrond Myklebust spin_unlock(&inode->i_lock); 159016e14375STrond Myklebust } 15910eecb214SAnna Schumaker return 0; 15920eecb214SAnna Schumaker } 15930eecb214SAnna Schumaker 15940eecb214SAnna Schumaker /* 15950eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15960eecb214SAnna Schumaker */ 1597d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1598d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15990eecb214SAnna Schumaker { 1600d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1601d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 16021f2edbe3STrond Myklebust 16031f2edbe3STrond Myklebust if (resp->count < argp->count) { 16041da177e4SLinus Torvalds static unsigned long complain; 16051da177e4SLinus Torvalds 16066c75dc0dSFred Isaman /* This a short write! */ 1607d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 160891d5b470SChuck Lever 16091da177e4SLinus Torvalds /* Has the server at least made some progress? */ 16106c75dc0dSFred Isaman if (resp->count == 0) { 16116c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 16126c75dc0dSFred Isaman printk(KERN_WARNING 16136c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16146c75dc0dSFred Isaman argp->count); 16156c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16166c75dc0dSFred Isaman } 1617d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16186c75dc0dSFred Isaman task->tk_status = -EIO; 16196c75dc0dSFred Isaman return; 16206c75dc0dSFred Isaman } 1621f8417b48SKinglong Mee 1622f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1623f8417b48SKinglong Mee if (!task->tk_ops) { 1624f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1625f8417b48SKinglong Mee return; 1626f8417b48SKinglong Mee } 1627f8417b48SKinglong Mee 16281da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16291da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16301da177e4SLinus Torvalds /* Resend from where the server left off */ 1631d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16321da177e4SLinus Torvalds argp->offset += resp->count; 16331da177e4SLinus Torvalds argp->pgbase += resp->count; 16341da177e4SLinus Torvalds argp->count -= resp->count; 16351da177e4SLinus Torvalds } else { 16361da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16371da177e4SLinus Torvalds * headaches in the case of a server crash. 16381da177e4SLinus Torvalds */ 16391da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16401da177e4SLinus Torvalds } 16418c9cb714STrond Myklebust resp->count = 0; 16428c9cb714STrond Myklebust resp->verf->committed = 0; 1643d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16441da177e4SLinus Torvalds } 16451da177e4SLinus Torvalds } 16461da177e4SLinus Torvalds 1647af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 164871d0a611STrond Myklebust { 1649723c921eSPeter Zijlstra return wait_var_event_killable(&cinfo->rpcs_out, 1650723c921eSPeter Zijlstra !atomic_read(&cinfo->rpcs_out)); 1651af7cf057STrond Myklebust } 1652af7cf057STrond Myklebust 1653af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1654af7cf057STrond Myklebust { 1655af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1656af7cf057STrond Myklebust } 1657af7cf057STrond Myklebust 1658133a48abSTrond Myklebust bool nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1659af7cf057STrond Myklebust { 1660133a48abSTrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) { 1661723c921eSPeter Zijlstra wake_up_var(&cinfo->rpcs_out); 1662133a48abSTrond Myklebust return true; 1663133a48abSTrond Myklebust } 1664133a48abSTrond Myklebust return false; 166571d0a611STrond Myklebust } 166671d0a611STrond Myklebust 16670b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16681da177e4SLinus Torvalds { 16690b7c0153SFred Isaman put_nfs_open_context(data->context); 16700b7c0153SFred Isaman nfs_commit_free(data); 16711da177e4SLinus Torvalds } 1672e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16731da177e4SLinus Torvalds 16740b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1675c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16769ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16779f0ec176SAndy Adamson int how, int flags) 16781da177e4SLinus Torvalds { 167907737691STrond Myklebust struct rpc_task *task; 16809ace33cdSFred Isaman int priority = flush_task_priority(how); 1681bdc7f021STrond Myklebust struct rpc_message msg = { 1682bdc7f021STrond Myklebust .rpc_argp = &data->args, 1683bdc7f021STrond Myklebust .rpc_resp = &data->res, 16849ace33cdSFred Isaman .rpc_cred = data->cred, 1685bdc7f021STrond Myklebust }; 168684115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 168707737691STrond Myklebust .task = &data->task, 16889ace33cdSFred Isaman .rpc_client = clnt, 1689bdc7f021STrond Myklebust .rpc_message = &msg, 16909ace33cdSFred Isaman .callback_ops = call_ops, 169184115e1cSTrond Myklebust .callback_data = data, 1692101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16934fa7ef69STrond Myklebust .flags = RPC_TASK_ASYNC | flags, 16943ff7576dSTrond Myklebust .priority = priority, 169584115e1cSTrond Myklebust }; 1696118f09edSOlga Kornievskaia 1697118f09edSOlga Kornievskaia if (nfs_server_capable(data->inode, NFS_CAP_MOVEABLE)) 1698118f09edSOlga Kornievskaia task_setup_data.flags |= RPC_TASK_MOVEABLE; 1699118f09edSOlga Kornievskaia 1700788e7a89STrond Myklebust /* Set up the initial task struct. */ 1701e9ae1ee2SAnna Schumaker nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); 17028224b273SChuck Lever trace_nfs_initiate_commit(data); 17031da177e4SLinus Torvalds 1704b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1705bdc7f021STrond Myklebust 170607737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1707dbae4c73STrond Myklebust if (IS_ERR(task)) 1708dbae4c73STrond Myklebust return PTR_ERR(task); 1709d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1710d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 171107737691STrond Myklebust rpc_put_task(task); 1712dbae4c73STrond Myklebust return 0; 17131da177e4SLinus Torvalds } 1714e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 17151da177e4SLinus Torvalds 1716378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1717378520b8SPeng Tao { 1718378520b8SPeng Tao loff_t lwb = 0; 1719378520b8SPeng Tao struct nfs_page *req; 1720378520b8SPeng Tao 1721378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1722378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1723378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1724378520b8SPeng Tao 1725378520b8SPeng Tao return lwb; 1726378520b8SPeng Tao } 1727378520b8SPeng Tao 17281da177e4SLinus Torvalds /* 17299ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17309ace33cdSFred Isaman */ 17310b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1732988b6dceSFred Isaman struct list_head *head, 1733f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1734f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17359ace33cdSFred Isaman { 173619573c93STrond Myklebust struct nfs_page *first; 173719573c93STrond Myklebust struct nfs_open_context *ctx; 173819573c93STrond Myklebust struct inode *inode; 17399ace33cdSFred Isaman 17409ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17419ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17429ace33cdSFred Isaman 174319573c93STrond Myklebust if (head) 17449ace33cdSFred Isaman list_splice_init(head, &data->pages); 17459ace33cdSFred Isaman 174619573c93STrond Myklebust first = nfs_list_entry(data->pages.next); 174719573c93STrond Myklebust ctx = nfs_req_openctx(first); 174819573c93STrond Myklebust inode = d_inode(ctx->dentry); 174919573c93STrond Myklebust 17509ace33cdSFred Isaman data->inode = inode; 17519fcd5960STrond Myklebust data->cred = ctx->cred; 1752988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1753378520b8SPeng Tao /* only set lwb for pnfs commit */ 1754378520b8SPeng Tao if (lseg) 1755378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17569ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1757f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1758b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17599ace33cdSFred Isaman 17609ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17619ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17629ace33cdSFred Isaman data->args.offset = 0; 17639ace33cdSFred Isaman data->args.count = 0; 17649fcd5960STrond Myklebust data->context = get_nfs_open_context(ctx); 17659ace33cdSFred Isaman data->res.fattr = &data->fattr; 17669ace33cdSFred Isaman data->res.verf = &data->verf; 17679ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 1768133a48abSTrond Myklebust nfs_commit_begin(cinfo->mds); 17699ace33cdSFred Isaman } 1770e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17719ace33cdSFred Isaman 1772e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1773ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1774b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1775b57ff130SWeston Andros Adamson u32 ds_commit_idx) 177664bfeb49SFred Isaman { 177764bfeb49SFred Isaman struct nfs_page *req; 177864bfeb49SFred Isaman 177964bfeb49SFred Isaman while (!list_empty(page_list)) { 178064bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 178164bfeb49SFred Isaman nfs_list_remove_request(req); 1782b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 17830c493b5cSTrond Myklebust nfs_folio_clear_commit(nfs_page_to_folio(req)); 17841d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 178564bfeb49SFred Isaman } 178664bfeb49SFred Isaman } 1787e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 178864bfeb49SFred Isaman 17890c493b5cSTrond Myklebust static void nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1790b20135d0STrond Myklebust struct nfs_page *req) 1791b20135d0STrond Myklebust { 17920c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 17930c493b5cSTrond Myklebust 17940c493b5cSTrond Myklebust filemap_dirty_folio(folio_mapping(folio), folio); 1795b20135d0STrond Myklebust } 1796b20135d0STrond Myklebust 17979ace33cdSFred Isaman /* 17981da177e4SLinus Torvalds * Commit dirty pages 17991da177e4SLinus Torvalds */ 18001da177e4SLinus Torvalds static int 1801ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1802ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 18031da177e4SLinus Torvalds { 18040b7c0153SFred Isaman struct nfs_commit_data *data; 180585e39feeSOlga Kornievskaia unsigned short task_flags = 0; 18061da177e4SLinus Torvalds 1807ade8febdSWeston Andros Adamson /* another commit raced with us */ 1808ade8febdSWeston Andros Adamson if (list_empty(head)) 1809ade8febdSWeston Andros Adamson return 0; 1810ade8febdSWeston Andros Adamson 1811515dcdcdSTrond Myklebust data = nfs_commitdata_alloc(); 1812515dcdcdSTrond Myklebust if (!data) { 1813515dcdcdSTrond Myklebust nfs_retry_commit(head, NULL, cinfo, -1); 1814515dcdcdSTrond Myklebust return -ENOMEM; 1815515dcdcdSTrond Myklebust } 18161da177e4SLinus Torvalds 18171da177e4SLinus Torvalds /* Set up the argument struct */ 1818f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 181985e39feeSOlga Kornievskaia if (NFS_SERVER(inode)->nfs_client->cl_minorversion) 182085e39feeSOlga Kornievskaia task_flags = RPC_TASK_MOVEABLE; 1821c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 182285e39feeSOlga Kornievskaia data->mds_ops, how, 182385e39feeSOlga Kornievskaia RPC_TASK_CRED_NOREF | task_flags); 18241da177e4SLinus Torvalds } 18251da177e4SLinus Torvalds 18261da177e4SLinus Torvalds /* 18271da177e4SLinus Torvalds * COMMIT call returned 18281da177e4SLinus Torvalds */ 1829788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 18301da177e4SLinus Torvalds { 18310b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18321da177e4SLinus Torvalds 1833788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1834c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 18357bdd297eSTrond Myklebust trace_nfs_commit_done(task, data); 1836c9d8f89dSTrond Myklebust } 1837c9d8f89dSTrond Myklebust 1838f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1839c9d8f89dSTrond Myklebust { 1840221203ceSTrond Myklebust const struct nfs_writeverf *verf = data->res.verf; 1841c9d8f89dSTrond Myklebust struct nfs_page *req; 1842c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1843f453a54aSFred Isaman struct nfs_commit_info cinfo; 1844353db796SNeilBrown struct nfs_server *nfss; 18450c493b5cSTrond Myklebust struct folio *folio; 1846788e7a89STrond Myklebust 18471da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18481da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18491da177e4SLinus Torvalds nfs_list_remove_request(req); 18500c493b5cSTrond Myklebust folio = nfs_page_to_folio(req); 18510c493b5cSTrond Myklebust nfs_folio_clear_commit(folio); 18521da177e4SLinus Torvalds 18531e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18549fcd5960STrond Myklebust nfs_req_openctx(req)->dentry->d_sb->s_id, 18559fcd5960STrond Myklebust (unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)), 18561da177e4SLinus Torvalds req->wb_bytes, 18571da177e4SLinus Torvalds (long long)req_offset(req)); 1858c9d8f89dSTrond Myklebust if (status < 0) { 18590c493b5cSTrond Myklebust if (folio) { 1860af887e43STrond Myklebust trace_nfs_commit_error(data->inode, req, 1861af887e43STrond Myklebust status); 18620c493b5cSTrond Myklebust nfs_mapping_set_error(folio, status); 18631da177e4SLinus Torvalds nfs_inode_remove_request(req); 18646fbda89bSTrond Myklebust } 1865ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18661da177e4SLinus Torvalds goto next; 18671da177e4SLinus Torvalds } 18681da177e4SLinus Torvalds 18691da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18701da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18711f28476dSTrond Myklebust if (nfs_write_match_verf(verf, req)) { 18721da177e4SLinus Torvalds /* We have a match */ 18730c493b5cSTrond Myklebust if (folio) 18741da177e4SLinus Torvalds nfs_inode_remove_request(req); 1875ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18761da177e4SLinus Torvalds goto next; 18771da177e4SLinus Torvalds } 18781da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1879ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18806d884e8fSFred nfs_mark_request_dirty(req); 188167f4b5dcSTrond Myklebust atomic_long_inc(&NFS_I(data->inode)->redirtied_pages); 18821da177e4SLinus Torvalds next: 18831d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18847f1bda44STrond Myklebust /* Latency breaker */ 18857f1bda44STrond Myklebust cond_resched(); 18861da177e4SLinus Torvalds } 1887353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1888353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18896df25e58SNeilBrown nfss->write_congested = 0; 1890353db796SNeilBrown 1891f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1892af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18935917ce84SFred Isaman } 18945917ce84SFred Isaman 18955917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18965917ce84SFred Isaman { 18970b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18985917ce84SFred Isaman 1899f453a54aSFred Isaman data->completion_ops->completion(data); 1900c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 19011da177e4SLinus Torvalds } 1902788e7a89STrond Myklebust 1903788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 19040b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1905788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1906788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1907788e7a89STrond Myklebust }; 19081da177e4SLinus Torvalds 1909f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1910f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1911b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1912f453a54aSFred Isaman }; 1913f453a54aSFred Isaman 19141763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1915ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 191684c53ab5SFred Isaman { 191784c53ab5SFred Isaman int status; 191884c53ab5SFred Isaman 1919ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 192084c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1921ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 192284c53ab5SFred Isaman return status; 192384c53ab5SFred Isaman } 192484c53ab5SFred Isaman 1925c4f24df9STrond Myklebust static int __nfs_commit_inode(struct inode *inode, int how, 1926c4f24df9STrond Myklebust struct writeback_control *wbc) 19271da177e4SLinus Torvalds { 19281da177e4SLinus Torvalds LIST_HEAD(head); 1929ea2cf228SFred Isaman struct nfs_commit_info cinfo; 193071d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1931c4f24df9STrond Myklebust int ret, nscan; 19321da177e4SLinus Torvalds 193364a93dbfSTrond Myklebust how &= ~FLUSH_SYNC; 1934ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1935af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1936c4f24df9STrond Myklebust for (;;) { 1937c4f24df9STrond Myklebust ret = nscan = nfs_scan_commit(inode, &head, &cinfo); 1938c4f24df9STrond Myklebust if (ret <= 0) 1939c4f24df9STrond Myklebust break; 1940c4f24df9STrond Myklebust ret = nfs_generic_commit_list(inode, &head, how, &cinfo); 1941c4f24df9STrond Myklebust if (ret < 0) 1942c4f24df9STrond Myklebust break; 1943c4f24df9STrond Myklebust ret = 0; 1944c4f24df9STrond Myklebust if (wbc && wbc->sync_mode == WB_SYNC_NONE) { 1945c4f24df9STrond Myklebust if (nscan < wbc->nr_to_write) 1946c4f24df9STrond Myklebust wbc->nr_to_write -= nscan; 1947c4f24df9STrond Myklebust else 1948c4f24df9STrond Myklebust wbc->nr_to_write = 0; 1949c4f24df9STrond Myklebust } 1950c4f24df9STrond Myklebust if (nscan < INT_MAX) 1951c4f24df9STrond Myklebust break; 1952c4f24df9STrond Myklebust cond_resched(); 1953c4f24df9STrond Myklebust } 1954af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 1955c4f24df9STrond Myklebust if (ret || !may_wait) 1956c4f24df9STrond Myklebust return ret; 1957c4f24df9STrond Myklebust return wait_on_commit(cinfo.mds); 1958c4f24df9STrond Myklebust } 1959c4f24df9STrond Myklebust 1960c4f24df9STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 1961c4f24df9STrond Myklebust { 1962c4f24df9STrond Myklebust return __nfs_commit_inode(inode, how, NULL); 19631da177e4SLinus Torvalds } 1964b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19658fc795f7STrond Myklebust 1966ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19678fc795f7STrond Myklebust { 1968420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1969420e3646STrond Myklebust int flags = FLUSH_SYNC; 1970420e3646STrond Myklebust int ret = 0; 19718fc795f7STrond Myklebust 1972c4f24df9STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 19733236c3e1SJeff Layton /* no commits means nothing needs to be done */ 19745cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 1975c4f24df9STrond Myklebust goto check_requests_outstanding; 19763236c3e1SJeff Layton 1977a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1978a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1979420e3646STrond Myklebust */ 19801a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1981420e3646STrond Myklebust goto out_mark_dirty; 1982420e3646STrond Myklebust 1983a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1984420e3646STrond Myklebust flags = 0; 1985a00dd6c0SJeff Layton } 1986a00dd6c0SJeff Layton 1987c4f24df9STrond Myklebust ret = __nfs_commit_inode(inode, flags, wbc); 1988c4f24df9STrond Myklebust if (!ret) { 1989c4f24df9STrond Myklebust if (flags & FLUSH_SYNC) 19908fc795f7STrond Myklebust return 0; 1991c4f24df9STrond Myklebust } else if (atomic_long_read(&nfsi->commit_info.ncommit)) 1992c4f24df9STrond Myklebust goto out_mark_dirty; 1993c4f24df9STrond Myklebust 1994c4f24df9STrond Myklebust check_requests_outstanding: 1995c4f24df9STrond Myklebust if (!atomic_read(&nfsi->commit_info.rpcs_out)) 1996c4f24df9STrond Myklebust return ret; 1997420e3646STrond Myklebust out_mark_dirty: 19988fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19998fc795f7STrond Myklebust return ret; 20008fc795f7STrond Myklebust } 200189d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 2002863a3c6cSAndy Adamson 2003acdc53b2STrond Myklebust /* 2004837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 2005837bb1d7STrond Myklebust * 2006837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 2007837bb1d7STrond Myklebust * client. 2008837bb1d7STrond Myklebust */ 2009837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 2010837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 2011837bb1d7STrond Myklebust { 2012837bb1d7STrond Myklebust int ret; 2013837bb1d7STrond Myklebust 2014837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 2015837bb1d7STrond Myklebust if (ret == 0) 2016837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 2017837bb1d7STrond Myklebust return ret; 2018837bb1d7STrond Myklebust } 2019837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 2020837bb1d7STrond Myklebust 2021837bb1d7STrond Myklebust /* 2022acdc53b2STrond Myklebust * flush the inode to disk. 2023acdc53b2STrond Myklebust */ 2024acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 202534901f70STrond Myklebust { 2026f4ce1299STrond Myklebust int ret; 202734901f70STrond Myklebust 2028f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 2029f4ce1299STrond Myklebust 20305bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 20316b196875SChuck Lever if (ret) 20326b196875SChuck Lever goto out; 20335bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20346b196875SChuck Lever if (ret < 0) 20356b196875SChuck Lever goto out; 20365bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 20376b196875SChuck Lever ret = 0; 2038f4ce1299STrond Myklebust 20396b196875SChuck Lever out: 2040f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 2041f4ce1299STrond Myklebust return ret; 20421c75950bSTrond Myklebust } 2043ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 20441c75950bSTrond Myklebust 20456d740c76SMatthew Wilcox (Oracle) int nfs_wb_folio_cancel(struct inode *inode, struct folio *folio) 20461b3b4a1aSTrond Myklebust { 20471b3b4a1aSTrond Myklebust struct nfs_page *req; 20481b3b4a1aSTrond Myklebust int ret = 0; 20491b3b4a1aSTrond Myklebust 20506d740c76SMatthew Wilcox (Oracle) folio_wait_writeback(folio); 20513e217045SWeston Andros Adamson 20523e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20533e217045SWeston Andros Adamson * request */ 20540c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 20553e217045SWeston Andros Adamson 20563e217045SWeston Andros Adamson if (IS_ERR(req)) { 20573e217045SWeston Andros Adamson ret = PTR_ERR(req); 20583e217045SWeston Andros Adamson } else if (req) { 20596d740c76SMatthew Wilcox (Oracle) /* all requests from this folio have been cancelled by 20603e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20613e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20623e217045SWeston Andros Adamson * release it */ 20631b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20641d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20651b3b4a1aSTrond Myklebust } 20663e217045SWeston Andros Adamson 20671b3b4a1aSTrond Myklebust return ret; 20681b3b4a1aSTrond Myklebust } 20691b3b4a1aSTrond Myklebust 20705241060eSTrond Myklebust /** 20715241060eSTrond Myklebust * nfs_wb_folio - Write back all requests on one page 20725241060eSTrond Myklebust * @inode: pointer to page 20735241060eSTrond Myklebust * @folio: pointer to folio 20745241060eSTrond Myklebust * 20755241060eSTrond Myklebust * Assumes that the folio has been locked by the caller, and will 20765241060eSTrond Myklebust * not unlock it. 20771c75950bSTrond Myklebust */ 20785241060eSTrond Myklebust int nfs_wb_folio(struct inode *inode, struct folio *folio) 20791c75950bSTrond Myklebust { 20805241060eSTrond Myklebust loff_t range_start = folio_file_pos(folio); 20815241060eSTrond Myklebust loff_t range_end = range_start + (loff_t)folio_size(folio) - 1; 20827f2f12d9STrond Myklebust struct writeback_control wbc = { 20837f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20847f2f12d9STrond Myklebust .nr_to_write = 0, 20857f2f12d9STrond Myklebust .range_start = range_start, 20867f2f12d9STrond Myklebust .range_end = range_end, 20877f2f12d9STrond Myklebust }; 20887f2f12d9STrond Myklebust int ret; 20897f2f12d9STrond Myklebust 2090256093feSTrond Myklebust trace_nfs_writeback_folio(inode, folio); 2091f4ce1299STrond Myklebust 20920522f6adSTrond Myklebust for (;;) { 20935241060eSTrond Myklebust folio_wait_writeback(folio); 20945241060eSTrond Myklebust if (folio_clear_dirty_for_io(folio)) { 20950c493b5cSTrond Myklebust ret = nfs_writepage_locked(folio, &wbc); 20967f2f12d9STrond Myklebust if (ret < 0) 20977f2f12d9STrond Myklebust goto out_error; 20980522f6adSTrond Myklebust continue; 20997f2f12d9STrond Myklebust } 2100f4ce1299STrond Myklebust ret = 0; 21015241060eSTrond Myklebust if (!folio_test_private(folio)) 21020522f6adSTrond Myklebust break; 21030522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 21047f2f12d9STrond Myklebust if (ret < 0) 21057f2f12d9STrond Myklebust goto out_error; 21067f2f12d9STrond Myklebust } 21077f2f12d9STrond Myklebust out_error: 2108256093feSTrond Myklebust trace_nfs_writeback_folio_done(inode, folio, ret); 21097f2f12d9STrond Myklebust return ret; 21101c75950bSTrond Myklebust } 21111c75950bSTrond Myklebust 2112074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 21134ae84a80SMatthew Wilcox (Oracle) int nfs_migrate_folio(struct address_space *mapping, struct folio *dst, 21144ae84a80SMatthew Wilcox (Oracle) struct folio *src, enum migrate_mode mode) 2115074cc1deSTrond Myklebust { 21162da95652SJeff Layton /* 21174ae84a80SMatthew Wilcox (Oracle) * If the private flag is set, the folio is currently associated with 21182da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 21192da95652SJeff Layton * 21202da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 21212da95652SJeff Layton * that we can safely release the inode reference while holding 21224ae84a80SMatthew Wilcox (Oracle) * the folio lock. 21232da95652SJeff Layton */ 21244ae84a80SMatthew Wilcox (Oracle) if (folio_test_private(src)) 21252da95652SJeff Layton return -EBUSY; 2126074cc1deSTrond Myklebust 21274ae84a80SMatthew Wilcox (Oracle) if (folio_test_fscache(src)) { 212816f2f4e6SDavid Howells if (mode == MIGRATE_ASYNC) 21298c209ce7SDavid Howells return -EBUSY; 21304ae84a80SMatthew Wilcox (Oracle) folio_wait_fscache(src); 213116f2f4e6SDavid Howells } 2132074cc1deSTrond Myklebust 213354184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 2134074cc1deSTrond Myklebust } 2135074cc1deSTrond Myklebust #endif 2136074cc1deSTrond Myklebust 2137f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 21381da177e4SLinus Torvalds { 21391da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 21401e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 21411da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 214220c2df83SPaul Mundt NULL); 21431da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 21441da177e4SLinus Torvalds return -ENOMEM; 21451da177e4SLinus Torvalds 214693d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 21471da177e4SLinus Torvalds nfs_wdata_cachep); 21481da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 21493dd4765fSJeff Layton goto out_destroy_write_cache; 21501da177e4SLinus Torvalds 21510b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 21520b7c0153SFred Isaman sizeof(struct nfs_commit_data), 21530b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21540b7c0153SFred Isaman NULL); 21550b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21563dd4765fSJeff Layton goto out_destroy_write_mempool; 21570b7c0153SFred Isaman 215893d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21594c100210SYanchuan Nian nfs_cdata_cachep); 21601da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21613dd4765fSJeff Layton goto out_destroy_commit_cache; 21621da177e4SLinus Torvalds 216389a09141SPeter Zijlstra /* 216489a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 216589a09141SPeter Zijlstra * 216689a09141SPeter Zijlstra * 64MB: 8192k 216789a09141SPeter Zijlstra * 128MB: 11585k 216889a09141SPeter Zijlstra * 256MB: 16384k 216989a09141SPeter Zijlstra * 512MB: 23170k 217089a09141SPeter Zijlstra * 1GB: 32768k 217189a09141SPeter Zijlstra * 2GB: 46340k 217289a09141SPeter Zijlstra * 4GB: 65536k 217389a09141SPeter Zijlstra * 8GB: 92681k 217489a09141SPeter Zijlstra * 16GB: 131072k 217589a09141SPeter Zijlstra * 217689a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 217789a09141SPeter Zijlstra * Limit the default to 256M 217889a09141SPeter Zijlstra */ 2179ca79b0c2SArun KS nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); 218089a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 218189a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 218289a09141SPeter Zijlstra 21831da177e4SLinus Torvalds return 0; 21843dd4765fSJeff Layton 21853dd4765fSJeff Layton out_destroy_commit_cache: 21863dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21873dd4765fSJeff Layton out_destroy_write_mempool: 21883dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21893dd4765fSJeff Layton out_destroy_write_cache: 21903dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21913dd4765fSJeff Layton return -ENOMEM; 21921da177e4SLinus Torvalds } 21931da177e4SLinus Torvalds 2194266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21951da177e4SLinus Torvalds { 21961da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21973dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21981da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21991a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 22001da177e4SLinus Torvalds } 22011da177e4SLinus Torvalds 22024a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 22034a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 22044a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 22050eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 22060eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 22071ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 22084a0de55cSAnna Schumaker }; 2209