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: 292*e12912d9STrond Myklebust /* Atomically update timestamps if they are delegated to us. */ 293*e12912d9STrond Myklebust nfs_update_delegated_mtime_locked(inode); 294a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 295a6b5a28eSDave Wysochanski nfs_fscache_invalidate(inode, 0); 2961da177e4SLinus Torvalds } 2971da177e4SLinus Torvalds 298a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 299d2ceb7e5SBenjamin Coddington static void nfs_set_pageerror(struct address_space *mapping) 300a301b777STrond Myklebust { 3010df68cedSTrond Myklebust struct inode *inode = mapping->host; 3020df68cedSTrond Myklebust 303d2ceb7e5SBenjamin Coddington nfs_zap_mapping(mapping->host, mapping); 3040df68cedSTrond Myklebust /* Force file size revalidation */ 3050df68cedSTrond Myklebust spin_lock(&inode->i_lock); 306ac46b3d7STrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_REVAL_FORCED | 30788a6099fSTrond Myklebust NFS_INO_INVALID_CHANGE | 308ac46b3d7STrond Myklebust NFS_INO_INVALID_SIZE); 3090df68cedSTrond Myklebust spin_unlock(&inode->i_lock); 310a301b777STrond Myklebust } 311a301b777STrond Myklebust 3120c493b5cSTrond Myklebust static void nfs_mapping_set_error(struct folio *folio, int error) 3136fbda89bSTrond Myklebust { 3140c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 315b8946d7bSTrond Myklebust 3160c493b5cSTrond Myklebust folio_set_error(folio); 3176c984083STrond Myklebust filemap_set_wb_err(mapping, error); 3186c984083STrond Myklebust if (mapping->host) 3196c984083STrond Myklebust errseq_set(&mapping->host->i_sb->s_wb_err, 3206c984083STrond Myklebust error == -ENOSPC ? -ENOSPC : -EIO); 321b8946d7bSTrond Myklebust nfs_set_pageerror(mapping); 3226fbda89bSTrond Myklebust } 3236fbda89bSTrond Myklebust 324d72ddcbaSWeston Andros Adamson /* 325d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 326d72ddcbaSWeston Andros Adamson * @head - head request of page group 327d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 328d72ddcbaSWeston Andros Adamson * 329d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 330d72ddcbaSWeston Andros Adamson * page offset @page_offset. 331d72ddcbaSWeston Andros Adamson * 332d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 333d72ddcbaSWeston Andros Adamson * match is found. 334d72ddcbaSWeston Andros Adamson * 335d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 336d72ddcbaSWeston Andros Adamson */ 337d72ddcbaSWeston Andros Adamson static struct nfs_page * 338d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 339d72ddcbaSWeston Andros Adamson { 340d72ddcbaSWeston Andros Adamson struct nfs_page *req; 341d72ddcbaSWeston Andros Adamson 342d72ddcbaSWeston Andros Adamson req = head; 343d72ddcbaSWeston Andros Adamson do { 344d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 345d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 346d72ddcbaSWeston Andros Adamson return req; 347d72ddcbaSWeston Andros Adamson 348d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 349d72ddcbaSWeston Andros Adamson } while (req != head); 350d72ddcbaSWeston Andros Adamson 351d72ddcbaSWeston Andros Adamson return NULL; 352d72ddcbaSWeston Andros Adamson } 353d72ddcbaSWeston Andros Adamson 354d72ddcbaSWeston Andros Adamson /* 355d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 356d72ddcbaSWeston Andros Adamson * @head - head request of page group 357d72ddcbaSWeston Andros Adamson * 358d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 359d72ddcbaSWeston Andros Adamson * returns false otherwise 360d72ddcbaSWeston Andros Adamson */ 361d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 362d72ddcbaSWeston Andros Adamson { 3630c493b5cSTrond Myklebust unsigned int len = nfs_folio_length(nfs_page_to_folio(req)); 364d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 365d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 366d72ddcbaSWeston Andros Adamson 3671344b7eaSTrond Myklebust nfs_page_group_lock(req); 368d72ddcbaSWeston Andros Adamson 3697e8a30f8STrond Myklebust for (;;) { 370d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 3717e8a30f8STrond Myklebust if (!tmp) 3727e8a30f8STrond Myklebust break; 3737e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 374d72ddcbaSWeston Andros Adamson } 375d72ddcbaSWeston Andros Adamson 376d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3777e8a30f8STrond Myklebust return pos >= len; 378d72ddcbaSWeston Andros Adamson } 379d72ddcbaSWeston Andros Adamson 3801da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3811da177e4SLinus Torvalds * covers the full page. 3821da177e4SLinus Torvalds */ 383d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3841da177e4SLinus Torvalds { 3850c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 3860c493b5cSTrond Myklebust 3870c493b5cSTrond Myklebust if (folio_test_uptodate(folio)) 3881da177e4SLinus Torvalds return; 389d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3901da177e4SLinus Torvalds return; 3910c493b5cSTrond Myklebust folio_mark_uptodate(folio); 3921da177e4SLinus Torvalds } 3931da177e4SLinus Torvalds 3941da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3951da177e4SLinus Torvalds { 396e87b4c7aSNeilBrown int ret = 0; 397cca588d6STrond Myklebust 398e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 399e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 400e87b4c7aSNeilBrown return ret; 4011da177e4SLinus Torvalds } 4021da177e4SLinus Torvalds 4031da177e4SLinus Torvalds /* 40489a09141SPeter Zijlstra * NFS congestion control 40589a09141SPeter Zijlstra */ 40689a09141SPeter Zijlstra 40789a09141SPeter Zijlstra int nfs_congestion_kb; 40889a09141SPeter Zijlstra 40989a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 41089a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 41189a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 41289a09141SPeter Zijlstra 4130c493b5cSTrond Myklebust static void nfs_folio_set_writeback(struct folio *folio) 41489a09141SPeter Zijlstra { 4150c493b5cSTrond Myklebust struct nfs_server *nfss = NFS_SERVER(folio_file_mapping(folio)->host); 4165a6d41b3STrond Myklebust 4170c493b5cSTrond Myklebust folio_start_writeback(folio); 4180c493b5cSTrond Myklebust if (atomic_long_inc_return(&nfss->writeback) > NFS_CONGESTION_ON_THRESH) 4196df25e58SNeilBrown nfss->write_congested = 1; 42089a09141SPeter Zijlstra } 42189a09141SPeter Zijlstra 4220c493b5cSTrond Myklebust static void nfs_folio_end_writeback(struct folio *folio) 42389a09141SPeter Zijlstra { 4240c493b5cSTrond Myklebust struct nfs_server *nfss = NFS_SERVER(folio_file_mapping(folio)->host); 42589a09141SPeter Zijlstra 4260c493b5cSTrond Myklebust folio_end_writeback(folio); 4270c493b5cSTrond Myklebust if (atomic_long_dec_return(&nfss->writeback) < 4280c493b5cSTrond Myklebust NFS_CONGESTION_OFF_THRESH) 4296df25e58SNeilBrown nfss->write_congested = 0; 43089a09141SPeter Zijlstra } 43189a09141SPeter Zijlstra 4320c493b5cSTrond Myklebust static void nfs_page_end_writeback(struct nfs_page *req) 4330c493b5cSTrond Myklebust { 4340c493b5cSTrond Myklebust if (nfs_page_group_sync_on_bit(req, PG_WB_END)) { 4350c493b5cSTrond Myklebust nfs_unlock_request(req); 4360c493b5cSTrond Myklebust nfs_folio_end_writeback(nfs_page_to_folio(req)); 4370c493b5cSTrond Myklebust } else 4380c493b5cSTrond Myklebust nfs_unlock_request(req); 4390c493b5cSTrond Myklebust } 4400c493b5cSTrond Myklebust 441d4581383SWeston Andros Adamson /* 442d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 443d4581383SWeston Andros Adamson * 444d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 445d4581383SWeston Andros Adamson * @old_head - the old head of the list 446d4581383SWeston Andros Adamson * 447d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 448d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 449d4581383SWeston Andros Adamson * with a reference held by some other context. 450d4581383SWeston Andros Adamson */ 451d4581383SWeston Andros Adamson static void 452d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 453b66aaa8dSTrond Myklebust struct nfs_page *old_head, 454b66aaa8dSTrond Myklebust struct inode *inode) 455d4581383SWeston Andros Adamson { 456d4581383SWeston Andros Adamson while (destroy_list) { 457d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 458d4581383SWeston Andros Adamson 459d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 460d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 461d4581383SWeston Andros Adamson 46208ca8b21STrond Myklebust /* Note: lock subreq in order to change subreq->wb_head */ 46308ca8b21STrond Myklebust nfs_page_set_headlock(subreq); 464d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 465d4581383SWeston Andros Adamson 466d4581383SWeston Andros Adamson /* make sure old group is not used */ 467d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 46808ca8b21STrond Myklebust subreq->wb_head = subreq; 469d4581383SWeston Andros Adamson 470902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 471902a4c00STrond Myklebust 4725b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4735b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4745b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 47508ca8b21STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) { 47608ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 4775b2b5187STrond Myklebust nfs_free_request(subreq); 47808ca8b21STrond Myklebust } else 47908ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 4805b2b5187STrond Myklebust continue; 4815b2b5187STrond Myklebust } 48208ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 483d4581383SWeston Andros Adamson 484add42de3STrond Myklebust nfs_release_request(old_head); 4855b2b5187STrond Myklebust 486b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 487d4581383SWeston Andros Adamson nfs_release_request(subreq); 488a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 489d4581383SWeston Andros Adamson } 4905b2b5187STrond Myklebust 4915b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4925b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4935b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 494d4581383SWeston Andros Adamson } 495d4581383SWeston Andros Adamson } 496d4581383SWeston Andros Adamson 497d4581383SWeston Andros Adamson /* 498e00ed89dSTrond Myklebust * nfs_join_page_group - destroy subrequests of the head req 499e00ed89dSTrond Myklebust * @head: the page used to lookup the "page group" of nfs_page structures 500e00ed89dSTrond Myklebust * @inode: Inode to which the request belongs. 501d4581383SWeston Andros Adamson * 502d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 503d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 504d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 505d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 506d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 507d4581383SWeston Andros Adamson */ 508b193a78dSTrond Myklebust void nfs_join_page_group(struct nfs_page *head, struct nfs_commit_info *cinfo, 509b193a78dSTrond Myklebust struct inode *inode) 510d4581383SWeston Andros Adamson { 511e00ed89dSTrond Myklebust struct nfs_page *subreq; 512d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 513a62f8e3bSTrond Myklebust unsigned int pgbase, off, bytes; 514a62f8e3bSTrond Myklebust 515a62f8e3bSTrond Myklebust pgbase = head->wb_pgbase; 516a62f8e3bSTrond Myklebust bytes = head->wb_bytes; 517a62f8e3bSTrond Myklebust off = head->wb_offset; 518a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 519a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 520a62f8e3bSTrond Myklebust /* Subrequests should always form a contiguous range */ 521a62f8e3bSTrond Myklebust if (pgbase > subreq->wb_pgbase) { 522a62f8e3bSTrond Myklebust off -= pgbase - subreq->wb_pgbase; 523a62f8e3bSTrond Myklebust bytes += pgbase - subreq->wb_pgbase; 524a62f8e3bSTrond Myklebust pgbase = subreq->wb_pgbase; 525a62f8e3bSTrond Myklebust } 526a62f8e3bSTrond Myklebust bytes = max(subreq->wb_pgbase + subreq->wb_bytes 527a62f8e3bSTrond Myklebust - pgbase, bytes); 5281bd5d6d0STrond Myklebust } 5291bd5d6d0STrond Myklebust 530a62f8e3bSTrond Myklebust /* Set the head request's range to cover the former page group */ 531a62f8e3bSTrond Myklebust head->wb_pgbase = pgbase; 532a62f8e3bSTrond Myklebust head->wb_bytes = bytes; 533a62f8e3bSTrond Myklebust head->wb_offset = off; 534d4581383SWeston Andros Adamson 535d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 536d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 537d4581383SWeston Andros Adamson subreq = head; 538d4581383SWeston Andros Adamson do { 539b193a78dSTrond Myklebust nfs_clear_request_commit(cinfo, subreq); 540d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 541d4581383SWeston Andros Adamson } while (subreq != head); 542d4581383SWeston Andros Adamson 543d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 544d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 545d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 546d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 547d4581383SWeston Andros Adamson head->wb_this_page = head; 548d4581383SWeston Andros Adamson } 549d4581383SWeston Andros Adamson 550b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 551b5bab9bfSTrond Myklebust } 552b5bab9bfSTrond Myklebust 553e00ed89dSTrond Myklebust /* 554e00ed89dSTrond Myklebust * nfs_lock_and_join_requests - join all subreqs to the head req 5550c493b5cSTrond Myklebust * @folio: the folio used to lookup the "page group" of nfs_page structures 556e00ed89dSTrond Myklebust * 557e00ed89dSTrond Myklebust * This function joins all sub requests to the head request by first 558e00ed89dSTrond Myklebust * locking all requests in the group, cancelling any pending operations 559e00ed89dSTrond Myklebust * and finally updating the head request to cover the whole range covered by 560e00ed89dSTrond Myklebust * the (former) group. All subrequests are removed from any write or commit 561e00ed89dSTrond Myklebust * lists, unlinked from the group and destroyed. 562e00ed89dSTrond Myklebust * 563e00ed89dSTrond Myklebust * Returns a locked, referenced pointer to the head request - which after 564e00ed89dSTrond Myklebust * this call is guaranteed to be the only request associated with the page. 5650c493b5cSTrond Myklebust * Returns NULL if no requests are found for @folio, or a ERR_PTR if an 566e00ed89dSTrond Myklebust * error was encountered. 567e00ed89dSTrond Myklebust */ 5680c493b5cSTrond Myklebust static struct nfs_page *nfs_lock_and_join_requests(struct folio *folio) 569e00ed89dSTrond Myklebust { 5700c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 571e00ed89dSTrond Myklebust struct nfs_page *head; 572b193a78dSTrond Myklebust struct nfs_commit_info cinfo; 573e00ed89dSTrond Myklebust int ret; 574e00ed89dSTrond Myklebust 575b193a78dSTrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 576e00ed89dSTrond Myklebust /* 577e00ed89dSTrond Myklebust * A reference is taken only on the head request which acts as a 578e00ed89dSTrond Myklebust * reference to the whole page group - the group will not be destroyed 579e00ed89dSTrond Myklebust * until the head reference is released. 580e00ed89dSTrond Myklebust */ 5810c493b5cSTrond Myklebust head = nfs_folio_find_and_lock_request(folio); 582e00ed89dSTrond Myklebust if (IS_ERR_OR_NULL(head)) 583d4581383SWeston Andros Adamson return head; 5840671d8f1SMarkus Elfring 585e00ed89dSTrond Myklebust /* lock each request in the page group */ 586e00ed89dSTrond Myklebust ret = nfs_page_group_lock_subrequests(head); 587e00ed89dSTrond Myklebust if (ret < 0) { 5880671d8f1SMarkus Elfring nfs_unlock_and_release_request(head); 5890671d8f1SMarkus Elfring return ERR_PTR(ret); 590612c9384STrond Myklebust } 591074cc1deSTrond Myklebust 592b193a78dSTrond Myklebust nfs_join_page_group(head, &cinfo, inode); 593e00ed89dSTrond Myklebust 594e00ed89dSTrond Myklebust return head; 595e00ed89dSTrond Myklebust } 596e00ed89dSTrond Myklebust 5976fbda89bSTrond Myklebust static void nfs_write_error(struct nfs_page *req, int error) 5980bcbf039SPeng Tao { 5996dd85e83STrond Myklebust trace_nfs_write_error(nfs_page_to_inode(req), req, error); 6000c493b5cSTrond Myklebust nfs_mapping_set_error(nfs_page_to_folio(req), error); 60106c9fdf3STrond Myklebust nfs_inode_remove_request(req); 6020c493b5cSTrond Myklebust nfs_page_end_writeback(req); 6031f84ccdfSFred Isaman nfs_release_request(req); 6040bcbf039SPeng Tao } 6050bcbf039SPeng Tao 606074cc1deSTrond Myklebust /* 607074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 608074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 609074cc1deSTrond Myklebust */ 6100c493b5cSTrond Myklebust static int nfs_page_async_flush(struct folio *folio, 611c6fd3511STrond Myklebust struct writeback_control *wbc, 612c6fd3511STrond Myklebust struct nfs_pageio_descriptor *pgio) 613074cc1deSTrond Myklebust { 614074cc1deSTrond Myklebust struct nfs_page *req; 615074cc1deSTrond Myklebust int ret = 0; 616074cc1deSTrond Myklebust 6170c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 618074cc1deSTrond Myklebust if (!req) 619074cc1deSTrond Myklebust goto out; 620074cc1deSTrond Myklebust ret = PTR_ERR(req); 621074cc1deSTrond Myklebust if (IS_ERR(req)) 622074cc1deSTrond Myklebust goto out; 623074cc1deSTrond Myklebust 6240c493b5cSTrond Myklebust nfs_folio_set_writeback(folio); 625deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 626074cc1deSTrond Myklebust 627a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 62896c41455STrond Myklebust ret = pgio->pg_error; 62996c41455STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 630a6598813STrond Myklebust goto out_launder; 631a6598813STrond Myklebust 63296c41455STrond Myklebust ret = 0; 633f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 634074cc1deSTrond Myklebust ret = pgio->pg_error; 6350bcbf039SPeng Tao /* 636c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6370bcbf039SPeng Tao */ 638c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 639a6598813STrond Myklebust goto out_launder; 640c6fd3511STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) 641c6fd3511STrond Myklebust ret = AOP_WRITEPAGE_ACTIVATE; 6420c493b5cSTrond Myklebust folio_redirty_for_writepage(wbc, folio); 6438fc75bedSTrond Myklebust nfs_redirty_request(req); 64496c41455STrond Myklebust pgio->pg_error = 0; 64540f90271STrond Myklebust } else 6460c493b5cSTrond Myklebust nfs_add_stats(folio_file_mapping(folio)->host, 64740f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 648074cc1deSTrond Myklebust out: 649074cc1deSTrond Myklebust return ret; 650a6598813STrond Myklebust out_launder: 6516fbda89bSTrond Myklebust nfs_write_error(req, ret); 65214bebe3cSTrond Myklebust return 0; 653e261f51fSTrond Myklebust } 654e261f51fSTrond Myklebust 6550c493b5cSTrond Myklebust static int nfs_do_writepage(struct folio *folio, struct writeback_control *wbc, 656c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 657f758c885STrond Myklebust { 6580c493b5cSTrond Myklebust nfs_pageio_cond_complete(pgio, folio_index(folio)); 6590c493b5cSTrond Myklebust return nfs_page_async_flush(folio, wbc, pgio); 660f758c885STrond Myklebust } 661f758c885STrond Myklebust 662e261f51fSTrond Myklebust /* 6631da177e4SLinus Torvalds * Write an mmapped page to the server. 6641da177e4SLinus Torvalds */ 6650c493b5cSTrond Myklebust static int nfs_writepage_locked(struct folio *folio, 666c373fff7STrond Myklebust struct writeback_control *wbc) 6671da177e4SLinus Torvalds { 668f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 6690c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 670e261f51fSTrond Myklebust int err; 6711da177e4SLinus Torvalds 67240f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 6730c493b5cSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, false, 6740c493b5cSTrond Myklebust &nfs_async_write_completion_ops); 6750c493b5cSTrond Myklebust err = nfs_do_writepage(folio, wbc, &pgio); 67696c41455STrond Myklebust pgio.pg_error = 0; 677f758c885STrond Myklebust nfs_pageio_complete(&pgio); 6784d770ccfSTrond Myklebust return err; 6794d770ccfSTrond Myklebust } 6804d770ccfSTrond Myklebust 681d585bdbeSMatthew Wilcox (Oracle) static int nfs_writepages_callback(struct folio *folio, 6820c493b5cSTrond Myklebust struct writeback_control *wbc, void *data) 683f758c885STrond Myklebust { 684f758c885STrond Myklebust int ret; 685f758c885STrond Myklebust 6860c493b5cSTrond Myklebust ret = nfs_do_writepage(folio, wbc, data); 68796c41455STrond Myklebust if (ret != AOP_WRITEPAGE_ACTIVATE) 688d585bdbeSMatthew Wilcox (Oracle) folio_unlock(folio); 689f758c885STrond Myklebust return ret; 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds 692919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 693919e3bd9STrond Myklebust { 694919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 695919e3bd9STrond Myklebust } 696919e3bd9STrond Myklebust 6971da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6981da177e4SLinus Torvalds { 6991da177e4SLinus Torvalds struct inode *inode = mapping->host; 700c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 701ed7bcdb3STrond Myklebust struct nfs_io_completion *ioc = NULL; 702ed7bcdb3STrond Myklebust unsigned int mntflags = NFS_SERVER(inode)->flags; 703ed7bcdb3STrond Myklebust int priority = 0; 7041da177e4SLinus Torvalds int err; 7051da177e4SLinus Torvalds 7066df25e58SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 7076df25e58SNeilBrown NFS_SERVER(inode)->write_congested) 7086df25e58SNeilBrown return 0; 7096df25e58SNeilBrown 71091d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 71191d5b470SChuck Lever 712ed7bcdb3STrond Myklebust if (!(mntflags & NFS_MOUNT_WRITE_EAGER) || wbc->for_kupdate || 713ed7bcdb3STrond Myklebust wbc->for_background || wbc->for_sync || wbc->for_reclaim) { 7142b17d725STrond Myklebust ioc = nfs_io_completion_alloc(GFP_KERNEL); 715919e3bd9STrond Myklebust if (ioc) 716ed7bcdb3STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, 717ed7bcdb3STrond Myklebust inode); 718ed7bcdb3STrond Myklebust priority = wb_priority(wbc); 719ed7bcdb3STrond Myklebust } 720919e3bd9STrond Myklebust 721c6fd3511STrond Myklebust do { 722ed7bcdb3STrond Myklebust nfs_pageio_init_write(&pgio, inode, priority, false, 723a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 724919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 725c6fd3511STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, 726c6fd3511STrond Myklebust &pgio); 72796c41455STrond Myklebust pgio.pg_error = 0; 728c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 7296e7434abSTrond Myklebust if (err == -EAGAIN && mntflags & NFS_MOUNT_SOFTERR) 7306e7434abSTrond Myklebust break; 731c6fd3511STrond Myklebust } while (err < 0 && !nfs_error_is_fatal(err)); 732919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 73372cb77f4STrond Myklebust 734f758c885STrond Myklebust if (err < 0) 73572cb77f4STrond Myklebust goto out_err; 736c63c7b05STrond Myklebust return 0; 73772cb77f4STrond Myklebust out_err: 73872cb77f4STrond Myklebust return err; 7391da177e4SLinus Torvalds } 7401da177e4SLinus Torvalds 7411da177e4SLinus Torvalds /* 7421da177e4SLinus Torvalds * Insert a write request into an inode 7431da177e4SLinus Torvalds */ 7440c493b5cSTrond Myklebust static void nfs_inode_add_request(struct nfs_page *req) 7451da177e4SLinus Torvalds { 7460c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 7470c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 7480c493b5cSTrond Myklebust struct nfs_inode *nfsi = NFS_I(mapping->host); 749e7d39069STrond Myklebust 7502bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7512bfc6e56SWeston Andros Adamson 752e7d39069STrond Myklebust /* Lock the request! */ 7537ad84aa9STrond Myklebust nfs_lock_request(req); 754e7d39069STrond Myklebust 75529418aa4SMel Gorman /* 75629418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 75729418aa4SMel Gorman * with invalidate/truncate. 75829418aa4SMel Gorman */ 759600f111eSMatthew Wilcox (Oracle) spin_lock(&mapping->i_private_lock); 7600c493b5cSTrond Myklebust if (likely(!folio_test_swapcache(folio))) { 7612df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 7620c493b5cSTrond Myklebust folio_set_private(folio); 7630c493b5cSTrond Myklebust folio->private = req; 76429418aa4SMel Gorman } 765600f111eSMatthew Wilcox (Oracle) spin_unlock(&mapping->i_private_lock); 766a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 76717089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 768cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 769cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 770cb1410c7SWeston Andros Adamson * adding subrequests. */ 77117089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 772c03b4024STrond Myklebust kref_get(&req->wb_kref); 7731da177e4SLinus Torvalds } 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds /* 77689a09141SPeter Zijlstra * Remove a write request from an inode 7771da177e4SLinus Torvalds */ 7781da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7791da177e4SLinus Torvalds { 7806a6d4644SScott Mayhew struct nfs_inode *nfsi = NFS_I(nfs_page_to_inode(req)); 7816a6d4644SScott Mayhew 78220633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 7830c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req->wb_head); 7840c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 7851da177e4SLinus Torvalds 786600f111eSMatthew Wilcox (Oracle) spin_lock(&mapping->i_private_lock); 7870c493b5cSTrond Myklebust if (likely(folio && !folio_test_swapcache(folio))) { 7880c493b5cSTrond Myklebust folio->private = NULL; 7890c493b5cSTrond Myklebust folio_clear_private(folio); 7900c493b5cSTrond Myklebust clear_bit(PG_MAPPED, &req->wb_head->wb_flags); 79129418aa4SMel Gorman } 792600f111eSMatthew Wilcox (Oracle) spin_unlock(&mapping->i_private_lock); 79320633f04SWeston Andros Adamson } 79417089a29SWeston Andros Adamson 79533ea5aaaSZhangXiaoxu if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) { 7966a6d4644SScott Mayhew atomic_long_dec(&nfsi->nrequests); 797dd1b2026SJeff Layton nfs_release_request(req); 79833ea5aaaSZhangXiaoxu } 7991da177e4SLinus Torvalds } 8001da177e4SLinus Torvalds 8010c493b5cSTrond Myklebust static void nfs_mark_request_dirty(struct nfs_page *req) 80261822ab5STrond Myklebust { 8030c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 8040c493b5cSTrond Myklebust if (folio) 8050c493b5cSTrond Myklebust filemap_dirty_folio(folio_mapping(folio), folio); 80661822ab5STrond Myklebust } 80761822ab5STrond Myklebust 8083a3908c8STrond Myklebust /* 8093a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8103a3908c8STrond Myklebust * 8110c493b5cSTrond Myklebust * Search through commit lists on @inode for the head request for @folio. 8123a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8133a3908c8STrond Myklebust * 8143a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8153a3908c8STrond Myklebust */ 8163a3908c8STrond Myklebust static struct nfs_page * 8173a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8180c493b5cSTrond Myklebust struct folio *folio) 8193a3908c8STrond Myklebust { 8203a3908c8STrond Myklebust struct nfs_page *freq, *t; 8213a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8223a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8233a3908c8STrond Myklebust 8243a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8253a3908c8STrond Myklebust 8263a3908c8STrond Myklebust /* search through pnfs commit lists */ 8270c493b5cSTrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, folio); 8283a3908c8STrond Myklebust if (freq) 8293a3908c8STrond Myklebust return freq->wb_head; 8303a3908c8STrond Myklebust 8313a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8323a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8330c493b5cSTrond Myklebust if (nfs_page_to_folio(freq) == folio) 8343a3908c8STrond Myklebust return freq->wb_head; 8353a3908c8STrond Myklebust } 8363a3908c8STrond Myklebust 8373a3908c8STrond Myklebust return NULL; 8383a3908c8STrond Myklebust } 8393a3908c8STrond Myklebust 8408dd37758STrond Myklebust /** 84186d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 84286d80f97STrond Myklebust * @req: pointer to a struct nfs_page 84386d80f97STrond Myklebust * @dst: commit list head 84486d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 84586d80f97STrond Myklebust * 84686d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 84786d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 84886d80f97STrond Myklebust * the MM page stats. 84986d80f97STrond Myklebust * 850e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 851e824f99aSTrond Myklebust * nfs_page lock. 85286d80f97STrond Myklebust */ 85386d80f97STrond Myklebust void 85486d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 85586d80f97STrond Myklebust struct nfs_commit_info *cinfo) 85686d80f97STrond Myklebust { 85786d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 85886d80f97STrond Myklebust nfs_list_add_request(req, dst); 8595cb953d4STrond Myklebust atomic_long_inc(&cinfo->mds->ncommit); 86086d80f97STrond Myklebust } 86186d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 86286d80f97STrond Myklebust 86386d80f97STrond Myklebust /** 8648dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8658dd37758STrond Myklebust * @req: pointer to a struct nfs_page 866ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8678dd37758STrond Myklebust * 868ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8698dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8708dd37758STrond Myklebust * the MM page stats. 8718dd37758STrond Myklebust * 872ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8738dd37758STrond Myklebust * holding the nfs_page lock. 8748dd37758STrond Myklebust */ 8758dd37758STrond Myklebust void 8766272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8778dd37758STrond Myklebust { 878e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8796272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 880e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 8810c493b5cSTrond Myklebust nfs_folio_mark_unstable(nfs_page_to_folio(req), cinfo); 8828dd37758STrond Myklebust } 8838dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8848dd37758STrond Myklebust 8858dd37758STrond Myklebust /** 8868dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8878dd37758STrond Myklebust * @req: pointer to a nfs_page 888ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8898dd37758STrond Myklebust * 890ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8918dd37758STrond Myklebust * number of outstanding requests requiring a commit 8928dd37758STrond Myklebust * It does not update the MM page stats. 8938dd37758STrond Myklebust * 894ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8958dd37758STrond Myklebust */ 8968dd37758STrond Myklebust void 897ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 898ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8998dd37758STrond Myklebust { 9008dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9018dd37758STrond Myklebust return; 9028dd37758STrond Myklebust nfs_list_remove_request(req); 9035cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9048dd37758STrond Myklebust } 9058dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9068dd37758STrond Myklebust 907ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 908ea2cf228SFred Isaman struct inode *inode) 909ea2cf228SFred Isaman { 910fe238e60SDave Wysochanski cinfo->inode = inode; 911ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 912ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 913b359f9d0SFred Isaman cinfo->dreq = NULL; 914f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 915ea2cf228SFred Isaman } 916ea2cf228SFred Isaman 917ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 918ea2cf228SFred Isaman struct inode *inode, 919ea2cf228SFred Isaman struct nfs_direct_req *dreq) 920ea2cf228SFred Isaman { 9211763da12SFred Isaman if (dreq) 9221763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9231763da12SFred Isaman else 924ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 925ea2cf228SFred Isaman } 926ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9278dd37758STrond Myklebust 9281da177e4SLinus Torvalds /* 9291da177e4SLinus Torvalds * Add a request to the inode's commit list. 9301da177e4SLinus Torvalds */ 9311763da12SFred Isaman void 932ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 933b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9341da177e4SLinus Torvalds { 935b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9368dd37758STrond Myklebust return; 9376272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9381da177e4SLinus Torvalds } 9398e821cadSTrond Myklebust 9400c493b5cSTrond Myklebust static void nfs_folio_clear_commit(struct folio *folio) 941e468bae9STrond Myklebust { 9420c493b5cSTrond Myklebust if (folio) { 9430c493b5cSTrond Myklebust long nr = folio_nr_pages(folio); 9440c493b5cSTrond Myklebust 9450c493b5cSTrond Myklebust node_stat_mod_folio(folio, NR_WRITEBACK, -nr); 9460c493b5cSTrond Myklebust wb_stat_mod(&inode_to_bdi(folio_file_mapping(folio)->host)->wb, 9470c493b5cSTrond Myklebust WB_WRITEBACK, -nr); 9480c493b5cSTrond Myklebust } 949e468bae9STrond Myklebust } 950d6d6dc7cSFred Isaman 951b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 952b193a78dSTrond Myklebust static void nfs_clear_request_commit(struct nfs_commit_info *cinfo, 953b193a78dSTrond Myklebust struct nfs_page *req) 954d6d6dc7cSFred Isaman { 9558dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9569fcd5960STrond Myklebust struct nfs_open_context *ctx = nfs_req_openctx(req); 9579fcd5960STrond Myklebust struct inode *inode = d_inode(ctx->dentry); 958d6d6dc7cSFred Isaman 959e824f99aSTrond Myklebust mutex_lock(&NFS_I(inode)->commit_mutex); 960b193a78dSTrond Myklebust if (!pnfs_clear_request_commit(req, cinfo)) { 961b193a78dSTrond Myklebust nfs_request_remove_commit_list(req, cinfo); 962d6d6dc7cSFred Isaman } 963e824f99aSTrond Myklebust mutex_unlock(&NFS_I(inode)->commit_mutex); 9640c493b5cSTrond Myklebust nfs_folio_clear_commit(nfs_page_to_folio(req)); 9658dd37758STrond Myklebust } 966e468bae9STrond Myklebust } 967e468bae9STrond Myklebust 968d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9698e821cadSTrond Myklebust { 970c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 971d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 972c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9738e821cadSTrond Myklebust } 9748e821cadSTrond Myklebust 975919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 976919e3bd9STrond Myklebust { 977919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 978919e3bd9STrond Myklebust } 979919e3bd9STrond Myklebust 980061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9816c75dc0dSFred Isaman { 982ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9836c75dc0dSFred Isaman unsigned long bytes = 0; 9846c75dc0dSFred Isaman 9856c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9866c75dc0dSFred Isaman goto out; 987ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9886c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9896c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9906c75dc0dSFred Isaman 9916c75dc0dSFred Isaman bytes += req->wb_bytes; 9926c75dc0dSFred Isaman nfs_list_remove_request(req); 9936c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9946c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 995af887e43STrond Myklebust trace_nfs_comp_error(hdr->inode, req, hdr->error); 9960c493b5cSTrond Myklebust nfs_mapping_set_error(nfs_page_to_folio(req), 9970c493b5cSTrond Myklebust hdr->error); 9986c75dc0dSFred Isaman goto remove_req; 9996c75dc0dSFred Isaman } 1000c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 100133344e0fSTrond Myklebust /* Reset wb_nio, since the write was successful. */ 100233344e0fSTrond Myklebust req->wb_nio = 0; 1003f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1004b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1005a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10066c75dc0dSFred Isaman goto next; 10076c75dc0dSFred Isaman } 10086c75dc0dSFred Isaman remove_req: 10096c75dc0dSFred Isaman nfs_inode_remove_request(req); 10106c75dc0dSFred Isaman next: 10110c493b5cSTrond Myklebust nfs_page_end_writeback(req); 10123aff4ebbSTrond Myklebust nfs_release_request(req); 10136c75dc0dSFred Isaman } 10146c75dc0dSFred Isaman out: 1015919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10166c75dc0dSFred Isaman hdr->release(hdr); 10176c75dc0dSFred Isaman } 10186c75dc0dSFred Isaman 1019ce59515cSAnna Schumaker unsigned long 1020ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1021fb8a1f11STrond Myklebust { 10225cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1023fb8a1f11STrond Myklebust } 1024fb8a1f11STrond Myklebust 1025e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10261763da12SFred Isaman int 1027ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1028ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1029d6d6dc7cSFred Isaman { 1030137da553STrond Myklebust struct nfs_page *req, *tmp; 1031d6d6dc7cSFred Isaman int ret = 0; 1032d6d6dc7cSFred Isaman 1033137da553STrond Myklebust list_for_each_entry_safe(req, tmp, src, wb_list) { 10347ad84aa9STrond Myklebust kref_get(&req->wb_kref); 10352ce209c4STrond Myklebust if (!nfs_lock_request(req)) { 1036137da553STrond Myklebust nfs_release_request(req); 1037137da553STrond Myklebust continue; 1038137da553STrond Myklebust } 1039ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10405d2a9d9dSTrond Myklebust clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); 10418dd37758STrond Myklebust nfs_list_add_request(req, dst); 1042d6d6dc7cSFred Isaman ret++; 10431763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1044d6d6dc7cSFred Isaman break; 1045e824f99aSTrond Myklebust cond_resched(); 1046d6d6dc7cSFred Isaman } 1047d6d6dc7cSFred Isaman return ret; 1048d6d6dc7cSFred Isaman } 10495d2a9d9dSTrond Myklebust EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 1050d6d6dc7cSFred Isaman 10511da177e4SLinus Torvalds /* 10521da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10531da177e4SLinus Torvalds * @inode: NFS inode to scan 1054ea2cf228SFred Isaman * @dst: mds destination list 1055ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10561da177e4SLinus Torvalds * 10571da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10581da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10591da177e4SLinus Torvalds */ 10601763da12SFred Isaman int 1061ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1062ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10631da177e4SLinus Torvalds { 1064d6d6dc7cSFred Isaman int ret = 0; 1065fb8a1f11STrond Myklebust 10665cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10675cb953d4STrond Myklebust return 0; 1068e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10695cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10708dd37758STrond Myklebust const int max = INT_MAX; 1071d6d6dc7cSFred Isaman 1072ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1073ea2cf228SFred Isaman cinfo, max); 1074ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1075d6d6dc7cSFred Isaman } 1076e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1077ff778d02STrond Myklebust return ret; 10781da177e4SLinus Torvalds } 1079d6d6dc7cSFred Isaman 10801da177e4SLinus Torvalds /* 1081e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1082e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10831da177e4SLinus Torvalds * 1084e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1085e7d39069STrond Myklebust * to disk. 10861da177e4SLinus Torvalds */ 10870c493b5cSTrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct folio *folio, 1088e7d39069STrond Myklebust unsigned int offset, 1089e7d39069STrond Myklebust unsigned int bytes) 10901da177e4SLinus Torvalds { 1091e7d39069STrond Myklebust struct nfs_page *req; 1092e7d39069STrond Myklebust unsigned int rqend; 1093e7d39069STrond Myklebust unsigned int end; 10941da177e4SLinus Torvalds int error; 1095277459d2STrond Myklebust 1096e7d39069STrond Myklebust end = offset + bytes; 1097e7d39069STrond Myklebust 10980c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 1099f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1100f6032f21STrond Myklebust return req; 11012bfc6e56SWeston Andros Adamson 1102e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1103e7d39069STrond Myklebust /* 1104e7d39069STrond Myklebust * Tell the caller to flush out the request if 1105e7d39069STrond Myklebust * the offsets are non-contiguous. 1106e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1107e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1108e7d39069STrond Myklebust */ 1109f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1110e7d39069STrond Myklebust goto out_flushme; 1111e7d39069STrond Myklebust 11121da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11131da177e4SLinus Torvalds if (offset < req->wb_offset) { 11141da177e4SLinus Torvalds req->wb_offset = offset; 11151da177e4SLinus Torvalds req->wb_pgbase = offset; 11161da177e4SLinus Torvalds } 11171da177e4SLinus Torvalds if (end > rqend) 11181da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1119e7d39069STrond Myklebust else 1120e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 112133344e0fSTrond Myklebust req->wb_nio = 0; 1122e7d39069STrond Myklebust return req; 1123e7d39069STrond Myklebust out_flushme: 1124f6032f21STrond Myklebust /* 1125f6032f21STrond Myklebust * Note: we mark the request dirty here because 1126f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1127f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1128f6032f21STrond Myklebust */ 1129f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1130f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 11310c493b5cSTrond Myklebust error = nfs_wb_folio(folio_file_mapping(folio)->host, folio); 1132f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1133e7d39069STrond Myklebust } 11341da177e4SLinus Torvalds 1135e7d39069STrond Myklebust /* 1136e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1137e7d39069STrond Myklebust * 1138e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1139e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1140e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1141e7d39069STrond Myklebust */ 1142e7d39069STrond Myklebust static struct nfs_page *nfs_setup_write_request(struct nfs_open_context *ctx, 11430c493b5cSTrond Myklebust struct folio *folio, 11440c493b5cSTrond Myklebust unsigned int offset, 11450c493b5cSTrond Myklebust unsigned int bytes) 1146e7d39069STrond Myklebust { 1147e7d39069STrond Myklebust struct nfs_page *req; 1148e7d39069STrond Myklebust 11490c493b5cSTrond Myklebust req = nfs_try_to_update_request(folio, offset, bytes); 1150e7d39069STrond Myklebust if (req != NULL) 1151e7d39069STrond Myklebust goto out; 11520c493b5cSTrond Myklebust req = nfs_page_create_from_folio(ctx, folio, offset, bytes); 1153e7d39069STrond Myklebust if (IS_ERR(req)) 1154e7d39069STrond Myklebust goto out; 11550c493b5cSTrond Myklebust nfs_inode_add_request(req); 1156efc91ed0STrond Myklebust out: 115761e930a9STrond Myklebust return req; 11581da177e4SLinus Torvalds } 11591da177e4SLinus Torvalds 11600c493b5cSTrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, 11610c493b5cSTrond Myklebust struct folio *folio, unsigned int offset, 11620c493b5cSTrond Myklebust unsigned int count) 1163e7d39069STrond Myklebust { 1164e7d39069STrond Myklebust struct nfs_page *req; 1165e7d39069STrond Myklebust 11660c493b5cSTrond Myklebust req = nfs_setup_write_request(ctx, folio, offset, count); 1167e7d39069STrond Myklebust if (IS_ERR(req)) 1168e7d39069STrond Myklebust return PTR_ERR(req); 1169e7d39069STrond Myklebust /* Update file length */ 11700c493b5cSTrond Myklebust nfs_grow_file(folio, offset, count); 1171d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1172a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11731d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1174e7d39069STrond Myklebust return 0; 1175e7d39069STrond Myklebust } 1176e7d39069STrond Myklebust 11770c493b5cSTrond Myklebust int nfs_flush_incompatible(struct file *file, struct folio *folio) 11781da177e4SLinus Torvalds { 1179cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11802a369153STrond Myklebust struct nfs_lock_context *l_ctx; 118117b985deSJeff Layton struct file_lock_context *flctx = locks_inode_context(file_inode(file)); 11821da177e4SLinus Torvalds struct nfs_page *req; 11831a54533eSTrond Myklebust int do_flush, status; 11841da177e4SLinus Torvalds /* 11851da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11861da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11871da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11881da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11891da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11901da177e4SLinus Torvalds * dropped page. 11911da177e4SLinus Torvalds */ 11921a54533eSTrond Myklebust do { 11930c493b5cSTrond Myklebust req = nfs_folio_find_head_request(folio); 11941a54533eSTrond Myklebust if (req == NULL) 11951a54533eSTrond Myklebust return 0; 11962a369153STrond Myklebust l_ctx = req->wb_lock_context; 11970c493b5cSTrond Myklebust do_flush = nfs_page_to_folio(req) != folio || 11989fcd5960STrond Myklebust !nfs_match_open_context(nfs_req_openctx(req), ctx); 1199bd61e0a9SJeff Layton if (l_ctx && flctx && 1200bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1201bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1202d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12035263e31eSJeff Layton } 12041da177e4SLinus Torvalds nfs_release_request(req); 12051a54533eSTrond Myklebust if (!do_flush) 12061a54533eSTrond Myklebust return 0; 12070c493b5cSTrond Myklebust status = nfs_wb_folio(folio_file_mapping(folio)->host, folio); 12081a54533eSTrond Myklebust } while (status == 0); 12091a54533eSTrond Myklebust return status; 12101da177e4SLinus Torvalds } 12111da177e4SLinus Torvalds 12121da177e4SLinus Torvalds /* 1213dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1214dc24826bSAndy Adamson * expire soon. 1215dc24826bSAndy Adamson * 1216dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1217dc24826bSAndy Adamson * 1218dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1219dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1220dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1221dc24826bSAndy Adamson */ 1222dc24826bSAndy Adamson int 1223dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1224dc24826bSAndy Adamson { 1225dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1226dc24826bSAndy Adamson 1227ddf529eeSNeilBrown if (nfs_ctx_key_to_expire(ctx, inode) && 1228ca05cbaeSTrond Myklebust !rcu_access_pointer(ctx->ll_cred)) 1229ddf529eeSNeilBrown /* Already expired! */ 1230ddf529eeSNeilBrown return -EACCES; 1231ddf529eeSNeilBrown return 0; 1232dc24826bSAndy Adamson } 1233dc24826bSAndy Adamson 1234dc24826bSAndy Adamson /* 1235dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1236dc24826bSAndy Adamson */ 1237ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1238dc24826bSAndy Adamson { 1239ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1240ca05cbaeSTrond Myklebust struct rpc_cred *cred, *new, *old = NULL; 1241ddf529eeSNeilBrown struct auth_cred acred = { 1242a52458b4SNeilBrown .cred = ctx->cred, 1243ddf529eeSNeilBrown }; 1244ca05cbaeSTrond Myklebust bool ret = false; 1245ce52914eSScott Mayhew 1246ca05cbaeSTrond Myklebust rcu_read_lock(); 1247ca05cbaeSTrond Myklebust cred = rcu_dereference(ctx->ll_cred); 1248ca05cbaeSTrond Myklebust if (cred && !(cred->cr_ops->crkey_timeout && 1249ca05cbaeSTrond Myklebust cred->cr_ops->crkey_timeout(cred))) 1250ca05cbaeSTrond Myklebust goto out; 1251ca05cbaeSTrond Myklebust rcu_read_unlock(); 1252ca05cbaeSTrond Myklebust 1253ca05cbaeSTrond Myklebust new = auth->au_ops->lookup_cred(auth, &acred, 0); 1254ca05cbaeSTrond Myklebust if (new == cred) { 1255ca05cbaeSTrond Myklebust put_rpccred(new); 1256ddf529eeSNeilBrown return true; 1257ca05cbaeSTrond Myklebust } 1258ca05cbaeSTrond Myklebust if (IS_ERR_OR_NULL(new)) { 1259ca05cbaeSTrond Myklebust new = NULL; 1260ca05cbaeSTrond Myklebust ret = true; 1261ca05cbaeSTrond Myklebust } else if (new->cr_ops->crkey_timeout && 1262ca05cbaeSTrond Myklebust new->cr_ops->crkey_timeout(new)) 1263ca05cbaeSTrond Myklebust ret = true; 1264ca05cbaeSTrond Myklebust 1265ca05cbaeSTrond Myklebust rcu_read_lock(); 1266ca05cbaeSTrond Myklebust old = rcu_dereference_protected(xchg(&ctx->ll_cred, 1267ca05cbaeSTrond Myklebust RCU_INITIALIZER(new)), 1); 1268ca05cbaeSTrond Myklebust out: 1269ca05cbaeSTrond Myklebust rcu_read_unlock(); 1270ca05cbaeSTrond Myklebust put_rpccred(old); 1271ca05cbaeSTrond Myklebust return ret; 1272dc24826bSAndy Adamson } 1273dc24826bSAndy Adamson 1274dc24826bSAndy Adamson /* 12755d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12765d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12775d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12785d47a356STrond Myklebust */ 12790c493b5cSTrond Myklebust static bool nfs_folio_write_uptodate(struct folio *folio, unsigned int pagelen) 12805d47a356STrond Myklebust { 12810c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 1282d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1283d529ef83SJeff Layton 12848d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12858d197a56STrond Myklebust goto out; 1286fc9dc401STrond Myklebust if (nfsi->cache_validity & 128713c0b082STrond Myklebust (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE)) 1288d529ef83SJeff Layton return false; 12894db72b40SJeff Layton smp_rmb(); 1290fc9dc401STrond Myklebust if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags) && pagelen != 0) 12918d197a56STrond Myklebust return false; 12928d197a56STrond Myklebust out: 1293fc9dc401STrond Myklebust if (nfsi->cache_validity & NFS_INO_INVALID_DATA && pagelen != 0) 129418dd78c4SScott Mayhew return false; 12950c493b5cSTrond Myklebust return folio_test_uptodate(folio) != 0; 12965d47a356STrond Myklebust } 12975d47a356STrond Myklebust 12985263e31eSJeff Layton static bool 12995263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 13005263e31eSJeff Layton { 13015263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 1302d7c9616bSJeff Layton lock_is_write(fl); 13035263e31eSJeff Layton } 13045263e31eSJeff Layton 1305c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1306c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1307c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1308c7559663SScott Mayhew * inefficiencies. 1309c7559663SScott Mayhew * 1310263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1311263b4509SScott Mayhew * of the checks. 1312c7559663SScott Mayhew */ 13130c493b5cSTrond Myklebust static int nfs_can_extend_write(struct file *file, struct folio *folio, 13140c493b5cSTrond Myklebust unsigned int pagelen) 1315c7559663SScott Mayhew { 13160c493b5cSTrond Myklebust struct inode *inode = file_inode(file); 131717b985deSJeff Layton struct file_lock_context *flctx = locks_inode_context(inode); 13185263e31eSJeff Layton struct file_lock *fl; 13190c493b5cSTrond Myklebust int ret; 13205263e31eSJeff Layton 1321c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1322c7559663SScott Mayhew return 0; 13230c493b5cSTrond Myklebust if (!nfs_folio_write_uptodate(folio, pagelen)) 1324263b4509SScott Mayhew return 0; 13254201916fSTrond Myklebust if (nfs_have_write_delegation(inode)) 1326c7559663SScott Mayhew return 1; 1327bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1328bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13298fa4592aSTrond Myklebust return 1; 13305263e31eSJeff Layton 13315263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13325263e31eSJeff Layton ret = 0; 13336109c850SJeff Layton spin_lock(&flctx->flc_lock); 1334bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1335bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1336dd1fac6aSJeff Layton c.flc_list); 1337bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13385263e31eSJeff Layton ret = 1; 1339bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13405263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 1341dd1fac6aSJeff Layton c.flc_list); 1342d7c9616bSJeff Layton if (lock_is_write(fl)) 13435263e31eSJeff Layton ret = 1; 13445263e31eSJeff Layton } 13456109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13465263e31eSJeff Layton return ret; 1347c7559663SScott Mayhew } 1348c7559663SScott Mayhew 13495d47a356STrond Myklebust /* 13501da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13511da177e4SLinus Torvalds * 13521da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13531da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13541da177e4SLinus Torvalds */ 13550c493b5cSTrond Myklebust int nfs_update_folio(struct file *file, struct folio *folio, 13561da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13571da177e4SLinus Torvalds { 1358cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 13590c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 1360d2ceb7e5SBenjamin Coddington struct inode *inode = mapping->host; 13610c493b5cSTrond Myklebust unsigned int pagelen = nfs_folio_length(folio); 13621da177e4SLinus Torvalds int status = 0; 13631da177e4SLinus Torvalds 136491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 136591d5b470SChuck Lever 13660c493b5cSTrond Myklebust dprintk("NFS: nfs_update_folio(%pD2 %d@%lld)\n", file, count, 13670c493b5cSTrond Myklebust (long long)(folio_file_pos(folio) + offset)); 13681da177e4SLinus Torvalds 1369149a4fddSBenjamin Coddington if (!count) 1370149a4fddSBenjamin Coddington goto out; 1371149a4fddSBenjamin Coddington 13720c493b5cSTrond Myklebust if (nfs_can_extend_write(file, folio, pagelen)) { 1373fc9dc401STrond Myklebust count = max(count + offset, pagelen); 13741da177e4SLinus Torvalds offset = 0; 13751da177e4SLinus Torvalds } 13761da177e4SLinus Torvalds 13770c493b5cSTrond Myklebust status = nfs_writepage_setup(ctx, folio, offset, count); 137803fa9e84STrond Myklebust if (status < 0) 1379d2ceb7e5SBenjamin Coddington nfs_set_pageerror(mapping); 1380149a4fddSBenjamin Coddington out: 13810c493b5cSTrond Myklebust dprintk("NFS: nfs_update_folio returns %d (isize %lld)\n", 13821da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13831da177e4SLinus Torvalds return status; 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds 13863ff7576dSTrond Myklebust static int flush_task_priority(int how) 13871da177e4SLinus Torvalds { 13881da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13891da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13901da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13911da177e4SLinus Torvalds case FLUSH_LOWPRI: 13921da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13931da177e4SLinus Torvalds } 13941da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds 1397d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1398d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1399abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 14001ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 14011da177e4SLinus Torvalds { 14023ff7576dSTrond Myklebust int priority = flush_task_priority(how); 14031da177e4SLinus Torvalds 14048db55a03SNeilBrown if (IS_SWAPFILE(hdr->inode)) 14058db55a03SNeilBrown task_setup_data->flags |= RPC_TASK_SWAPPER; 14061ed26f33SAnna Schumaker task_setup_data->priority = priority; 1407fb91fb0eSAnna Schumaker rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); 14085bb2a7cbSTrond Myklebust trace_nfs_initiate_write(hdr); 1409275acaafSTrond Myklebust } 1410275acaafSTrond Myklebust 14116d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 14126d884e8fSFred * call this on each, which will prepare them to be retried on next 14136d884e8fSFred * writeback using standard nfs. 14146d884e8fSFred */ 14156d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14166d884e8fSFred { 14176dd85e83STrond Myklebust struct nfs_inode *nfsi = NFS_I(nfs_page_to_inode(req)); 141867f4b5dcSTrond Myklebust 141933344e0fSTrond Myklebust /* Bump the transmission count */ 142033344e0fSTrond Myklebust req->wb_nio++; 14216d884e8fSFred nfs_mark_request_dirty(req); 142267f4b5dcSTrond Myklebust atomic_long_inc(&nfsi->redirtied_pages); 14230c493b5cSTrond Myklebust nfs_page_end_writeback(req); 14243aff4ebbSTrond Myklebust nfs_release_request(req); 14256d884e8fSFred } 14266d884e8fSFred 1427df3accb8STrond Myklebust static void nfs_async_write_error(struct list_head *head, int error) 14286c75dc0dSFred Isaman { 14296c75dc0dSFred Isaman struct nfs_page *req; 14306c75dc0dSFred Isaman 14316c75dc0dSFred Isaman while (!list_empty(head)) { 14326c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14336c75dc0dSFred Isaman nfs_list_remove_request(req); 1434cea9ba72STrond Myklebust if (nfs_error_is_fatal_on_server(error)) 14356fbda89bSTrond Myklebust nfs_write_error(req, error); 14366fbda89bSTrond Myklebust else 14376c75dc0dSFred Isaman nfs_redirty_request(req); 14386c75dc0dSFred Isaman } 14396c75dc0dSFred Isaman } 14406c75dc0dSFred Isaman 1441dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1442dc602dd7STrond Myklebust { 1443df3accb8STrond Myklebust nfs_async_write_error(&hdr->pages, 0); 1444dc602dd7STrond Myklebust } 1445dc602dd7STrond Myklebust 1446061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1447919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1448061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1449061ae2edSFred Isaman .completion = nfs_write_completion, 1450dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1451061ae2edSFred Isaman }; 1452061ae2edSFred Isaman 145357208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1454a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1455061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14561751c363STrond Myklebust { 1457a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 145841d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1459a20c93e3SChristoph Hellwig 1460a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1461a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1462a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1463a20c93e3SChristoph Hellwig #endif 14644a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 14653bde7afdSTrond Myklebust server->wsize, ioflags); 14661751c363STrond Myklebust } 1467ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14681751c363STrond Myklebust 1469dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1470dce81290STrond Myklebust { 1471a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1472a7d42ddbSWeston Andros Adamson 14736f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14746f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14756f29b9bbSKinglong Mee 147641d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1477a7d42ddbSWeston Andros Adamson 1478a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1479a7d42ddbSWeston Andros Adamson 1480a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1481a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1482dce81290STrond Myklebust } 14831f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1484dce81290STrond Myklebust 14851da177e4SLinus Torvalds 14860b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14870b7c0153SFred Isaman { 14880b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14890b7c0153SFred Isaman 14900b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14910b7c0153SFred Isaman } 14920b7c0153SFred Isaman 1493a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1494a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1495a08a8cd3STrond Myklebust { 1496a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1497a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14982b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1499a08a8cd3STrond Myklebust 1500a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15012b83d3deSTrond Myklebust fattr->size = size; 15022b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15032b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1504a08a8cd3STrond Myklebust return; 15052b83d3deSTrond Myklebust } 15062b83d3deSTrond Myklebust if (size != fattr->size) 1507a08a8cd3STrond Myklebust return; 1508a08a8cd3STrond Myklebust /* Set attribute barrier */ 1509a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15102b83d3deSTrond Myklebust /* ...and update size */ 15112b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1512a08a8cd3STrond Myklebust } 1513a08a8cd3STrond Myklebust 1514a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1515a08a8cd3STrond Myklebust { 15162b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1517a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1518a08a8cd3STrond Myklebust 1519*e12912d9STrond Myklebust if (nfs_have_delegated_mtime(inode)) { 1520*e12912d9STrond Myklebust spin_lock(&inode->i_lock); 1521*e12912d9STrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS); 1522*e12912d9STrond Myklebust spin_unlock(&inode->i_lock); 1523*e12912d9STrond Myklebust return; 1524*e12912d9STrond Myklebust } 1525*e12912d9STrond Myklebust 1526a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1527a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1528a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1529a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1530a08a8cd3STrond Myklebust } 1531a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1532a08a8cd3STrond Myklebust 15331da177e4SLinus Torvalds /* 15341da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15351da177e4SLinus Torvalds */ 1536d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1537d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15380eecb214SAnna Schumaker struct inode *inode) 15391da177e4SLinus Torvalds { 1540788e7a89STrond Myklebust int status; 15411da177e4SLinus Torvalds 1542f551e44fSChuck Lever /* 1543f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1544f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1545f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1546f551e44fSChuck Lever * another writer had changed the file, but some applications 1547f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1548f551e44fSChuck Lever */ 1549d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1550788e7a89STrond Myklebust if (status != 0) 15510eecb214SAnna Schumaker return status; 15528224b273SChuck Lever 1553d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 15545bb2a7cbSTrond Myklebust trace_nfs_writeback_done(task, hdr); 155591d5b470SChuck Lever 155669d96651SJeff Layton if (task->tk_status >= 0) { 155769d96651SJeff Layton enum nfs3_stable_how committed = hdr->res.verf->committed; 155869d96651SJeff Layton 155969d96651SJeff Layton if (committed == NFS_UNSTABLE) { 156069d96651SJeff Layton /* 156169d96651SJeff Layton * We have some uncommitted data on the server at 156269d96651SJeff Layton * this point, so ensure that we keep track of that 156369d96651SJeff Layton * fact irrespective of what later writes do. 156469d96651SJeff Layton */ 156569d96651SJeff Layton set_bit(NFS_IOHDR_UNSTABLE_WRITES, &hdr->flags); 156669d96651SJeff Layton } 156769d96651SJeff Layton 156869d96651SJeff Layton if (committed < hdr->args.stable) { 15691da177e4SLinus Torvalds /* We tried a write call, but the server did not 15701da177e4SLinus Torvalds * commit data to stable storage even though we 15711da177e4SLinus Torvalds * requested it. 15721da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15731da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15741da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15751da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15761da177e4SLinus Torvalds */ 15771da177e4SLinus Torvalds static unsigned long complain; 15781da177e4SLinus Torvalds 1579a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15801da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15811da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15821da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1583cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 158469d96651SJeff Layton committed, hdr->args.stable); 15851da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15861da177e4SLinus Torvalds } 15871da177e4SLinus Torvalds } 158869d96651SJeff Layton } 15891f2edbe3STrond Myklebust 15901f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 159116e14375STrond Myklebust if (nfs_should_remove_suid(inode)) { 159216e14375STrond Myklebust spin_lock(&inode->i_lock); 1593720869ebSTrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); 159416e14375STrond Myklebust spin_unlock(&inode->i_lock); 159516e14375STrond Myklebust } 15960eecb214SAnna Schumaker return 0; 15970eecb214SAnna Schumaker } 15980eecb214SAnna Schumaker 15990eecb214SAnna Schumaker /* 16000eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 16010eecb214SAnna Schumaker */ 1602d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1603d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 16040eecb214SAnna Schumaker { 1605d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1606d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 16071f2edbe3STrond Myklebust 16081f2edbe3STrond Myklebust if (resp->count < argp->count) { 16091da177e4SLinus Torvalds static unsigned long complain; 16101da177e4SLinus Torvalds 16116c75dc0dSFred Isaman /* This a short write! */ 1612d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 161391d5b470SChuck Lever 16141da177e4SLinus Torvalds /* Has the server at least made some progress? */ 16156c75dc0dSFred Isaman if (resp->count == 0) { 16166c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 16176c75dc0dSFred Isaman printk(KERN_WARNING 16186c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16196c75dc0dSFred Isaman argp->count); 16206c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16216c75dc0dSFred Isaman } 1622d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16236c75dc0dSFred Isaman task->tk_status = -EIO; 16246c75dc0dSFred Isaman return; 16256c75dc0dSFred Isaman } 1626f8417b48SKinglong Mee 1627f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1628f8417b48SKinglong Mee if (!task->tk_ops) { 1629f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1630f8417b48SKinglong Mee return; 1631f8417b48SKinglong Mee } 1632f8417b48SKinglong Mee 16331da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16341da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16351da177e4SLinus Torvalds /* Resend from where the server left off */ 1636d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16371da177e4SLinus Torvalds argp->offset += resp->count; 16381da177e4SLinus Torvalds argp->pgbase += resp->count; 16391da177e4SLinus Torvalds argp->count -= resp->count; 16401da177e4SLinus Torvalds } else { 16411da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16421da177e4SLinus Torvalds * headaches in the case of a server crash. 16431da177e4SLinus Torvalds */ 16441da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16451da177e4SLinus Torvalds } 16468c9cb714STrond Myklebust resp->count = 0; 16478c9cb714STrond Myklebust resp->verf->committed = 0; 1648d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16491da177e4SLinus Torvalds } 16501da177e4SLinus Torvalds } 16511da177e4SLinus Torvalds 1652af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 165371d0a611STrond Myklebust { 1654723c921eSPeter Zijlstra return wait_var_event_killable(&cinfo->rpcs_out, 1655723c921eSPeter Zijlstra !atomic_read(&cinfo->rpcs_out)); 1656af7cf057STrond Myklebust } 1657af7cf057STrond Myklebust 165817f46b80SJosef Bacik void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1659af7cf057STrond Myklebust { 1660af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1661af7cf057STrond Myklebust } 1662af7cf057STrond Myklebust 1663133a48abSTrond Myklebust bool nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1664af7cf057STrond Myklebust { 1665133a48abSTrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) { 1666723c921eSPeter Zijlstra wake_up_var(&cinfo->rpcs_out); 1667133a48abSTrond Myklebust return true; 1668133a48abSTrond Myklebust } 1669133a48abSTrond Myklebust return false; 167071d0a611STrond Myklebust } 167171d0a611STrond Myklebust 16720b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16731da177e4SLinus Torvalds { 16740b7c0153SFred Isaman put_nfs_open_context(data->context); 16750b7c0153SFred Isaman nfs_commit_free(data); 16761da177e4SLinus Torvalds } 1677e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16781da177e4SLinus Torvalds 16790b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1680c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16819ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16829f0ec176SAndy Adamson int how, int flags) 16831da177e4SLinus Torvalds { 168407737691STrond Myklebust struct rpc_task *task; 16859ace33cdSFred Isaman int priority = flush_task_priority(how); 1686bdc7f021STrond Myklebust struct rpc_message msg = { 1687bdc7f021STrond Myklebust .rpc_argp = &data->args, 1688bdc7f021STrond Myklebust .rpc_resp = &data->res, 16899ace33cdSFred Isaman .rpc_cred = data->cred, 1690bdc7f021STrond Myklebust }; 169184115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 169207737691STrond Myklebust .task = &data->task, 16939ace33cdSFred Isaman .rpc_client = clnt, 1694bdc7f021STrond Myklebust .rpc_message = &msg, 16959ace33cdSFred Isaman .callback_ops = call_ops, 169684115e1cSTrond Myklebust .callback_data = data, 1697101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16984fa7ef69STrond Myklebust .flags = RPC_TASK_ASYNC | flags, 16993ff7576dSTrond Myklebust .priority = priority, 170084115e1cSTrond Myklebust }; 1701118f09edSOlga Kornievskaia 1702118f09edSOlga Kornievskaia if (nfs_server_capable(data->inode, NFS_CAP_MOVEABLE)) 1703118f09edSOlga Kornievskaia task_setup_data.flags |= RPC_TASK_MOVEABLE; 1704118f09edSOlga Kornievskaia 1705788e7a89STrond Myklebust /* Set up the initial task struct. */ 1706e9ae1ee2SAnna Schumaker nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); 17078224b273SChuck Lever trace_nfs_initiate_commit(data); 17081da177e4SLinus Torvalds 1709b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1710bdc7f021STrond Myklebust 171107737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1712dbae4c73STrond Myklebust if (IS_ERR(task)) 1713dbae4c73STrond Myklebust return PTR_ERR(task); 1714d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1715d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 171607737691STrond Myklebust rpc_put_task(task); 1717dbae4c73STrond Myklebust return 0; 17181da177e4SLinus Torvalds } 1719e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 17201da177e4SLinus Torvalds 1721378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1722378520b8SPeng Tao { 1723378520b8SPeng Tao loff_t lwb = 0; 1724378520b8SPeng Tao struct nfs_page *req; 1725378520b8SPeng Tao 1726378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1727378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1728378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1729378520b8SPeng Tao 1730378520b8SPeng Tao return lwb; 1731378520b8SPeng Tao } 1732378520b8SPeng Tao 17331da177e4SLinus Torvalds /* 17349ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17359ace33cdSFred Isaman */ 17360b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1737988b6dceSFred Isaman struct list_head *head, 1738f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1739f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17409ace33cdSFred Isaman { 174119573c93STrond Myklebust struct nfs_page *first; 174219573c93STrond Myklebust struct nfs_open_context *ctx; 174319573c93STrond Myklebust struct inode *inode; 17449ace33cdSFred Isaman 17459ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17469ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17479ace33cdSFred Isaman 174819573c93STrond Myklebust if (head) 17499ace33cdSFred Isaman list_splice_init(head, &data->pages); 17509ace33cdSFred Isaman 175119573c93STrond Myklebust first = nfs_list_entry(data->pages.next); 175219573c93STrond Myklebust ctx = nfs_req_openctx(first); 175319573c93STrond Myklebust inode = d_inode(ctx->dentry); 175419573c93STrond Myklebust 17559ace33cdSFred Isaman data->inode = inode; 17569fcd5960STrond Myklebust data->cred = ctx->cred; 1757988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1758378520b8SPeng Tao /* only set lwb for pnfs commit */ 1759378520b8SPeng Tao if (lseg) 1760378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17619ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1762f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1763b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17649ace33cdSFred Isaman 17659ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17669ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17679ace33cdSFred Isaman data->args.offset = 0; 17689ace33cdSFred Isaman data->args.count = 0; 17699fcd5960STrond Myklebust data->context = get_nfs_open_context(ctx); 17709ace33cdSFred Isaman data->res.fattr = &data->fattr; 17719ace33cdSFred Isaman data->res.verf = &data->verf; 17729ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 1773133a48abSTrond Myklebust nfs_commit_begin(cinfo->mds); 17749ace33cdSFred Isaman } 1775e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17769ace33cdSFred Isaman 1777e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1778ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1779b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1780b57ff130SWeston Andros Adamson u32 ds_commit_idx) 178164bfeb49SFred Isaman { 178264bfeb49SFred Isaman struct nfs_page *req; 178364bfeb49SFred Isaman 178464bfeb49SFred Isaman while (!list_empty(page_list)) { 178564bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 178664bfeb49SFred Isaman nfs_list_remove_request(req); 1787b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 17880c493b5cSTrond Myklebust nfs_folio_clear_commit(nfs_page_to_folio(req)); 17891d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 179064bfeb49SFred Isaman } 179164bfeb49SFred Isaman } 1792e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 179364bfeb49SFred Isaman 17940c493b5cSTrond Myklebust static void nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1795b20135d0STrond Myklebust struct nfs_page *req) 1796b20135d0STrond Myklebust { 17970c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 17980c493b5cSTrond Myklebust 17990c493b5cSTrond Myklebust filemap_dirty_folio(folio_mapping(folio), folio); 1800b20135d0STrond Myklebust } 1801b20135d0STrond Myklebust 18029ace33cdSFred Isaman /* 18031da177e4SLinus Torvalds * Commit dirty pages 18041da177e4SLinus Torvalds */ 18051da177e4SLinus Torvalds static int 1806ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1807ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 18081da177e4SLinus Torvalds { 18090b7c0153SFred Isaman struct nfs_commit_data *data; 181085e39feeSOlga Kornievskaia unsigned short task_flags = 0; 18111da177e4SLinus Torvalds 1812ade8febdSWeston Andros Adamson /* another commit raced with us */ 1813ade8febdSWeston Andros Adamson if (list_empty(head)) 1814ade8febdSWeston Andros Adamson return 0; 1815ade8febdSWeston Andros Adamson 1816515dcdcdSTrond Myklebust data = nfs_commitdata_alloc(); 1817515dcdcdSTrond Myklebust if (!data) { 1818515dcdcdSTrond Myklebust nfs_retry_commit(head, NULL, cinfo, -1); 1819515dcdcdSTrond Myklebust return -ENOMEM; 1820515dcdcdSTrond Myklebust } 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds /* Set up the argument struct */ 1823f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 182485e39feeSOlga Kornievskaia if (NFS_SERVER(inode)->nfs_client->cl_minorversion) 182585e39feeSOlga Kornievskaia task_flags = RPC_TASK_MOVEABLE; 1826c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 182785e39feeSOlga Kornievskaia data->mds_ops, how, 182885e39feeSOlga Kornievskaia RPC_TASK_CRED_NOREF | task_flags); 18291da177e4SLinus Torvalds } 18301da177e4SLinus Torvalds 18311da177e4SLinus Torvalds /* 18321da177e4SLinus Torvalds * COMMIT call returned 18331da177e4SLinus Torvalds */ 1834788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 18351da177e4SLinus Torvalds { 18360b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18371da177e4SLinus Torvalds 1838788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1839c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 18407bdd297eSTrond Myklebust trace_nfs_commit_done(task, data); 1841c9d8f89dSTrond Myklebust } 1842c9d8f89dSTrond Myklebust 1843f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1844c9d8f89dSTrond Myklebust { 1845221203ceSTrond Myklebust const struct nfs_writeverf *verf = data->res.verf; 1846c9d8f89dSTrond Myklebust struct nfs_page *req; 1847c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1848f453a54aSFred Isaman struct nfs_commit_info cinfo; 1849353db796SNeilBrown struct nfs_server *nfss; 18500c493b5cSTrond Myklebust struct folio *folio; 1851788e7a89STrond Myklebust 18521da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18531da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18541da177e4SLinus Torvalds nfs_list_remove_request(req); 18550c493b5cSTrond Myklebust folio = nfs_page_to_folio(req); 18560c493b5cSTrond Myklebust nfs_folio_clear_commit(folio); 18571da177e4SLinus Torvalds 18581e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18599fcd5960STrond Myklebust nfs_req_openctx(req)->dentry->d_sb->s_id, 18609fcd5960STrond Myklebust (unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)), 18611da177e4SLinus Torvalds req->wb_bytes, 18621da177e4SLinus Torvalds (long long)req_offset(req)); 1863c9d8f89dSTrond Myklebust if (status < 0) { 18640c493b5cSTrond Myklebust if (folio) { 1865af887e43STrond Myklebust trace_nfs_commit_error(data->inode, req, 1866af887e43STrond Myklebust status); 18670c493b5cSTrond Myklebust nfs_mapping_set_error(folio, status); 18681da177e4SLinus Torvalds nfs_inode_remove_request(req); 18696fbda89bSTrond Myklebust } 1870ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18711da177e4SLinus Torvalds goto next; 18721da177e4SLinus Torvalds } 18731da177e4SLinus Torvalds 18741da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18751da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18761f28476dSTrond Myklebust if (nfs_write_match_verf(verf, req)) { 18771da177e4SLinus Torvalds /* We have a match */ 18780c493b5cSTrond Myklebust if (folio) 18791da177e4SLinus Torvalds nfs_inode_remove_request(req); 1880ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18811da177e4SLinus Torvalds goto next; 18821da177e4SLinus Torvalds } 18831da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1884ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18856d884e8fSFred nfs_mark_request_dirty(req); 188667f4b5dcSTrond Myklebust atomic_long_inc(&NFS_I(data->inode)->redirtied_pages); 18871da177e4SLinus Torvalds next: 18881d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18897f1bda44STrond Myklebust /* Latency breaker */ 18907f1bda44STrond Myklebust cond_resched(); 18911da177e4SLinus Torvalds } 1892353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1893353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18946df25e58SNeilBrown nfss->write_congested = 0; 1895353db796SNeilBrown 1896f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1897af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18985917ce84SFred Isaman } 18995917ce84SFred Isaman 19005917ce84SFred Isaman static void nfs_commit_release(void *calldata) 19015917ce84SFred Isaman { 19020b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 19035917ce84SFred Isaman 1904f453a54aSFred Isaman data->completion_ops->completion(data); 1905c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 19061da177e4SLinus Torvalds } 1907788e7a89STrond Myklebust 1908788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 19090b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1910788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1911788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1912788e7a89STrond Myklebust }; 19131da177e4SLinus Torvalds 1914f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1915f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1916b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1917f453a54aSFred Isaman }; 1918f453a54aSFred Isaman 19191763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1920ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 192184c53ab5SFred Isaman { 192284c53ab5SFred Isaman int status; 192384c53ab5SFred Isaman 1924ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 192584c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1926ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 192784c53ab5SFred Isaman return status; 192884c53ab5SFred Isaman } 192984c53ab5SFred Isaman 1930c4f24df9STrond Myklebust static int __nfs_commit_inode(struct inode *inode, int how, 1931c4f24df9STrond Myklebust struct writeback_control *wbc) 19321da177e4SLinus Torvalds { 19331da177e4SLinus Torvalds LIST_HEAD(head); 1934ea2cf228SFred Isaman struct nfs_commit_info cinfo; 193571d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1936c4f24df9STrond Myklebust int ret, nscan; 19371da177e4SLinus Torvalds 193864a93dbfSTrond Myklebust how &= ~FLUSH_SYNC; 1939ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1940af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1941c4f24df9STrond Myklebust for (;;) { 1942c4f24df9STrond Myklebust ret = nscan = nfs_scan_commit(inode, &head, &cinfo); 1943c4f24df9STrond Myklebust if (ret <= 0) 1944c4f24df9STrond Myklebust break; 1945c4f24df9STrond Myklebust ret = nfs_generic_commit_list(inode, &head, how, &cinfo); 1946c4f24df9STrond Myklebust if (ret < 0) 1947c4f24df9STrond Myklebust break; 1948c4f24df9STrond Myklebust ret = 0; 1949c4f24df9STrond Myklebust if (wbc && wbc->sync_mode == WB_SYNC_NONE) { 1950c4f24df9STrond Myklebust if (nscan < wbc->nr_to_write) 1951c4f24df9STrond Myklebust wbc->nr_to_write -= nscan; 1952c4f24df9STrond Myklebust else 1953c4f24df9STrond Myklebust wbc->nr_to_write = 0; 1954c4f24df9STrond Myklebust } 1955c4f24df9STrond Myklebust if (nscan < INT_MAX) 1956c4f24df9STrond Myklebust break; 1957c4f24df9STrond Myklebust cond_resched(); 1958c4f24df9STrond Myklebust } 1959af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 1960c4f24df9STrond Myklebust if (ret || !may_wait) 1961c4f24df9STrond Myklebust return ret; 1962c4f24df9STrond Myklebust return wait_on_commit(cinfo.mds); 1963c4f24df9STrond Myklebust } 1964c4f24df9STrond Myklebust 1965c4f24df9STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 1966c4f24df9STrond Myklebust { 1967c4f24df9STrond Myklebust return __nfs_commit_inode(inode, how, NULL); 19681da177e4SLinus Torvalds } 1969b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19708fc795f7STrond Myklebust 1971ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19728fc795f7STrond Myklebust { 1973420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1974420e3646STrond Myklebust int flags = FLUSH_SYNC; 1975420e3646STrond Myklebust int ret = 0; 19768fc795f7STrond Myklebust 1977c4f24df9STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 19783236c3e1SJeff Layton /* no commits means nothing needs to be done */ 19795cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 1980c4f24df9STrond Myklebust goto check_requests_outstanding; 19813236c3e1SJeff Layton 1982a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1983a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1984420e3646STrond Myklebust */ 19851a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1986420e3646STrond Myklebust goto out_mark_dirty; 1987420e3646STrond Myklebust 1988a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1989420e3646STrond Myklebust flags = 0; 1990a00dd6c0SJeff Layton } 1991a00dd6c0SJeff Layton 1992c4f24df9STrond Myklebust ret = __nfs_commit_inode(inode, flags, wbc); 1993c4f24df9STrond Myklebust if (!ret) { 1994c4f24df9STrond Myklebust if (flags & FLUSH_SYNC) 19958fc795f7STrond Myklebust return 0; 1996c4f24df9STrond Myklebust } else if (atomic_long_read(&nfsi->commit_info.ncommit)) 1997c4f24df9STrond Myklebust goto out_mark_dirty; 1998c4f24df9STrond Myklebust 1999c4f24df9STrond Myklebust check_requests_outstanding: 2000c4f24df9STrond Myklebust if (!atomic_read(&nfsi->commit_info.rpcs_out)) 2001c4f24df9STrond Myklebust return ret; 2002420e3646STrond Myklebust out_mark_dirty: 20038fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 20048fc795f7STrond Myklebust return ret; 20058fc795f7STrond Myklebust } 200689d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 2007863a3c6cSAndy Adamson 2008acdc53b2STrond Myklebust /* 2009837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 2010837bb1d7STrond Myklebust * 2011837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 2012837bb1d7STrond Myklebust * client. 2013837bb1d7STrond Myklebust */ 2014837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 2015837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 2016837bb1d7STrond Myklebust { 2017837bb1d7STrond Myklebust int ret; 2018837bb1d7STrond Myklebust 2019837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 2020837bb1d7STrond Myklebust if (ret == 0) 2021837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 2022837bb1d7STrond Myklebust return ret; 2023837bb1d7STrond Myklebust } 2024837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 2025837bb1d7STrond Myklebust 2026837bb1d7STrond Myklebust /* 2027acdc53b2STrond Myklebust * flush the inode to disk. 2028acdc53b2STrond Myklebust */ 2029acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 203034901f70STrond Myklebust { 2031f4ce1299STrond Myklebust int ret; 203234901f70STrond Myklebust 2033f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 2034f4ce1299STrond Myklebust 20355bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 20366b196875SChuck Lever if (ret) 20376b196875SChuck Lever goto out; 20385bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20396b196875SChuck Lever if (ret < 0) 20406b196875SChuck Lever goto out; 20415bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 20426b196875SChuck Lever ret = 0; 2043f4ce1299STrond Myklebust 20446b196875SChuck Lever out: 2045f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 2046f4ce1299STrond Myklebust return ret; 20471c75950bSTrond Myklebust } 2048ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 20491c75950bSTrond Myklebust 20506d740c76SMatthew Wilcox (Oracle) int nfs_wb_folio_cancel(struct inode *inode, struct folio *folio) 20511b3b4a1aSTrond Myklebust { 20521b3b4a1aSTrond Myklebust struct nfs_page *req; 20531b3b4a1aSTrond Myklebust int ret = 0; 20541b3b4a1aSTrond Myklebust 20556d740c76SMatthew Wilcox (Oracle) folio_wait_writeback(folio); 20563e217045SWeston Andros Adamson 20573e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20583e217045SWeston Andros Adamson * request */ 20590c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 20603e217045SWeston Andros Adamson 20613e217045SWeston Andros Adamson if (IS_ERR(req)) { 20623e217045SWeston Andros Adamson ret = PTR_ERR(req); 20633e217045SWeston Andros Adamson } else if (req) { 20646d740c76SMatthew Wilcox (Oracle) /* all requests from this folio have been cancelled by 20653e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20663e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20673e217045SWeston Andros Adamson * release it */ 20681b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20691d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20701b3b4a1aSTrond Myklebust } 20713e217045SWeston Andros Adamson 20721b3b4a1aSTrond Myklebust return ret; 20731b3b4a1aSTrond Myklebust } 20741b3b4a1aSTrond Myklebust 20755241060eSTrond Myklebust /** 20765241060eSTrond Myklebust * nfs_wb_folio - Write back all requests on one page 20775241060eSTrond Myklebust * @inode: pointer to page 20785241060eSTrond Myklebust * @folio: pointer to folio 20795241060eSTrond Myklebust * 20805241060eSTrond Myklebust * Assumes that the folio has been locked by the caller, and will 20815241060eSTrond Myklebust * not unlock it. 20821c75950bSTrond Myklebust */ 20835241060eSTrond Myklebust int nfs_wb_folio(struct inode *inode, struct folio *folio) 20841c75950bSTrond Myklebust { 20855241060eSTrond Myklebust loff_t range_start = folio_file_pos(folio); 20865241060eSTrond Myklebust loff_t range_end = range_start + (loff_t)folio_size(folio) - 1; 20877f2f12d9STrond Myklebust struct writeback_control wbc = { 20887f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20897f2f12d9STrond Myklebust .nr_to_write = 0, 20907f2f12d9STrond Myklebust .range_start = range_start, 20917f2f12d9STrond Myklebust .range_end = range_end, 20927f2f12d9STrond Myklebust }; 20937f2f12d9STrond Myklebust int ret; 20947f2f12d9STrond Myklebust 2095256093feSTrond Myklebust trace_nfs_writeback_folio(inode, folio); 2096f4ce1299STrond Myklebust 20970522f6adSTrond Myklebust for (;;) { 20985241060eSTrond Myklebust folio_wait_writeback(folio); 20995241060eSTrond Myklebust if (folio_clear_dirty_for_io(folio)) { 21000c493b5cSTrond Myklebust ret = nfs_writepage_locked(folio, &wbc); 21017f2f12d9STrond Myklebust if (ret < 0) 21027f2f12d9STrond Myklebust goto out_error; 21030522f6adSTrond Myklebust continue; 21047f2f12d9STrond Myklebust } 2105f4ce1299STrond Myklebust ret = 0; 21065241060eSTrond Myklebust if (!folio_test_private(folio)) 21070522f6adSTrond Myklebust break; 21080522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 21097f2f12d9STrond Myklebust if (ret < 0) 21107f2f12d9STrond Myklebust goto out_error; 21117f2f12d9STrond Myklebust } 21127f2f12d9STrond Myklebust out_error: 2113256093feSTrond Myklebust trace_nfs_writeback_folio_done(inode, folio, ret); 21147f2f12d9STrond Myklebust return ret; 21151c75950bSTrond Myklebust } 21161c75950bSTrond Myklebust 2117074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 21184ae84a80SMatthew Wilcox (Oracle) int nfs_migrate_folio(struct address_space *mapping, struct folio *dst, 21194ae84a80SMatthew Wilcox (Oracle) struct folio *src, enum migrate_mode mode) 2120074cc1deSTrond Myklebust { 21212da95652SJeff Layton /* 21224ae84a80SMatthew Wilcox (Oracle) * If the private flag is set, the folio is currently associated with 21232da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 21242da95652SJeff Layton * 21252da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 21262da95652SJeff Layton * that we can safely release the inode reference while holding 21274ae84a80SMatthew Wilcox (Oracle) * the folio lock. 21282da95652SJeff Layton */ 21294ae84a80SMatthew Wilcox (Oracle) if (folio_test_private(src)) 21302da95652SJeff Layton return -EBUSY; 2131074cc1deSTrond Myklebust 21322e9d7e4bSDavid Howells if (folio_test_private_2(src)) { /* [DEPRECATED] */ 213316f2f4e6SDavid Howells if (mode == MIGRATE_ASYNC) 21348c209ce7SDavid Howells return -EBUSY; 21352e9d7e4bSDavid Howells folio_wait_private_2(src); 213616f2f4e6SDavid Howells } 2137074cc1deSTrond Myklebust 213854184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 2139074cc1deSTrond Myklebust } 2140074cc1deSTrond Myklebust #endif 2141074cc1deSTrond Myklebust 2142f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 21431da177e4SLinus Torvalds { 21441da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 21451e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 21461da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 214720c2df83SPaul Mundt NULL); 21481da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 21491da177e4SLinus Torvalds return -ENOMEM; 21501da177e4SLinus Torvalds 215193d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 21521da177e4SLinus Torvalds nfs_wdata_cachep); 21531da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 21543dd4765fSJeff Layton goto out_destroy_write_cache; 21551da177e4SLinus Torvalds 21560b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 21570b7c0153SFred Isaman sizeof(struct nfs_commit_data), 21580b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21590b7c0153SFred Isaman NULL); 21600b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21613dd4765fSJeff Layton goto out_destroy_write_mempool; 21620b7c0153SFred Isaman 216393d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21644c100210SYanchuan Nian nfs_cdata_cachep); 21651da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21663dd4765fSJeff Layton goto out_destroy_commit_cache; 21671da177e4SLinus Torvalds 216889a09141SPeter Zijlstra /* 216989a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 217089a09141SPeter Zijlstra * 217189a09141SPeter Zijlstra * 64MB: 8192k 217289a09141SPeter Zijlstra * 128MB: 11585k 217389a09141SPeter Zijlstra * 256MB: 16384k 217489a09141SPeter Zijlstra * 512MB: 23170k 217589a09141SPeter Zijlstra * 1GB: 32768k 217689a09141SPeter Zijlstra * 2GB: 46340k 217789a09141SPeter Zijlstra * 4GB: 65536k 217889a09141SPeter Zijlstra * 8GB: 92681k 217989a09141SPeter Zijlstra * 16GB: 131072k 218089a09141SPeter Zijlstra * 218189a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 218289a09141SPeter Zijlstra * Limit the default to 256M 218389a09141SPeter Zijlstra */ 2184ca79b0c2SArun KS nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); 218589a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 218689a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 218789a09141SPeter Zijlstra 21881da177e4SLinus Torvalds return 0; 21893dd4765fSJeff Layton 21903dd4765fSJeff Layton out_destroy_commit_cache: 21913dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21923dd4765fSJeff Layton out_destroy_write_mempool: 21933dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21943dd4765fSJeff Layton out_destroy_write_cache: 21953dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21963dd4765fSJeff Layton return -ENOMEM; 21971da177e4SLinus Torvalds } 21981da177e4SLinus Torvalds 2199266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 22001da177e4SLinus Torvalds { 22011da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 22023dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 22031da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 22041a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 22051da177e4SLinus Torvalds } 22061da177e4SLinus Torvalds 22074a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 22084a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 22094a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 22100eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 22110eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 22121ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 22134a0de55cSAnna Schumaker }; 2214