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> 283fcfab16SAndrew Morton 297c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 30875bc3fbSTrond Myklebust #include <linux/sched/mm.h> 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds #include "delegation.h" 3349a70f27STrond Myklebust #include "internal.h" 3491d5b470SChuck Lever #include "iostat.h" 35def6ed7eSAndy Adamson #include "nfs4_fs.h" 36074cc1deSTrond Myklebust #include "fscache.h" 3794ad1c80SFred Isaman #include "pnfs.h" 381da177e4SLinus Torvalds 39f4ce1299STrond Myklebust #include "nfstrace.h" 40f4ce1299STrond Myklebust 411da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 421da177e4SLinus Torvalds 431da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 441da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 451da177e4SLinus Torvalds 46919e3bd9STrond Myklebust struct nfs_io_completion { 47919e3bd9STrond Myklebust void (*complete)(void *data); 48919e3bd9STrond Myklebust void *data; 49919e3bd9STrond Myklebust struct kref refcount; 50919e3bd9STrond Myklebust }; 51919e3bd9STrond Myklebust 521da177e4SLinus Torvalds /* 531da177e4SLinus Torvalds * Local function declarations 541da177e4SLinus Torvalds */ 55f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 56788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 57061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 58f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 594a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 6006c9fdf3STrond Myklebust static void nfs_inode_remove_request(struct nfs_page *req); 61d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 6202d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 6302d1426cSWeston Andros Adamson struct inode *inode); 643a3908c8STrond Myklebust static struct nfs_page * 653a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 660c493b5cSTrond Myklebust struct folio *folio); 671da177e4SLinus Torvalds 68e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 693feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 700b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 711da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 721da177e4SLinus Torvalds 73515dcdcdSTrond Myklebust struct nfs_commit_data *nfs_commitdata_alloc(void) 741da177e4SLinus Torvalds { 75518662e0SNeilBrown struct nfs_commit_data *p; 7640859d7eSChuck Lever 77515dcdcdSTrond Myklebust p = kmem_cache_zalloc(nfs_cdata_cachep, nfs_io_gfp_mask()); 78515dcdcdSTrond Myklebust if (!p) { 79518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); 80518662e0SNeilBrown if (!p) 81518662e0SNeilBrown return NULL; 821da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 83515dcdcdSTrond Myklebust } 841da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 851da177e4SLinus Torvalds return p; 861da177e4SLinus Torvalds } 87e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 881da177e4SLinus Torvalds 890b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 901da177e4SLinus Torvalds { 911da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 921da177e4SLinus Torvalds } 93e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 941da177e4SLinus Torvalds 951e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 963feb2d49STrond Myklebust { 970bae835bSTrond Myklebust struct nfs_pgio_header *p; 983feb2d49STrond Myklebust 990bae835bSTrond Myklebust p = kmem_cache_zalloc(nfs_wdata_cachep, nfs_io_gfp_mask()); 1000bae835bSTrond Myklebust if (!p) { 1010bae835bSTrond Myklebust p = mempool_alloc(nfs_wdata_mempool, GFP_NOWAIT); 1020bae835bSTrond Myklebust if (!p) 1030bae835bSTrond Myklebust return NULL; 1043feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 1050bae835bSTrond Myklebust } 106fbe77c30SBenjamin Coddington p->rw_mode = FMODE_WRITE; 1073feb2d49STrond Myklebust return p; 1083feb2d49STrond Myklebust } 1093feb2d49STrond Myklebust 1101e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 1116c75dc0dSFred Isaman { 1121e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 1133feb2d49STrond Myklebust } 1141da177e4SLinus Torvalds 115919e3bd9STrond Myklebust static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) 116919e3bd9STrond Myklebust { 117919e3bd9STrond Myklebust return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); 118919e3bd9STrond Myklebust } 119919e3bd9STrond Myklebust 120919e3bd9STrond Myklebust static void nfs_io_completion_init(struct nfs_io_completion *ioc, 121919e3bd9STrond Myklebust void (*complete)(void *), void *data) 122919e3bd9STrond Myklebust { 123919e3bd9STrond Myklebust ioc->complete = complete; 124919e3bd9STrond Myklebust ioc->data = data; 125919e3bd9STrond Myklebust kref_init(&ioc->refcount); 126919e3bd9STrond Myklebust } 127919e3bd9STrond Myklebust 128919e3bd9STrond Myklebust static void nfs_io_completion_release(struct kref *kref) 129919e3bd9STrond Myklebust { 130919e3bd9STrond Myklebust struct nfs_io_completion *ioc = container_of(kref, 131919e3bd9STrond Myklebust struct nfs_io_completion, refcount); 132919e3bd9STrond Myklebust ioc->complete(ioc->data); 133919e3bd9STrond Myklebust kfree(ioc); 134919e3bd9STrond Myklebust } 135919e3bd9STrond Myklebust 136919e3bd9STrond Myklebust static void nfs_io_completion_get(struct nfs_io_completion *ioc) 137919e3bd9STrond Myklebust { 138919e3bd9STrond Myklebust if (ioc != NULL) 139919e3bd9STrond Myklebust kref_get(&ioc->refcount); 140919e3bd9STrond Myklebust } 141919e3bd9STrond Myklebust 142919e3bd9STrond Myklebust static void nfs_io_completion_put(struct nfs_io_completion *ioc) 143919e3bd9STrond Myklebust { 144919e3bd9STrond Myklebust if (ioc != NULL) 145919e3bd9STrond Myklebust kref_put(&ioc->refcount, nfs_io_completion_release); 146919e3bd9STrond Myklebust } 147919e3bd9STrond Myklebust 148e00ed89dSTrond Myklebust static void 149e00ed89dSTrond Myklebust nfs_page_set_inode_ref(struct nfs_page *req, struct inode *inode) 150e00ed89dSTrond Myklebust { 151e00ed89dSTrond Myklebust if (!test_and_set_bit(PG_INODE_REF, &req->wb_flags)) { 152e00ed89dSTrond Myklebust kref_get(&req->wb_kref); 153e00ed89dSTrond Myklebust atomic_long_inc(&NFS_I(inode)->nrequests); 154e00ed89dSTrond Myklebust } 155e00ed89dSTrond Myklebust } 156e00ed89dSTrond Myklebust 157e00ed89dSTrond Myklebust static int 158e00ed89dSTrond Myklebust nfs_cancel_remove_inode(struct nfs_page *req, struct inode *inode) 159e00ed89dSTrond Myklebust { 160e00ed89dSTrond Myklebust int ret; 161e00ed89dSTrond Myklebust 162e00ed89dSTrond Myklebust if (!test_bit(PG_REMOVE, &req->wb_flags)) 163e00ed89dSTrond Myklebust return 0; 164e00ed89dSTrond Myklebust ret = nfs_page_group_lock(req); 165e00ed89dSTrond Myklebust if (ret) 166e00ed89dSTrond Myklebust return ret; 167e00ed89dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &req->wb_flags)) 168e00ed89dSTrond Myklebust nfs_page_set_inode_ref(req, inode); 169e00ed89dSTrond Myklebust nfs_page_group_unlock(req); 170e00ed89dSTrond Myklebust return 0; 171e00ed89dSTrond Myklebust } 172e00ed89dSTrond Myklebust 1730c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_private_request(struct folio *folio) 174bd37d6fcSTrond Myklebust { 1750c493b5cSTrond Myklebust return folio_get_private(folio); 176bd37d6fcSTrond Myklebust } 177bd37d6fcSTrond Myklebust 1780c493b5cSTrond Myklebust /** 1790c493b5cSTrond Myklebust * nfs_folio_find_private_request - find head request associated with a folio 1800c493b5cSTrond Myklebust * @folio: pointer to folio 18184d3a9a9SWeston Andros Adamson * 18284d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 18384d3a9a9SWeston Andros Adamson * 18484d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 18584d3a9a9SWeston Andros Adamson */ 1860c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_find_private_request(struct folio *folio) 187277459d2STrond Myklebust { 1880c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 189bd37d6fcSTrond Myklebust struct nfs_page *req; 190277459d2STrond Myklebust 1910c493b5cSTrond Myklebust if (!folio_test_private(folio)) 192b30d2f04STrond Myklebust return NULL; 1934b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 1940c493b5cSTrond Myklebust req = nfs_folio_private_request(folio); 19584d3a9a9SWeston Andros Adamson if (req) { 19684d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 19729418aa4SMel Gorman kref_get(&req->wb_kref); 19884d3a9a9SWeston Andros Adamson } 1994b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 200b30d2f04STrond Myklebust return req; 201b30d2f04STrond Myklebust } 20229418aa4SMel Gorman 2030c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_find_swap_request(struct folio *folio) 204b30d2f04STrond Myklebust { 2050c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 206b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 207b30d2f04STrond Myklebust struct nfs_page *req = NULL; 2080c493b5cSTrond Myklebust if (!folio_test_swapcache(folio)) 209b30d2f04STrond Myklebust return NULL; 210e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 2110c493b5cSTrond Myklebust if (folio_test_swapcache(folio)) { 212b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 2130c493b5cSTrond Myklebust folio); 214b30d2f04STrond Myklebust if (req) { 215b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 216b30d2f04STrond Myklebust kref_get(&req->wb_kref); 217b30d2f04STrond Myklebust } 218b30d2f04STrond Myklebust } 219e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 220277459d2STrond Myklebust return req; 221277459d2STrond Myklebust } 222277459d2STrond Myklebust 2230c493b5cSTrond Myklebust /** 2240c493b5cSTrond Myklebust * nfs_folio_find_head_request - find head request associated with a folio 2250c493b5cSTrond Myklebust * @folio: pointer to folio 22684d3a9a9SWeston Andros Adamson * 22784d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 22884d3a9a9SWeston Andros Adamson */ 2290c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_find_head_request(struct folio *folio) 230277459d2STrond Myklebust { 231b30d2f04STrond Myklebust struct nfs_page *req; 232277459d2STrond Myklebust 2330c493b5cSTrond Myklebust req = nfs_folio_find_private_request(folio); 234b30d2f04STrond Myklebust if (!req) 2350c493b5cSTrond Myklebust req = nfs_folio_find_swap_request(folio); 236277459d2STrond Myklebust return req; 237277459d2STrond Myklebust } 238277459d2STrond Myklebust 2390c493b5cSTrond Myklebust static struct nfs_page *nfs_folio_find_and_lock_request(struct folio *folio) 240e00ed89dSTrond Myklebust { 2410c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 242e00ed89dSTrond Myklebust struct nfs_page *req, *head; 243e00ed89dSTrond Myklebust int ret; 244e00ed89dSTrond Myklebust 245e00ed89dSTrond Myklebust for (;;) { 2460c493b5cSTrond Myklebust req = nfs_folio_find_head_request(folio); 247e00ed89dSTrond Myklebust if (!req) 248e00ed89dSTrond Myklebust return req; 249e00ed89dSTrond Myklebust head = nfs_page_group_lock_head(req); 250e00ed89dSTrond Myklebust if (head != req) 251e00ed89dSTrond Myklebust nfs_release_request(req); 252e00ed89dSTrond Myklebust if (IS_ERR(head)) 253e00ed89dSTrond Myklebust return head; 254e00ed89dSTrond Myklebust ret = nfs_cancel_remove_inode(head, inode); 255e00ed89dSTrond Myklebust if (ret < 0) { 256e00ed89dSTrond Myklebust nfs_unlock_and_release_request(head); 257e00ed89dSTrond Myklebust return ERR_PTR(ret); 258e00ed89dSTrond Myklebust } 259e00ed89dSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 2600c493b5cSTrond Myklebust if (head == nfs_folio_private_request(folio)) 261e00ed89dSTrond Myklebust break; 2620c493b5cSTrond Myklebust if (folio_test_swapcache(folio)) 263e00ed89dSTrond Myklebust break; 264e00ed89dSTrond Myklebust nfs_unlock_and_release_request(head); 265e00ed89dSTrond Myklebust } 266e00ed89dSTrond Myklebust return head; 267e00ed89dSTrond Myklebust } 268e00ed89dSTrond Myklebust 2691da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2700c493b5cSTrond Myklebust static void nfs_grow_file(struct folio *folio, unsigned int offset, 2710c493b5cSTrond Myklebust unsigned int count) 2721da177e4SLinus Torvalds { 2730c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 274a3d01454STrond Myklebust loff_t end, i_size; 275a3d01454STrond Myklebust pgoff_t end_index; 2761da177e4SLinus Torvalds 277a3d01454STrond Myklebust spin_lock(&inode->i_lock); 278a3d01454STrond Myklebust i_size = i_size_read(inode); 2790c493b5cSTrond Myklebust end_index = ((i_size - 1) >> folio_shift(folio)) << folio_order(folio); 2800c493b5cSTrond Myklebust if (i_size > 0 && folio_index(folio) < end_index) 281a3d01454STrond Myklebust goto out; 2820c493b5cSTrond Myklebust end = folio_file_pos(folio) + (loff_t)offset + (loff_t)count; 2831da177e4SLinus Torvalds if (i_size >= end) 284a3d01454STrond Myklebust goto out; 285110cb2d2SChuck Lever trace_nfs_size_grow(inode, end); 2861da177e4SLinus Torvalds i_size_write(inode, end); 287f6cdfa6dSTrond Myklebust NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 288a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 289a3d01454STrond Myklebust out: 290a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 291a6b5a28eSDave Wysochanski nfs_fscache_invalidate(inode, 0); 2921da177e4SLinus Torvalds } 2931da177e4SLinus Torvalds 294a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 295d2ceb7e5SBenjamin Coddington static void nfs_set_pageerror(struct address_space *mapping) 296a301b777STrond Myklebust { 2970df68cedSTrond Myklebust struct inode *inode = mapping->host; 2980df68cedSTrond Myklebust 299d2ceb7e5SBenjamin Coddington nfs_zap_mapping(mapping->host, mapping); 3000df68cedSTrond Myklebust /* Force file size revalidation */ 3010df68cedSTrond Myklebust spin_lock(&inode->i_lock); 302ac46b3d7STrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_REVAL_FORCED | 30388a6099fSTrond Myklebust NFS_INO_INVALID_CHANGE | 304ac46b3d7STrond Myklebust NFS_INO_INVALID_SIZE); 3050df68cedSTrond Myklebust spin_unlock(&inode->i_lock); 306a301b777STrond Myklebust } 307a301b777STrond Myklebust 3080c493b5cSTrond Myklebust static void nfs_mapping_set_error(struct folio *folio, int error) 3096fbda89bSTrond Myklebust { 3100c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 311b8946d7bSTrond Myklebust 3120c493b5cSTrond Myklebust folio_set_error(folio); 3136c984083STrond Myklebust filemap_set_wb_err(mapping, error); 3146c984083STrond Myklebust if (mapping->host) 3156c984083STrond Myklebust errseq_set(&mapping->host->i_sb->s_wb_err, 3166c984083STrond Myklebust error == -ENOSPC ? -ENOSPC : -EIO); 317b8946d7bSTrond Myklebust nfs_set_pageerror(mapping); 3186fbda89bSTrond Myklebust } 3196fbda89bSTrond Myklebust 320d72ddcbaSWeston Andros Adamson /* 321d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 322d72ddcbaSWeston Andros Adamson * @head - head request of page group 323d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 324d72ddcbaSWeston Andros Adamson * 325d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 326d72ddcbaSWeston Andros Adamson * page offset @page_offset. 327d72ddcbaSWeston Andros Adamson * 328d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 329d72ddcbaSWeston Andros Adamson * match is found. 330d72ddcbaSWeston Andros Adamson * 331d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 332d72ddcbaSWeston Andros Adamson */ 333d72ddcbaSWeston Andros Adamson static struct nfs_page * 334d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 335d72ddcbaSWeston Andros Adamson { 336d72ddcbaSWeston Andros Adamson struct nfs_page *req; 337d72ddcbaSWeston Andros Adamson 338d72ddcbaSWeston Andros Adamson req = head; 339d72ddcbaSWeston Andros Adamson do { 340d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 341d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 342d72ddcbaSWeston Andros Adamson return req; 343d72ddcbaSWeston Andros Adamson 344d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 345d72ddcbaSWeston Andros Adamson } while (req != head); 346d72ddcbaSWeston Andros Adamson 347d72ddcbaSWeston Andros Adamson return NULL; 348d72ddcbaSWeston Andros Adamson } 349d72ddcbaSWeston Andros Adamson 350d72ddcbaSWeston Andros Adamson /* 351d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 352d72ddcbaSWeston Andros Adamson * @head - head request of page group 353d72ddcbaSWeston Andros Adamson * 354d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 355d72ddcbaSWeston Andros Adamson * returns false otherwise 356d72ddcbaSWeston Andros Adamson */ 357d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 358d72ddcbaSWeston Andros Adamson { 3590c493b5cSTrond Myklebust unsigned int len = nfs_folio_length(nfs_page_to_folio(req)); 360d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 361d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 362d72ddcbaSWeston Andros Adamson 3631344b7eaSTrond Myklebust nfs_page_group_lock(req); 364d72ddcbaSWeston Andros Adamson 3657e8a30f8STrond Myklebust for (;;) { 366d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 3677e8a30f8STrond Myklebust if (!tmp) 3687e8a30f8STrond Myklebust break; 3697e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 370d72ddcbaSWeston Andros Adamson } 371d72ddcbaSWeston Andros Adamson 372d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3737e8a30f8STrond Myklebust return pos >= len; 374d72ddcbaSWeston Andros Adamson } 375d72ddcbaSWeston Andros Adamson 3761da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3771da177e4SLinus Torvalds * covers the full page. 3781da177e4SLinus Torvalds */ 379d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3801da177e4SLinus Torvalds { 3810c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 3820c493b5cSTrond Myklebust 3830c493b5cSTrond Myklebust if (folio_test_uptodate(folio)) 3841da177e4SLinus Torvalds return; 385d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3861da177e4SLinus Torvalds return; 3870c493b5cSTrond Myklebust folio_mark_uptodate(folio); 3881da177e4SLinus Torvalds } 3891da177e4SLinus Torvalds 3901da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3911da177e4SLinus Torvalds { 392e87b4c7aSNeilBrown int ret = 0; 393cca588d6STrond Myklebust 394e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 395e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 396e87b4c7aSNeilBrown return ret; 3971da177e4SLinus Torvalds } 3981da177e4SLinus Torvalds 3991da177e4SLinus Torvalds /* 40089a09141SPeter Zijlstra * NFS congestion control 40189a09141SPeter Zijlstra */ 40289a09141SPeter Zijlstra 40389a09141SPeter Zijlstra int nfs_congestion_kb; 40489a09141SPeter Zijlstra 40589a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 40689a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 40789a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 40889a09141SPeter Zijlstra 4090c493b5cSTrond Myklebust static void nfs_folio_set_writeback(struct folio *folio) 41089a09141SPeter Zijlstra { 4110c493b5cSTrond Myklebust struct nfs_server *nfss = NFS_SERVER(folio_file_mapping(folio)->host); 4125a6d41b3STrond Myklebust 4130c493b5cSTrond Myklebust folio_start_writeback(folio); 4140c493b5cSTrond Myklebust if (atomic_long_inc_return(&nfss->writeback) > NFS_CONGESTION_ON_THRESH) 4156df25e58SNeilBrown nfss->write_congested = 1; 41689a09141SPeter Zijlstra } 41789a09141SPeter Zijlstra 4180c493b5cSTrond Myklebust static void nfs_folio_end_writeback(struct folio *folio) 41989a09141SPeter Zijlstra { 4200c493b5cSTrond Myklebust struct nfs_server *nfss = NFS_SERVER(folio_file_mapping(folio)->host); 42189a09141SPeter Zijlstra 4220c493b5cSTrond Myklebust folio_end_writeback(folio); 4230c493b5cSTrond Myklebust if (atomic_long_dec_return(&nfss->writeback) < 4240c493b5cSTrond Myklebust NFS_CONGESTION_OFF_THRESH) 4256df25e58SNeilBrown nfss->write_congested = 0; 42689a09141SPeter Zijlstra } 42789a09141SPeter Zijlstra 4280c493b5cSTrond Myklebust static void nfs_page_end_writeback(struct nfs_page *req) 4290c493b5cSTrond Myklebust { 4300c493b5cSTrond Myklebust if (nfs_page_group_sync_on_bit(req, PG_WB_END)) { 4310c493b5cSTrond Myklebust nfs_unlock_request(req); 4320c493b5cSTrond Myklebust nfs_folio_end_writeback(nfs_page_to_folio(req)); 4330c493b5cSTrond Myklebust } else 4340c493b5cSTrond Myklebust nfs_unlock_request(req); 4350c493b5cSTrond Myklebust } 4360c493b5cSTrond Myklebust 437d4581383SWeston Andros Adamson /* 438d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 439d4581383SWeston Andros Adamson * 440d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 441d4581383SWeston Andros Adamson * @old_head - the old head of the list 442d4581383SWeston Andros Adamson * 443d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 444d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 445d4581383SWeston Andros Adamson * with a reference held by some other context. 446d4581383SWeston Andros Adamson */ 447d4581383SWeston Andros Adamson static void 448d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 449b66aaa8dSTrond Myklebust struct nfs_page *old_head, 450b66aaa8dSTrond Myklebust struct inode *inode) 451d4581383SWeston Andros Adamson { 452d4581383SWeston Andros Adamson while (destroy_list) { 453d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 454d4581383SWeston Andros Adamson 455d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 456d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 457d4581383SWeston Andros Adamson 45808ca8b21STrond Myklebust /* Note: lock subreq in order to change subreq->wb_head */ 45908ca8b21STrond Myklebust nfs_page_set_headlock(subreq); 460d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 461d4581383SWeston Andros Adamson 462d4581383SWeston Andros Adamson /* make sure old group is not used */ 463d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 46408ca8b21STrond Myklebust subreq->wb_head = subreq; 465d4581383SWeston Andros Adamson 466902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 467902a4c00STrond Myklebust 4685b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4695b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4705b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 47108ca8b21STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) { 47208ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 4735b2b5187STrond Myklebust nfs_free_request(subreq); 47408ca8b21STrond Myklebust } else 47508ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 4765b2b5187STrond Myklebust continue; 4775b2b5187STrond Myklebust } 47808ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 479d4581383SWeston Andros Adamson 480add42de3STrond Myklebust nfs_release_request(old_head); 4815b2b5187STrond Myklebust 482b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 483d4581383SWeston Andros Adamson nfs_release_request(subreq); 484a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 485d4581383SWeston Andros Adamson } 4865b2b5187STrond Myklebust 4875b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4885b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4895b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 490d4581383SWeston Andros Adamson } 491d4581383SWeston Andros Adamson } 492d4581383SWeston Andros Adamson 493d4581383SWeston Andros Adamson /* 494e00ed89dSTrond Myklebust * nfs_join_page_group - destroy subrequests of the head req 495e00ed89dSTrond Myklebust * @head: the page used to lookup the "page group" of nfs_page structures 496e00ed89dSTrond Myklebust * @inode: Inode to which the request belongs. 497d4581383SWeston Andros Adamson * 498d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 499d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 500d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 501d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 502d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 503d4581383SWeston Andros Adamson */ 504ed5d588fSTrond Myklebust void 505e00ed89dSTrond Myklebust nfs_join_page_group(struct nfs_page *head, struct inode *inode) 506d4581383SWeston Andros Adamson { 507e00ed89dSTrond Myklebust struct nfs_page *subreq; 508d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 509a62f8e3bSTrond Myklebust unsigned int pgbase, off, bytes; 510a62f8e3bSTrond Myklebust 511a62f8e3bSTrond Myklebust pgbase = head->wb_pgbase; 512a62f8e3bSTrond Myklebust bytes = head->wb_bytes; 513a62f8e3bSTrond Myklebust off = head->wb_offset; 514a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 515a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 516a62f8e3bSTrond Myklebust /* Subrequests should always form a contiguous range */ 517a62f8e3bSTrond Myklebust if (pgbase > subreq->wb_pgbase) { 518a62f8e3bSTrond Myklebust off -= pgbase - subreq->wb_pgbase; 519a62f8e3bSTrond Myklebust bytes += pgbase - subreq->wb_pgbase; 520a62f8e3bSTrond Myklebust pgbase = subreq->wb_pgbase; 521a62f8e3bSTrond Myklebust } 522a62f8e3bSTrond Myklebust bytes = max(subreq->wb_pgbase + subreq->wb_bytes 523a62f8e3bSTrond Myklebust - pgbase, bytes); 5241bd5d6d0STrond Myklebust } 5251bd5d6d0STrond Myklebust 526a62f8e3bSTrond Myklebust /* Set the head request's range to cover the former page group */ 527a62f8e3bSTrond Myklebust head->wb_pgbase = pgbase; 528a62f8e3bSTrond Myklebust head->wb_bytes = bytes; 529a62f8e3bSTrond Myklebust head->wb_offset = off; 530d4581383SWeston Andros Adamson 531d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 532d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 533d4581383SWeston Andros Adamson subreq = head; 534d4581383SWeston Andros Adamson do { 535411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 536d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 537d4581383SWeston Andros Adamson } while (subreq != head); 538d4581383SWeston Andros Adamson 539d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 540d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 541d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 542d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 543d4581383SWeston Andros Adamson head->wb_this_page = head; 544d4581383SWeston Andros Adamson } 545d4581383SWeston Andros Adamson 546b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 547b5bab9bfSTrond Myklebust } 548b5bab9bfSTrond Myklebust 549e00ed89dSTrond Myklebust /* 550e00ed89dSTrond Myklebust * nfs_lock_and_join_requests - join all subreqs to the head req 5510c493b5cSTrond Myklebust * @folio: the folio used to lookup the "page group" of nfs_page structures 552e00ed89dSTrond Myklebust * 553e00ed89dSTrond Myklebust * This function joins all sub requests to the head request by first 554e00ed89dSTrond Myklebust * locking all requests in the group, cancelling any pending operations 555e00ed89dSTrond Myklebust * and finally updating the head request to cover the whole range covered by 556e00ed89dSTrond Myklebust * the (former) group. All subrequests are removed from any write or commit 557e00ed89dSTrond Myklebust * lists, unlinked from the group and destroyed. 558e00ed89dSTrond Myklebust * 559e00ed89dSTrond Myklebust * Returns a locked, referenced pointer to the head request - which after 560e00ed89dSTrond Myklebust * this call is guaranteed to be the only request associated with the page. 5610c493b5cSTrond Myklebust * Returns NULL if no requests are found for @folio, or a ERR_PTR if an 562e00ed89dSTrond Myklebust * error was encountered. 563e00ed89dSTrond Myklebust */ 5640c493b5cSTrond Myklebust static struct nfs_page *nfs_lock_and_join_requests(struct folio *folio) 565e00ed89dSTrond Myklebust { 5660c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 567e00ed89dSTrond Myklebust struct nfs_page *head; 568e00ed89dSTrond Myklebust int ret; 569e00ed89dSTrond Myklebust 570e00ed89dSTrond Myklebust /* 571e00ed89dSTrond Myklebust * A reference is taken only on the head request which acts as a 572e00ed89dSTrond Myklebust * reference to the whole page group - the group will not be destroyed 573e00ed89dSTrond Myklebust * until the head reference is released. 574e00ed89dSTrond Myklebust */ 5750c493b5cSTrond Myklebust head = nfs_folio_find_and_lock_request(folio); 576e00ed89dSTrond Myklebust if (IS_ERR_OR_NULL(head)) 577d4581383SWeston Andros Adamson return head; 5780671d8f1SMarkus Elfring 579e00ed89dSTrond Myklebust /* lock each request in the page group */ 580e00ed89dSTrond Myklebust ret = nfs_page_group_lock_subrequests(head); 581e00ed89dSTrond Myklebust if (ret < 0) { 5820671d8f1SMarkus Elfring nfs_unlock_and_release_request(head); 5830671d8f1SMarkus Elfring return ERR_PTR(ret); 584612c9384STrond Myklebust } 585074cc1deSTrond Myklebust 586e00ed89dSTrond Myklebust nfs_join_page_group(head, inode); 587e00ed89dSTrond Myklebust 588e00ed89dSTrond Myklebust return head; 589e00ed89dSTrond Myklebust } 590e00ed89dSTrond Myklebust 5916fbda89bSTrond Myklebust static void nfs_write_error(struct nfs_page *req, int error) 5920bcbf039SPeng Tao { 5936dd85e83STrond Myklebust trace_nfs_write_error(nfs_page_to_inode(req), req, error); 5940c493b5cSTrond Myklebust nfs_mapping_set_error(nfs_page_to_folio(req), error); 59506c9fdf3STrond Myklebust nfs_inode_remove_request(req); 5960c493b5cSTrond Myklebust nfs_page_end_writeback(req); 5971f84ccdfSFred Isaman nfs_release_request(req); 5980bcbf039SPeng Tao } 5990bcbf039SPeng Tao 600074cc1deSTrond Myklebust /* 601074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 602074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 603074cc1deSTrond Myklebust */ 6040c493b5cSTrond Myklebust static int nfs_page_async_flush(struct folio *folio, 605c6fd3511STrond Myklebust struct writeback_control *wbc, 606c6fd3511STrond Myklebust struct nfs_pageio_descriptor *pgio) 607074cc1deSTrond Myklebust { 608074cc1deSTrond Myklebust struct nfs_page *req; 609074cc1deSTrond Myklebust int ret = 0; 610074cc1deSTrond Myklebust 6110c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 612074cc1deSTrond Myklebust if (!req) 613074cc1deSTrond Myklebust goto out; 614074cc1deSTrond Myklebust ret = PTR_ERR(req); 615074cc1deSTrond Myklebust if (IS_ERR(req)) 616074cc1deSTrond Myklebust goto out; 617074cc1deSTrond Myklebust 6180c493b5cSTrond Myklebust nfs_folio_set_writeback(folio); 619deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 620074cc1deSTrond Myklebust 621a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 62296c41455STrond Myklebust ret = pgio->pg_error; 62396c41455STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 624a6598813STrond Myklebust goto out_launder; 625a6598813STrond Myklebust 62696c41455STrond Myklebust ret = 0; 627f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 628074cc1deSTrond Myklebust ret = pgio->pg_error; 6290bcbf039SPeng Tao /* 630c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6310bcbf039SPeng Tao */ 632c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 633a6598813STrond Myklebust goto out_launder; 634c6fd3511STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) 635c6fd3511STrond Myklebust ret = AOP_WRITEPAGE_ACTIVATE; 6360c493b5cSTrond Myklebust folio_redirty_for_writepage(wbc, folio); 6378fc75bedSTrond Myklebust nfs_redirty_request(req); 63896c41455STrond Myklebust pgio->pg_error = 0; 63940f90271STrond Myklebust } else 6400c493b5cSTrond Myklebust nfs_add_stats(folio_file_mapping(folio)->host, 64140f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 642074cc1deSTrond Myklebust out: 643074cc1deSTrond Myklebust return ret; 644a6598813STrond Myklebust out_launder: 6456fbda89bSTrond Myklebust nfs_write_error(req, ret); 64614bebe3cSTrond Myklebust return 0; 647e261f51fSTrond Myklebust } 648e261f51fSTrond Myklebust 6490c493b5cSTrond Myklebust static int nfs_do_writepage(struct folio *folio, struct writeback_control *wbc, 650c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 651f758c885STrond Myklebust { 6520c493b5cSTrond Myklebust nfs_pageio_cond_complete(pgio, folio_index(folio)); 6530c493b5cSTrond Myklebust return nfs_page_async_flush(folio, wbc, pgio); 654f758c885STrond Myklebust } 655f758c885STrond Myklebust 656e261f51fSTrond Myklebust /* 6571da177e4SLinus Torvalds * Write an mmapped page to the server. 6581da177e4SLinus Torvalds */ 6590c493b5cSTrond Myklebust static int nfs_writepage_locked(struct folio *folio, 660c373fff7STrond Myklebust struct writeback_control *wbc) 6611da177e4SLinus Torvalds { 662f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 6630c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 664e261f51fSTrond Myklebust int err; 6651da177e4SLinus Torvalds 6666df25e58SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 6676df25e58SNeilBrown NFS_SERVER(inode)->write_congested) 6686df25e58SNeilBrown return AOP_WRITEPAGE_ACTIVATE; 6696df25e58SNeilBrown 67040f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 6710c493b5cSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, false, 6720c493b5cSTrond Myklebust &nfs_async_write_completion_ops); 6730c493b5cSTrond Myklebust err = nfs_do_writepage(folio, wbc, &pgio); 67496c41455STrond Myklebust pgio.pg_error = 0; 675f758c885STrond Myklebust nfs_pageio_complete(&pgio); 6764d770ccfSTrond Myklebust return err; 6774d770ccfSTrond Myklebust } 6784d770ccfSTrond Myklebust 6794d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6804d770ccfSTrond Myklebust { 6810c493b5cSTrond Myklebust struct folio *folio = page_folio(page); 682f758c885STrond Myklebust int ret; 6834d770ccfSTrond Myklebust 6840c493b5cSTrond Myklebust ret = nfs_writepage_locked(folio, wbc); 68596c41455STrond Myklebust if (ret != AOP_WRITEPAGE_ACTIVATE) 6861da177e4SLinus Torvalds unlock_page(page); 687f758c885STrond Myklebust return ret; 688f758c885STrond Myklebust } 689f758c885STrond Myklebust 6900c493b5cSTrond Myklebust static int nfs_writepages_callback(struct page *page, 6910c493b5cSTrond Myklebust struct writeback_control *wbc, void *data) 692f758c885STrond Myklebust { 6930c493b5cSTrond Myklebust struct folio *folio = page_folio(page); 694f758c885STrond Myklebust int ret; 695f758c885STrond Myklebust 6960c493b5cSTrond Myklebust ret = nfs_do_writepage(folio, wbc, data); 69796c41455STrond Myklebust if (ret != AOP_WRITEPAGE_ACTIVATE) 698f758c885STrond Myklebust unlock_page(page); 699f758c885STrond Myklebust return ret; 7001da177e4SLinus Torvalds } 7011da177e4SLinus Torvalds 702919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 703919e3bd9STrond Myklebust { 704919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 705919e3bd9STrond Myklebust } 706919e3bd9STrond Myklebust 7071da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7081da177e4SLinus Torvalds { 7091da177e4SLinus Torvalds struct inode *inode = mapping->host; 710c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 711ed7bcdb3STrond Myklebust struct nfs_io_completion *ioc = NULL; 712ed7bcdb3STrond Myklebust unsigned int mntflags = NFS_SERVER(inode)->flags; 713ed7bcdb3STrond Myklebust int priority = 0; 7141da177e4SLinus Torvalds int err; 7151da177e4SLinus Torvalds 7166df25e58SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 7176df25e58SNeilBrown NFS_SERVER(inode)->write_congested) 7186df25e58SNeilBrown return 0; 7196df25e58SNeilBrown 72091d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 72191d5b470SChuck Lever 722ed7bcdb3STrond Myklebust if (!(mntflags & NFS_MOUNT_WRITE_EAGER) || wbc->for_kupdate || 723ed7bcdb3STrond Myklebust wbc->for_background || wbc->for_sync || wbc->for_reclaim) { 7242b17d725STrond Myklebust ioc = nfs_io_completion_alloc(GFP_KERNEL); 725919e3bd9STrond Myklebust if (ioc) 726ed7bcdb3STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, 727ed7bcdb3STrond Myklebust inode); 728ed7bcdb3STrond Myklebust priority = wb_priority(wbc); 729ed7bcdb3STrond Myklebust } 730919e3bd9STrond Myklebust 731c6fd3511STrond Myklebust do { 732ed7bcdb3STrond Myklebust nfs_pageio_init_write(&pgio, inode, priority, false, 733a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 734919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 735c6fd3511STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, 736c6fd3511STrond Myklebust &pgio); 73796c41455STrond Myklebust pgio.pg_error = 0; 738c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 739c6fd3511STrond Myklebust } while (err < 0 && !nfs_error_is_fatal(err)); 740919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 74172cb77f4STrond Myklebust 742f758c885STrond Myklebust if (err < 0) 74372cb77f4STrond Myklebust goto out_err; 744c63c7b05STrond Myklebust return 0; 74572cb77f4STrond Myklebust out_err: 74672cb77f4STrond Myklebust return err; 7471da177e4SLinus Torvalds } 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds /* 7501da177e4SLinus Torvalds * Insert a write request into an inode 7511da177e4SLinus Torvalds */ 7520c493b5cSTrond Myklebust static void nfs_inode_add_request(struct nfs_page *req) 7531da177e4SLinus Torvalds { 7540c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 7550c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 7560c493b5cSTrond Myklebust struct nfs_inode *nfsi = NFS_I(mapping->host); 757e7d39069STrond Myklebust 7582bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7592bfc6e56SWeston Andros Adamson 760e7d39069STrond Myklebust /* Lock the request! */ 7617ad84aa9STrond Myklebust nfs_lock_request(req); 762e7d39069STrond Myklebust 76329418aa4SMel Gorman /* 76429418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 76529418aa4SMel Gorman * with invalidate/truncate. 76629418aa4SMel Gorman */ 7674b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 7680c493b5cSTrond Myklebust if (likely(!folio_test_swapcache(folio))) { 7692df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 7700c493b5cSTrond Myklebust folio_set_private(folio); 7710c493b5cSTrond Myklebust folio->private = req; 77229418aa4SMel Gorman } 7734b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 774a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 77517089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 776cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 777cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 778cb1410c7SWeston Andros Adamson * adding subrequests. */ 77917089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 780c03b4024STrond Myklebust kref_get(&req->wb_kref); 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds 7831da177e4SLinus Torvalds /* 78489a09141SPeter Zijlstra * Remove a write request from an inode 7851da177e4SLinus Torvalds */ 7861da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7871da177e4SLinus Torvalds { 78820633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 7890c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req->wb_head); 7900c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 7911da177e4SLinus Torvalds 7924b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 7930c493b5cSTrond Myklebust if (likely(folio && !folio_test_swapcache(folio))) { 7940c493b5cSTrond Myklebust folio->private = NULL; 7950c493b5cSTrond Myklebust folio_clear_private(folio); 7960c493b5cSTrond Myklebust clear_bit(PG_MAPPED, &req->wb_head->wb_flags); 79729418aa4SMel Gorman } 7984b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 79920633f04SWeston Andros Adamson } 80017089a29SWeston Andros Adamson 80133ea5aaaSZhangXiaoxu if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) { 8021da177e4SLinus Torvalds nfs_release_request(req); 8030c493b5cSTrond Myklebust atomic_long_dec(&NFS_I(nfs_page_to_inode(req))->nrequests); 80433ea5aaaSZhangXiaoxu } 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 8070c493b5cSTrond Myklebust static void nfs_mark_request_dirty(struct nfs_page *req) 80861822ab5STrond Myklebust { 8090c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 8100c493b5cSTrond Myklebust if (folio) 8110c493b5cSTrond Myklebust filemap_dirty_folio(folio_mapping(folio), folio); 81261822ab5STrond Myklebust } 81361822ab5STrond Myklebust 8143a3908c8STrond Myklebust /* 8153a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8163a3908c8STrond Myklebust * 8170c493b5cSTrond Myklebust * Search through commit lists on @inode for the head request for @folio. 8183a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8193a3908c8STrond Myklebust * 8203a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8213a3908c8STrond Myklebust */ 8223a3908c8STrond Myklebust static struct nfs_page * 8233a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8240c493b5cSTrond Myklebust struct folio *folio) 8253a3908c8STrond Myklebust { 8263a3908c8STrond Myklebust struct nfs_page *freq, *t; 8273a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8283a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8293a3908c8STrond Myklebust 8303a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8313a3908c8STrond Myklebust 8323a3908c8STrond Myklebust /* search through pnfs commit lists */ 8330c493b5cSTrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, folio); 8343a3908c8STrond Myklebust if (freq) 8353a3908c8STrond Myklebust return freq->wb_head; 8363a3908c8STrond Myklebust 8373a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8383a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8390c493b5cSTrond Myklebust if (nfs_page_to_folio(freq) == folio) 8403a3908c8STrond Myklebust return freq->wb_head; 8413a3908c8STrond Myklebust } 8423a3908c8STrond Myklebust 8433a3908c8STrond Myklebust return NULL; 8443a3908c8STrond Myklebust } 8453a3908c8STrond Myklebust 8468dd37758STrond Myklebust /** 84786d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 84886d80f97STrond Myklebust * @req: pointer to a struct nfs_page 84986d80f97STrond Myklebust * @dst: commit list head 85086d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 85186d80f97STrond Myklebust * 85286d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 85386d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 85486d80f97STrond Myklebust * the MM page stats. 85586d80f97STrond Myklebust * 856e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 857e824f99aSTrond Myklebust * nfs_page lock. 85886d80f97STrond Myklebust */ 85986d80f97STrond Myklebust void 86086d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 86186d80f97STrond Myklebust struct nfs_commit_info *cinfo) 86286d80f97STrond Myklebust { 86386d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 86486d80f97STrond Myklebust nfs_list_add_request(req, dst); 8655cb953d4STrond Myklebust atomic_long_inc(&cinfo->mds->ncommit); 86686d80f97STrond Myklebust } 86786d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 86886d80f97STrond Myklebust 86986d80f97STrond Myklebust /** 8708dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8718dd37758STrond Myklebust * @req: pointer to a struct nfs_page 872ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8738dd37758STrond Myklebust * 874ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8758dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8768dd37758STrond Myklebust * the MM page stats. 8778dd37758STrond Myklebust * 878ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8798dd37758STrond Myklebust * holding the nfs_page lock. 8808dd37758STrond Myklebust */ 8818dd37758STrond Myklebust void 8826272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8838dd37758STrond Myklebust { 884e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8856272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 886e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 8870c493b5cSTrond Myklebust nfs_folio_mark_unstable(nfs_page_to_folio(req), cinfo); 8888dd37758STrond Myklebust } 8898dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8908dd37758STrond Myklebust 8918dd37758STrond Myklebust /** 8928dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8938dd37758STrond Myklebust * @req: pointer to a nfs_page 894ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8958dd37758STrond Myklebust * 896ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8978dd37758STrond Myklebust * number of outstanding requests requiring a commit 8988dd37758STrond Myklebust * It does not update the MM page stats. 8998dd37758STrond Myklebust * 900ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9018dd37758STrond Myklebust */ 9028dd37758STrond Myklebust void 903ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 904ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9058dd37758STrond Myklebust { 9068dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9078dd37758STrond Myklebust return; 9088dd37758STrond Myklebust nfs_list_remove_request(req); 9095cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9108dd37758STrond Myklebust } 9118dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9128dd37758STrond Myklebust 913ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 914ea2cf228SFred Isaman struct inode *inode) 915ea2cf228SFred Isaman { 916fe238e60SDave Wysochanski cinfo->inode = inode; 917ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 918ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 919b359f9d0SFred Isaman cinfo->dreq = NULL; 920f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 921ea2cf228SFred Isaman } 922ea2cf228SFred Isaman 923ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 924ea2cf228SFred Isaman struct inode *inode, 925ea2cf228SFred Isaman struct nfs_direct_req *dreq) 926ea2cf228SFred Isaman { 9271763da12SFred Isaman if (dreq) 9281763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9291763da12SFred Isaman else 930ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 931ea2cf228SFred Isaman } 932ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9338dd37758STrond Myklebust 9341da177e4SLinus Torvalds /* 9351da177e4SLinus Torvalds * Add a request to the inode's commit list. 9361da177e4SLinus Torvalds */ 9371763da12SFred Isaman void 938ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 939b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9401da177e4SLinus Torvalds { 941b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9428dd37758STrond Myklebust return; 9436272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9441da177e4SLinus Torvalds } 9458e821cadSTrond Myklebust 9460c493b5cSTrond Myklebust static void nfs_folio_clear_commit(struct folio *folio) 947e468bae9STrond Myklebust { 9480c493b5cSTrond Myklebust if (folio) { 9490c493b5cSTrond Myklebust long nr = folio_nr_pages(folio); 9500c493b5cSTrond Myklebust 9510c493b5cSTrond Myklebust node_stat_mod_folio(folio, NR_WRITEBACK, -nr); 9520c493b5cSTrond Myklebust wb_stat_mod(&inode_to_bdi(folio_file_mapping(folio)->host)->wb, 9530c493b5cSTrond Myklebust WB_WRITEBACK, -nr); 9540c493b5cSTrond Myklebust } 955e468bae9STrond Myklebust } 956d6d6dc7cSFred Isaman 957b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 9588dd37758STrond Myklebust static void 959d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 960d6d6dc7cSFred Isaman { 9618dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9629fcd5960STrond Myklebust struct nfs_open_context *ctx = nfs_req_openctx(req); 9639fcd5960STrond Myklebust struct inode *inode = d_inode(ctx->dentry); 964ea2cf228SFred Isaman struct nfs_commit_info cinfo; 965d6d6dc7cSFred Isaman 966ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 967e824f99aSTrond Myklebust mutex_lock(&NFS_I(inode)->commit_mutex); 968ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 969ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 970d6d6dc7cSFred Isaman } 971e824f99aSTrond Myklebust mutex_unlock(&NFS_I(inode)->commit_mutex); 9720c493b5cSTrond Myklebust nfs_folio_clear_commit(nfs_page_to_folio(req)); 9738dd37758STrond Myklebust } 974e468bae9STrond Myklebust } 975e468bae9STrond Myklebust 976d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9778e821cadSTrond Myklebust { 978c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 979d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 980c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9818e821cadSTrond Myklebust } 9828e821cadSTrond Myklebust 983919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 984919e3bd9STrond Myklebust { 985919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 986919e3bd9STrond Myklebust } 987919e3bd9STrond Myklebust 988061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9896c75dc0dSFred Isaman { 990ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9916c75dc0dSFred Isaman unsigned long bytes = 0; 9926c75dc0dSFred Isaman 9936c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9946c75dc0dSFred Isaman goto out; 995ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9966c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9976c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9986c75dc0dSFred Isaman 9996c75dc0dSFred Isaman bytes += req->wb_bytes; 10006c75dc0dSFred Isaman nfs_list_remove_request(req); 10016c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 10026c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 1003af887e43STrond Myklebust trace_nfs_comp_error(hdr->inode, req, hdr->error); 10040c493b5cSTrond Myklebust nfs_mapping_set_error(nfs_page_to_folio(req), 10050c493b5cSTrond Myklebust hdr->error); 10066c75dc0dSFred Isaman goto remove_req; 10076c75dc0dSFred Isaman } 1008c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 100933344e0fSTrond Myklebust /* Reset wb_nio, since the write was successful. */ 101033344e0fSTrond Myklebust req->wb_nio = 0; 1011f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1012b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1013a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10146c75dc0dSFred Isaman goto next; 10156c75dc0dSFred Isaman } 10166c75dc0dSFred Isaman remove_req: 10176c75dc0dSFred Isaman nfs_inode_remove_request(req); 10186c75dc0dSFred Isaman next: 10190c493b5cSTrond Myklebust nfs_page_end_writeback(req); 10203aff4ebbSTrond Myklebust nfs_release_request(req); 10216c75dc0dSFred Isaman } 10226c75dc0dSFred Isaman out: 1023919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10246c75dc0dSFred Isaman hdr->release(hdr); 10256c75dc0dSFred Isaman } 10266c75dc0dSFred Isaman 1027ce59515cSAnna Schumaker unsigned long 1028ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1029fb8a1f11STrond Myklebust { 10305cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1031fb8a1f11STrond Myklebust } 1032fb8a1f11STrond Myklebust 1033e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10341763da12SFred Isaman int 1035ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1036ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1037d6d6dc7cSFred Isaman { 1038137da553STrond Myklebust struct nfs_page *req, *tmp; 1039d6d6dc7cSFred Isaman int ret = 0; 1040d6d6dc7cSFred Isaman 1041137da553STrond Myklebust list_for_each_entry_safe(req, tmp, src, wb_list) { 10427ad84aa9STrond Myklebust kref_get(&req->wb_kref); 10432ce209c4STrond Myklebust if (!nfs_lock_request(req)) { 1044137da553STrond Myklebust nfs_release_request(req); 1045137da553STrond Myklebust continue; 1046137da553STrond Myklebust } 1047ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10485d2a9d9dSTrond Myklebust clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); 10498dd37758STrond Myklebust nfs_list_add_request(req, dst); 1050d6d6dc7cSFred Isaman ret++; 10511763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1052d6d6dc7cSFred Isaman break; 1053e824f99aSTrond Myklebust cond_resched(); 1054d6d6dc7cSFred Isaman } 1055d6d6dc7cSFred Isaman return ret; 1056d6d6dc7cSFred Isaman } 10575d2a9d9dSTrond Myklebust EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 1058d6d6dc7cSFred Isaman 10591da177e4SLinus Torvalds /* 10601da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10611da177e4SLinus Torvalds * @inode: NFS inode to scan 1062ea2cf228SFred Isaman * @dst: mds destination list 1063ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10641da177e4SLinus Torvalds * 10651da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10661da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10671da177e4SLinus Torvalds */ 10681763da12SFred Isaman int 1069ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1070ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10711da177e4SLinus Torvalds { 1072d6d6dc7cSFred Isaman int ret = 0; 1073fb8a1f11STrond Myklebust 10745cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10755cb953d4STrond Myklebust return 0; 1076e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10775cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10788dd37758STrond Myklebust const int max = INT_MAX; 1079d6d6dc7cSFred Isaman 1080ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1081ea2cf228SFred Isaman cinfo, max); 1082ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1083d6d6dc7cSFred Isaman } 1084e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1085ff778d02STrond Myklebust return ret; 10861da177e4SLinus Torvalds } 1087d6d6dc7cSFred Isaman 10881da177e4SLinus Torvalds /* 1089e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1090e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10911da177e4SLinus Torvalds * 1092e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1093e7d39069STrond Myklebust * to disk. 10941da177e4SLinus Torvalds */ 10950c493b5cSTrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct folio *folio, 1096e7d39069STrond Myklebust unsigned int offset, 1097e7d39069STrond Myklebust unsigned int bytes) 10981da177e4SLinus Torvalds { 1099e7d39069STrond Myklebust struct nfs_page *req; 1100e7d39069STrond Myklebust unsigned int rqend; 1101e7d39069STrond Myklebust unsigned int end; 11021da177e4SLinus Torvalds int error; 1103277459d2STrond Myklebust 1104e7d39069STrond Myklebust end = offset + bytes; 1105e7d39069STrond Myklebust 11060c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 1107f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1108f6032f21STrond Myklebust return req; 11092bfc6e56SWeston Andros Adamson 1110e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1111e7d39069STrond Myklebust /* 1112e7d39069STrond Myklebust * Tell the caller to flush out the request if 1113e7d39069STrond Myklebust * the offsets are non-contiguous. 1114e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1115e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1116e7d39069STrond Myklebust */ 1117f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1118e7d39069STrond Myklebust goto out_flushme; 1119e7d39069STrond Myklebust 11201da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11211da177e4SLinus Torvalds if (offset < req->wb_offset) { 11221da177e4SLinus Torvalds req->wb_offset = offset; 11231da177e4SLinus Torvalds req->wb_pgbase = offset; 11241da177e4SLinus Torvalds } 11251da177e4SLinus Torvalds if (end > rqend) 11261da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1127e7d39069STrond Myklebust else 1128e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 112933344e0fSTrond Myklebust req->wb_nio = 0; 1130e7d39069STrond Myklebust return req; 1131e7d39069STrond Myklebust out_flushme: 1132f6032f21STrond Myklebust /* 1133f6032f21STrond Myklebust * Note: we mark the request dirty here because 1134f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1135f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1136f6032f21STrond Myklebust */ 1137f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1138f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 11390c493b5cSTrond Myklebust error = nfs_wb_folio(folio_file_mapping(folio)->host, folio); 1140f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1141e7d39069STrond Myklebust } 11421da177e4SLinus Torvalds 1143e7d39069STrond Myklebust /* 1144e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1145e7d39069STrond Myklebust * 1146e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1147e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1148e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1149e7d39069STrond Myklebust */ 1150e7d39069STrond Myklebust static struct nfs_page *nfs_setup_write_request(struct nfs_open_context *ctx, 11510c493b5cSTrond Myklebust struct folio *folio, 11520c493b5cSTrond Myklebust unsigned int offset, 11530c493b5cSTrond Myklebust unsigned int bytes) 1154e7d39069STrond Myklebust { 1155e7d39069STrond Myklebust struct nfs_page *req; 1156e7d39069STrond Myklebust 11570c493b5cSTrond Myklebust req = nfs_try_to_update_request(folio, offset, bytes); 1158e7d39069STrond Myklebust if (req != NULL) 1159e7d39069STrond Myklebust goto out; 11600c493b5cSTrond Myklebust req = nfs_page_create_from_folio(ctx, folio, offset, bytes); 1161e7d39069STrond Myklebust if (IS_ERR(req)) 1162e7d39069STrond Myklebust goto out; 11630c493b5cSTrond Myklebust nfs_inode_add_request(req); 1164efc91ed0STrond Myklebust out: 116561e930a9STrond Myklebust return req; 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds 11680c493b5cSTrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, 11690c493b5cSTrond Myklebust struct folio *folio, unsigned int offset, 11700c493b5cSTrond Myklebust unsigned int count) 1171e7d39069STrond Myklebust { 1172e7d39069STrond Myklebust struct nfs_page *req; 1173e7d39069STrond Myklebust 11740c493b5cSTrond Myklebust req = nfs_setup_write_request(ctx, folio, offset, count); 1175e7d39069STrond Myklebust if (IS_ERR(req)) 1176e7d39069STrond Myklebust return PTR_ERR(req); 1177e7d39069STrond Myklebust /* Update file length */ 11780c493b5cSTrond Myklebust nfs_grow_file(folio, offset, count); 1179d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1180a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11811d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1182e7d39069STrond Myklebust return 0; 1183e7d39069STrond Myklebust } 1184e7d39069STrond Myklebust 11850c493b5cSTrond Myklebust int nfs_flush_incompatible(struct file *file, struct folio *folio) 11861da177e4SLinus Torvalds { 1187cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11882a369153STrond Myklebust struct nfs_lock_context *l_ctx; 118917b985deSJeff Layton struct file_lock_context *flctx = locks_inode_context(file_inode(file)); 11901da177e4SLinus Torvalds struct nfs_page *req; 11911a54533eSTrond Myklebust int do_flush, status; 11921da177e4SLinus Torvalds /* 11931da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11941da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11951da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11961da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11971da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11981da177e4SLinus Torvalds * dropped page. 11991da177e4SLinus Torvalds */ 12001a54533eSTrond Myklebust do { 12010c493b5cSTrond Myklebust req = nfs_folio_find_head_request(folio); 12021a54533eSTrond Myklebust if (req == NULL) 12031a54533eSTrond Myklebust return 0; 12042a369153STrond Myklebust l_ctx = req->wb_lock_context; 12050c493b5cSTrond Myklebust do_flush = nfs_page_to_folio(req) != folio || 12069fcd5960STrond Myklebust !nfs_match_open_context(nfs_req_openctx(req), ctx); 1207bd61e0a9SJeff Layton if (l_ctx && flctx && 1208bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1209bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1210d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12115263e31eSJeff Layton } 12121da177e4SLinus Torvalds nfs_release_request(req); 12131a54533eSTrond Myklebust if (!do_flush) 12141a54533eSTrond Myklebust return 0; 12150c493b5cSTrond Myklebust status = nfs_wb_folio(folio_file_mapping(folio)->host, folio); 12161a54533eSTrond Myklebust } while (status == 0); 12171a54533eSTrond Myklebust return status; 12181da177e4SLinus Torvalds } 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds /* 1221dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1222dc24826bSAndy Adamson * expire soon. 1223dc24826bSAndy Adamson * 1224dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1225dc24826bSAndy Adamson * 1226dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1227dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1228dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1229dc24826bSAndy Adamson */ 1230dc24826bSAndy Adamson int 1231dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1232dc24826bSAndy Adamson { 1233dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1234dc24826bSAndy Adamson 1235ddf529eeSNeilBrown if (nfs_ctx_key_to_expire(ctx, inode) && 1236ca05cbaeSTrond Myklebust !rcu_access_pointer(ctx->ll_cred)) 1237ddf529eeSNeilBrown /* Already expired! */ 1238ddf529eeSNeilBrown return -EACCES; 1239ddf529eeSNeilBrown return 0; 1240dc24826bSAndy Adamson } 1241dc24826bSAndy Adamson 1242dc24826bSAndy Adamson /* 1243dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1244dc24826bSAndy Adamson */ 1245ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1246dc24826bSAndy Adamson { 1247ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1248ca05cbaeSTrond Myklebust struct rpc_cred *cred, *new, *old = NULL; 1249ddf529eeSNeilBrown struct auth_cred acred = { 1250a52458b4SNeilBrown .cred = ctx->cred, 1251ddf529eeSNeilBrown }; 1252ca05cbaeSTrond Myklebust bool ret = false; 1253ce52914eSScott Mayhew 1254ca05cbaeSTrond Myklebust rcu_read_lock(); 1255ca05cbaeSTrond Myklebust cred = rcu_dereference(ctx->ll_cred); 1256ca05cbaeSTrond Myklebust if (cred && !(cred->cr_ops->crkey_timeout && 1257ca05cbaeSTrond Myklebust cred->cr_ops->crkey_timeout(cred))) 1258ca05cbaeSTrond Myklebust goto out; 1259ca05cbaeSTrond Myklebust rcu_read_unlock(); 1260ca05cbaeSTrond Myklebust 1261ca05cbaeSTrond Myklebust new = auth->au_ops->lookup_cred(auth, &acred, 0); 1262ca05cbaeSTrond Myklebust if (new == cred) { 1263ca05cbaeSTrond Myklebust put_rpccred(new); 1264ddf529eeSNeilBrown return true; 1265ca05cbaeSTrond Myklebust } 1266ca05cbaeSTrond Myklebust if (IS_ERR_OR_NULL(new)) { 1267ca05cbaeSTrond Myklebust new = NULL; 1268ca05cbaeSTrond Myklebust ret = true; 1269ca05cbaeSTrond Myklebust } else if (new->cr_ops->crkey_timeout && 1270ca05cbaeSTrond Myklebust new->cr_ops->crkey_timeout(new)) 1271ca05cbaeSTrond Myklebust ret = true; 1272ca05cbaeSTrond Myklebust 1273ca05cbaeSTrond Myklebust rcu_read_lock(); 1274ca05cbaeSTrond Myklebust old = rcu_dereference_protected(xchg(&ctx->ll_cred, 1275ca05cbaeSTrond Myklebust RCU_INITIALIZER(new)), 1); 1276ca05cbaeSTrond Myklebust out: 1277ca05cbaeSTrond Myklebust rcu_read_unlock(); 1278ca05cbaeSTrond Myklebust put_rpccred(old); 1279ca05cbaeSTrond Myklebust return ret; 1280dc24826bSAndy Adamson } 1281dc24826bSAndy Adamson 1282dc24826bSAndy Adamson /* 12835d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12845d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12855d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12865d47a356STrond Myklebust */ 12870c493b5cSTrond Myklebust static bool nfs_folio_write_uptodate(struct folio *folio, unsigned int pagelen) 12885d47a356STrond Myklebust { 12890c493b5cSTrond Myklebust struct inode *inode = folio_file_mapping(folio)->host; 1290d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1291d529ef83SJeff Layton 12928d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12938d197a56STrond Myklebust goto out; 1294fc9dc401STrond Myklebust if (nfsi->cache_validity & 129513c0b082STrond Myklebust (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE)) 1296d529ef83SJeff Layton return false; 12974db72b40SJeff Layton smp_rmb(); 1298fc9dc401STrond Myklebust if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags) && pagelen != 0) 12998d197a56STrond Myklebust return false; 13008d197a56STrond Myklebust out: 1301fc9dc401STrond Myklebust if (nfsi->cache_validity & NFS_INO_INVALID_DATA && pagelen != 0) 130218dd78c4SScott Mayhew return false; 13030c493b5cSTrond Myklebust return folio_test_uptodate(folio) != 0; 13045d47a356STrond Myklebust } 13055d47a356STrond Myklebust 13065263e31eSJeff Layton static bool 13075263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 13085263e31eSJeff Layton { 13095263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 13105263e31eSJeff Layton fl->fl_type == F_WRLCK; 13115263e31eSJeff Layton } 13125263e31eSJeff Layton 1313c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1314c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1315c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1316c7559663SScott Mayhew * inefficiencies. 1317c7559663SScott Mayhew * 1318263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1319263b4509SScott Mayhew * of the checks. 1320c7559663SScott Mayhew */ 13210c493b5cSTrond Myklebust static int nfs_can_extend_write(struct file *file, struct folio *folio, 13220c493b5cSTrond Myklebust unsigned int pagelen) 1323c7559663SScott Mayhew { 13240c493b5cSTrond Myklebust struct inode *inode = file_inode(file); 132517b985deSJeff Layton struct file_lock_context *flctx = locks_inode_context(inode); 13265263e31eSJeff Layton struct file_lock *fl; 13270c493b5cSTrond Myklebust int ret; 13285263e31eSJeff Layton 1329c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1330c7559663SScott Mayhew return 0; 13310c493b5cSTrond Myklebust if (!nfs_folio_write_uptodate(folio, pagelen)) 1332263b4509SScott Mayhew return 0; 1333c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1334c7559663SScott Mayhew return 1; 1335bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1336bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13378fa4592aSTrond Myklebust return 1; 13385263e31eSJeff Layton 13395263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13405263e31eSJeff Layton ret = 0; 13416109c850SJeff Layton spin_lock(&flctx->flc_lock); 1342bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1343bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1344bd61e0a9SJeff Layton fl_list); 1345bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13465263e31eSJeff Layton ret = 1; 1347bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13485263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13495263e31eSJeff Layton fl_list); 13505263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13515263e31eSJeff Layton ret = 1; 13525263e31eSJeff Layton } 13536109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13545263e31eSJeff Layton return ret; 1355c7559663SScott Mayhew } 1356c7559663SScott Mayhew 13575d47a356STrond Myklebust /* 13581da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13591da177e4SLinus Torvalds * 13601da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13611da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13621da177e4SLinus Torvalds */ 13630c493b5cSTrond Myklebust int nfs_update_folio(struct file *file, struct folio *folio, 13641da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13651da177e4SLinus Torvalds { 1366cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 13670c493b5cSTrond Myklebust struct address_space *mapping = folio_file_mapping(folio); 1368d2ceb7e5SBenjamin Coddington struct inode *inode = mapping->host; 13690c493b5cSTrond Myklebust unsigned int pagelen = nfs_folio_length(folio); 13701da177e4SLinus Torvalds int status = 0; 13711da177e4SLinus Torvalds 137291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 137391d5b470SChuck Lever 13740c493b5cSTrond Myklebust dprintk("NFS: nfs_update_folio(%pD2 %d@%lld)\n", file, count, 13750c493b5cSTrond Myklebust (long long)(folio_file_pos(folio) + offset)); 13761da177e4SLinus Torvalds 1377149a4fddSBenjamin Coddington if (!count) 1378149a4fddSBenjamin Coddington goto out; 1379149a4fddSBenjamin Coddington 13800c493b5cSTrond Myklebust if (nfs_can_extend_write(file, folio, pagelen)) { 1381fc9dc401STrond Myklebust count = max(count + offset, pagelen); 13821da177e4SLinus Torvalds offset = 0; 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds 13850c493b5cSTrond Myklebust status = nfs_writepage_setup(ctx, folio, offset, count); 138603fa9e84STrond Myklebust if (status < 0) 1387d2ceb7e5SBenjamin Coddington nfs_set_pageerror(mapping); 1388149a4fddSBenjamin Coddington out: 13890c493b5cSTrond Myklebust dprintk("NFS: nfs_update_folio returns %d (isize %lld)\n", 13901da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13911da177e4SLinus Torvalds return status; 13921da177e4SLinus Torvalds } 13931da177e4SLinus Torvalds 13943ff7576dSTrond Myklebust static int flush_task_priority(int how) 13951da177e4SLinus Torvalds { 13961da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13971da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13981da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13991da177e4SLinus Torvalds case FLUSH_LOWPRI: 14001da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 14011da177e4SLinus Torvalds } 14021da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 14031da177e4SLinus Torvalds } 14041da177e4SLinus Torvalds 1405d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1406d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1407abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 14081ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 14091da177e4SLinus Torvalds { 14103ff7576dSTrond Myklebust int priority = flush_task_priority(how); 14111da177e4SLinus Torvalds 14128db55a03SNeilBrown if (IS_SWAPFILE(hdr->inode)) 14138db55a03SNeilBrown task_setup_data->flags |= RPC_TASK_SWAPPER; 14141ed26f33SAnna Schumaker task_setup_data->priority = priority; 1415fb91fb0eSAnna Schumaker rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); 14165bb2a7cbSTrond Myklebust trace_nfs_initiate_write(hdr); 1417275acaafSTrond Myklebust } 1418275acaafSTrond Myklebust 14196d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 14206d884e8fSFred * call this on each, which will prepare them to be retried on next 14216d884e8fSFred * writeback using standard nfs. 14226d884e8fSFred */ 14236d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14246d884e8fSFred { 14256dd85e83STrond Myklebust struct nfs_inode *nfsi = NFS_I(nfs_page_to_inode(req)); 142667f4b5dcSTrond Myklebust 142733344e0fSTrond Myklebust /* Bump the transmission count */ 142833344e0fSTrond Myklebust req->wb_nio++; 14296d884e8fSFred nfs_mark_request_dirty(req); 143067f4b5dcSTrond Myklebust atomic_long_inc(&nfsi->redirtied_pages); 14310c493b5cSTrond Myklebust nfs_page_end_writeback(req); 14323aff4ebbSTrond Myklebust nfs_release_request(req); 14336d884e8fSFred } 14346d884e8fSFred 1435df3accb8STrond Myklebust static void nfs_async_write_error(struct list_head *head, int error) 14366c75dc0dSFred Isaman { 14376c75dc0dSFred Isaman struct nfs_page *req; 14386c75dc0dSFred Isaman 14396c75dc0dSFred Isaman while (!list_empty(head)) { 14406c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14416c75dc0dSFred Isaman nfs_list_remove_request(req); 1442cea9ba72STrond Myklebust if (nfs_error_is_fatal_on_server(error)) 14436fbda89bSTrond Myklebust nfs_write_error(req, error); 14446fbda89bSTrond Myklebust else 14456c75dc0dSFred Isaman nfs_redirty_request(req); 14466c75dc0dSFred Isaman } 14476c75dc0dSFred Isaman } 14486c75dc0dSFred Isaman 1449dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1450dc602dd7STrond Myklebust { 1451df3accb8STrond Myklebust nfs_async_write_error(&hdr->pages, 0); 1452dc602dd7STrond Myklebust } 1453dc602dd7STrond Myklebust 1454061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1455919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1456061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1457061ae2edSFred Isaman .completion = nfs_write_completion, 1458dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1459061ae2edSFred Isaman }; 1460061ae2edSFred Isaman 146157208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1462a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1463061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14641751c363STrond Myklebust { 1465a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 146641d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1467a20c93e3SChristoph Hellwig 1468a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1469a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1470a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1471a20c93e3SChristoph Hellwig #endif 14724a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 14733bde7afdSTrond Myklebust server->wsize, ioflags); 14741751c363STrond Myklebust } 1475ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14761751c363STrond Myklebust 1477dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1478dce81290STrond Myklebust { 1479a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1480a7d42ddbSWeston Andros Adamson 14816f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14826f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14836f29b9bbSKinglong Mee 148441d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1485a7d42ddbSWeston Andros Adamson 1486a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1487a7d42ddbSWeston Andros Adamson 1488a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1489a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1490dce81290STrond Myklebust } 14911f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1492dce81290STrond Myklebust 14931da177e4SLinus Torvalds 14940b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14950b7c0153SFred Isaman { 14960b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14970b7c0153SFred Isaman 14980b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14990b7c0153SFred Isaman } 15000b7c0153SFred Isaman 1501a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1502a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1503a08a8cd3STrond Myklebust { 1504a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1505a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 15062b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1507a08a8cd3STrond Myklebust 1508a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15092b83d3deSTrond Myklebust fattr->size = size; 15102b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15112b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1512a08a8cd3STrond Myklebust return; 15132b83d3deSTrond Myklebust } 15142b83d3deSTrond Myklebust if (size != fattr->size) 1515a08a8cd3STrond Myklebust return; 1516a08a8cd3STrond Myklebust /* Set attribute barrier */ 1517a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15182b83d3deSTrond Myklebust /* ...and update size */ 15192b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1520a08a8cd3STrond Myklebust } 1521a08a8cd3STrond Myklebust 1522a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1523a08a8cd3STrond Myklebust { 15242b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1525a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1526a08a8cd3STrond Myklebust 1527a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1528a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1529a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1530a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1531a08a8cd3STrond Myklebust } 1532a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1533a08a8cd3STrond Myklebust 15341da177e4SLinus Torvalds /* 15351da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15361da177e4SLinus Torvalds */ 1537d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1538d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15390eecb214SAnna Schumaker struct inode *inode) 15401da177e4SLinus Torvalds { 1541788e7a89STrond Myklebust int status; 15421da177e4SLinus Torvalds 1543f551e44fSChuck Lever /* 1544f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1545f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1546f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1547f551e44fSChuck Lever * another writer had changed the file, but some applications 1548f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1549f551e44fSChuck Lever */ 1550d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1551788e7a89STrond Myklebust if (status != 0) 15520eecb214SAnna Schumaker return status; 15538224b273SChuck Lever 1554d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 15555bb2a7cbSTrond Myklebust trace_nfs_writeback_done(task, hdr); 155691d5b470SChuck Lever 155769d96651SJeff Layton if (task->tk_status >= 0) { 155869d96651SJeff Layton enum nfs3_stable_how committed = hdr->res.verf->committed; 155969d96651SJeff Layton 156069d96651SJeff Layton if (committed == NFS_UNSTABLE) { 156169d96651SJeff Layton /* 156269d96651SJeff Layton * We have some uncommitted data on the server at 156369d96651SJeff Layton * this point, so ensure that we keep track of that 156469d96651SJeff Layton * fact irrespective of what later writes do. 156569d96651SJeff Layton */ 156669d96651SJeff Layton set_bit(NFS_IOHDR_UNSTABLE_WRITES, &hdr->flags); 156769d96651SJeff Layton } 156869d96651SJeff Layton 156969d96651SJeff Layton if (committed < hdr->args.stable) { 15701da177e4SLinus Torvalds /* We tried a write call, but the server did not 15711da177e4SLinus Torvalds * commit data to stable storage even though we 15721da177e4SLinus Torvalds * requested it. 15731da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15741da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15751da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15761da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15771da177e4SLinus Torvalds */ 15781da177e4SLinus Torvalds static unsigned long complain; 15791da177e4SLinus Torvalds 1580a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15811da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15821da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15831da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1584cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 158569d96651SJeff Layton committed, hdr->args.stable); 15861da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15871da177e4SLinus Torvalds } 15881da177e4SLinus Torvalds } 158969d96651SJeff Layton } 15901f2edbe3STrond Myklebust 15911f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 159216e14375STrond Myklebust if (nfs_should_remove_suid(inode)) { 159316e14375STrond Myklebust spin_lock(&inode->i_lock); 1594720869ebSTrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); 159516e14375STrond Myklebust spin_unlock(&inode->i_lock); 159616e14375STrond Myklebust } 15970eecb214SAnna Schumaker return 0; 15980eecb214SAnna Schumaker } 15990eecb214SAnna Schumaker 16000eecb214SAnna Schumaker /* 16010eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 16020eecb214SAnna Schumaker */ 1603d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1604d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 16050eecb214SAnna Schumaker { 1606d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1607d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 16081f2edbe3STrond Myklebust 16091f2edbe3STrond Myklebust if (resp->count < argp->count) { 16101da177e4SLinus Torvalds static unsigned long complain; 16111da177e4SLinus Torvalds 16126c75dc0dSFred Isaman /* This a short write! */ 1613d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 161491d5b470SChuck Lever 16151da177e4SLinus Torvalds /* Has the server at least made some progress? */ 16166c75dc0dSFred Isaman if (resp->count == 0) { 16176c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 16186c75dc0dSFred Isaman printk(KERN_WARNING 16196c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16206c75dc0dSFred Isaman argp->count); 16216c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16226c75dc0dSFred Isaman } 1623d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16246c75dc0dSFred Isaman task->tk_status = -EIO; 16256c75dc0dSFred Isaman return; 16266c75dc0dSFred Isaman } 1627f8417b48SKinglong Mee 1628f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1629f8417b48SKinglong Mee if (!task->tk_ops) { 1630f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1631f8417b48SKinglong Mee return; 1632f8417b48SKinglong Mee } 1633f8417b48SKinglong Mee 16341da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16351da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16361da177e4SLinus Torvalds /* Resend from where the server left off */ 1637d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16381da177e4SLinus Torvalds argp->offset += resp->count; 16391da177e4SLinus Torvalds argp->pgbase += resp->count; 16401da177e4SLinus Torvalds argp->count -= resp->count; 16411da177e4SLinus Torvalds } else { 16421da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16431da177e4SLinus Torvalds * headaches in the case of a server crash. 16441da177e4SLinus Torvalds */ 16451da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16461da177e4SLinus Torvalds } 16478c9cb714STrond Myklebust resp->count = 0; 16488c9cb714STrond Myklebust resp->verf->committed = 0; 1649d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16501da177e4SLinus Torvalds } 16511da177e4SLinus Torvalds } 16521da177e4SLinus Torvalds 1653af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 165471d0a611STrond Myklebust { 1655723c921eSPeter Zijlstra return wait_var_event_killable(&cinfo->rpcs_out, 1656723c921eSPeter Zijlstra !atomic_read(&cinfo->rpcs_out)); 1657af7cf057STrond Myklebust } 1658af7cf057STrond Myklebust 1659af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1660af7cf057STrond Myklebust { 1661af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1662af7cf057STrond Myklebust } 1663af7cf057STrond Myklebust 1664133a48abSTrond Myklebust bool nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1665af7cf057STrond Myklebust { 1666133a48abSTrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) { 1667723c921eSPeter Zijlstra wake_up_var(&cinfo->rpcs_out); 1668133a48abSTrond Myklebust return true; 1669133a48abSTrond Myklebust } 1670133a48abSTrond Myklebust return false; 167171d0a611STrond Myklebust } 167271d0a611STrond Myklebust 16730b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16741da177e4SLinus Torvalds { 16750b7c0153SFred Isaman put_nfs_open_context(data->context); 16760b7c0153SFred Isaman nfs_commit_free(data); 16771da177e4SLinus Torvalds } 1678e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16791da177e4SLinus Torvalds 16800b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1681c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16829ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16839f0ec176SAndy Adamson int how, int flags) 16841da177e4SLinus Torvalds { 168507737691STrond Myklebust struct rpc_task *task; 16869ace33cdSFred Isaman int priority = flush_task_priority(how); 1687bdc7f021STrond Myklebust struct rpc_message msg = { 1688bdc7f021STrond Myklebust .rpc_argp = &data->args, 1689bdc7f021STrond Myklebust .rpc_resp = &data->res, 16909ace33cdSFred Isaman .rpc_cred = data->cred, 1691bdc7f021STrond Myklebust }; 169284115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 169307737691STrond Myklebust .task = &data->task, 16949ace33cdSFred Isaman .rpc_client = clnt, 1695bdc7f021STrond Myklebust .rpc_message = &msg, 16969ace33cdSFred Isaman .callback_ops = call_ops, 169784115e1cSTrond Myklebust .callback_data = data, 1698101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16994fa7ef69STrond Myklebust .flags = RPC_TASK_ASYNC | flags, 17003ff7576dSTrond Myklebust .priority = priority, 170184115e1cSTrond Myklebust }; 1702118f09edSOlga Kornievskaia 1703118f09edSOlga Kornievskaia if (nfs_server_capable(data->inode, NFS_CAP_MOVEABLE)) 1704118f09edSOlga Kornievskaia task_setup_data.flags |= RPC_TASK_MOVEABLE; 1705118f09edSOlga Kornievskaia 1706788e7a89STrond Myklebust /* Set up the initial task struct. */ 1707e9ae1ee2SAnna Schumaker nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); 17088224b273SChuck Lever trace_nfs_initiate_commit(data); 17091da177e4SLinus Torvalds 1710b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1711bdc7f021STrond Myklebust 171207737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1713dbae4c73STrond Myklebust if (IS_ERR(task)) 1714dbae4c73STrond Myklebust return PTR_ERR(task); 1715d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1716d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 171707737691STrond Myklebust rpc_put_task(task); 1718dbae4c73STrond Myklebust return 0; 17191da177e4SLinus Torvalds } 1720e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 17211da177e4SLinus Torvalds 1722378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1723378520b8SPeng Tao { 1724378520b8SPeng Tao loff_t lwb = 0; 1725378520b8SPeng Tao struct nfs_page *req; 1726378520b8SPeng Tao 1727378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1728378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1729378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1730378520b8SPeng Tao 1731378520b8SPeng Tao return lwb; 1732378520b8SPeng Tao } 1733378520b8SPeng Tao 17341da177e4SLinus Torvalds /* 17359ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17369ace33cdSFred Isaman */ 17370b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1738988b6dceSFred Isaman struct list_head *head, 1739f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1740f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17419ace33cdSFred Isaman { 174219573c93STrond Myklebust struct nfs_page *first; 174319573c93STrond Myklebust struct nfs_open_context *ctx; 174419573c93STrond Myklebust struct inode *inode; 17459ace33cdSFred Isaman 17469ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17479ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17489ace33cdSFred Isaman 174919573c93STrond Myklebust if (head) 17509ace33cdSFred Isaman list_splice_init(head, &data->pages); 17519ace33cdSFred Isaman 175219573c93STrond Myklebust first = nfs_list_entry(data->pages.next); 175319573c93STrond Myklebust ctx = nfs_req_openctx(first); 175419573c93STrond Myklebust inode = d_inode(ctx->dentry); 175519573c93STrond Myklebust 17569ace33cdSFred Isaman data->inode = inode; 17579fcd5960STrond Myklebust data->cred = ctx->cred; 1758988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1759378520b8SPeng Tao /* only set lwb for pnfs commit */ 1760378520b8SPeng Tao if (lseg) 1761378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17629ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1763f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1764b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17659ace33cdSFred Isaman 17669ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17679ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17689ace33cdSFred Isaman data->args.offset = 0; 17699ace33cdSFred Isaman data->args.count = 0; 17709fcd5960STrond Myklebust data->context = get_nfs_open_context(ctx); 17719ace33cdSFred Isaman data->res.fattr = &data->fattr; 17729ace33cdSFred Isaman data->res.verf = &data->verf; 17739ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 1774133a48abSTrond Myklebust nfs_commit_begin(cinfo->mds); 17759ace33cdSFred Isaman } 1776e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17779ace33cdSFred Isaman 1778e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1779ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1780b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1781b57ff130SWeston Andros Adamson u32 ds_commit_idx) 178264bfeb49SFred Isaman { 178364bfeb49SFred Isaman struct nfs_page *req; 178464bfeb49SFred Isaman 178564bfeb49SFred Isaman while (!list_empty(page_list)) { 178664bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 178764bfeb49SFred Isaman nfs_list_remove_request(req); 1788b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 17890c493b5cSTrond Myklebust nfs_folio_clear_commit(nfs_page_to_folio(req)); 17901d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 179164bfeb49SFred Isaman } 179264bfeb49SFred Isaman } 1793e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 179464bfeb49SFred Isaman 17950c493b5cSTrond Myklebust static void nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1796b20135d0STrond Myklebust struct nfs_page *req) 1797b20135d0STrond Myklebust { 17980c493b5cSTrond Myklebust struct folio *folio = nfs_page_to_folio(req); 17990c493b5cSTrond Myklebust 18000c493b5cSTrond Myklebust filemap_dirty_folio(folio_mapping(folio), folio); 1801b20135d0STrond Myklebust } 1802b20135d0STrond Myklebust 18039ace33cdSFred Isaman /* 18041da177e4SLinus Torvalds * Commit dirty pages 18051da177e4SLinus Torvalds */ 18061da177e4SLinus Torvalds static int 1807ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1808ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 18091da177e4SLinus Torvalds { 18100b7c0153SFred Isaman struct nfs_commit_data *data; 181185e39feeSOlga Kornievskaia unsigned short task_flags = 0; 18121da177e4SLinus Torvalds 1813ade8febdSWeston Andros Adamson /* another commit raced with us */ 1814ade8febdSWeston Andros Adamson if (list_empty(head)) 1815ade8febdSWeston Andros Adamson return 0; 1816ade8febdSWeston Andros Adamson 1817515dcdcdSTrond Myklebust data = nfs_commitdata_alloc(); 1818515dcdcdSTrond Myklebust if (!data) { 1819515dcdcdSTrond Myklebust nfs_retry_commit(head, NULL, cinfo, -1); 1820515dcdcdSTrond Myklebust return -ENOMEM; 1821515dcdcdSTrond Myklebust } 18221da177e4SLinus Torvalds 18231da177e4SLinus Torvalds /* Set up the argument struct */ 1824f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 182585e39feeSOlga Kornievskaia if (NFS_SERVER(inode)->nfs_client->cl_minorversion) 182685e39feeSOlga Kornievskaia task_flags = RPC_TASK_MOVEABLE; 1827c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 182885e39feeSOlga Kornievskaia data->mds_ops, how, 182985e39feeSOlga Kornievskaia RPC_TASK_CRED_NOREF | task_flags); 18301da177e4SLinus Torvalds } 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds /* 18331da177e4SLinus Torvalds * COMMIT call returned 18341da177e4SLinus Torvalds */ 1835788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 18361da177e4SLinus Torvalds { 18370b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18381da177e4SLinus Torvalds 1839788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1840c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 18417bdd297eSTrond Myklebust trace_nfs_commit_done(task, data); 1842c9d8f89dSTrond Myklebust } 1843c9d8f89dSTrond Myklebust 1844f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1845c9d8f89dSTrond Myklebust { 1846221203ceSTrond Myklebust const struct nfs_writeverf *verf = data->res.verf; 1847c9d8f89dSTrond Myklebust struct nfs_page *req; 1848c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1849f453a54aSFred Isaman struct nfs_commit_info cinfo; 1850353db796SNeilBrown struct nfs_server *nfss; 18510c493b5cSTrond Myklebust struct folio *folio; 1852788e7a89STrond Myklebust 18531da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18541da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18551da177e4SLinus Torvalds nfs_list_remove_request(req); 18560c493b5cSTrond Myklebust folio = nfs_page_to_folio(req); 18570c493b5cSTrond Myklebust nfs_folio_clear_commit(folio); 18581da177e4SLinus Torvalds 18591e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18609fcd5960STrond Myklebust nfs_req_openctx(req)->dentry->d_sb->s_id, 18619fcd5960STrond Myklebust (unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)), 18621da177e4SLinus Torvalds req->wb_bytes, 18631da177e4SLinus Torvalds (long long)req_offset(req)); 1864c9d8f89dSTrond Myklebust if (status < 0) { 18650c493b5cSTrond Myklebust if (folio) { 1866af887e43STrond Myklebust trace_nfs_commit_error(data->inode, req, 1867af887e43STrond Myklebust status); 18680c493b5cSTrond Myklebust nfs_mapping_set_error(folio, status); 18691da177e4SLinus Torvalds nfs_inode_remove_request(req); 18706fbda89bSTrond Myklebust } 1871ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18721da177e4SLinus Torvalds goto next; 18731da177e4SLinus Torvalds } 18741da177e4SLinus Torvalds 18751da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18761da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18771f28476dSTrond Myklebust if (nfs_write_match_verf(verf, req)) { 18781da177e4SLinus Torvalds /* We have a match */ 18790c493b5cSTrond Myklebust if (folio) 18801da177e4SLinus Torvalds nfs_inode_remove_request(req); 1881ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18821da177e4SLinus Torvalds goto next; 18831da177e4SLinus Torvalds } 18841da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1885ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18866d884e8fSFred nfs_mark_request_dirty(req); 188767f4b5dcSTrond Myklebust atomic_long_inc(&NFS_I(data->inode)->redirtied_pages); 18881da177e4SLinus Torvalds next: 18891d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18907f1bda44STrond Myklebust /* Latency breaker */ 18917f1bda44STrond Myklebust cond_resched(); 18921da177e4SLinus Torvalds } 1893353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1894353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18956df25e58SNeilBrown nfss->write_congested = 0; 1896353db796SNeilBrown 1897f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1898af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18995917ce84SFred Isaman } 19005917ce84SFred Isaman 19015917ce84SFred Isaman static void nfs_commit_release(void *calldata) 19025917ce84SFred Isaman { 19030b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 19045917ce84SFred Isaman 1905f453a54aSFred Isaman data->completion_ops->completion(data); 1906c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 19071da177e4SLinus Torvalds } 1908788e7a89STrond Myklebust 1909788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 19100b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1911788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1912788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1913788e7a89STrond Myklebust }; 19141da177e4SLinus Torvalds 1915f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1916f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1917b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1918f453a54aSFred Isaman }; 1919f453a54aSFred Isaman 19201763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1921ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 192284c53ab5SFred Isaman { 192384c53ab5SFred Isaman int status; 192484c53ab5SFred Isaman 1925ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 192684c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1927ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 192884c53ab5SFred Isaman return status; 192984c53ab5SFred Isaman } 193084c53ab5SFred Isaman 1931c4f24df9STrond Myklebust static int __nfs_commit_inode(struct inode *inode, int how, 1932c4f24df9STrond Myklebust struct writeback_control *wbc) 19331da177e4SLinus Torvalds { 19341da177e4SLinus Torvalds LIST_HEAD(head); 1935ea2cf228SFred Isaman struct nfs_commit_info cinfo; 193671d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1937c4f24df9STrond Myklebust int ret, nscan; 19381da177e4SLinus Torvalds 193964a93dbfSTrond Myklebust how &= ~FLUSH_SYNC; 1940ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1941af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1942c4f24df9STrond Myklebust for (;;) { 1943c4f24df9STrond Myklebust ret = nscan = nfs_scan_commit(inode, &head, &cinfo); 1944c4f24df9STrond Myklebust if (ret <= 0) 1945c4f24df9STrond Myklebust break; 1946c4f24df9STrond Myklebust ret = nfs_generic_commit_list(inode, &head, how, &cinfo); 1947c4f24df9STrond Myklebust if (ret < 0) 1948c4f24df9STrond Myklebust break; 1949c4f24df9STrond Myklebust ret = 0; 1950c4f24df9STrond Myklebust if (wbc && wbc->sync_mode == WB_SYNC_NONE) { 1951c4f24df9STrond Myklebust if (nscan < wbc->nr_to_write) 1952c4f24df9STrond Myklebust wbc->nr_to_write -= nscan; 1953c4f24df9STrond Myklebust else 1954c4f24df9STrond Myklebust wbc->nr_to_write = 0; 1955c4f24df9STrond Myklebust } 1956c4f24df9STrond Myklebust if (nscan < INT_MAX) 1957c4f24df9STrond Myklebust break; 1958c4f24df9STrond Myklebust cond_resched(); 1959c4f24df9STrond Myklebust } 1960af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 1961c4f24df9STrond Myklebust if (ret || !may_wait) 1962c4f24df9STrond Myklebust return ret; 1963c4f24df9STrond Myklebust return wait_on_commit(cinfo.mds); 1964c4f24df9STrond Myklebust } 1965c4f24df9STrond Myklebust 1966c4f24df9STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 1967c4f24df9STrond Myklebust { 1968c4f24df9STrond Myklebust return __nfs_commit_inode(inode, how, NULL); 19691da177e4SLinus Torvalds } 1970b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19718fc795f7STrond Myklebust 1972ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19738fc795f7STrond Myklebust { 1974420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1975420e3646STrond Myklebust int flags = FLUSH_SYNC; 1976420e3646STrond Myklebust int ret = 0; 19778fc795f7STrond Myklebust 1978c4f24df9STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 19793236c3e1SJeff Layton /* no commits means nothing needs to be done */ 19805cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 1981c4f24df9STrond Myklebust goto check_requests_outstanding; 19823236c3e1SJeff Layton 1983a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1984a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1985420e3646STrond Myklebust */ 19861a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1987420e3646STrond Myklebust goto out_mark_dirty; 1988420e3646STrond Myklebust 1989a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1990420e3646STrond Myklebust flags = 0; 1991a00dd6c0SJeff Layton } 1992a00dd6c0SJeff Layton 1993c4f24df9STrond Myklebust ret = __nfs_commit_inode(inode, flags, wbc); 1994c4f24df9STrond Myklebust if (!ret) { 1995c4f24df9STrond Myklebust if (flags & FLUSH_SYNC) 19968fc795f7STrond Myklebust return 0; 1997c4f24df9STrond Myklebust } else if (atomic_long_read(&nfsi->commit_info.ncommit)) 1998c4f24df9STrond Myklebust goto out_mark_dirty; 1999c4f24df9STrond Myklebust 2000c4f24df9STrond Myklebust check_requests_outstanding: 2001c4f24df9STrond Myklebust if (!atomic_read(&nfsi->commit_info.rpcs_out)) 2002c4f24df9STrond Myklebust return ret; 2003420e3646STrond Myklebust out_mark_dirty: 20048fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 20058fc795f7STrond Myklebust return ret; 20068fc795f7STrond Myklebust } 200789d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 2008863a3c6cSAndy Adamson 2009acdc53b2STrond Myklebust /* 2010837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 2011837bb1d7STrond Myklebust * 2012837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 2013837bb1d7STrond Myklebust * client. 2014837bb1d7STrond Myklebust */ 2015837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 2016837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 2017837bb1d7STrond Myklebust { 2018837bb1d7STrond Myklebust int ret; 2019837bb1d7STrond Myklebust 2020837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 2021837bb1d7STrond Myklebust if (ret == 0) 2022837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 2023837bb1d7STrond Myklebust return ret; 2024837bb1d7STrond Myklebust } 2025837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 2026837bb1d7STrond Myklebust 2027837bb1d7STrond Myklebust /* 2028acdc53b2STrond Myklebust * flush the inode to disk. 2029acdc53b2STrond Myklebust */ 2030acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 203134901f70STrond Myklebust { 2032f4ce1299STrond Myklebust int ret; 203334901f70STrond Myklebust 2034f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 2035f4ce1299STrond Myklebust 20365bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 20376b196875SChuck Lever if (ret) 20386b196875SChuck Lever goto out; 20395bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20406b196875SChuck Lever if (ret < 0) 20416b196875SChuck Lever goto out; 20425bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 20436b196875SChuck Lever ret = 0; 2044f4ce1299STrond Myklebust 20456b196875SChuck Lever out: 2046f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 2047f4ce1299STrond Myklebust return ret; 20481c75950bSTrond Myklebust } 2049ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 20501c75950bSTrond Myklebust 20516d740c76SMatthew Wilcox (Oracle) int nfs_wb_folio_cancel(struct inode *inode, struct folio *folio) 20521b3b4a1aSTrond Myklebust { 20531b3b4a1aSTrond Myklebust struct nfs_page *req; 20541b3b4a1aSTrond Myklebust int ret = 0; 20551b3b4a1aSTrond Myklebust 20566d740c76SMatthew Wilcox (Oracle) folio_wait_writeback(folio); 20573e217045SWeston Andros Adamson 20583e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20593e217045SWeston Andros Adamson * request */ 20600c493b5cSTrond Myklebust req = nfs_lock_and_join_requests(folio); 20613e217045SWeston Andros Adamson 20623e217045SWeston Andros Adamson if (IS_ERR(req)) { 20633e217045SWeston Andros Adamson ret = PTR_ERR(req); 20643e217045SWeston Andros Adamson } else if (req) { 20656d740c76SMatthew Wilcox (Oracle) /* all requests from this folio have been cancelled by 20663e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20673e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20683e217045SWeston Andros Adamson * release it */ 20691b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20701d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20711b3b4a1aSTrond Myklebust } 20723e217045SWeston Andros Adamson 20731b3b4a1aSTrond Myklebust return ret; 20741b3b4a1aSTrond Myklebust } 20751b3b4a1aSTrond Myklebust 20765241060eSTrond Myklebust /** 20775241060eSTrond Myklebust * nfs_wb_folio - Write back all requests on one page 20785241060eSTrond Myklebust * @inode: pointer to page 20795241060eSTrond Myklebust * @folio: pointer to folio 20805241060eSTrond Myklebust * 20815241060eSTrond Myklebust * Assumes that the folio has been locked by the caller, and will 20825241060eSTrond Myklebust * not unlock it. 20831c75950bSTrond Myklebust */ 20845241060eSTrond Myklebust int nfs_wb_folio(struct inode *inode, struct folio *folio) 20851c75950bSTrond Myklebust { 20865241060eSTrond Myklebust loff_t range_start = folio_file_pos(folio); 20875241060eSTrond Myklebust loff_t range_end = range_start + (loff_t)folio_size(folio) - 1; 20887f2f12d9STrond Myklebust struct writeback_control wbc = { 20897f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20907f2f12d9STrond Myklebust .nr_to_write = 0, 20917f2f12d9STrond Myklebust .range_start = range_start, 20927f2f12d9STrond Myklebust .range_end = range_end, 20937f2f12d9STrond Myklebust }; 20947f2f12d9STrond Myklebust int ret; 20957f2f12d9STrond Myklebust 2096*256093feSTrond Myklebust trace_nfs_writeback_folio(inode, folio); 2097f4ce1299STrond Myklebust 20980522f6adSTrond Myklebust for (;;) { 20995241060eSTrond Myklebust folio_wait_writeback(folio); 21005241060eSTrond Myklebust if (folio_clear_dirty_for_io(folio)) { 21010c493b5cSTrond Myklebust ret = nfs_writepage_locked(folio, &wbc); 21027f2f12d9STrond Myklebust if (ret < 0) 21037f2f12d9STrond Myklebust goto out_error; 21040522f6adSTrond Myklebust continue; 21057f2f12d9STrond Myklebust } 2106f4ce1299STrond Myklebust ret = 0; 21075241060eSTrond Myklebust if (!folio_test_private(folio)) 21080522f6adSTrond Myklebust break; 21090522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 21107f2f12d9STrond Myklebust if (ret < 0) 21117f2f12d9STrond Myklebust goto out_error; 21127f2f12d9STrond Myklebust } 21137f2f12d9STrond Myklebust out_error: 2114*256093feSTrond Myklebust trace_nfs_writeback_folio_done(inode, folio, ret); 21157f2f12d9STrond Myklebust return ret; 21161c75950bSTrond Myklebust } 21171c75950bSTrond Myklebust 2118074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 21194ae84a80SMatthew Wilcox (Oracle) int nfs_migrate_folio(struct address_space *mapping, struct folio *dst, 21204ae84a80SMatthew Wilcox (Oracle) struct folio *src, enum migrate_mode mode) 2121074cc1deSTrond Myklebust { 21222da95652SJeff Layton /* 21234ae84a80SMatthew Wilcox (Oracle) * If the private flag is set, the folio is currently associated with 21242da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 21252da95652SJeff Layton * 21262da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 21272da95652SJeff Layton * that we can safely release the inode reference while holding 21284ae84a80SMatthew Wilcox (Oracle) * the folio lock. 21292da95652SJeff Layton */ 21304ae84a80SMatthew Wilcox (Oracle) if (folio_test_private(src)) 21312da95652SJeff Layton return -EBUSY; 2132074cc1deSTrond Myklebust 21334ae84a80SMatthew Wilcox (Oracle) if (folio_test_fscache(src)) { 213416f2f4e6SDavid Howells if (mode == MIGRATE_ASYNC) 21358c209ce7SDavid Howells return -EBUSY; 21364ae84a80SMatthew Wilcox (Oracle) folio_wait_fscache(src); 213716f2f4e6SDavid Howells } 2138074cc1deSTrond Myklebust 213954184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 2140074cc1deSTrond Myklebust } 2141074cc1deSTrond Myklebust #endif 2142074cc1deSTrond Myklebust 2143f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 21441da177e4SLinus Torvalds { 21451da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 21461e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 21471da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 214820c2df83SPaul Mundt NULL); 21491da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 21501da177e4SLinus Torvalds return -ENOMEM; 21511da177e4SLinus Torvalds 215293d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 21531da177e4SLinus Torvalds nfs_wdata_cachep); 21541da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 21553dd4765fSJeff Layton goto out_destroy_write_cache; 21561da177e4SLinus Torvalds 21570b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 21580b7c0153SFred Isaman sizeof(struct nfs_commit_data), 21590b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21600b7c0153SFred Isaman NULL); 21610b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21623dd4765fSJeff Layton goto out_destroy_write_mempool; 21630b7c0153SFred Isaman 216493d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21654c100210SYanchuan Nian nfs_cdata_cachep); 21661da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21673dd4765fSJeff Layton goto out_destroy_commit_cache; 21681da177e4SLinus Torvalds 216989a09141SPeter Zijlstra /* 217089a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 217189a09141SPeter Zijlstra * 217289a09141SPeter Zijlstra * 64MB: 8192k 217389a09141SPeter Zijlstra * 128MB: 11585k 217489a09141SPeter Zijlstra * 256MB: 16384k 217589a09141SPeter Zijlstra * 512MB: 23170k 217689a09141SPeter Zijlstra * 1GB: 32768k 217789a09141SPeter Zijlstra * 2GB: 46340k 217889a09141SPeter Zijlstra * 4GB: 65536k 217989a09141SPeter Zijlstra * 8GB: 92681k 218089a09141SPeter Zijlstra * 16GB: 131072k 218189a09141SPeter Zijlstra * 218289a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 218389a09141SPeter Zijlstra * Limit the default to 256M 218489a09141SPeter Zijlstra */ 2185ca79b0c2SArun KS nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); 218689a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 218789a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 218889a09141SPeter Zijlstra 21891da177e4SLinus Torvalds return 0; 21903dd4765fSJeff Layton 21913dd4765fSJeff Layton out_destroy_commit_cache: 21923dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21933dd4765fSJeff Layton out_destroy_write_mempool: 21943dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21953dd4765fSJeff Layton out_destroy_write_cache: 21963dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21973dd4765fSJeff Layton return -ENOMEM; 21981da177e4SLinus Torvalds } 21991da177e4SLinus Torvalds 2200266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 22011da177e4SLinus Torvalds { 22021da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 22033dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 22041da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 22051a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 22061da177e4SLinus Torvalds } 22071da177e4SLinus Torvalds 22084a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 22094a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 22104a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 22110eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 22120eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 22131ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 22144a0de55cSAnna Schumaker }; 2215