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, 663a3908c8STrond Myklebust struct page *page); 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 173bd37d6fcSTrond Myklebust static struct nfs_page * 174bd37d6fcSTrond Myklebust nfs_page_private_request(struct page *page) 175bd37d6fcSTrond Myklebust { 176bd37d6fcSTrond Myklebust if (!PagePrivate(page)) 177bd37d6fcSTrond Myklebust return NULL; 178bd37d6fcSTrond Myklebust return (struct nfs_page *)page_private(page); 179bd37d6fcSTrond Myklebust } 180bd37d6fcSTrond Myklebust 18184d3a9a9SWeston Andros Adamson /* 18284d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 18384d3a9a9SWeston Andros Adamson * 18484d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 18584d3a9a9SWeston Andros Adamson * 18684d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 18784d3a9a9SWeston Andros Adamson */ 18829418aa4SMel Gorman static struct nfs_page * 189b30d2f04STrond Myklebust nfs_page_find_private_request(struct page *page) 190277459d2STrond Myklebust { 1914b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(page); 192bd37d6fcSTrond Myklebust struct nfs_page *req; 193277459d2STrond Myklebust 194b30d2f04STrond Myklebust if (!PagePrivate(page)) 195b30d2f04STrond Myklebust return NULL; 1964b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 197bd37d6fcSTrond Myklebust req = nfs_page_private_request(page); 19884d3a9a9SWeston Andros Adamson if (req) { 19984d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 20029418aa4SMel Gorman kref_get(&req->wb_kref); 20184d3a9a9SWeston Andros Adamson } 2024b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 203b30d2f04STrond Myklebust return req; 204b30d2f04STrond Myklebust } 20529418aa4SMel Gorman 206b30d2f04STrond Myklebust static struct nfs_page * 207b30d2f04STrond Myklebust nfs_page_find_swap_request(struct page *page) 208b30d2f04STrond Myklebust { 209b30d2f04STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 210b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 211b30d2f04STrond Myklebust struct nfs_page *req = NULL; 212b30d2f04STrond Myklebust if (!PageSwapCache(page)) 213b30d2f04STrond Myklebust return NULL; 214e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 215b30d2f04STrond Myklebust if (PageSwapCache(page)) { 216b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 217b30d2f04STrond Myklebust page); 218b30d2f04STrond Myklebust if (req) { 219b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 220b30d2f04STrond Myklebust kref_get(&req->wb_kref); 221b30d2f04STrond Myklebust } 222b30d2f04STrond Myklebust } 223e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 224277459d2STrond Myklebust return req; 225277459d2STrond Myklebust } 226277459d2STrond Myklebust 22784d3a9a9SWeston Andros Adamson /* 22884d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 22984d3a9a9SWeston Andros Adamson * 23084d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 23184d3a9a9SWeston Andros Adamson */ 23284d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 233277459d2STrond Myklebust { 234b30d2f04STrond Myklebust struct nfs_page *req; 235277459d2STrond Myklebust 236b30d2f04STrond Myklebust req = nfs_page_find_private_request(page); 237b30d2f04STrond Myklebust if (!req) 238b30d2f04STrond Myklebust req = nfs_page_find_swap_request(page); 239277459d2STrond Myklebust return req; 240277459d2STrond Myklebust } 241277459d2STrond Myklebust 242e00ed89dSTrond Myklebust static struct nfs_page *nfs_find_and_lock_page_request(struct page *page) 243e00ed89dSTrond Myklebust { 244e00ed89dSTrond Myklebust struct inode *inode = page_file_mapping(page)->host; 245e00ed89dSTrond Myklebust struct nfs_page *req, *head; 246e00ed89dSTrond Myklebust int ret; 247e00ed89dSTrond Myklebust 248e00ed89dSTrond Myklebust for (;;) { 249e00ed89dSTrond Myklebust req = nfs_page_find_head_request(page); 250e00ed89dSTrond Myklebust if (!req) 251e00ed89dSTrond Myklebust return req; 252e00ed89dSTrond Myklebust head = nfs_page_group_lock_head(req); 253e00ed89dSTrond Myklebust if (head != req) 254e00ed89dSTrond Myklebust nfs_release_request(req); 255e00ed89dSTrond Myklebust if (IS_ERR(head)) 256e00ed89dSTrond Myklebust return head; 257e00ed89dSTrond Myklebust ret = nfs_cancel_remove_inode(head, inode); 258e00ed89dSTrond Myklebust if (ret < 0) { 259e00ed89dSTrond Myklebust nfs_unlock_and_release_request(head); 260e00ed89dSTrond Myklebust return ERR_PTR(ret); 261e00ed89dSTrond Myklebust } 262e00ed89dSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 263e00ed89dSTrond Myklebust if (head == nfs_page_private_request(page)) 264e00ed89dSTrond Myklebust break; 265e00ed89dSTrond Myklebust if (PageSwapCache(page)) 266e00ed89dSTrond Myklebust break; 267e00ed89dSTrond Myklebust nfs_unlock_and_release_request(head); 268e00ed89dSTrond Myklebust } 269e00ed89dSTrond Myklebust return head; 270e00ed89dSTrond Myklebust } 271e00ed89dSTrond Myklebust 2721da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2731da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2741da177e4SLinus Torvalds { 275d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 276a3d01454STrond Myklebust loff_t end, i_size; 277a3d01454STrond Myklebust pgoff_t end_index; 2781da177e4SLinus Torvalds 279a3d01454STrond Myklebust spin_lock(&inode->i_lock); 280a3d01454STrond Myklebust i_size = i_size_read(inode); 28109cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2828cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 283a3d01454STrond Myklebust goto out; 284d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2851da177e4SLinus Torvalds if (i_size >= end) 286a3d01454STrond Myklebust goto out; 287110cb2d2SChuck Lever trace_nfs_size_grow(inode, end); 2881da177e4SLinus Torvalds i_size_write(inode, end); 289f6cdfa6dSTrond Myklebust NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 290a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 291a3d01454STrond Myklebust out: 292a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 293a6b5a28eSDave Wysochanski nfs_fscache_invalidate(inode, 0); 2941da177e4SLinus Torvalds } 2951da177e4SLinus Torvalds 296a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 297d2ceb7e5SBenjamin Coddington static void nfs_set_pageerror(struct address_space *mapping) 298a301b777STrond Myklebust { 2990df68cedSTrond Myklebust struct inode *inode = mapping->host; 3000df68cedSTrond Myklebust 301d2ceb7e5SBenjamin Coddington nfs_zap_mapping(mapping->host, mapping); 3020df68cedSTrond Myklebust /* Force file size revalidation */ 3030df68cedSTrond Myklebust spin_lock(&inode->i_lock); 304ac46b3d7STrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_REVAL_FORCED | 30588a6099fSTrond Myklebust NFS_INO_INVALID_CHANGE | 306ac46b3d7STrond Myklebust NFS_INO_INVALID_SIZE); 3070df68cedSTrond Myklebust spin_unlock(&inode->i_lock); 308a301b777STrond Myklebust } 309a301b777STrond Myklebust 3106fbda89bSTrond Myklebust static void nfs_mapping_set_error(struct page *page, int error) 3116fbda89bSTrond Myklebust { 312b8946d7bSTrond Myklebust struct address_space *mapping = page_file_mapping(page); 313b8946d7bSTrond Myklebust 3146fbda89bSTrond Myklebust SetPageError(page); 3156c984083STrond Myklebust filemap_set_wb_err(mapping, error); 3166c984083STrond Myklebust if (mapping->host) 3176c984083STrond Myklebust errseq_set(&mapping->host->i_sb->s_wb_err, 3186c984083STrond Myklebust error == -ENOSPC ? -ENOSPC : -EIO); 319b8946d7bSTrond Myklebust nfs_set_pageerror(mapping); 3206fbda89bSTrond Myklebust } 3216fbda89bSTrond Myklebust 322d72ddcbaSWeston Andros Adamson /* 323d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 324d72ddcbaSWeston Andros Adamson * @head - head request of page group 325d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 326d72ddcbaSWeston Andros Adamson * 327d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 328d72ddcbaSWeston Andros Adamson * page offset @page_offset. 329d72ddcbaSWeston Andros Adamson * 330d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 331d72ddcbaSWeston Andros Adamson * match is found. 332d72ddcbaSWeston Andros Adamson * 333d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 334d72ddcbaSWeston Andros Adamson */ 335d72ddcbaSWeston Andros Adamson static struct nfs_page * 336d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 337d72ddcbaSWeston Andros Adamson { 338d72ddcbaSWeston Andros Adamson struct nfs_page *req; 339d72ddcbaSWeston Andros Adamson 340d72ddcbaSWeston Andros Adamson req = head; 341d72ddcbaSWeston Andros Adamson do { 342d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 343d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 344d72ddcbaSWeston Andros Adamson return req; 345d72ddcbaSWeston Andros Adamson 346d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 347d72ddcbaSWeston Andros Adamson } while (req != head); 348d72ddcbaSWeston Andros Adamson 349d72ddcbaSWeston Andros Adamson return NULL; 350d72ddcbaSWeston Andros Adamson } 351d72ddcbaSWeston Andros Adamson 352d72ddcbaSWeston Andros Adamson /* 353d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 354d72ddcbaSWeston Andros Adamson * @head - head request of page group 355d72ddcbaSWeston Andros Adamson * 356d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 357d72ddcbaSWeston Andros Adamson * returns false otherwise 358d72ddcbaSWeston Andros Adamson */ 359d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 360d72ddcbaSWeston Andros Adamson { 361d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 362d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 363d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 364d72ddcbaSWeston Andros Adamson 3651344b7eaSTrond Myklebust nfs_page_group_lock(req); 366d72ddcbaSWeston Andros Adamson 3677e8a30f8STrond Myklebust for (;;) { 368d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 3697e8a30f8STrond Myklebust if (!tmp) 3707e8a30f8STrond Myklebust break; 3717e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 372d72ddcbaSWeston Andros Adamson } 373d72ddcbaSWeston Andros Adamson 374d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3757e8a30f8STrond Myklebust return pos >= len; 376d72ddcbaSWeston Andros Adamson } 377d72ddcbaSWeston Andros Adamson 3781da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3791da177e4SLinus Torvalds * covers the full page. 3801da177e4SLinus Torvalds */ 381d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3821da177e4SLinus Torvalds { 383d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 3841da177e4SLinus Torvalds return; 385d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3861da177e4SLinus Torvalds return; 387d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 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 409deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 41089a09141SPeter Zijlstra { 4110db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 4120db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 4135a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 4145a6d41b3STrond Myklebust 415deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 41689a09141SPeter Zijlstra 417277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 4180db10944SJan Kara NFS_CONGESTION_ON_THRESH) 4196df25e58SNeilBrown nfss->write_congested = 1; 42089a09141SPeter Zijlstra } 42189a09141SPeter Zijlstra 42220633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 42389a09141SPeter Zijlstra { 42420633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 42589a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 42631a01f09STrond Myklebust bool is_done; 42789a09141SPeter Zijlstra 42831a01f09STrond Myklebust is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); 42931a01f09STrond Myklebust nfs_unlock_request(req); 43031a01f09STrond Myklebust if (!is_done) 43120633f04SWeston Andros Adamson return; 43220633f04SWeston Andros Adamson 43320633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 434c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 4356df25e58SNeilBrown nfss->write_congested = 0; 43689a09141SPeter Zijlstra } 43789a09141SPeter Zijlstra 438d4581383SWeston Andros Adamson /* 439d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 440d4581383SWeston Andros Adamson * 441d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 442d4581383SWeston Andros Adamson * @old_head - the old head of the list 443d4581383SWeston Andros Adamson * 444d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 445d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 446d4581383SWeston Andros Adamson * with a reference held by some other context. 447d4581383SWeston Andros Adamson */ 448d4581383SWeston Andros Adamson static void 449d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 450b66aaa8dSTrond Myklebust struct nfs_page *old_head, 451b66aaa8dSTrond Myklebust struct inode *inode) 452d4581383SWeston Andros Adamson { 453d4581383SWeston Andros Adamson while (destroy_list) { 454d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 455d4581383SWeston Andros Adamson 456d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 457d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 458d4581383SWeston Andros Adamson 45908ca8b21STrond Myklebust /* Note: lock subreq in order to change subreq->wb_head */ 46008ca8b21STrond Myklebust nfs_page_set_headlock(subreq); 461d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 462d4581383SWeston Andros Adamson 463d4581383SWeston Andros Adamson /* make sure old group is not used */ 464d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 46508ca8b21STrond Myklebust subreq->wb_head = subreq; 466d4581383SWeston Andros Adamson 467902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 468902a4c00STrond Myklebust 4695b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4705b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4715b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 47208ca8b21STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) { 47308ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 4745b2b5187STrond Myklebust nfs_free_request(subreq); 47508ca8b21STrond Myklebust } else 47608ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 4775b2b5187STrond Myklebust continue; 4785b2b5187STrond Myklebust } 47908ca8b21STrond Myklebust nfs_page_clear_headlock(subreq); 480d4581383SWeston Andros Adamson 481add42de3STrond Myklebust nfs_release_request(old_head); 4825b2b5187STrond Myklebust 483b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 484d4581383SWeston Andros Adamson nfs_release_request(subreq); 485a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 486d4581383SWeston Andros Adamson } 4875b2b5187STrond Myklebust 4885b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4895b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4905b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 491d4581383SWeston Andros Adamson } 492d4581383SWeston Andros Adamson } 493d4581383SWeston Andros Adamson 494d4581383SWeston Andros Adamson /* 495e00ed89dSTrond Myklebust * nfs_join_page_group - destroy subrequests of the head req 496e00ed89dSTrond Myklebust * @head: the page used to lookup the "page group" of nfs_page structures 497e00ed89dSTrond Myklebust * @inode: Inode to which the request belongs. 498d4581383SWeston Andros Adamson * 499d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 500d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 501d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 502d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 503d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 504d4581383SWeston Andros Adamson */ 505ed5d588fSTrond Myklebust void 506e00ed89dSTrond Myklebust nfs_join_page_group(struct nfs_page *head, struct inode *inode) 507d4581383SWeston Andros Adamson { 508e00ed89dSTrond Myklebust struct nfs_page *subreq; 509d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 510a62f8e3bSTrond Myklebust unsigned int pgbase, off, bytes; 511a62f8e3bSTrond Myklebust 512a62f8e3bSTrond Myklebust pgbase = head->wb_pgbase; 513a62f8e3bSTrond Myklebust bytes = head->wb_bytes; 514a62f8e3bSTrond Myklebust off = head->wb_offset; 515a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 516a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 517a62f8e3bSTrond Myklebust /* Subrequests should always form a contiguous range */ 518a62f8e3bSTrond Myklebust if (pgbase > subreq->wb_pgbase) { 519a62f8e3bSTrond Myklebust off -= pgbase - subreq->wb_pgbase; 520a62f8e3bSTrond Myklebust bytes += pgbase - subreq->wb_pgbase; 521a62f8e3bSTrond Myklebust pgbase = subreq->wb_pgbase; 522a62f8e3bSTrond Myklebust } 523a62f8e3bSTrond Myklebust bytes = max(subreq->wb_pgbase + subreq->wb_bytes 524a62f8e3bSTrond Myklebust - pgbase, bytes); 5251bd5d6d0STrond Myklebust } 5261bd5d6d0STrond Myklebust 527a62f8e3bSTrond Myklebust /* Set the head request's range to cover the former page group */ 528a62f8e3bSTrond Myklebust head->wb_pgbase = pgbase; 529a62f8e3bSTrond Myklebust head->wb_bytes = bytes; 530a62f8e3bSTrond Myklebust head->wb_offset = off; 531d4581383SWeston Andros Adamson 532d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 533d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 534d4581383SWeston Andros Adamson subreq = head; 535d4581383SWeston Andros Adamson do { 536411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 537d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 538d4581383SWeston Andros Adamson } while (subreq != head); 539d4581383SWeston Andros Adamson 540d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 541d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 542d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 543d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 544d4581383SWeston Andros Adamson head->wb_this_page = head; 545d4581383SWeston Andros Adamson } 546d4581383SWeston Andros Adamson 547b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 548b5bab9bfSTrond Myklebust } 549b5bab9bfSTrond Myklebust 550e00ed89dSTrond Myklebust /* 551e00ed89dSTrond Myklebust * nfs_lock_and_join_requests - join all subreqs to the head req 552e00ed89dSTrond Myklebust * @page: the page used to lookup the "page group" of nfs_page structures 553e00ed89dSTrond Myklebust * 554e00ed89dSTrond Myklebust * This function joins all sub requests to the head request by first 555e00ed89dSTrond Myklebust * locking all requests in the group, cancelling any pending operations 556e00ed89dSTrond Myklebust * and finally updating the head request to cover the whole range covered by 557e00ed89dSTrond Myklebust * the (former) group. All subrequests are removed from any write or commit 558e00ed89dSTrond Myklebust * lists, unlinked from the group and destroyed. 559e00ed89dSTrond Myklebust * 560e00ed89dSTrond Myklebust * Returns a locked, referenced pointer to the head request - which after 561e00ed89dSTrond Myklebust * this call is guaranteed to be the only request associated with the page. 562e00ed89dSTrond Myklebust * Returns NULL if no requests are found for @page, or a ERR_PTR if an 563e00ed89dSTrond Myklebust * error was encountered. 564e00ed89dSTrond Myklebust */ 565e00ed89dSTrond Myklebust static struct nfs_page * 566e00ed89dSTrond Myklebust nfs_lock_and_join_requests(struct page *page) 567e00ed89dSTrond Myklebust { 568e00ed89dSTrond Myklebust struct inode *inode = page_file_mapping(page)->host; 569e00ed89dSTrond Myklebust struct nfs_page *head; 570e00ed89dSTrond Myklebust int ret; 571e00ed89dSTrond Myklebust 572e00ed89dSTrond Myklebust /* 573e00ed89dSTrond Myklebust * A reference is taken only on the head request which acts as a 574e00ed89dSTrond Myklebust * reference to the whole page group - the group will not be destroyed 575e00ed89dSTrond Myklebust * until the head reference is released. 576e00ed89dSTrond Myklebust */ 577e00ed89dSTrond Myklebust head = nfs_find_and_lock_page_request(page); 578e00ed89dSTrond Myklebust if (IS_ERR_OR_NULL(head)) 579d4581383SWeston Andros Adamson return head; 5800671d8f1SMarkus Elfring 581e00ed89dSTrond Myklebust /* lock each request in the page group */ 582e00ed89dSTrond Myklebust ret = nfs_page_group_lock_subrequests(head); 583e00ed89dSTrond Myklebust if (ret < 0) { 5840671d8f1SMarkus Elfring nfs_unlock_and_release_request(head); 5850671d8f1SMarkus Elfring return ERR_PTR(ret); 586612c9384STrond Myklebust } 587074cc1deSTrond Myklebust 588e00ed89dSTrond Myklebust nfs_join_page_group(head, inode); 589e00ed89dSTrond Myklebust 590e00ed89dSTrond Myklebust return head; 591e00ed89dSTrond Myklebust } 592e00ed89dSTrond Myklebust 5936fbda89bSTrond Myklebust static void nfs_write_error(struct nfs_page *req, int error) 5940bcbf039SPeng Tao { 595861e1671SChuck Lever trace_nfs_write_error(req, error); 5966fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, error); 59706c9fdf3STrond Myklebust nfs_inode_remove_request(req); 5980bcbf039SPeng Tao nfs_end_page_writeback(req); 5991f84ccdfSFred Isaman nfs_release_request(req); 6000bcbf039SPeng Tao } 6010bcbf039SPeng Tao 602074cc1deSTrond Myklebust /* 603074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 604074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 605074cc1deSTrond Myklebust */ 606c6fd3511STrond Myklebust static int nfs_page_async_flush(struct page *page, 607c6fd3511STrond Myklebust struct writeback_control *wbc, 608c6fd3511STrond Myklebust struct nfs_pageio_descriptor *pgio) 609074cc1deSTrond Myklebust { 610074cc1deSTrond Myklebust struct nfs_page *req; 611074cc1deSTrond Myklebust int ret = 0; 612074cc1deSTrond Myklebust 6136d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 614074cc1deSTrond Myklebust if (!req) 615074cc1deSTrond Myklebust goto out; 616074cc1deSTrond Myklebust ret = PTR_ERR(req); 617074cc1deSTrond Myklebust if (IS_ERR(req)) 618074cc1deSTrond Myklebust goto out; 619074cc1deSTrond Myklebust 620deed85e7STrond Myklebust nfs_set_page_writeback(page); 621deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 622074cc1deSTrond Myklebust 623a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 62496c41455STrond Myklebust ret = pgio->pg_error; 62596c41455STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 626a6598813STrond Myklebust goto out_launder; 627a6598813STrond Myklebust 62896c41455STrond Myklebust ret = 0; 629f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 630074cc1deSTrond Myklebust ret = pgio->pg_error; 6310bcbf039SPeng Tao /* 632c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6330bcbf039SPeng Tao */ 634c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 635a6598813STrond Myklebust goto out_launder; 636c6fd3511STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) 637c6fd3511STrond Myklebust ret = AOP_WRITEPAGE_ACTIVATE; 638c6fd3511STrond Myklebust redirty_page_for_writepage(wbc, page); 6398fc75bedSTrond Myklebust nfs_redirty_request(req); 64096c41455STrond Myklebust pgio->pg_error = 0; 64140f90271STrond Myklebust } else 64240f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 64340f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 644074cc1deSTrond Myklebust out: 645074cc1deSTrond Myklebust return ret; 646a6598813STrond Myklebust out_launder: 6476fbda89bSTrond Myklebust nfs_write_error(req, ret); 64814bebe3cSTrond Myklebust return 0; 649e261f51fSTrond Myklebust } 650e261f51fSTrond Myklebust 651d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 652c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 653f758c885STrond Myklebust { 6548cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 655c6fd3511STrond Myklebust return nfs_page_async_flush(page, wbc, pgio); 656f758c885STrond Myklebust } 657f758c885STrond Myklebust 658e261f51fSTrond Myklebust /* 6591da177e4SLinus Torvalds * Write an mmapped page to the server. 6601da177e4SLinus Torvalds */ 661d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 662c373fff7STrond Myklebust struct writeback_control *wbc) 6631da177e4SLinus Torvalds { 664f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 66540f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 666e261f51fSTrond Myklebust int err; 6671da177e4SLinus Torvalds 6686df25e58SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 6696df25e58SNeilBrown NFS_SERVER(inode)->write_congested) 6706df25e58SNeilBrown return AOP_WRITEPAGE_ACTIVATE; 6716df25e58SNeilBrown 67240f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 673811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 674a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 675c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 67696c41455STrond Myklebust pgio.pg_error = 0; 677f758c885STrond Myklebust nfs_pageio_complete(&pgio); 6784d770ccfSTrond Myklebust return err; 6794d770ccfSTrond Myklebust } 6804d770ccfSTrond Myklebust 6814d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6824d770ccfSTrond Myklebust { 683f758c885STrond Myklebust int ret; 6844d770ccfSTrond Myklebust 685c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 68696c41455STrond Myklebust if (ret != AOP_WRITEPAGE_ACTIVATE) 6871da177e4SLinus Torvalds unlock_page(page); 688f758c885STrond Myklebust return ret; 689f758c885STrond Myklebust } 690f758c885STrond Myklebust 691f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 692f758c885STrond Myklebust { 693f758c885STrond Myklebust int ret; 694f758c885STrond Myklebust 695c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 69696c41455STrond Myklebust if (ret != AOP_WRITEPAGE_ACTIVATE) 697f758c885STrond Myklebust unlock_page(page); 698f758c885STrond Myklebust return ret; 6991da177e4SLinus Torvalds } 7001da177e4SLinus Torvalds 701919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 702919e3bd9STrond Myklebust { 703919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 704919e3bd9STrond Myklebust } 705919e3bd9STrond Myklebust 7061da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7071da177e4SLinus Torvalds { 7081da177e4SLinus Torvalds struct inode *inode = mapping->host; 709c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 710ed7bcdb3STrond Myklebust struct nfs_io_completion *ioc = NULL; 711ed7bcdb3STrond Myklebust unsigned int mntflags = NFS_SERVER(inode)->flags; 712ed7bcdb3STrond Myklebust int priority = 0; 7131da177e4SLinus Torvalds int err; 7141da177e4SLinus Torvalds 7156df25e58SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 7166df25e58SNeilBrown NFS_SERVER(inode)->write_congested) 7176df25e58SNeilBrown return 0; 7186df25e58SNeilBrown 71991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 72091d5b470SChuck Lever 721ed7bcdb3STrond Myklebust if (!(mntflags & NFS_MOUNT_WRITE_EAGER) || wbc->for_kupdate || 722ed7bcdb3STrond Myklebust wbc->for_background || wbc->for_sync || wbc->for_reclaim) { 7232b17d725STrond Myklebust ioc = nfs_io_completion_alloc(GFP_KERNEL); 724919e3bd9STrond Myklebust if (ioc) 725ed7bcdb3STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, 726ed7bcdb3STrond Myklebust inode); 727ed7bcdb3STrond Myklebust priority = wb_priority(wbc); 728ed7bcdb3STrond Myklebust } 729919e3bd9STrond Myklebust 730c6fd3511STrond Myklebust do { 731ed7bcdb3STrond Myklebust nfs_pageio_init_write(&pgio, inode, priority, false, 732a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 733919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 734c6fd3511STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, 735c6fd3511STrond Myklebust &pgio); 73696c41455STrond Myklebust pgio.pg_error = 0; 737c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 738c6fd3511STrond Myklebust } while (err < 0 && !nfs_error_is_fatal(err)); 739919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 74072cb77f4STrond Myklebust 741f758c885STrond Myklebust if (err < 0) 74272cb77f4STrond Myklebust goto out_err; 743c63c7b05STrond Myklebust return 0; 74472cb77f4STrond Myklebust out_err: 74572cb77f4STrond Myklebust return err; 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds 7481da177e4SLinus Torvalds /* 7491da177e4SLinus Torvalds * Insert a write request into an inode 7501da177e4SLinus Torvalds */ 751d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7521da177e4SLinus Torvalds { 7534b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7541da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 755e7d39069STrond Myklebust 7562bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7572bfc6e56SWeston Andros Adamson 758e7d39069STrond Myklebust /* Lock the request! */ 7597ad84aa9STrond Myklebust nfs_lock_request(req); 760e7d39069STrond Myklebust 76129418aa4SMel Gorman /* 76229418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 76329418aa4SMel Gorman * with invalidate/truncate. 76429418aa4SMel Gorman */ 7654b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 76629418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7672df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 768deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 769277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 77029418aa4SMel Gorman } 7714b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 772a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 77317089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 774cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 775cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 776cb1410c7SWeston Andros Adamson * adding subrequests. */ 77717089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 778c03b4024STrond Myklebust kref_get(&req->wb_kref); 7791da177e4SLinus Torvalds } 7801da177e4SLinus Torvalds 7811da177e4SLinus Torvalds /* 78289a09141SPeter Zijlstra * Remove a write request from an inode 7831da177e4SLinus Torvalds */ 7841da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7851da177e4SLinus Torvalds { 7864b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7874b9bb25bSTrond Myklebust struct inode *inode = mapping->host; 7881da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 78920633f04SWeston Andros Adamson struct nfs_page *head; 79020633f04SWeston Andros Adamson 79120633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 79220633f04SWeston Andros Adamson head = req->wb_head; 7931da177e4SLinus Torvalds 7944b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 79567911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 79620633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 79720633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 79820633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 79929418aa4SMel Gorman } 8004b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 80120633f04SWeston Andros Adamson } 80217089a29SWeston Andros Adamson 80333ea5aaaSZhangXiaoxu if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) { 8041da177e4SLinus Torvalds nfs_release_request(req); 80533ea5aaaSZhangXiaoxu atomic_long_dec(&nfsi->nrequests); 80633ea5aaaSZhangXiaoxu } 8071da177e4SLinus Torvalds } 8081da177e4SLinus Torvalds 80961822ab5STrond Myklebust static void 8106d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 81161822ab5STrond Myklebust { 81267911c8fSAnna Schumaker if (req->wb_page) 81361822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 81461822ab5STrond Myklebust } 81561822ab5STrond Myklebust 8163a3908c8STrond Myklebust /* 8173a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8183a3908c8STrond Myklebust * 8193a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8203a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8213a3908c8STrond Myklebust * 8223a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8233a3908c8STrond Myklebust */ 8243a3908c8STrond Myklebust static struct nfs_page * 8253a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8263a3908c8STrond Myklebust struct page *page) 8273a3908c8STrond Myklebust { 8283a3908c8STrond Myklebust struct nfs_page *freq, *t; 8293a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8303a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8313a3908c8STrond Myklebust 8323a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8333a3908c8STrond Myklebust 8343a3908c8STrond Myklebust /* search through pnfs commit lists */ 8353a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8363a3908c8STrond Myklebust if (freq) 8373a3908c8STrond Myklebust return freq->wb_head; 8383a3908c8STrond Myklebust 8393a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8403a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8413a3908c8STrond Myklebust if (freq->wb_page == page) 8423a3908c8STrond Myklebust return freq->wb_head; 8433a3908c8STrond Myklebust } 8443a3908c8STrond Myklebust 8453a3908c8STrond Myklebust return NULL; 8463a3908c8STrond Myklebust } 8473a3908c8STrond Myklebust 8488dd37758STrond Myklebust /** 84986d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 85086d80f97STrond Myklebust * @req: pointer to a struct nfs_page 85186d80f97STrond Myklebust * @dst: commit list head 85286d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 85386d80f97STrond Myklebust * 85486d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 85586d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 85686d80f97STrond Myklebust * the MM page stats. 85786d80f97STrond Myklebust * 858e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 859e824f99aSTrond Myklebust * nfs_page lock. 86086d80f97STrond Myklebust */ 86186d80f97STrond Myklebust void 86286d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 86386d80f97STrond Myklebust struct nfs_commit_info *cinfo) 86486d80f97STrond Myklebust { 86586d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 86686d80f97STrond Myklebust nfs_list_add_request(req, dst); 8675cb953d4STrond Myklebust atomic_long_inc(&cinfo->mds->ncommit); 86886d80f97STrond Myklebust } 86986d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 87086d80f97STrond Myklebust 87186d80f97STrond Myklebust /** 8728dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8738dd37758STrond Myklebust * @req: pointer to a struct nfs_page 874ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8758dd37758STrond Myklebust * 876ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8778dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8788dd37758STrond Myklebust * the MM page stats. 8798dd37758STrond Myklebust * 880ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8818dd37758STrond Myklebust * holding the nfs_page lock. 8828dd37758STrond Myklebust */ 8838dd37758STrond Myklebust void 8846272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8858dd37758STrond Myklebust { 886e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8876272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 888e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 88967911c8fSAnna Schumaker if (req->wb_page) 89086d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8918dd37758STrond Myklebust } 8928dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8938dd37758STrond Myklebust 8948dd37758STrond Myklebust /** 8958dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8968dd37758STrond Myklebust * @req: pointer to a nfs_page 897ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8988dd37758STrond Myklebust * 899ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 9008dd37758STrond Myklebust * number of outstanding requests requiring a commit 9018dd37758STrond Myklebust * It does not update the MM page stats. 9028dd37758STrond Myklebust * 903ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9048dd37758STrond Myklebust */ 9058dd37758STrond Myklebust void 906ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 907ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9088dd37758STrond Myklebust { 9098dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9108dd37758STrond Myklebust return; 9118dd37758STrond Myklebust nfs_list_remove_request(req); 9125cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9138dd37758STrond Myklebust } 9148dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9158dd37758STrond Myklebust 916ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 917ea2cf228SFred Isaman struct inode *inode) 918ea2cf228SFred Isaman { 919fe238e60SDave Wysochanski cinfo->inode = inode; 920ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 921ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 922b359f9d0SFred Isaman cinfo->dreq = NULL; 923f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 924ea2cf228SFred Isaman } 925ea2cf228SFred Isaman 926ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 927ea2cf228SFred Isaman struct inode *inode, 928ea2cf228SFred Isaman struct nfs_direct_req *dreq) 929ea2cf228SFred Isaman { 9301763da12SFred Isaman if (dreq) 9311763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9321763da12SFred Isaman else 933ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 934ea2cf228SFred Isaman } 935ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9368dd37758STrond Myklebust 9371da177e4SLinus Torvalds /* 9381da177e4SLinus Torvalds * Add a request to the inode's commit list. 9391da177e4SLinus Torvalds */ 9401763da12SFred Isaman void 941ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 942b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9431da177e4SLinus Torvalds { 944b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9458dd37758STrond Myklebust return; 9466272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9471da177e4SLinus Torvalds } 9488e821cadSTrond Myklebust 949d6d6dc7cSFred Isaman static void 950d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 951e468bae9STrond Myklebust { 9528d92890bSNeilBrown dec_node_page_state(page, NR_WRITEBACK); 95393f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 9548d92890bSNeilBrown WB_WRITEBACK); 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); 9728dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 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)) { 1003861e1671SChuck Lever trace_nfs_comp_error(req, hdr->error); 10046fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, hdr->error); 10056c75dc0dSFred Isaman goto remove_req; 10066c75dc0dSFred Isaman } 1007c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 100833344e0fSTrond Myklebust /* Reset wb_nio, since the write was successful. */ 100933344e0fSTrond Myklebust req->wb_nio = 0; 1010f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1011b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1012a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10136c75dc0dSFred Isaman goto next; 10146c75dc0dSFred Isaman } 10156c75dc0dSFred Isaman remove_req: 10166c75dc0dSFred Isaman nfs_inode_remove_request(req); 10176c75dc0dSFred Isaman next: 101820633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10193aff4ebbSTrond Myklebust nfs_release_request(req); 10206c75dc0dSFred Isaman } 10216c75dc0dSFred Isaman out: 1022919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10236c75dc0dSFred Isaman hdr->release(hdr); 10246c75dc0dSFred Isaman } 10256c75dc0dSFred Isaman 1026ce59515cSAnna Schumaker unsigned long 1027ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1028fb8a1f11STrond Myklebust { 10295cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1030fb8a1f11STrond Myklebust } 1031fb8a1f11STrond Myklebust 1032e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10331763da12SFred Isaman int 1034ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1035ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1036d6d6dc7cSFred Isaman { 1037137da553STrond Myklebust struct nfs_page *req, *tmp; 1038d6d6dc7cSFred Isaman int ret = 0; 1039d6d6dc7cSFred Isaman 1040137da553STrond Myklebust list_for_each_entry_safe(req, tmp, src, wb_list) { 10417ad84aa9STrond Myklebust kref_get(&req->wb_kref); 10422ce209c4STrond Myklebust if (!nfs_lock_request(req)) { 1043137da553STrond Myklebust nfs_release_request(req); 1044137da553STrond Myklebust continue; 1045137da553STrond Myklebust } 1046ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10475d2a9d9dSTrond Myklebust clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); 10488dd37758STrond Myklebust nfs_list_add_request(req, dst); 1049d6d6dc7cSFred Isaman ret++; 10501763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1051d6d6dc7cSFred Isaman break; 1052e824f99aSTrond Myklebust cond_resched(); 1053d6d6dc7cSFred Isaman } 1054d6d6dc7cSFred Isaman return ret; 1055d6d6dc7cSFred Isaman } 10565d2a9d9dSTrond Myklebust EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 1057d6d6dc7cSFred Isaman 10581da177e4SLinus Torvalds /* 10591da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10601da177e4SLinus Torvalds * @inode: NFS inode to scan 1061ea2cf228SFred Isaman * @dst: mds destination list 1062ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10631da177e4SLinus Torvalds * 10641da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10651da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10661da177e4SLinus Torvalds */ 10671763da12SFred Isaman int 1068ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1069ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10701da177e4SLinus Torvalds { 1071d6d6dc7cSFred Isaman int ret = 0; 1072fb8a1f11STrond Myklebust 10735cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10745cb953d4STrond Myklebust return 0; 1075e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10765cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10778dd37758STrond Myklebust const int max = INT_MAX; 1078d6d6dc7cSFred Isaman 1079ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1080ea2cf228SFred Isaman cinfo, max); 1081ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1082d6d6dc7cSFred Isaman } 1083e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1084ff778d02STrond Myklebust return ret; 10851da177e4SLinus Torvalds } 1086d6d6dc7cSFred Isaman 10871da177e4SLinus Torvalds /* 1088e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1089e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10901da177e4SLinus Torvalds * 1091e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1092e7d39069STrond Myklebust * to disk. 10931da177e4SLinus Torvalds */ 1094e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1095e7d39069STrond Myklebust struct page *page, 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 1106f6032f21STrond Myklebust req = nfs_lock_and_join_requests(page); 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); 1139e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 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, 1151e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1152e7d39069STrond Myklebust { 1153d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1154e7d39069STrond Myklebust struct nfs_page *req; 1155e7d39069STrond Myklebust 1156e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1157e7d39069STrond Myklebust if (req != NULL) 1158e7d39069STrond Myklebust goto out; 115928b1d3f5STrond Myklebust req = nfs_create_request(ctx, page, offset, bytes); 1160e7d39069STrond Myklebust if (IS_ERR(req)) 1161e7d39069STrond Myklebust goto out; 1162d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1163efc91ed0STrond Myklebust out: 116461e930a9STrond Myklebust return req; 11651da177e4SLinus Torvalds } 11661da177e4SLinus Torvalds 1167e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1168e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1169e7d39069STrond Myklebust { 1170e7d39069STrond Myklebust struct nfs_page *req; 1171e7d39069STrond Myklebust 1172e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1173e7d39069STrond Myklebust if (IS_ERR(req)) 1174e7d39069STrond Myklebust return PTR_ERR(req); 1175e7d39069STrond Myklebust /* Update file length */ 1176e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1177d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1178a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11791d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1180e7d39069STrond Myklebust return 0; 1181e7d39069STrond Myklebust } 1182e7d39069STrond Myklebust 11831da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11841da177e4SLinus Torvalds { 1185cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11862a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1187bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11881da177e4SLinus Torvalds struct nfs_page *req; 11891a54533eSTrond Myklebust int do_flush, status; 11901da177e4SLinus Torvalds /* 11911da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11921da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11931da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11941da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11951da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11961da177e4SLinus Torvalds * dropped page. 11971da177e4SLinus Torvalds */ 11981a54533eSTrond Myklebust do { 119984d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 12001a54533eSTrond Myklebust if (req == NULL) 12011a54533eSTrond Myklebust return 0; 12022a369153STrond Myklebust l_ctx = req->wb_lock_context; 1203138a2935STrond Myklebust do_flush = req->wb_page != page || 12049fcd5960STrond Myklebust !nfs_match_open_context(nfs_req_openctx(req), ctx); 1205bd61e0a9SJeff Layton if (l_ctx && flctx && 1206bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1207bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1208d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12095263e31eSJeff Layton } 12101da177e4SLinus Torvalds nfs_release_request(req); 12111a54533eSTrond Myklebust if (!do_flush) 12121a54533eSTrond Myklebust return 0; 1213d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12141a54533eSTrond Myklebust } while (status == 0); 12151a54533eSTrond Myklebust return status; 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds 12181da177e4SLinus Torvalds /* 1219dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1220dc24826bSAndy Adamson * expire soon. 1221dc24826bSAndy Adamson * 1222dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1223dc24826bSAndy Adamson * 1224dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1225dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1226dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1227dc24826bSAndy Adamson */ 1228dc24826bSAndy Adamson int 1229dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1230dc24826bSAndy Adamson { 1231dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1232dc24826bSAndy Adamson 1233ddf529eeSNeilBrown if (nfs_ctx_key_to_expire(ctx, inode) && 1234ca05cbaeSTrond Myklebust !rcu_access_pointer(ctx->ll_cred)) 1235ddf529eeSNeilBrown /* Already expired! */ 1236ddf529eeSNeilBrown return -EACCES; 1237ddf529eeSNeilBrown return 0; 1238dc24826bSAndy Adamson } 1239dc24826bSAndy Adamson 1240dc24826bSAndy Adamson /* 1241dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1242dc24826bSAndy Adamson */ 1243ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1244dc24826bSAndy Adamson { 1245ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1246ca05cbaeSTrond Myklebust struct rpc_cred *cred, *new, *old = NULL; 1247ddf529eeSNeilBrown struct auth_cred acred = { 1248a52458b4SNeilBrown .cred = ctx->cred, 1249ddf529eeSNeilBrown }; 1250ca05cbaeSTrond Myklebust bool ret = false; 1251ce52914eSScott Mayhew 1252ca05cbaeSTrond Myklebust rcu_read_lock(); 1253ca05cbaeSTrond Myklebust cred = rcu_dereference(ctx->ll_cred); 1254ca05cbaeSTrond Myklebust if (cred && !(cred->cr_ops->crkey_timeout && 1255ca05cbaeSTrond Myklebust cred->cr_ops->crkey_timeout(cred))) 1256ca05cbaeSTrond Myklebust goto out; 1257ca05cbaeSTrond Myklebust rcu_read_unlock(); 1258ca05cbaeSTrond Myklebust 1259ca05cbaeSTrond Myklebust new = auth->au_ops->lookup_cred(auth, &acred, 0); 1260ca05cbaeSTrond Myklebust if (new == cred) { 1261ca05cbaeSTrond Myklebust put_rpccred(new); 1262ddf529eeSNeilBrown return true; 1263ca05cbaeSTrond Myklebust } 1264ca05cbaeSTrond Myklebust if (IS_ERR_OR_NULL(new)) { 1265ca05cbaeSTrond Myklebust new = NULL; 1266ca05cbaeSTrond Myklebust ret = true; 1267ca05cbaeSTrond Myklebust } else if (new->cr_ops->crkey_timeout && 1268ca05cbaeSTrond Myklebust new->cr_ops->crkey_timeout(new)) 1269ca05cbaeSTrond Myklebust ret = true; 1270ca05cbaeSTrond Myklebust 1271ca05cbaeSTrond Myklebust rcu_read_lock(); 1272ca05cbaeSTrond Myklebust old = rcu_dereference_protected(xchg(&ctx->ll_cred, 1273ca05cbaeSTrond Myklebust RCU_INITIALIZER(new)), 1); 1274ca05cbaeSTrond Myklebust out: 1275ca05cbaeSTrond Myklebust rcu_read_unlock(); 1276ca05cbaeSTrond Myklebust put_rpccred(old); 1277ca05cbaeSTrond Myklebust return ret; 1278dc24826bSAndy Adamson } 1279dc24826bSAndy Adamson 1280dc24826bSAndy Adamson /* 12815d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12825d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12835d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12845d47a356STrond Myklebust */ 1285fc9dc401STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode, 1286fc9dc401STrond Myklebust unsigned int pagelen) 12875d47a356STrond Myklebust { 1288d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1289d529ef83SJeff Layton 12908d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12918d197a56STrond Myklebust goto out; 1292fc9dc401STrond Myklebust if (nfsi->cache_validity & 129313c0b082STrond Myklebust (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE)) 1294d529ef83SJeff Layton return false; 12954db72b40SJeff Layton smp_rmb(); 1296fc9dc401STrond Myklebust if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags) && pagelen != 0) 12978d197a56STrond Myklebust return false; 12988d197a56STrond Myklebust out: 1299fc9dc401STrond Myklebust if (nfsi->cache_validity & NFS_INO_INVALID_DATA && pagelen != 0) 130018dd78c4SScott Mayhew return false; 13018d197a56STrond Myklebust return PageUptodate(page) != 0; 13025d47a356STrond Myklebust } 13035d47a356STrond Myklebust 13045263e31eSJeff Layton static bool 13055263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 13065263e31eSJeff Layton { 13075263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 13085263e31eSJeff Layton fl->fl_type == F_WRLCK; 13095263e31eSJeff Layton } 13105263e31eSJeff Layton 1311c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1312c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1313c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1314c7559663SScott Mayhew * inefficiencies. 1315c7559663SScott Mayhew * 1316263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1317263b4509SScott Mayhew * of the checks. 1318c7559663SScott Mayhew */ 1319fc9dc401STrond Myklebust static int nfs_can_extend_write(struct file *file, struct page *page, 1320fc9dc401STrond Myklebust struct inode *inode, unsigned int pagelen) 1321c7559663SScott Mayhew { 13225263e31eSJeff Layton int ret; 13235263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 13245263e31eSJeff Layton struct file_lock *fl; 13255263e31eSJeff Layton 1326c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1327c7559663SScott Mayhew return 0; 1328fc9dc401STrond Myklebust if (!nfs_write_pageuptodate(page, inode, pagelen)) 1329263b4509SScott Mayhew return 0; 1330c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1331c7559663SScott Mayhew return 1; 1332bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1333bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13348fa4592aSTrond Myklebust return 1; 13355263e31eSJeff Layton 13365263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13375263e31eSJeff Layton ret = 0; 13386109c850SJeff Layton spin_lock(&flctx->flc_lock); 1339bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1340bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1341bd61e0a9SJeff Layton fl_list); 1342bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13435263e31eSJeff Layton ret = 1; 1344bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13455263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13465263e31eSJeff Layton fl_list); 13475263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13485263e31eSJeff Layton ret = 1; 13495263e31eSJeff Layton } 13506109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13515263e31eSJeff Layton return ret; 1352c7559663SScott Mayhew } 1353c7559663SScott Mayhew 13545d47a356STrond Myklebust /* 13551da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13561da177e4SLinus Torvalds * 13571da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13581da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13591da177e4SLinus Torvalds */ 13601da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13611da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13621da177e4SLinus Torvalds { 1363cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1364d2ceb7e5SBenjamin Coddington struct address_space *mapping = page_file_mapping(page); 1365d2ceb7e5SBenjamin Coddington struct inode *inode = mapping->host; 1366fc9dc401STrond Myklebust unsigned int pagelen = nfs_page_length(page); 13671da177e4SLinus Torvalds int status = 0; 13681da177e4SLinus Torvalds 136991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 137091d5b470SChuck Lever 13716de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13726de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13731da177e4SLinus Torvalds 1374149a4fddSBenjamin Coddington if (!count) 1375149a4fddSBenjamin Coddington goto out; 1376149a4fddSBenjamin Coddington 1377fc9dc401STrond Myklebust if (nfs_can_extend_write(file, page, inode, pagelen)) { 1378fc9dc401STrond Myklebust count = max(count + offset, pagelen); 13791da177e4SLinus Torvalds offset = 0; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds 1382e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 138303fa9e84STrond Myklebust if (status < 0) 1384d2ceb7e5SBenjamin Coddington nfs_set_pageerror(mapping); 1385149a4fddSBenjamin Coddington out: 138648186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13871da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13881da177e4SLinus Torvalds return status; 13891da177e4SLinus Torvalds } 13901da177e4SLinus Torvalds 13913ff7576dSTrond Myklebust static int flush_task_priority(int how) 13921da177e4SLinus Torvalds { 13931da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13941da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13951da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13961da177e4SLinus Torvalds case FLUSH_LOWPRI: 13971da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13981da177e4SLinus Torvalds } 13991da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 14001da177e4SLinus Torvalds } 14011da177e4SLinus Torvalds 1402d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1403d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1404abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 14051ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 14061da177e4SLinus Torvalds { 14073ff7576dSTrond Myklebust int priority = flush_task_priority(how); 14081da177e4SLinus Torvalds 14098db55a03SNeilBrown if (IS_SWAPFILE(hdr->inode)) 14108db55a03SNeilBrown task_setup_data->flags |= RPC_TASK_SWAPPER; 14111ed26f33SAnna Schumaker task_setup_data->priority = priority; 1412fb91fb0eSAnna Schumaker rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); 14135bb2a7cbSTrond Myklebust trace_nfs_initiate_write(hdr); 1414275acaafSTrond Myklebust } 1415275acaafSTrond Myklebust 14166d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 14176d884e8fSFred * call this on each, which will prepare them to be retried on next 14186d884e8fSFred * writeback using standard nfs. 14196d884e8fSFred */ 14206d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14216d884e8fSFred { 142233344e0fSTrond Myklebust /* Bump the transmission count */ 142333344e0fSTrond Myklebust req->wb_nio++; 14246d884e8fSFred nfs_mark_request_dirty(req); 14259fcd5960STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); 142620633f04SWeston Andros Adamson nfs_end_page_writeback(req); 14273aff4ebbSTrond Myklebust nfs_release_request(req); 14286d884e8fSFred } 14296d884e8fSFred 1430df3accb8STrond Myklebust static void nfs_async_write_error(struct list_head *head, int error) 14316c75dc0dSFred Isaman { 14326c75dc0dSFred Isaman struct nfs_page *req; 14336c75dc0dSFred Isaman 14346c75dc0dSFred Isaman while (!list_empty(head)) { 14356c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14366c75dc0dSFred Isaman nfs_list_remove_request(req); 1437cea9ba72STrond Myklebust if (nfs_error_is_fatal_on_server(error)) 14386fbda89bSTrond Myklebust nfs_write_error(req, error); 14396fbda89bSTrond Myklebust else 14406c75dc0dSFred Isaman nfs_redirty_request(req); 14416c75dc0dSFred Isaman } 14426c75dc0dSFred Isaman } 14436c75dc0dSFred Isaman 1444dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1445dc602dd7STrond Myklebust { 1446df3accb8STrond Myklebust nfs_async_write_error(&hdr->pages, 0); 14477be7b3caSTrond Myklebust filemap_fdatawrite_range(hdr->inode->i_mapping, hdr->args.offset, 14487be7b3caSTrond Myklebust hdr->args.offset + hdr->args.count - 1); 1449dc602dd7STrond Myklebust } 1450dc602dd7STrond Myklebust 1451061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1452919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1453061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1454061ae2edSFred Isaman .completion = nfs_write_completion, 1455dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1456061ae2edSFred Isaman }; 1457061ae2edSFred Isaman 145857208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1459a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1460061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14611751c363STrond Myklebust { 1462a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 146341d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1464a20c93e3SChristoph Hellwig 1465a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1466a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1467a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1468a20c93e3SChristoph Hellwig #endif 14694a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 14703bde7afdSTrond Myklebust server->wsize, ioflags); 14711751c363STrond Myklebust } 1472ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14731751c363STrond Myklebust 1474dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1475dce81290STrond Myklebust { 1476a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1477a7d42ddbSWeston Andros Adamson 14786f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14796f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14806f29b9bbSKinglong Mee 148141d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1482a7d42ddbSWeston Andros Adamson 1483a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1484a7d42ddbSWeston Andros Adamson 1485a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1486a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1487dce81290STrond Myklebust } 14881f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1489dce81290STrond Myklebust 14901da177e4SLinus Torvalds 14910b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14920b7c0153SFred Isaman { 14930b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14940b7c0153SFred Isaman 14950b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14960b7c0153SFred Isaman } 14970b7c0153SFred Isaman 14981f2edbe3STrond Myklebust /* 14991f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 15001f2edbe3STrond Myklebust */ 15011f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 15021f2edbe3STrond Myklebust { 15031f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 15041f2edbe3STrond Myklebust int kill = 0; 1505788e7a89STrond Myklebust 15061f2edbe3STrond Myklebust /* suid always must be killed */ 15071f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 15081f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 15091f2edbe3STrond Myklebust 15101f2edbe3STrond Myklebust /* 15111f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 15121f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 15131f2edbe3STrond Myklebust */ 15141f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 15151f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 15161f2edbe3STrond Myklebust 15171f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 15181f2edbe3STrond Myklebust return kill; 15191f2edbe3STrond Myklebust 15201f2edbe3STrond Myklebust return 0; 15211f2edbe3STrond Myklebust } 1522788e7a89STrond Myklebust 1523a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1524a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1525a08a8cd3STrond Myklebust { 1526a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1527a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 15282b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1529a08a8cd3STrond Myklebust 1530a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15312b83d3deSTrond Myklebust fattr->size = size; 15322b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15332b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1534a08a8cd3STrond Myklebust return; 15352b83d3deSTrond Myklebust } 15362b83d3deSTrond Myklebust if (size != fattr->size) 1537a08a8cd3STrond Myklebust return; 1538a08a8cd3STrond Myklebust /* Set attribute barrier */ 1539a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15402b83d3deSTrond Myklebust /* ...and update size */ 15412b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1542a08a8cd3STrond Myklebust } 1543a08a8cd3STrond Myklebust 1544a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1545a08a8cd3STrond Myklebust { 15462b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1547a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1548a08a8cd3STrond Myklebust 1549a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1550a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1551a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1552a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1553a08a8cd3STrond Myklebust } 1554a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1555a08a8cd3STrond Myklebust 15561da177e4SLinus Torvalds /* 15571da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15581da177e4SLinus Torvalds */ 1559d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1560d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15610eecb214SAnna Schumaker struct inode *inode) 15621da177e4SLinus Torvalds { 1563788e7a89STrond Myklebust int status; 15641da177e4SLinus Torvalds 1565f551e44fSChuck Lever /* 1566f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1567f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1568f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1569f551e44fSChuck Lever * another writer had changed the file, but some applications 1570f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1571f551e44fSChuck Lever */ 1572d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1573788e7a89STrond Myklebust if (status != 0) 15740eecb214SAnna Schumaker return status; 15758224b273SChuck Lever 1576d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 15775bb2a7cbSTrond Myklebust trace_nfs_writeback_done(task, hdr); 157891d5b470SChuck Lever 1579*69d96651SJeff Layton if (task->tk_status >= 0) { 1580*69d96651SJeff Layton enum nfs3_stable_how committed = hdr->res.verf->committed; 1581*69d96651SJeff Layton 1582*69d96651SJeff Layton if (committed == NFS_UNSTABLE) { 1583*69d96651SJeff Layton /* 1584*69d96651SJeff Layton * We have some uncommitted data on the server at 1585*69d96651SJeff Layton * this point, so ensure that we keep track of that 1586*69d96651SJeff Layton * fact irrespective of what later writes do. 1587*69d96651SJeff Layton */ 1588*69d96651SJeff Layton set_bit(NFS_IOHDR_UNSTABLE_WRITES, &hdr->flags); 1589*69d96651SJeff Layton } 1590*69d96651SJeff Layton 1591*69d96651SJeff Layton if (committed < hdr->args.stable) { 15921da177e4SLinus Torvalds /* We tried a write call, but the server did not 15931da177e4SLinus Torvalds * commit data to stable storage even though we 15941da177e4SLinus Torvalds * requested it. 15951da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15961da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15971da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15981da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15991da177e4SLinus Torvalds */ 16001da177e4SLinus Torvalds static unsigned long complain; 16011da177e4SLinus Torvalds 1602a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 16031da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 16041da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 16051da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1606cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1607*69d96651SJeff Layton committed, hdr->args.stable); 16081da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 16091da177e4SLinus Torvalds } 16101da177e4SLinus Torvalds } 1611*69d96651SJeff Layton } 16121f2edbe3STrond Myklebust 16131f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 161416e14375STrond Myklebust if (nfs_should_remove_suid(inode)) { 161516e14375STrond Myklebust spin_lock(&inode->i_lock); 1616720869ebSTrond Myklebust nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); 161716e14375STrond Myklebust spin_unlock(&inode->i_lock); 161816e14375STrond Myklebust } 16190eecb214SAnna Schumaker return 0; 16200eecb214SAnna Schumaker } 16210eecb214SAnna Schumaker 16220eecb214SAnna Schumaker /* 16230eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 16240eecb214SAnna Schumaker */ 1625d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1626d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 16270eecb214SAnna Schumaker { 1628d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1629d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 16301f2edbe3STrond Myklebust 16311f2edbe3STrond Myklebust if (resp->count < argp->count) { 16321da177e4SLinus Torvalds static unsigned long complain; 16331da177e4SLinus Torvalds 16346c75dc0dSFred Isaman /* This a short write! */ 1635d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 163691d5b470SChuck Lever 16371da177e4SLinus Torvalds /* Has the server at least made some progress? */ 16386c75dc0dSFred Isaman if (resp->count == 0) { 16396c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 16406c75dc0dSFred Isaman printk(KERN_WARNING 16416c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16426c75dc0dSFred Isaman argp->count); 16436c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16446c75dc0dSFred Isaman } 1645d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16466c75dc0dSFred Isaman task->tk_status = -EIO; 16476c75dc0dSFred Isaman return; 16486c75dc0dSFred Isaman } 1649f8417b48SKinglong Mee 1650f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1651f8417b48SKinglong Mee if (!task->tk_ops) { 1652f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1653f8417b48SKinglong Mee return; 1654f8417b48SKinglong Mee } 1655f8417b48SKinglong Mee 16561da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16571da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16581da177e4SLinus Torvalds /* Resend from where the server left off */ 1659d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16601da177e4SLinus Torvalds argp->offset += resp->count; 16611da177e4SLinus Torvalds argp->pgbase += resp->count; 16621da177e4SLinus Torvalds argp->count -= resp->count; 16631da177e4SLinus Torvalds } else { 16641da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16651da177e4SLinus Torvalds * headaches in the case of a server crash. 16661da177e4SLinus Torvalds */ 16671da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16681da177e4SLinus Torvalds } 16698c9cb714STrond Myklebust resp->count = 0; 16708c9cb714STrond Myklebust resp->verf->committed = 0; 1671d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16721da177e4SLinus Torvalds } 16731da177e4SLinus Torvalds } 16741da177e4SLinus Torvalds 1675af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 167671d0a611STrond Myklebust { 1677723c921eSPeter Zijlstra return wait_var_event_killable(&cinfo->rpcs_out, 1678723c921eSPeter Zijlstra !atomic_read(&cinfo->rpcs_out)); 1679af7cf057STrond Myklebust } 1680af7cf057STrond Myklebust 1681af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1682af7cf057STrond Myklebust { 1683af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1684af7cf057STrond Myklebust } 1685af7cf057STrond Myklebust 1686133a48abSTrond Myklebust bool nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1687af7cf057STrond Myklebust { 1688133a48abSTrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) { 1689723c921eSPeter Zijlstra wake_up_var(&cinfo->rpcs_out); 1690133a48abSTrond Myklebust return true; 1691133a48abSTrond Myklebust } 1692133a48abSTrond Myklebust return false; 169371d0a611STrond Myklebust } 169471d0a611STrond Myklebust 16950b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16961da177e4SLinus Torvalds { 16970b7c0153SFred Isaman put_nfs_open_context(data->context); 16980b7c0153SFred Isaman nfs_commit_free(data); 16991da177e4SLinus Torvalds } 1700e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 17011da177e4SLinus Torvalds 17020b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1703c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 17049ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 17059f0ec176SAndy Adamson int how, int flags) 17061da177e4SLinus Torvalds { 170707737691STrond Myklebust struct rpc_task *task; 17089ace33cdSFred Isaman int priority = flush_task_priority(how); 1709bdc7f021STrond Myklebust struct rpc_message msg = { 1710bdc7f021STrond Myklebust .rpc_argp = &data->args, 1711bdc7f021STrond Myklebust .rpc_resp = &data->res, 17129ace33cdSFred Isaman .rpc_cred = data->cred, 1713bdc7f021STrond Myklebust }; 171484115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 171507737691STrond Myklebust .task = &data->task, 17169ace33cdSFred Isaman .rpc_client = clnt, 1717bdc7f021STrond Myklebust .rpc_message = &msg, 17189ace33cdSFred Isaman .callback_ops = call_ops, 171984115e1cSTrond Myklebust .callback_data = data, 1720101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 17214fa7ef69STrond Myklebust .flags = RPC_TASK_ASYNC | flags, 17223ff7576dSTrond Myklebust .priority = priority, 172384115e1cSTrond Myklebust }; 1724118f09edSOlga Kornievskaia 1725118f09edSOlga Kornievskaia if (nfs_server_capable(data->inode, NFS_CAP_MOVEABLE)) 1726118f09edSOlga Kornievskaia task_setup_data.flags |= RPC_TASK_MOVEABLE; 1727118f09edSOlga Kornievskaia 1728788e7a89STrond Myklebust /* Set up the initial task struct. */ 1729e9ae1ee2SAnna Schumaker nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); 17308224b273SChuck Lever trace_nfs_initiate_commit(data); 17311da177e4SLinus Torvalds 1732b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1733bdc7f021STrond Myklebust 173407737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1735dbae4c73STrond Myklebust if (IS_ERR(task)) 1736dbae4c73STrond Myklebust return PTR_ERR(task); 1737d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1738d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 173907737691STrond Myklebust rpc_put_task(task); 1740dbae4c73STrond Myklebust return 0; 17411da177e4SLinus Torvalds } 1742e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 17431da177e4SLinus Torvalds 1744378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1745378520b8SPeng Tao { 1746378520b8SPeng Tao loff_t lwb = 0; 1747378520b8SPeng Tao struct nfs_page *req; 1748378520b8SPeng Tao 1749378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1750378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1751378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1752378520b8SPeng Tao 1753378520b8SPeng Tao return lwb; 1754378520b8SPeng Tao } 1755378520b8SPeng Tao 17561da177e4SLinus Torvalds /* 17579ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17589ace33cdSFred Isaman */ 17590b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1760988b6dceSFred Isaman struct list_head *head, 1761f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1762f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17639ace33cdSFred Isaman { 176419573c93STrond Myklebust struct nfs_page *first; 176519573c93STrond Myklebust struct nfs_open_context *ctx; 176619573c93STrond Myklebust struct inode *inode; 17679ace33cdSFred Isaman 17689ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17699ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17709ace33cdSFred Isaman 177119573c93STrond Myklebust if (head) 17729ace33cdSFred Isaman list_splice_init(head, &data->pages); 17739ace33cdSFred Isaman 177419573c93STrond Myklebust first = nfs_list_entry(data->pages.next); 177519573c93STrond Myklebust ctx = nfs_req_openctx(first); 177619573c93STrond Myklebust inode = d_inode(ctx->dentry); 177719573c93STrond Myklebust 17789ace33cdSFred Isaman data->inode = inode; 17799fcd5960STrond Myklebust data->cred = ctx->cred; 1780988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1781378520b8SPeng Tao /* only set lwb for pnfs commit */ 1782378520b8SPeng Tao if (lseg) 1783378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17849ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1785f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1786b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17879ace33cdSFred Isaman 17889ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17899ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17909ace33cdSFred Isaman data->args.offset = 0; 17919ace33cdSFred Isaman data->args.count = 0; 17929fcd5960STrond Myklebust data->context = get_nfs_open_context(ctx); 17939ace33cdSFred Isaman data->res.fattr = &data->fattr; 17949ace33cdSFred Isaman data->res.verf = &data->verf; 17959ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 1796133a48abSTrond Myklebust nfs_commit_begin(cinfo->mds); 17979ace33cdSFred Isaman } 1798e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17999ace33cdSFred Isaman 1800e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1801ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1802b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1803b57ff130SWeston Andros Adamson u32 ds_commit_idx) 180464bfeb49SFred Isaman { 180564bfeb49SFred Isaman struct nfs_page *req; 180664bfeb49SFred Isaman 180764bfeb49SFred Isaman while (!list_empty(page_list)) { 180864bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 180964bfeb49SFred Isaman nfs_list_remove_request(req); 1810b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1811487b9b8aSTom Haynes if (!cinfo->dreq) 1812487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 18131d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 181464bfeb49SFred Isaman } 181564bfeb49SFred Isaman } 1816e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 181764bfeb49SFred Isaman 1818b20135d0STrond Myklebust static void 1819b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1820b20135d0STrond Myklebust struct nfs_page *req) 1821b20135d0STrond Myklebust { 1822b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1823b20135d0STrond Myklebust } 1824b20135d0STrond Myklebust 18259ace33cdSFred Isaman /* 18261da177e4SLinus Torvalds * Commit dirty pages 18271da177e4SLinus Torvalds */ 18281da177e4SLinus Torvalds static int 1829ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1830ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 18311da177e4SLinus Torvalds { 18320b7c0153SFred Isaman struct nfs_commit_data *data; 183385e39feeSOlga Kornievskaia unsigned short task_flags = 0; 18341da177e4SLinus Torvalds 1835ade8febdSWeston Andros Adamson /* another commit raced with us */ 1836ade8febdSWeston Andros Adamson if (list_empty(head)) 1837ade8febdSWeston Andros Adamson return 0; 1838ade8febdSWeston Andros Adamson 1839515dcdcdSTrond Myklebust data = nfs_commitdata_alloc(); 1840515dcdcdSTrond Myklebust if (!data) { 1841515dcdcdSTrond Myklebust nfs_retry_commit(head, NULL, cinfo, -1); 1842515dcdcdSTrond Myklebust return -ENOMEM; 1843515dcdcdSTrond Myklebust } 18441da177e4SLinus Torvalds 18451da177e4SLinus Torvalds /* Set up the argument struct */ 1846f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 184785e39feeSOlga Kornievskaia if (NFS_SERVER(inode)->nfs_client->cl_minorversion) 184885e39feeSOlga Kornievskaia task_flags = RPC_TASK_MOVEABLE; 1849c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 185085e39feeSOlga Kornievskaia data->mds_ops, how, 185185e39feeSOlga Kornievskaia RPC_TASK_CRED_NOREF | task_flags); 18521da177e4SLinus Torvalds } 18531da177e4SLinus Torvalds 18541da177e4SLinus Torvalds /* 18551da177e4SLinus Torvalds * COMMIT call returned 18561da177e4SLinus Torvalds */ 1857788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 18581da177e4SLinus Torvalds { 18590b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18601da177e4SLinus Torvalds 1861788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1862c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 18637bdd297eSTrond Myklebust trace_nfs_commit_done(task, data); 1864c9d8f89dSTrond Myklebust } 1865c9d8f89dSTrond Myklebust 1866f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1867c9d8f89dSTrond Myklebust { 1868221203ceSTrond Myklebust const struct nfs_writeverf *verf = data->res.verf; 1869c9d8f89dSTrond Myklebust struct nfs_page *req; 1870c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1871f453a54aSFred Isaman struct nfs_commit_info cinfo; 1872353db796SNeilBrown struct nfs_server *nfss; 1873788e7a89STrond Myklebust 18741da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18751da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18761da177e4SLinus Torvalds nfs_list_remove_request(req); 187767911c8fSAnna Schumaker if (req->wb_page) 1878d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 18791da177e4SLinus Torvalds 18801e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18819fcd5960STrond Myklebust nfs_req_openctx(req)->dentry->d_sb->s_id, 18829fcd5960STrond Myklebust (unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)), 18831da177e4SLinus Torvalds req->wb_bytes, 18841da177e4SLinus Torvalds (long long)req_offset(req)); 1885c9d8f89dSTrond Myklebust if (status < 0) { 18866fbda89bSTrond Myklebust if (req->wb_page) { 1887861e1671SChuck Lever trace_nfs_commit_error(req, status); 18886fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, status); 18891da177e4SLinus Torvalds nfs_inode_remove_request(req); 18906fbda89bSTrond Myklebust } 1891ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18921da177e4SLinus Torvalds goto next; 18931da177e4SLinus Torvalds } 18941da177e4SLinus Torvalds 18951da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18961da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18971f28476dSTrond Myklebust if (nfs_write_match_verf(verf, req)) { 18981da177e4SLinus Torvalds /* We have a match */ 189938a33101SKinglong Mee if (req->wb_page) 19001da177e4SLinus Torvalds nfs_inode_remove_request(req); 1901ddeaa637SJoe Perches dprintk_cont(" OK\n"); 19021da177e4SLinus Torvalds goto next; 19031da177e4SLinus Torvalds } 19041da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1905ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 19066d884e8fSFred nfs_mark_request_dirty(req); 19079fcd5960STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); 19081da177e4SLinus Torvalds next: 19091d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19107f1bda44STrond Myklebust /* Latency breaker */ 19117f1bda44STrond Myklebust cond_resched(); 19121da177e4SLinus Torvalds } 1913353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1914353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 19156df25e58SNeilBrown nfss->write_congested = 0; 1916353db796SNeilBrown 1917f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1918af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 19195917ce84SFred Isaman } 19205917ce84SFred Isaman 19215917ce84SFred Isaman static void nfs_commit_release(void *calldata) 19225917ce84SFred Isaman { 19230b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 19245917ce84SFred Isaman 1925f453a54aSFred Isaman data->completion_ops->completion(data); 1926c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 19271da177e4SLinus Torvalds } 1928788e7a89STrond Myklebust 1929788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 19300b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1931788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1932788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1933788e7a89STrond Myklebust }; 19341da177e4SLinus Torvalds 1935f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1936f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1937b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1938f453a54aSFred Isaman }; 1939f453a54aSFred Isaman 19401763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1941ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 194284c53ab5SFred Isaman { 194384c53ab5SFred Isaman int status; 194484c53ab5SFred Isaman 1945ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 194684c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1947ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 194884c53ab5SFred Isaman return status; 194984c53ab5SFred Isaman } 195084c53ab5SFred Isaman 1951c4f24df9STrond Myklebust static int __nfs_commit_inode(struct inode *inode, int how, 1952c4f24df9STrond Myklebust struct writeback_control *wbc) 19531da177e4SLinus Torvalds { 19541da177e4SLinus Torvalds LIST_HEAD(head); 1955ea2cf228SFred Isaman struct nfs_commit_info cinfo; 195671d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1957c4f24df9STrond Myklebust int ret, nscan; 19581da177e4SLinus Torvalds 195964a93dbfSTrond Myklebust how &= ~FLUSH_SYNC; 1960ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1961af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1962c4f24df9STrond Myklebust for (;;) { 1963c4f24df9STrond Myklebust ret = nscan = nfs_scan_commit(inode, &head, &cinfo); 1964c4f24df9STrond Myklebust if (ret <= 0) 1965c4f24df9STrond Myklebust break; 1966c4f24df9STrond Myklebust ret = nfs_generic_commit_list(inode, &head, how, &cinfo); 1967c4f24df9STrond Myklebust if (ret < 0) 1968c4f24df9STrond Myklebust break; 1969c4f24df9STrond Myklebust ret = 0; 1970c4f24df9STrond Myklebust if (wbc && wbc->sync_mode == WB_SYNC_NONE) { 1971c4f24df9STrond Myklebust if (nscan < wbc->nr_to_write) 1972c4f24df9STrond Myklebust wbc->nr_to_write -= nscan; 1973c4f24df9STrond Myklebust else 1974c4f24df9STrond Myklebust wbc->nr_to_write = 0; 1975c4f24df9STrond Myklebust } 1976c4f24df9STrond Myklebust if (nscan < INT_MAX) 1977c4f24df9STrond Myklebust break; 1978c4f24df9STrond Myklebust cond_resched(); 1979c4f24df9STrond Myklebust } 1980af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 1981c4f24df9STrond Myklebust if (ret || !may_wait) 1982c4f24df9STrond Myklebust return ret; 1983c4f24df9STrond Myklebust return wait_on_commit(cinfo.mds); 1984c4f24df9STrond Myklebust } 1985c4f24df9STrond Myklebust 1986c4f24df9STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 1987c4f24df9STrond Myklebust { 1988c4f24df9STrond Myklebust return __nfs_commit_inode(inode, how, NULL); 19891da177e4SLinus Torvalds } 1990b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19918fc795f7STrond Myklebust 1992ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19938fc795f7STrond Myklebust { 1994420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1995420e3646STrond Myklebust int flags = FLUSH_SYNC; 1996420e3646STrond Myklebust int ret = 0; 19978fc795f7STrond Myklebust 1998c4f24df9STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 19993236c3e1SJeff Layton /* no commits means nothing needs to be done */ 20005cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 2001c4f24df9STrond Myklebust goto check_requests_outstanding; 20023236c3e1SJeff Layton 2003a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 2004a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 2005420e3646STrond Myklebust */ 20061a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 2007420e3646STrond Myklebust goto out_mark_dirty; 2008420e3646STrond Myklebust 2009a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 2010420e3646STrond Myklebust flags = 0; 2011a00dd6c0SJeff Layton } 2012a00dd6c0SJeff Layton 2013c4f24df9STrond Myklebust ret = __nfs_commit_inode(inode, flags, wbc); 2014c4f24df9STrond Myklebust if (!ret) { 2015c4f24df9STrond Myklebust if (flags & FLUSH_SYNC) 20168fc795f7STrond Myklebust return 0; 2017c4f24df9STrond Myklebust } else if (atomic_long_read(&nfsi->commit_info.ncommit)) 2018c4f24df9STrond Myklebust goto out_mark_dirty; 2019c4f24df9STrond Myklebust 2020c4f24df9STrond Myklebust check_requests_outstanding: 2021c4f24df9STrond Myklebust if (!atomic_read(&nfsi->commit_info.rpcs_out)) 2022c4f24df9STrond Myklebust return ret; 2023420e3646STrond Myklebust out_mark_dirty: 20248fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 20258fc795f7STrond Myklebust return ret; 20268fc795f7STrond Myklebust } 202789d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 2028863a3c6cSAndy Adamson 2029acdc53b2STrond Myklebust /* 2030837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 2031837bb1d7STrond Myklebust * 2032837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 2033837bb1d7STrond Myklebust * client. 2034837bb1d7STrond Myklebust */ 2035837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 2036837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 2037837bb1d7STrond Myklebust { 2038837bb1d7STrond Myklebust int ret; 2039837bb1d7STrond Myklebust 2040837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 2041837bb1d7STrond Myklebust if (ret == 0) 2042837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 2043837bb1d7STrond Myklebust return ret; 2044837bb1d7STrond Myklebust } 2045837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 2046837bb1d7STrond Myklebust 2047837bb1d7STrond Myklebust /* 2048acdc53b2STrond Myklebust * flush the inode to disk. 2049acdc53b2STrond Myklebust */ 2050acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 205134901f70STrond Myklebust { 2052f4ce1299STrond Myklebust int ret; 205334901f70STrond Myklebust 2054f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 2055f4ce1299STrond Myklebust 20565bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 20576b196875SChuck Lever if (ret) 20586b196875SChuck Lever goto out; 20595bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20606b196875SChuck Lever if (ret < 0) 20616b196875SChuck Lever goto out; 20625bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 20636b196875SChuck Lever ret = 0; 2064f4ce1299STrond Myklebust 20656b196875SChuck Lever out: 2066f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 2067f4ce1299STrond Myklebust return ret; 20681c75950bSTrond Myklebust } 2069ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 20701c75950bSTrond Myklebust 20716d740c76SMatthew Wilcox (Oracle) int nfs_wb_folio_cancel(struct inode *inode, struct folio *folio) 20721b3b4a1aSTrond Myklebust { 20731b3b4a1aSTrond Myklebust struct nfs_page *req; 20741b3b4a1aSTrond Myklebust int ret = 0; 20751b3b4a1aSTrond Myklebust 20766d740c76SMatthew Wilcox (Oracle) folio_wait_writeback(folio); 20773e217045SWeston Andros Adamson 20783e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20793e217045SWeston Andros Adamson * request */ 20806d740c76SMatthew Wilcox (Oracle) req = nfs_lock_and_join_requests(&folio->page); 20813e217045SWeston Andros Adamson 20823e217045SWeston Andros Adamson if (IS_ERR(req)) { 20833e217045SWeston Andros Adamson ret = PTR_ERR(req); 20843e217045SWeston Andros Adamson } else if (req) { 20856d740c76SMatthew Wilcox (Oracle) /* all requests from this folio have been cancelled by 20863e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20873e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20883e217045SWeston Andros Adamson * release it */ 20891b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20901d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20911b3b4a1aSTrond Myklebust } 20923e217045SWeston Andros Adamson 20931b3b4a1aSTrond Myklebust return ret; 20941b3b4a1aSTrond Myklebust } 20951b3b4a1aSTrond Myklebust 20961c75950bSTrond Myklebust /* 20971c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 20981c75950bSTrond Myklebust */ 2099c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 21001c75950bSTrond Myklebust { 210129418aa4SMel Gorman loff_t range_start = page_file_offset(page); 210209cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 21037f2f12d9STrond Myklebust struct writeback_control wbc = { 21047f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 21057f2f12d9STrond Myklebust .nr_to_write = 0, 21067f2f12d9STrond Myklebust .range_start = range_start, 21077f2f12d9STrond Myklebust .range_end = range_end, 21087f2f12d9STrond Myklebust }; 21097f2f12d9STrond Myklebust int ret; 21107f2f12d9STrond Myklebust 2111f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2112f4ce1299STrond Myklebust 21130522f6adSTrond Myklebust for (;;) { 2114ba8b06e6STrond Myklebust wait_on_page_writeback(page); 21157f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2116c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 21177f2f12d9STrond Myklebust if (ret < 0) 21187f2f12d9STrond Myklebust goto out_error; 21190522f6adSTrond Myklebust continue; 21207f2f12d9STrond Myklebust } 2121f4ce1299STrond Myklebust ret = 0; 21220522f6adSTrond Myklebust if (!PagePrivate(page)) 21230522f6adSTrond Myklebust break; 21240522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 21257f2f12d9STrond Myklebust if (ret < 0) 21267f2f12d9STrond Myklebust goto out_error; 21277f2f12d9STrond Myklebust } 21287f2f12d9STrond Myklebust out_error: 2129f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 21307f2f12d9STrond Myklebust return ret; 21311c75950bSTrond Myklebust } 21321c75950bSTrond Myklebust 2133074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2134074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2135a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2136074cc1deSTrond Myklebust { 21372da95652SJeff Layton /* 21382da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 21392da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 21402da95652SJeff Layton * 21412da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 21422da95652SJeff Layton * that we can safely release the inode reference while holding 21432da95652SJeff Layton * the page lock. 21442da95652SJeff Layton */ 21452da95652SJeff Layton if (PagePrivate(page)) 21462da95652SJeff Layton return -EBUSY; 2147074cc1deSTrond Myklebust 214816f2f4e6SDavid Howells if (PageFsCache(page)) { 214916f2f4e6SDavid Howells if (mode == MIGRATE_ASYNC) 21508c209ce7SDavid Howells return -EBUSY; 215116f2f4e6SDavid Howells wait_on_page_fscache(page); 215216f2f4e6SDavid Howells } 2153074cc1deSTrond Myklebust 2154a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2155074cc1deSTrond Myklebust } 2156074cc1deSTrond Myklebust #endif 2157074cc1deSTrond Myklebust 2158f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 21591da177e4SLinus Torvalds { 21601da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 21611e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 21621da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 216320c2df83SPaul Mundt NULL); 21641da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 21651da177e4SLinus Torvalds return -ENOMEM; 21661da177e4SLinus Torvalds 216793d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 21681da177e4SLinus Torvalds nfs_wdata_cachep); 21691da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 21703dd4765fSJeff Layton goto out_destroy_write_cache; 21711da177e4SLinus Torvalds 21720b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 21730b7c0153SFred Isaman sizeof(struct nfs_commit_data), 21740b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21750b7c0153SFred Isaman NULL); 21760b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21773dd4765fSJeff Layton goto out_destroy_write_mempool; 21780b7c0153SFred Isaman 217993d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21804c100210SYanchuan Nian nfs_cdata_cachep); 21811da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21823dd4765fSJeff Layton goto out_destroy_commit_cache; 21831da177e4SLinus Torvalds 218489a09141SPeter Zijlstra /* 218589a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 218689a09141SPeter Zijlstra * 218789a09141SPeter Zijlstra * 64MB: 8192k 218889a09141SPeter Zijlstra * 128MB: 11585k 218989a09141SPeter Zijlstra * 256MB: 16384k 219089a09141SPeter Zijlstra * 512MB: 23170k 219189a09141SPeter Zijlstra * 1GB: 32768k 219289a09141SPeter Zijlstra * 2GB: 46340k 219389a09141SPeter Zijlstra * 4GB: 65536k 219489a09141SPeter Zijlstra * 8GB: 92681k 219589a09141SPeter Zijlstra * 16GB: 131072k 219689a09141SPeter Zijlstra * 219789a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 219889a09141SPeter Zijlstra * Limit the default to 256M 219989a09141SPeter Zijlstra */ 2200ca79b0c2SArun KS nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); 220189a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 220289a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 220389a09141SPeter Zijlstra 22041da177e4SLinus Torvalds return 0; 22053dd4765fSJeff Layton 22063dd4765fSJeff Layton out_destroy_commit_cache: 22073dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 22083dd4765fSJeff Layton out_destroy_write_mempool: 22093dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 22103dd4765fSJeff Layton out_destroy_write_cache: 22113dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 22123dd4765fSJeff Layton return -ENOMEM; 22131da177e4SLinus Torvalds } 22141da177e4SLinus Torvalds 2215266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 22161da177e4SLinus Torvalds { 22171da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 22183dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 22191da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 22201a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 22211da177e4SLinus Torvalds } 22221da177e4SLinus Torvalds 22234a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 22244a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 22254a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 22260eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 22270eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 22281ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 22294a0de55cSAnna Schumaker }; 2230