11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 16074cc1deSTrond Myklebust #include <linux/migrate.h> 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 191da177e4SLinus Torvalds #include <linux/nfs_fs.h> 201da177e4SLinus Torvalds #include <linux/nfs_mount.h> 211da177e4SLinus Torvalds #include <linux/nfs_page.h> 223fcfab16SAndrew Morton #include <linux/backing-dev.h> 23afeacc8cSPaul Gortmaker #include <linux/export.h> 24af7cf057STrond Myklebust #include <linux/freezer.h> 25af7cf057STrond Myklebust #include <linux/wait.h> 263fcfab16SAndrew Morton 277c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 281da177e4SLinus Torvalds 291da177e4SLinus Torvalds #include "delegation.h" 3049a70f27STrond Myklebust #include "internal.h" 3191d5b470SChuck Lever #include "iostat.h" 32def6ed7eSAndy Adamson #include "nfs4_fs.h" 33074cc1deSTrond Myklebust #include "fscache.h" 3494ad1c80SFred Isaman #include "pnfs.h" 351da177e4SLinus Torvalds 36f4ce1299STrond Myklebust #include "nfstrace.h" 37f4ce1299STrond Myklebust 381da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 391da177e4SLinus Torvalds 401da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 411da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 421da177e4SLinus Torvalds 43919e3bd9STrond Myklebust struct nfs_io_completion { 44919e3bd9STrond Myklebust void (*complete)(void *data); 45919e3bd9STrond Myklebust void *data; 46919e3bd9STrond Myklebust struct kref refcount; 47919e3bd9STrond Myklebust }; 48919e3bd9STrond Myklebust 491da177e4SLinus Torvalds /* 501da177e4SLinus Torvalds * Local function declarations 511da177e4SLinus Torvalds */ 52f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 53788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 54061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 55f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 564a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 57d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 5802d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 5902d1426cSWeston Andros Adamson struct inode *inode); 603a3908c8STrond Myklebust static struct nfs_page * 613a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 623a3908c8STrond Myklebust struct page *page); 631da177e4SLinus Torvalds 64e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 653feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 660b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 671da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 681da177e4SLinus Torvalds 69518662e0SNeilBrown struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) 701da177e4SLinus Torvalds { 71518662e0SNeilBrown struct nfs_commit_data *p; 7240859d7eSChuck Lever 73518662e0SNeilBrown if (never_fail) 74518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 75518662e0SNeilBrown else { 76518662e0SNeilBrown /* It is OK to do some reclaim, not no safe to wait 77518662e0SNeilBrown * for anything to be returned to the pool. 78518662e0SNeilBrown * mempool_alloc() cannot handle that particular combination, 79518662e0SNeilBrown * so we need two separate attempts. 80518662e0SNeilBrown */ 81518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); 82518662e0SNeilBrown if (!p) 83518662e0SNeilBrown p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | 84518662e0SNeilBrown __GFP_NOWARN | __GFP_NORETRY); 85518662e0SNeilBrown if (!p) 86518662e0SNeilBrown return NULL; 87518662e0SNeilBrown } 88518662e0SNeilBrown 891da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 901da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 911da177e4SLinus Torvalds return p; 921da177e4SLinus Torvalds } 93e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 941da177e4SLinus Torvalds 950b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 961da177e4SLinus Torvalds { 971da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 981da177e4SLinus Torvalds } 99e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 1001da177e4SLinus Torvalds 1011e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 1023feb2d49STrond Myklebust { 1031e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 1043feb2d49STrond Myklebust 1053feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 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 1487b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1497b159fc1STrond Myklebust { 1507b159fc1STrond Myklebust ctx->error = error; 1517b159fc1STrond Myklebust smp_wmb(); 1527b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1537b159fc1STrond Myklebust } 1547b159fc1STrond Myklebust 155bd37d6fcSTrond Myklebust static struct nfs_page * 156bd37d6fcSTrond Myklebust nfs_page_private_request(struct page *page) 157bd37d6fcSTrond Myklebust { 158bd37d6fcSTrond Myklebust if (!PagePrivate(page)) 159bd37d6fcSTrond Myklebust return NULL; 160bd37d6fcSTrond Myklebust return (struct nfs_page *)page_private(page); 161bd37d6fcSTrond Myklebust } 162bd37d6fcSTrond Myklebust 16384d3a9a9SWeston Andros Adamson /* 16484d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 16584d3a9a9SWeston Andros Adamson * 16684d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 16784d3a9a9SWeston Andros Adamson * 16884d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 16984d3a9a9SWeston Andros Adamson */ 17029418aa4SMel Gorman static struct nfs_page * 171b30d2f04STrond Myklebust nfs_page_find_private_request(struct page *page) 172277459d2STrond Myklebust { 1734b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(page); 174bd37d6fcSTrond Myklebust struct nfs_page *req; 175277459d2STrond Myklebust 176b30d2f04STrond Myklebust if (!PagePrivate(page)) 177b30d2f04STrond Myklebust return NULL; 1784b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 179bd37d6fcSTrond Myklebust req = nfs_page_private_request(page); 18084d3a9a9SWeston Andros Adamson if (req) { 18184d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 18229418aa4SMel Gorman kref_get(&req->wb_kref); 18384d3a9a9SWeston Andros Adamson } 1844b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 185b30d2f04STrond Myklebust return req; 186b30d2f04STrond Myklebust } 18729418aa4SMel Gorman 188b30d2f04STrond Myklebust static struct nfs_page * 189b30d2f04STrond Myklebust nfs_page_find_swap_request(struct page *page) 190b30d2f04STrond Myklebust { 191b30d2f04STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 192b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 193b30d2f04STrond Myklebust struct nfs_page *req = NULL; 194b30d2f04STrond Myklebust if (!PageSwapCache(page)) 195b30d2f04STrond Myklebust return NULL; 196e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 197b30d2f04STrond Myklebust if (PageSwapCache(page)) { 198b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 199b30d2f04STrond Myklebust page); 200b30d2f04STrond Myklebust if (req) { 201b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 202b30d2f04STrond Myklebust kref_get(&req->wb_kref); 203b30d2f04STrond Myklebust } 204b30d2f04STrond Myklebust } 205e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 206277459d2STrond Myklebust return req; 207277459d2STrond Myklebust } 208277459d2STrond Myklebust 20984d3a9a9SWeston Andros Adamson /* 21084d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 21184d3a9a9SWeston Andros Adamson * 21284d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 21384d3a9a9SWeston Andros Adamson */ 21484d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 215277459d2STrond Myklebust { 216b30d2f04STrond Myklebust struct nfs_page *req; 217277459d2STrond Myklebust 218b30d2f04STrond Myklebust req = nfs_page_find_private_request(page); 219b30d2f04STrond Myklebust if (!req) 220b30d2f04STrond Myklebust req = nfs_page_find_swap_request(page); 221277459d2STrond Myklebust return req; 222277459d2STrond Myklebust } 223277459d2STrond Myklebust 2241da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2251da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2261da177e4SLinus Torvalds { 227d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 228a3d01454STrond Myklebust loff_t end, i_size; 229a3d01454STrond Myklebust pgoff_t end_index; 2301da177e4SLinus Torvalds 231a3d01454STrond Myklebust spin_lock(&inode->i_lock); 232a3d01454STrond Myklebust i_size = i_size_read(inode); 23309cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2348cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 235a3d01454STrond Myklebust goto out; 236d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2371da177e4SLinus Torvalds if (i_size >= end) 238a3d01454STrond Myklebust goto out; 2391da177e4SLinus Torvalds i_size_write(inode, end); 240a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 241a3d01454STrond Myklebust out: 242a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2431da177e4SLinus Torvalds } 2441da177e4SLinus Torvalds 245a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 246a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 247a301b777STrond Myklebust { 248d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 249a301b777STrond Myklebust } 250a301b777STrond Myklebust 251d72ddcbaSWeston Andros Adamson /* 252d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 253d72ddcbaSWeston Andros Adamson * @head - head request of page group 254d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 255d72ddcbaSWeston Andros Adamson * 256d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 257d72ddcbaSWeston Andros Adamson * page offset @page_offset. 258d72ddcbaSWeston Andros Adamson * 259d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 260d72ddcbaSWeston Andros Adamson * match is found. 261d72ddcbaSWeston Andros Adamson * 262d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 263d72ddcbaSWeston Andros Adamson */ 264d72ddcbaSWeston Andros Adamson static struct nfs_page * 265d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 266d72ddcbaSWeston Andros Adamson { 267d72ddcbaSWeston Andros Adamson struct nfs_page *req; 268d72ddcbaSWeston Andros Adamson 269d72ddcbaSWeston Andros Adamson req = head; 270d72ddcbaSWeston Andros Adamson do { 271d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 272d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 273d72ddcbaSWeston Andros Adamson return req; 274d72ddcbaSWeston Andros Adamson 275d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 276d72ddcbaSWeston Andros Adamson } while (req != head); 277d72ddcbaSWeston Andros Adamson 278d72ddcbaSWeston Andros Adamson return NULL; 279d72ddcbaSWeston Andros Adamson } 280d72ddcbaSWeston Andros Adamson 281d72ddcbaSWeston Andros Adamson /* 282d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 283d72ddcbaSWeston Andros Adamson * @head - head request of page group 284d72ddcbaSWeston Andros Adamson * 285d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 286d72ddcbaSWeston Andros Adamson * returns false otherwise 287d72ddcbaSWeston Andros Adamson */ 288d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 289d72ddcbaSWeston Andros Adamson { 290d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 291d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 292d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 293d72ddcbaSWeston Andros Adamson 2941344b7eaSTrond Myklebust nfs_page_group_lock(req); 295d72ddcbaSWeston Andros Adamson 2967e8a30f8STrond Myklebust for (;;) { 297d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 2987e8a30f8STrond Myklebust if (!tmp) 2997e8a30f8STrond Myklebust break; 3007e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 301d72ddcbaSWeston Andros Adamson } 302d72ddcbaSWeston Andros Adamson 303d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3047e8a30f8STrond Myklebust return pos >= len; 305d72ddcbaSWeston Andros Adamson } 306d72ddcbaSWeston Andros Adamson 3071da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3081da177e4SLinus Torvalds * covers the full page. 3091da177e4SLinus Torvalds */ 310d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3111da177e4SLinus Torvalds { 312d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 3131da177e4SLinus Torvalds return; 314d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3151da177e4SLinus Torvalds return; 316d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 3171da177e4SLinus Torvalds } 3181da177e4SLinus Torvalds 3191da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3201da177e4SLinus Torvalds { 321e87b4c7aSNeilBrown int ret = 0; 322cca588d6STrond Myklebust 323e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 324e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 325e87b4c7aSNeilBrown return ret; 3261da177e4SLinus Torvalds } 3271da177e4SLinus Torvalds 3281da177e4SLinus Torvalds /* 32989a09141SPeter Zijlstra * NFS congestion control 33089a09141SPeter Zijlstra */ 33189a09141SPeter Zijlstra 33289a09141SPeter Zijlstra int nfs_congestion_kb; 33389a09141SPeter Zijlstra 33489a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 33589a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 33689a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 33789a09141SPeter Zijlstra 338deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 33989a09141SPeter Zijlstra { 3400db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3410db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3425a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3435a6d41b3STrond Myklebust 344deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 34589a09141SPeter Zijlstra 346277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3470db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3480db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 34989a09141SPeter Zijlstra } 35089a09141SPeter Zijlstra 35120633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 35289a09141SPeter Zijlstra { 35320633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 35489a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 35531a01f09STrond Myklebust bool is_done; 35689a09141SPeter Zijlstra 35731a01f09STrond Myklebust is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); 35831a01f09STrond Myklebust nfs_unlock_request(req); 35931a01f09STrond Myklebust if (!is_done) 36020633f04SWeston Andros Adamson return; 36120633f04SWeston Andros Adamson 36220633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 363c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3640db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 36589a09141SPeter Zijlstra } 36689a09141SPeter Zijlstra 367d4581383SWeston Andros Adamson /* 368d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 369d4581383SWeston Andros Adamson * 370d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 371d4581383SWeston Andros Adamson * 372d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 373d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 374d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 375d4581383SWeston Andros Adamson * 376b5bab9bfSTrond Myklebust * NOTE: this must be called holding page_group bit lock 377b5bab9bfSTrond Myklebust * which will be released before returning. 378d4581383SWeston Andros Adamson * 379d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 380d4581383SWeston Andros Adamson */ 38174a6d4b5STrond Myklebust static void 38274a6d4b5STrond Myklebust nfs_unroll_locks(struct inode *inode, struct nfs_page *head, 3836d17d653STrond Myklebust struct nfs_page *req) 384d4581383SWeston Andros Adamson { 385d4581383SWeston Andros Adamson struct nfs_page *tmp; 386e261f51fSTrond Myklebust 387d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 3885b2b5187STrond Myklebust for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { 3895b2b5187STrond Myklebust if (!kref_read(&tmp->wb_kref)) 3905b2b5187STrond Myklebust continue; 3915b2b5187STrond Myklebust nfs_unlock_and_release_request(tmp); 3925b2b5187STrond Myklebust } 393e261f51fSTrond Myklebust } 394d4581383SWeston Andros Adamson 395d4581383SWeston Andros Adamson /* 396d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 397d4581383SWeston Andros Adamson * 398d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 399d4581383SWeston Andros Adamson * @old_head - the old head of the list 400d4581383SWeston Andros Adamson * 401d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 402d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 403d4581383SWeston Andros Adamson * with a reference held by some other context. 404d4581383SWeston Andros Adamson */ 405d4581383SWeston Andros Adamson static void 406d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 407b66aaa8dSTrond Myklebust struct nfs_page *old_head, 408b66aaa8dSTrond Myklebust struct inode *inode) 409d4581383SWeston Andros Adamson { 410d4581383SWeston Andros Adamson while (destroy_list) { 411d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 412d4581383SWeston Andros Adamson 413d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 414d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 415d4581383SWeston Andros Adamson 416d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 417d4581383SWeston Andros Adamson 418d4581383SWeston Andros Adamson /* make sure old group is not used */ 419d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 420d4581383SWeston Andros Adamson 421902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 422902a4c00STrond Myklebust 4235b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4245b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4255b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 426902a4c00STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) 4275b2b5187STrond Myklebust nfs_free_request(subreq); 4285b2b5187STrond Myklebust continue; 4295b2b5187STrond Myklebust } 430d4581383SWeston Andros Adamson 4315b2b5187STrond Myklebust subreq->wb_head = subreq; 4325b2b5187STrond Myklebust 433b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 434d4581383SWeston Andros Adamson nfs_release_request(subreq); 435a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 436d4581383SWeston Andros Adamson } 4375b2b5187STrond Myklebust 4385b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4395b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4405b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 441d4581383SWeston Andros Adamson } 442d4581383SWeston Andros Adamson } 443d4581383SWeston Andros Adamson 444d4581383SWeston Andros Adamson /* 445d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 446d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 447d4581383SWeston Andros Adamson * operations for this page. 448d4581383SWeston Andros Adamson * 449d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 450d4581383SWeston Andros Adamson * 451d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 452d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 453d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 454d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 455d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 456d4581383SWeston Andros Adamson * 457d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 458d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 459d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 460d4581383SWeston Andros Adamson * error was encountered. 461d4581383SWeston Andros Adamson */ 462d4581383SWeston Andros Adamson static struct nfs_page * 4636d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 464d4581383SWeston Andros Adamson { 465d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 466d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 467d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 468d4581383SWeston Andros Adamson unsigned int total_bytes; 469d4581383SWeston Andros Adamson int ret; 470d4581383SWeston Andros Adamson 471d4581383SWeston Andros Adamson try_again: 472d4581383SWeston Andros Adamson /* 473d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 474d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 475d4581383SWeston Andros Adamson * until the head reference is released. 476d4581383SWeston Andros Adamson */ 477bd37d6fcSTrond Myklebust head = nfs_page_find_head_request(page); 478bd37d6fcSTrond Myklebust if (!head) 479d4581383SWeston Andros Adamson return NULL; 480d4581383SWeston Andros Adamson 481a0e265bcSTrond Myklebust /* lock the page head first in order to avoid an ABBA inefficiency */ 482a0e265bcSTrond Myklebust if (!nfs_lock_request(head)) { 483a0e265bcSTrond Myklebust ret = nfs_wait_on_request(head); 484a0e265bcSTrond Myklebust nfs_release_request(head); 485a0e265bcSTrond Myklebust if (ret < 0) 486a0e265bcSTrond Myklebust return ERR_PTR(ret); 487a0e265bcSTrond Myklebust goto try_again; 488a0e265bcSTrond Myklebust } 489bd37d6fcSTrond Myklebust 490bd37d6fcSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 491bd37d6fcSTrond Myklebust if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { 492bd37d6fcSTrond Myklebust nfs_unlock_and_release_request(head); 493bd37d6fcSTrond Myklebust goto try_again; 494bd37d6fcSTrond Myklebust } 4957c3af975SWeston Andros Adamson 4961344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 497b5bab9bfSTrond Myklebust if (ret < 0) { 498a0e265bcSTrond Myklebust nfs_unlock_and_release_request(head); 499b5bab9bfSTrond Myklebust return ERR_PTR(ret); 5007c3af975SWeston Andros Adamson } 5017c3af975SWeston Andros Adamson 5027c3af975SWeston Andros Adamson /* lock each request in the page group */ 503a0e265bcSTrond Myklebust total_bytes = head->wb_bytes; 504a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 505a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 50674a6d4b5STrond Myklebust 5071bd5d6d0STrond Myklebust if (!kref_get_unless_zero(&subreq->wb_kref)) { 5081bd5d6d0STrond Myklebust if (subreq->wb_offset == head->wb_offset + total_bytes) 5091bd5d6d0STrond Myklebust total_bytes += subreq->wb_bytes; 5105b2b5187STrond Myklebust continue; 5111bd5d6d0STrond Myklebust } 5121bd5d6d0STrond Myklebust 51374a6d4b5STrond Myklebust while (!nfs_lock_request(subreq)) { 514b5bab9bfSTrond Myklebust /* 51574a6d4b5STrond Myklebust * Unlock page to allow nfs_page_group_sync_on_bit() 51674a6d4b5STrond Myklebust * to succeed 517b5bab9bfSTrond Myklebust */ 51874a6d4b5STrond Myklebust nfs_page_group_unlock(head); 51974a6d4b5STrond Myklebust ret = nfs_wait_on_request(subreq); 52074a6d4b5STrond Myklebust if (!ret) 5211344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 52274a6d4b5STrond Myklebust if (ret < 0) { 52374a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5245b2b5187STrond Myklebust nfs_release_request(subreq); 52574a6d4b5STrond Myklebust nfs_unlock_and_release_request(head); 526d4581383SWeston Andros Adamson return ERR_PTR(ret); 527d4581383SWeston Andros Adamson } 52874a6d4b5STrond Myklebust } 529e14bebf6STrond Myklebust /* 530e14bebf6STrond Myklebust * Subrequests are always contiguous, non overlapping 531e14bebf6STrond Myklebust * and in order - but may be repeated (mirrored writes). 532e14bebf6STrond Myklebust */ 533e14bebf6STrond Myklebust if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 534e14bebf6STrond Myklebust /* keep track of how many bytes this group covers */ 535e14bebf6STrond Myklebust total_bytes += subreq->wb_bytes; 536e14bebf6STrond Myklebust } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 537e14bebf6STrond Myklebust ((subreq->wb_offset + subreq->wb_bytes) > 538e14bebf6STrond Myklebust (head->wb_offset + total_bytes)))) { 5398b77484fSTrond Myklebust nfs_page_group_unlock(head); 54074a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5415b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 54274a6d4b5STrond Myklebust nfs_unlock_and_release_request(head); 543e14bebf6STrond Myklebust return ERR_PTR(-EIO); 544e14bebf6STrond Myklebust } 545a0e265bcSTrond Myklebust } 546d4581383SWeston Andros Adamson 547d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 548d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 549d4581383SWeston Andros Adamson subreq = head; 550d4581383SWeston Andros Adamson do { 551411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 552d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 553d4581383SWeston Andros Adamson } while (subreq != head); 554d4581383SWeston Andros Adamson 555d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 556d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 557d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 558d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 559d4581383SWeston Andros Adamson head->wb_this_page = head; 560d4581383SWeston Andros Adamson 561d4581383SWeston Andros Adamson /* change head request to cover whole range that 562d4581383SWeston Andros Adamson * the former page group covered */ 563d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 564d4581383SWeston Andros Adamson } 565d4581383SWeston Andros Adamson 566b66aaa8dSTrond Myklebust /* Postpone destruction of this request */ 567b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { 568b66aaa8dSTrond Myklebust set_bit(PG_INODE_REF, &head->wb_flags); 569b66aaa8dSTrond Myklebust kref_get(&head->wb_kref); 570a6b6d5b8STrond Myklebust atomic_long_inc(&NFS_I(inode)->nrequests); 571b66aaa8dSTrond Myklebust } 572b66aaa8dSTrond Myklebust 573d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 574d4581383SWeston Andros Adamson 575b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 576d4581383SWeston Andros Adamson 577b5bab9bfSTrond Myklebust /* Did we lose a race with nfs_inode_remove_request()? */ 578b5bab9bfSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) { 579b5bab9bfSTrond Myklebust nfs_unlock_and_release_request(head); 580b5bab9bfSTrond Myklebust return NULL; 581b5bab9bfSTrond Myklebust } 582b5bab9bfSTrond Myklebust 583bd37d6fcSTrond Myklebust /* still holds ref on head from nfs_page_find_head_request 584d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 585d4581383SWeston Andros Adamson return head; 586612c9384STrond Myklebust } 587074cc1deSTrond Myklebust 5880bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5890bcbf039SPeng Tao { 5900bcbf039SPeng Tao nfs_end_page_writeback(req); 5910bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5920bcbf039SPeng Tao req->wb_page); 5931f84ccdfSFred Isaman nfs_release_request(req); 5940bcbf039SPeng Tao } 5950bcbf039SPeng Tao 596a6598813STrond Myklebust static bool 597a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 598a6598813STrond Myklebust { 599a6598813STrond Myklebust switch (err) { 600a6598813STrond Myklebust case 0: 601a6598813STrond Myklebust case -ERESTARTSYS: 602a6598813STrond Myklebust case -EINTR: 603a6598813STrond Myklebust return false; 604a6598813STrond Myklebust } 605a6598813STrond Myklebust return nfs_error_is_fatal(err); 606e261f51fSTrond Myklebust } 607074cc1deSTrond Myklebust 608074cc1deSTrond Myklebust /* 609074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 610074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 611074cc1deSTrond Myklebust */ 612074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 6136d17d653STrond Myklebust struct page *page) 614074cc1deSTrond Myklebust { 615074cc1deSTrond Myklebust struct nfs_page *req; 616074cc1deSTrond Myklebust int ret = 0; 617074cc1deSTrond Myklebust 6186d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 619074cc1deSTrond Myklebust if (!req) 620074cc1deSTrond Myklebust goto out; 621074cc1deSTrond Myklebust ret = PTR_ERR(req); 622074cc1deSTrond Myklebust if (IS_ERR(req)) 623074cc1deSTrond Myklebust goto out; 624074cc1deSTrond Myklebust 625deed85e7STrond Myklebust nfs_set_page_writeback(page); 626deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 627074cc1deSTrond Myklebust 628deed85e7STrond Myklebust ret = 0; 629a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 630a6598813STrond Myklebust if (nfs_error_is_fatal_on_server(req->wb_context->error)) 631a6598813STrond Myklebust goto out_launder; 632a6598813STrond Myklebust 633f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 634074cc1deSTrond Myklebust ret = pgio->pg_error; 6350bcbf039SPeng Tao /* 636c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6370bcbf039SPeng Tao */ 6380bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 6390bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 640c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 641a6598813STrond Myklebust goto out_launder; 642d6c843b9SPeng Tao } 6430bcbf039SPeng Tao nfs_redirty_request(req); 6440bcbf039SPeng Tao ret = -EAGAIN; 64540f90271STrond Myklebust } else 64640f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 64740f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 648074cc1deSTrond Myklebust out: 649074cc1deSTrond Myklebust return ret; 650a6598813STrond Myklebust out_launder: 651a6598813STrond Myklebust nfs_write_error_remove_page(req); 652a6598813STrond Myklebust return ret; 653e261f51fSTrond Myklebust } 654e261f51fSTrond Myklebust 655d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 656c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 657f758c885STrond Myklebust { 658cfb506e1STrond Myklebust int ret; 659f758c885STrond Myklebust 6608cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 6616d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 662cfb506e1STrond Myklebust if (ret == -EAGAIN) { 663cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 664cfb506e1STrond Myklebust ret = 0; 665cfb506e1STrond Myklebust } 666cfb506e1STrond Myklebust return ret; 667f758c885STrond Myklebust } 668f758c885STrond Myklebust 669e261f51fSTrond Myklebust /* 6701da177e4SLinus Torvalds * Write an mmapped page to the server. 6711da177e4SLinus Torvalds */ 672d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 673c373fff7STrond Myklebust struct writeback_control *wbc) 6741da177e4SLinus Torvalds { 675f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 67640f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 677e261f51fSTrond Myklebust int err; 6781da177e4SLinus Torvalds 67940f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 680811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 681a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 682c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 683f758c885STrond Myklebust nfs_pageio_complete(&pgio); 684f758c885STrond Myklebust if (err < 0) 6854d770ccfSTrond Myklebust return err; 686f758c885STrond Myklebust if (pgio.pg_error < 0) 687f758c885STrond Myklebust return pgio.pg_error; 688f758c885STrond Myklebust return 0; 6894d770ccfSTrond Myklebust } 6904d770ccfSTrond Myklebust 6914d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6924d770ccfSTrond Myklebust { 693f758c885STrond Myklebust int ret; 6944d770ccfSTrond Myklebust 695c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6961da177e4SLinus Torvalds unlock_page(page); 697f758c885STrond Myklebust return ret; 698f758c885STrond Myklebust } 699f758c885STrond Myklebust 700f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 701f758c885STrond Myklebust { 702f758c885STrond Myklebust int ret; 703f758c885STrond Myklebust 704c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 705f758c885STrond Myklebust unlock_page(page); 706f758c885STrond Myklebust return ret; 7071da177e4SLinus Torvalds } 7081da177e4SLinus Torvalds 709919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 710919e3bd9STrond Myklebust { 711919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 712919e3bd9STrond Myklebust } 713919e3bd9STrond Myklebust 7141da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7151da177e4SLinus Torvalds { 7161da177e4SLinus Torvalds struct inode *inode = mapping->host; 717c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 718919e3bd9STrond Myklebust struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); 7191da177e4SLinus Torvalds int err; 7201da177e4SLinus Torvalds 72191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 72291d5b470SChuck Lever 723919e3bd9STrond Myklebust if (ioc) 724919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 725919e3bd9STrond Myklebust 726a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 727a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 728919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 729f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 730c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 731919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 73272cb77f4STrond Myklebust 733f758c885STrond Myklebust if (err < 0) 73472cb77f4STrond Myklebust goto out_err; 73572cb77f4STrond Myklebust err = pgio.pg_error; 73672cb77f4STrond Myklebust if (err < 0) 73772cb77f4STrond Myklebust goto out_err; 738c63c7b05STrond Myklebust return 0; 73972cb77f4STrond Myklebust out_err: 74072cb77f4STrond Myklebust return err; 7411da177e4SLinus Torvalds } 7421da177e4SLinus Torvalds 7431da177e4SLinus Torvalds /* 7441da177e4SLinus Torvalds * Insert a write request into an inode 7451da177e4SLinus Torvalds */ 746d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7471da177e4SLinus Torvalds { 7484b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7491da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 750e7d39069STrond Myklebust 7512bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7522bfc6e56SWeston Andros Adamson 753e7d39069STrond Myklebust /* Lock the request! */ 7547ad84aa9STrond Myklebust nfs_lock_request(req); 755e7d39069STrond Myklebust 75629418aa4SMel Gorman /* 75729418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 75829418aa4SMel Gorman * with invalidate/truncate. 75929418aa4SMel Gorman */ 7604b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 7614b9bb25bSTrond Myklebust if (!nfs_have_writebacks(inode) && 7624b9bb25bSTrond Myklebust NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) { 7634b9bb25bSTrond Myklebust spin_lock(&inode->i_lock); 7644b9bb25bSTrond Myklebust inode->i_version++; 7654b9bb25bSTrond Myklebust spin_unlock(&inode->i_lock); 7664b9bb25bSTrond Myklebust } 76729418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7682df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 769deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 770277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 77129418aa4SMel Gorman } 7724b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 773a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 77417089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 775cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 776cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 777cb1410c7SWeston Andros Adamson * adding subrequests. */ 77817089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 779c03b4024STrond Myklebust kref_get(&req->wb_kref); 7801da177e4SLinus Torvalds } 7811da177e4SLinus Torvalds 7821da177e4SLinus Torvalds /* 78389a09141SPeter Zijlstra * Remove a write request from an inode 7841da177e4SLinus Torvalds */ 7851da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7861da177e4SLinus Torvalds { 7874b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7884b9bb25bSTrond Myklebust struct inode *inode = mapping->host; 7891da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 79020633f04SWeston Andros Adamson struct nfs_page *head; 79120633f04SWeston Andros Adamson 792a6b6d5b8STrond Myklebust atomic_long_dec(&nfsi->nrequests); 79320633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 79420633f04SWeston Andros Adamson head = req->wb_head; 7951da177e4SLinus Torvalds 7964b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 79767911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 79820633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 79920633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 80020633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 80129418aa4SMel Gorman } 8024b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 80320633f04SWeston Andros Adamson } 80417089a29SWeston Andros Adamson 80517089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 8061da177e4SLinus Torvalds nfs_release_request(req); 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 * @dst: commit list head 875ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8768dd37758STrond Myklebust * 877ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8788dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8798dd37758STrond Myklebust * the MM page stats. 8808dd37758STrond Myklebust * 881ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8828dd37758STrond Myklebust * holding the nfs_page lock. 8838dd37758STrond Myklebust */ 8848dd37758STrond Myklebust void 8856272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8868dd37758STrond Myklebust { 887e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8886272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 889e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 89067911c8fSAnna Schumaker if (req->wb_page) 89186d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8928dd37758STrond Myklebust } 8938dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8948dd37758STrond Myklebust 8958dd37758STrond Myklebust /** 8968dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8978dd37758STrond Myklebust * @req: pointer to a nfs_page 898ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8998dd37758STrond Myklebust * 900ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 9018dd37758STrond Myklebust * number of outstanding requests requiring a commit 9028dd37758STrond Myklebust * It does not update the MM page stats. 9038dd37758STrond Myklebust * 904ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9058dd37758STrond Myklebust */ 9068dd37758STrond Myklebust void 907ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 908ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9098dd37758STrond Myklebust { 9108dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9118dd37758STrond Myklebust return; 9128dd37758STrond Myklebust nfs_list_remove_request(req); 9135cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9148dd37758STrond Myklebust } 9158dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9168dd37758STrond Myklebust 917ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 918ea2cf228SFred Isaman struct inode *inode) 919ea2cf228SFred Isaman { 920fe238e60SDave Wysochanski cinfo->inode = inode; 921ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 922ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 923b359f9d0SFred Isaman cinfo->dreq = NULL; 924f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 925ea2cf228SFred Isaman } 926ea2cf228SFred Isaman 927ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 928ea2cf228SFred Isaman struct inode *inode, 929ea2cf228SFred Isaman struct nfs_direct_req *dreq) 930ea2cf228SFred Isaman { 9311763da12SFred Isaman if (dreq) 9321763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9331763da12SFred Isaman else 934ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 935ea2cf228SFred Isaman } 936ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9378dd37758STrond Myklebust 9381da177e4SLinus Torvalds /* 9391da177e4SLinus Torvalds * Add a request to the inode's commit list. 9401da177e4SLinus Torvalds */ 9411763da12SFred Isaman void 942ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 943b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9441da177e4SLinus Torvalds { 945b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9468dd37758STrond Myklebust return; 9476272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9481da177e4SLinus Torvalds } 9498e821cadSTrond Myklebust 950d6d6dc7cSFred Isaman static void 951d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 952e468bae9STrond Myklebust { 95311fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 95493f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 95593f78d88STejun Heo WB_RECLAIMABLE); 956e468bae9STrond Myklebust } 957d6d6dc7cSFred Isaman 958b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 9598dd37758STrond Myklebust static void 960d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 961d6d6dc7cSFred Isaman { 9628dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9632b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->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)) { 1003d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 10046c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 10056c75dc0dSFred Isaman goto remove_req; 10066c75dc0dSFred Isaman } 1007c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 1008f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1009b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1010a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10116c75dc0dSFred Isaman goto next; 10126c75dc0dSFred Isaman } 10136c75dc0dSFred Isaman remove_req: 10146c75dc0dSFred Isaman nfs_inode_remove_request(req); 10156c75dc0dSFred Isaman next: 101620633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10173aff4ebbSTrond Myklebust nfs_release_request(req); 10186c75dc0dSFred Isaman } 10196c75dc0dSFred Isaman out: 1020919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10216c75dc0dSFred Isaman hdr->release(hdr); 10226c75dc0dSFred Isaman } 10236c75dc0dSFred Isaman 1024ce59515cSAnna Schumaker unsigned long 1025ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1026fb8a1f11STrond Myklebust { 10275cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1028fb8a1f11STrond Myklebust } 1029fb8a1f11STrond Myklebust 1030e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10311763da12SFred Isaman int 1032ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1033ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1034d6d6dc7cSFred Isaman { 1035137da553STrond Myklebust struct nfs_page *req, *tmp; 1036d6d6dc7cSFred Isaman int ret = 0; 1037d6d6dc7cSFred Isaman 1038137da553STrond Myklebust restart: 1039137da553STrond Myklebust list_for_each_entry_safe(req, tmp, src, wb_list) { 10407ad84aa9STrond Myklebust kref_get(&req->wb_kref); 10412ce209c4STrond Myklebust if (!nfs_lock_request(req)) { 10422ce209c4STrond Myklebust int status; 1043137da553STrond Myklebust 1044137da553STrond Myklebust /* Prevent deadlock with nfs_lock_and_join_requests */ 1045137da553STrond Myklebust if (!list_empty(dst)) { 1046137da553STrond Myklebust nfs_release_request(req); 1047137da553STrond Myklebust continue; 1048137da553STrond Myklebust } 1049137da553STrond Myklebust /* Ensure we make progress to prevent livelock */ 10502ce209c4STrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 10512ce209c4STrond Myklebust status = nfs_wait_on_request(req); 10522ce209c4STrond Myklebust nfs_release_request(req); 10532ce209c4STrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10542ce209c4STrond Myklebust if (status < 0) 10552ce209c4STrond Myklebust break; 1056137da553STrond Myklebust goto restart; 10572ce209c4STrond Myklebust } 1058ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10595d2a9d9dSTrond Myklebust clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); 10608dd37758STrond Myklebust nfs_list_add_request(req, dst); 1061d6d6dc7cSFred Isaman ret++; 10621763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1063d6d6dc7cSFred Isaman break; 1064e824f99aSTrond Myklebust cond_resched(); 1065d6d6dc7cSFred Isaman } 1066d6d6dc7cSFred Isaman return ret; 1067d6d6dc7cSFred Isaman } 10685d2a9d9dSTrond Myklebust EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 1069d6d6dc7cSFred Isaman 10701da177e4SLinus Torvalds /* 10711da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10721da177e4SLinus Torvalds * @inode: NFS inode to scan 1073ea2cf228SFred Isaman * @dst: mds destination list 1074ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10751da177e4SLinus Torvalds * 10761da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10771da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10781da177e4SLinus Torvalds */ 10791763da12SFred Isaman int 1080ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1081ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10821da177e4SLinus Torvalds { 1083d6d6dc7cSFred Isaman int ret = 0; 1084fb8a1f11STrond Myklebust 10855cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10865cb953d4STrond Myklebust return 0; 1087e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10885cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10898dd37758STrond Myklebust const int max = INT_MAX; 1090d6d6dc7cSFred Isaman 1091ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1092ea2cf228SFred Isaman cinfo, max); 1093ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1094d6d6dc7cSFred Isaman } 1095e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1096ff778d02STrond Myklebust return ret; 10971da177e4SLinus Torvalds } 1098d6d6dc7cSFred Isaman 10991da177e4SLinus Torvalds /* 1100e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1101e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 11021da177e4SLinus Torvalds * 1103e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1104e7d39069STrond Myklebust * to disk. 11051da177e4SLinus Torvalds */ 1106e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1107e7d39069STrond Myklebust struct page *page, 1108e7d39069STrond Myklebust unsigned int offset, 1109e7d39069STrond Myklebust unsigned int bytes) 11101da177e4SLinus Torvalds { 1111e7d39069STrond Myklebust struct nfs_page *req; 1112e7d39069STrond Myklebust unsigned int rqend; 1113e7d39069STrond Myklebust unsigned int end; 11141da177e4SLinus Torvalds int error; 1115277459d2STrond Myklebust 1116e7d39069STrond Myklebust end = offset + bytes; 1117e7d39069STrond Myklebust 1118f6032f21STrond Myklebust req = nfs_lock_and_join_requests(page); 1119f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1120f6032f21STrond Myklebust return req; 11212bfc6e56SWeston Andros Adamson 1122e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1123e7d39069STrond Myklebust /* 1124e7d39069STrond Myklebust * Tell the caller to flush out the request if 1125e7d39069STrond Myklebust * the offsets are non-contiguous. 1126e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1127e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1128e7d39069STrond Myklebust */ 1129f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1130e7d39069STrond Myklebust goto out_flushme; 1131e7d39069STrond Myklebust 11321da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11331da177e4SLinus Torvalds if (offset < req->wb_offset) { 11341da177e4SLinus Torvalds req->wb_offset = offset; 11351da177e4SLinus Torvalds req->wb_pgbase = offset; 11361da177e4SLinus Torvalds } 11371da177e4SLinus Torvalds if (end > rqend) 11381da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1139e7d39069STrond Myklebust else 1140e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1141e7d39069STrond Myklebust return req; 1142e7d39069STrond Myklebust out_flushme: 1143f6032f21STrond Myklebust /* 1144f6032f21STrond Myklebust * Note: we mark the request dirty here because 1145f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1146f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1147f6032f21STrond Myklebust */ 1148f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1149f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 1150e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1151f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1152e7d39069STrond Myklebust } 11531da177e4SLinus Torvalds 1154e7d39069STrond Myklebust /* 1155e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1156e7d39069STrond Myklebust * 1157e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1158e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1159e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1160e7d39069STrond Myklebust */ 1161e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1162e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1163e7d39069STrond Myklebust { 1164d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1165e7d39069STrond Myklebust struct nfs_page *req; 1166e7d39069STrond Myklebust 1167e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1168e7d39069STrond Myklebust if (req != NULL) 1169e7d39069STrond Myklebust goto out; 11702bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1171e7d39069STrond Myklebust if (IS_ERR(req)) 1172e7d39069STrond Myklebust goto out; 1173d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1174efc91ed0STrond Myklebust out: 117561e930a9STrond Myklebust return req; 11761da177e4SLinus Torvalds } 11771da177e4SLinus Torvalds 1178e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1179e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1180e7d39069STrond Myklebust { 1181e7d39069STrond Myklebust struct nfs_page *req; 1182e7d39069STrond Myklebust 1183e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1184e7d39069STrond Myklebust if (IS_ERR(req)) 1185e7d39069STrond Myklebust return PTR_ERR(req); 1186e7d39069STrond Myklebust /* Update file length */ 1187e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1188d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1189a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11901d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1191e7d39069STrond Myklebust return 0; 1192e7d39069STrond Myklebust } 1193e7d39069STrond Myklebust 11941da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11951da177e4SLinus Torvalds { 1196cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11972a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1198bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11991da177e4SLinus Torvalds struct nfs_page *req; 12001a54533eSTrond Myklebust int do_flush, status; 12011da177e4SLinus Torvalds /* 12021da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 12031da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 12041da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 12051da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 12061da177e4SLinus Torvalds * Also do the same if we find a request from an existing 12071da177e4SLinus Torvalds * dropped page. 12081da177e4SLinus Torvalds */ 12091a54533eSTrond Myklebust do { 121084d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 12111a54533eSTrond Myklebust if (req == NULL) 12121a54533eSTrond Myklebust return 0; 12132a369153STrond Myklebust l_ctx = req->wb_lock_context; 1214138a2935STrond Myklebust do_flush = req->wb_page != page || 1215138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 1216bd61e0a9SJeff Layton if (l_ctx && flctx && 1217bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1218bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1219d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12205263e31eSJeff Layton } 12211da177e4SLinus Torvalds nfs_release_request(req); 12221a54533eSTrond Myklebust if (!do_flush) 12231a54533eSTrond Myklebust return 0; 1224d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12251a54533eSTrond Myklebust } while (status == 0); 12261a54533eSTrond Myklebust return status; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds /* 1230dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1231dc24826bSAndy Adamson * expire soon. 1232dc24826bSAndy Adamson * 1233dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1234dc24826bSAndy Adamson * 1235dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1236dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1237dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1238dc24826bSAndy Adamson */ 1239dc24826bSAndy Adamson int 1240dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1241dc24826bSAndy Adamson { 1242dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1243dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1244dc24826bSAndy Adamson 1245dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1246dc24826bSAndy Adamson } 1247dc24826bSAndy Adamson 1248dc24826bSAndy Adamson /* 1249dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1250dc24826bSAndy Adamson */ 1251ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1252dc24826bSAndy Adamson { 1253ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1254ce52914eSScott Mayhew 1255ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1256dc24826bSAndy Adamson } 1257dc24826bSAndy Adamson 1258dc24826bSAndy Adamson /* 12595d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12605d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12615d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12625d47a356STrond Myklebust */ 12638d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12645d47a356STrond Myklebust { 1265d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1266d529ef83SJeff Layton 12678d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12688d197a56STrond Myklebust goto out; 126918dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1270d529ef83SJeff Layton return false; 12714db72b40SJeff Layton smp_rmb(); 1272d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12738d197a56STrond Myklebust return false; 12748d197a56STrond Myklebust out: 127518dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 127618dd78c4SScott Mayhew return false; 12778d197a56STrond Myklebust return PageUptodate(page) != 0; 12785d47a356STrond Myklebust } 12795d47a356STrond Myklebust 12805263e31eSJeff Layton static bool 12815263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12825263e31eSJeff Layton { 12835263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12845263e31eSJeff Layton fl->fl_type == F_WRLCK; 12855263e31eSJeff Layton } 12865263e31eSJeff Layton 1287c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1288c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1289c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1290c7559663SScott Mayhew * inefficiencies. 1291c7559663SScott Mayhew * 1292263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1293263b4509SScott Mayhew * of the checks. 1294c7559663SScott Mayhew */ 1295c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1296c7559663SScott Mayhew { 12975263e31eSJeff Layton int ret; 12985263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12995263e31eSJeff Layton struct file_lock *fl; 13005263e31eSJeff Layton 1301c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1302c7559663SScott Mayhew return 0; 1303263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1304263b4509SScott Mayhew return 0; 1305c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1306c7559663SScott Mayhew return 1; 1307bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1308bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13098fa4592aSTrond Myklebust return 1; 13105263e31eSJeff Layton 13115263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13125263e31eSJeff Layton ret = 0; 13136109c850SJeff Layton spin_lock(&flctx->flc_lock); 1314bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1315bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1316bd61e0a9SJeff Layton fl_list); 1317bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13185263e31eSJeff Layton ret = 1; 1319bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13205263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13215263e31eSJeff Layton fl_list); 13225263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13235263e31eSJeff Layton ret = 1; 13245263e31eSJeff Layton } 13256109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13265263e31eSJeff Layton return ret; 1327c7559663SScott Mayhew } 1328c7559663SScott Mayhew 13295d47a356STrond Myklebust /* 13301da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13311da177e4SLinus Torvalds * 13321da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13331da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13341da177e4SLinus Torvalds */ 13351da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13361da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13371da177e4SLinus Torvalds { 1338cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1339d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 13401da177e4SLinus Torvalds int status = 0; 13411da177e4SLinus Torvalds 134291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 134391d5b470SChuck Lever 13446de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13456de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13461da177e4SLinus Torvalds 1347149a4fddSBenjamin Coddington if (!count) 1348149a4fddSBenjamin Coddington goto out; 1349149a4fddSBenjamin Coddington 1350c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 135149a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13521da177e4SLinus Torvalds offset = 0; 13531da177e4SLinus Torvalds } 13541da177e4SLinus Torvalds 1355e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 135603fa9e84STrond Myklebust if (status < 0) 135703fa9e84STrond Myklebust nfs_set_pageerror(page); 135859b7c05fSTrond Myklebust else 135959b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1360149a4fddSBenjamin Coddington out: 136148186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13621da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13631da177e4SLinus Torvalds return status; 13641da177e4SLinus Torvalds } 13651da177e4SLinus Torvalds 13663ff7576dSTrond Myklebust static int flush_task_priority(int how) 13671da177e4SLinus Torvalds { 13681da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13691da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13701da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13711da177e4SLinus Torvalds case FLUSH_LOWPRI: 13721da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13751da177e4SLinus Torvalds } 13761da177e4SLinus Torvalds 1377d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1378d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1379abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13801ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13811da177e4SLinus Torvalds { 13823ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13831da177e4SLinus Torvalds 13841ed26f33SAnna Schumaker task_setup_data->priority = priority; 1385abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1386d138d5d1SAndy Adamson 1387abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1388d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1389275acaafSTrond Myklebust } 1390275acaafSTrond Myklebust 13916d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13926d884e8fSFred * call this on each, which will prepare them to be retried on next 13936d884e8fSFred * writeback using standard nfs. 13946d884e8fSFred */ 13956d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13966d884e8fSFred { 13976d884e8fSFred nfs_mark_request_dirty(req); 1398c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 139920633f04SWeston Andros Adamson nfs_end_page_writeback(req); 14003aff4ebbSTrond Myklebust nfs_release_request(req); 14016d884e8fSFred } 14026d884e8fSFred 1403061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 14046c75dc0dSFred Isaman { 14056c75dc0dSFred Isaman struct nfs_page *req; 14066c75dc0dSFred Isaman 14076c75dc0dSFred Isaman while (!list_empty(head)) { 14086c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14096c75dc0dSFred Isaman nfs_list_remove_request(req); 14106c75dc0dSFred Isaman nfs_redirty_request(req); 14116c75dc0dSFred Isaman } 14126c75dc0dSFred Isaman } 14136c75dc0dSFred Isaman 1414dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1415dc602dd7STrond Myklebust { 1416dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1417dc602dd7STrond Myklebust } 1418dc602dd7STrond Myklebust 1419061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1420919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1421061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1422061ae2edSFred Isaman .completion = nfs_write_completion, 1423dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1424061ae2edSFred Isaman }; 1425061ae2edSFred Isaman 142657208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1427a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1428061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14291751c363STrond Myklebust { 1430a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 143141d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1432a20c93e3SChristoph Hellwig 1433a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1434a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1435a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1436a20c93e3SChristoph Hellwig #endif 14374a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 14383bde7afdSTrond Myklebust server->wsize, ioflags); 14391751c363STrond Myklebust } 1440ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14411751c363STrond Myklebust 1442dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1443dce81290STrond Myklebust { 1444a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1445a7d42ddbSWeston Andros Adamson 14466f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14476f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14486f29b9bbSKinglong Mee 144941d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1450a7d42ddbSWeston Andros Adamson 1451a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1452a7d42ddbSWeston Andros Adamson 1453a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1454a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1455dce81290STrond Myklebust } 14561f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1457dce81290STrond Myklebust 14581da177e4SLinus Torvalds 14590b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14600b7c0153SFred Isaman { 14610b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14620b7c0153SFred Isaman 14630b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14640b7c0153SFred Isaman } 14650b7c0153SFred Isaman 14661f2edbe3STrond Myklebust /* 14671f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14681f2edbe3STrond Myklebust */ 14691f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14701f2edbe3STrond Myklebust { 14711f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14721f2edbe3STrond Myklebust int kill = 0; 1473788e7a89STrond Myklebust 14741f2edbe3STrond Myklebust /* suid always must be killed */ 14751f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14761f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14771f2edbe3STrond Myklebust 14781f2edbe3STrond Myklebust /* 14791f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14801f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14811f2edbe3STrond Myklebust */ 14821f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14831f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14841f2edbe3STrond Myklebust 14851f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14861f2edbe3STrond Myklebust return kill; 14871f2edbe3STrond Myklebust 14881f2edbe3STrond Myklebust return 0; 14891f2edbe3STrond Myklebust } 1490788e7a89STrond Myklebust 1491a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1492a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1493a08a8cd3STrond Myklebust { 1494a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1495a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14962b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1497a08a8cd3STrond Myklebust 1498a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14992b83d3deSTrond Myklebust fattr->size = size; 15002b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15012b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1502a08a8cd3STrond Myklebust return; 15032b83d3deSTrond Myklebust } 15042b83d3deSTrond Myklebust if (size != fattr->size) 1505a08a8cd3STrond Myklebust return; 1506a08a8cd3STrond Myklebust /* Set attribute barrier */ 1507a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15082b83d3deSTrond Myklebust /* ...and update size */ 15092b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1510a08a8cd3STrond Myklebust } 1511a08a8cd3STrond Myklebust 1512a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1513a08a8cd3STrond Myklebust { 15142b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1515a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1516a08a8cd3STrond Myklebust 1517a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1518a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1519a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1520a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1521a08a8cd3STrond Myklebust } 1522a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1523a08a8cd3STrond Myklebust 15241da177e4SLinus Torvalds /* 15251da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15261da177e4SLinus Torvalds */ 1527d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1528d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15290eecb214SAnna Schumaker struct inode *inode) 15301da177e4SLinus Torvalds { 1531788e7a89STrond Myklebust int status; 15321da177e4SLinus Torvalds 1533f551e44fSChuck Lever /* 1534f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1535f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1536f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1537f551e44fSChuck Lever * another writer had changed the file, but some applications 1538f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1539f551e44fSChuck Lever */ 1540d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1541788e7a89STrond Myklebust if (status != 0) 15420eecb214SAnna Schumaker return status; 1543d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 154491d5b470SChuck Lever 1545d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1546d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15471da177e4SLinus Torvalds /* We tried a write call, but the server did not 15481da177e4SLinus Torvalds * commit data to stable storage even though we 15491da177e4SLinus Torvalds * requested it. 15501da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15511da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15521da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15531da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15541da177e4SLinus Torvalds */ 15551da177e4SLinus Torvalds static unsigned long complain; 15561da177e4SLinus Torvalds 1557a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15581da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15591da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15601da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1561cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1562d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15631da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15641da177e4SLinus Torvalds } 15651da177e4SLinus Torvalds } 15661f2edbe3STrond Myklebust 15671f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15681f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15691f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15700eecb214SAnna Schumaker return 0; 15710eecb214SAnna Schumaker } 15720eecb214SAnna Schumaker 15730eecb214SAnna Schumaker /* 15740eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15750eecb214SAnna Schumaker */ 1576d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1577d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15780eecb214SAnna Schumaker { 1579d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1580d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15811f2edbe3STrond Myklebust 15821f2edbe3STrond Myklebust if (resp->count < argp->count) { 15831da177e4SLinus Torvalds static unsigned long complain; 15841da177e4SLinus Torvalds 15856c75dc0dSFred Isaman /* This a short write! */ 1586d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 158791d5b470SChuck Lever 15881da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15896c75dc0dSFred Isaman if (resp->count == 0) { 15906c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15916c75dc0dSFred Isaman printk(KERN_WARNING 15926c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 15936c75dc0dSFred Isaman argp->count); 15946c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15956c75dc0dSFred Isaman } 1596d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15976c75dc0dSFred Isaman task->tk_status = -EIO; 15986c75dc0dSFred Isaman return; 15996c75dc0dSFred Isaman } 1600f8417b48SKinglong Mee 1601f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1602f8417b48SKinglong Mee if (!task->tk_ops) { 1603f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1604f8417b48SKinglong Mee return; 1605f8417b48SKinglong Mee } 1606f8417b48SKinglong Mee 16071da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16081da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16091da177e4SLinus Torvalds /* Resend from where the server left off */ 1610d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16111da177e4SLinus Torvalds argp->offset += resp->count; 16121da177e4SLinus Torvalds argp->pgbase += resp->count; 16131da177e4SLinus Torvalds argp->count -= resp->count; 16141da177e4SLinus Torvalds } else { 16151da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16161da177e4SLinus Torvalds * headaches in the case of a server crash. 16171da177e4SLinus Torvalds */ 16181da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16191da177e4SLinus Torvalds } 1620d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16211da177e4SLinus Torvalds } 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds 1624af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 162571d0a611STrond Myklebust { 1626af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1627af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1628af7cf057STrond Myklebust } 1629af7cf057STrond Myklebust 1630af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1631af7cf057STrond Myklebust { 1632af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1633af7cf057STrond Myklebust } 1634af7cf057STrond Myklebust 1635af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1636af7cf057STrond Myklebust { 1637af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1638af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 163971d0a611STrond Myklebust } 164071d0a611STrond Myklebust 16410b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16421da177e4SLinus Torvalds { 16430b7c0153SFred Isaman put_nfs_open_context(data->context); 16440b7c0153SFred Isaman nfs_commit_free(data); 16451da177e4SLinus Torvalds } 1646e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16471da177e4SLinus Torvalds 16480b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1649c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16509ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16519f0ec176SAndy Adamson int how, int flags) 16521da177e4SLinus Torvalds { 165307737691STrond Myklebust struct rpc_task *task; 16549ace33cdSFred Isaman int priority = flush_task_priority(how); 1655bdc7f021STrond Myklebust struct rpc_message msg = { 1656bdc7f021STrond Myklebust .rpc_argp = &data->args, 1657bdc7f021STrond Myklebust .rpc_resp = &data->res, 16589ace33cdSFred Isaman .rpc_cred = data->cred, 1659bdc7f021STrond Myklebust }; 166084115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 166107737691STrond Myklebust .task = &data->task, 16629ace33cdSFred Isaman .rpc_client = clnt, 1663bdc7f021STrond Myklebust .rpc_message = &msg, 16649ace33cdSFred Isaman .callback_ops = call_ops, 166584115e1cSTrond Myklebust .callback_data = data, 1666101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16679f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16683ff7576dSTrond Myklebust .priority = priority, 166984115e1cSTrond Myklebust }; 1670788e7a89STrond Myklebust /* Set up the initial task struct. */ 1671c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16721da177e4SLinus Torvalds 1673b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1674bdc7f021STrond Myklebust 16758c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16768c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16778c21c62cSWeston Andros Adamson 167807737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1679dbae4c73STrond Myklebust if (IS_ERR(task)) 1680dbae4c73STrond Myklebust return PTR_ERR(task); 1681d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1682d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 168307737691STrond Myklebust rpc_put_task(task); 1684dbae4c73STrond Myklebust return 0; 16851da177e4SLinus Torvalds } 1686e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16871da177e4SLinus Torvalds 1688378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1689378520b8SPeng Tao { 1690378520b8SPeng Tao loff_t lwb = 0; 1691378520b8SPeng Tao struct nfs_page *req; 1692378520b8SPeng Tao 1693378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1694378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1695378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1696378520b8SPeng Tao 1697378520b8SPeng Tao return lwb; 1698378520b8SPeng Tao } 1699378520b8SPeng Tao 17001da177e4SLinus Torvalds /* 17019ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17029ace33cdSFred Isaman */ 17030b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1704988b6dceSFred Isaman struct list_head *head, 1705f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1706f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17079ace33cdSFred Isaman { 17089ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 17092b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 17109ace33cdSFred Isaman 17119ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17129ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17139ace33cdSFred Isaman 17149ace33cdSFred Isaman list_splice_init(head, &data->pages); 17159ace33cdSFred Isaman 17169ace33cdSFred Isaman data->inode = inode; 17179ace33cdSFred Isaman data->cred = first->wb_context->cred; 1718988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1719378520b8SPeng Tao /* only set lwb for pnfs commit */ 1720378520b8SPeng Tao if (lseg) 1721378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17229ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1723f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1724b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17259ace33cdSFred Isaman 17269ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17279ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17289ace33cdSFred Isaman data->args.offset = 0; 17299ace33cdSFred Isaman data->args.count = 0; 17300b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 17319ace33cdSFred Isaman data->res.fattr = &data->fattr; 17329ace33cdSFred Isaman data->res.verf = &data->verf; 17339ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17349ace33cdSFred Isaman } 1735e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17369ace33cdSFred Isaman 1737e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1738ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1739b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1740b57ff130SWeston Andros Adamson u32 ds_commit_idx) 174164bfeb49SFred Isaman { 174264bfeb49SFred Isaman struct nfs_page *req; 174364bfeb49SFred Isaman 174464bfeb49SFred Isaman while (!list_empty(page_list)) { 174564bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 174664bfeb49SFred Isaman nfs_list_remove_request(req); 1747b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1748487b9b8aSTom Haynes if (!cinfo->dreq) 1749487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17501d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 175164bfeb49SFred Isaman } 175264bfeb49SFred Isaman } 1753e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 175464bfeb49SFred Isaman 1755b20135d0STrond Myklebust static void 1756b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1757b20135d0STrond Myklebust struct nfs_page *req) 1758b20135d0STrond Myklebust { 1759b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1760b20135d0STrond Myklebust } 1761b20135d0STrond Myklebust 17629ace33cdSFred Isaman /* 17631da177e4SLinus Torvalds * Commit dirty pages 17641da177e4SLinus Torvalds */ 17651da177e4SLinus Torvalds static int 1766ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1767ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17681da177e4SLinus Torvalds { 17690b7c0153SFred Isaman struct nfs_commit_data *data; 17701da177e4SLinus Torvalds 1771ade8febdSWeston Andros Adamson /* another commit raced with us */ 1772ade8febdSWeston Andros Adamson if (list_empty(head)) 1773ade8febdSWeston Andros Adamson return 0; 1774ade8febdSWeston Andros Adamson 1775518662e0SNeilBrown data = nfs_commitdata_alloc(true); 17761da177e4SLinus Torvalds 17771da177e4SLinus Torvalds /* Set up the argument struct */ 1778f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1779f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1780c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1781c36aae9aSPeng Tao data->mds_ops, how, 0); 17821da177e4SLinus Torvalds } 17831da177e4SLinus Torvalds 17841da177e4SLinus Torvalds /* 17851da177e4SLinus Torvalds * COMMIT call returned 17861da177e4SLinus Torvalds */ 1787788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17881da177e4SLinus Torvalds { 17890b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17901da177e4SLinus Torvalds 1791a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17921da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17931da177e4SLinus Torvalds 1794788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1795c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1796c9d8f89dSTrond Myklebust } 1797c9d8f89dSTrond Myklebust 1798f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1799c9d8f89dSTrond Myklebust { 1800c9d8f89dSTrond Myklebust struct nfs_page *req; 1801c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1802f453a54aSFred Isaman struct nfs_commit_info cinfo; 1803353db796SNeilBrown struct nfs_server *nfss; 1804788e7a89STrond Myklebust 18051da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18061da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18071da177e4SLinus Torvalds nfs_list_remove_request(req); 180867911c8fSAnna Schumaker if (req->wb_page) 1809d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 18101da177e4SLinus Torvalds 18111e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18123d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 18132b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 18141da177e4SLinus Torvalds req->wb_bytes, 18151da177e4SLinus Torvalds (long long)req_offset(req)); 1816c9d8f89dSTrond Myklebust if (status < 0) { 1817c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 181838a33101SKinglong Mee if (req->wb_page) 18191da177e4SLinus Torvalds nfs_inode_remove_request(req); 1820ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18211da177e4SLinus Torvalds goto next; 18221da177e4SLinus Torvalds } 18231da177e4SLinus Torvalds 18241da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18251da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18268fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18271da177e4SLinus Torvalds /* We have a match */ 182838a33101SKinglong Mee if (req->wb_page) 18291da177e4SLinus Torvalds nfs_inode_remove_request(req); 1830ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18311da177e4SLinus Torvalds goto next; 18321da177e4SLinus Torvalds } 18331da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1834ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18356d884e8fSFred nfs_mark_request_dirty(req); 183605990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18371da177e4SLinus Torvalds next: 18381d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18391da177e4SLinus Torvalds } 1840353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1841353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18420db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1843353db796SNeilBrown 1844f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1845af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18465917ce84SFred Isaman } 18475917ce84SFred Isaman 18485917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18495917ce84SFred Isaman { 18500b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18515917ce84SFred Isaman 1852f453a54aSFred Isaman data->completion_ops->completion(data); 1853c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18541da177e4SLinus Torvalds } 1855788e7a89STrond Myklebust 1856788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18570b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1858788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1859788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1860788e7a89STrond Myklebust }; 18611da177e4SLinus Torvalds 1862f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1863f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1864b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1865f453a54aSFred Isaman }; 1866f453a54aSFred Isaman 18671763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1868ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 186984c53ab5SFred Isaman { 187084c53ab5SFred Isaman int status; 187184c53ab5SFred Isaman 1872ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 187384c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1874ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 187584c53ab5SFred Isaman return status; 187684c53ab5SFred Isaman } 187784c53ab5SFred Isaman 1878b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18791da177e4SLinus Torvalds { 18801da177e4SLinus Torvalds LIST_HEAD(head); 1881ea2cf228SFred Isaman struct nfs_commit_info cinfo; 188271d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1883af7cf057STrond Myklebust int error = 0; 1884b8413f98STrond Myklebust int res; 18851da177e4SLinus Torvalds 1886ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1887af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1888ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1889af7cf057STrond Myklebust if (res) 1890ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1891af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18921da177e4SLinus Torvalds if (error < 0) 1893af7cf057STrond Myklebust goto out_error; 1894b8413f98STrond Myklebust if (!may_wait) 1895b8413f98STrond Myklebust goto out_mark_dirty; 1896af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1897b8413f98STrond Myklebust if (error < 0) 1898b8413f98STrond Myklebust return error; 1899c5efa5fcSTrond Myklebust return res; 1900af7cf057STrond Myklebust out_error: 1901af7cf057STrond Myklebust res = error; 1902c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1903c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1904c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1905c5efa5fcSTrond Myklebust * that the data is on the disk. 1906c5efa5fcSTrond Myklebust */ 1907c5efa5fcSTrond Myklebust out_mark_dirty: 1908c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19091da177e4SLinus Torvalds return res; 19101da177e4SLinus Torvalds } 1911b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19128fc795f7STrond Myklebust 1913ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19148fc795f7STrond Myklebust { 1915420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1916420e3646STrond Myklebust int flags = FLUSH_SYNC; 1917420e3646STrond Myklebust int ret = 0; 19188fc795f7STrond Myklebust 19193236c3e1SJeff Layton /* no commits means nothing needs to be done */ 19205cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 19213236c3e1SJeff Layton return ret; 19223236c3e1SJeff Layton 1923a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1924a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1925a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1926420e3646STrond Myklebust */ 19271a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1928420e3646STrond Myklebust goto out_mark_dirty; 1929420e3646STrond Myklebust 1930a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1931420e3646STrond Myklebust flags = 0; 1932a00dd6c0SJeff Layton } 1933a00dd6c0SJeff Layton 1934420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1935420e3646STrond Myklebust if (ret >= 0) { 1936420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1937420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1938420e3646STrond Myklebust wbc->nr_to_write -= ret; 1939420e3646STrond Myklebust else 1940420e3646STrond Myklebust wbc->nr_to_write = 0; 1941420e3646STrond Myklebust } 19428fc795f7STrond Myklebust return 0; 1943420e3646STrond Myklebust } 1944420e3646STrond Myklebust out_mark_dirty: 19458fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19468fc795f7STrond Myklebust return ret; 19478fc795f7STrond Myklebust } 194889d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1949863a3c6cSAndy Adamson 1950acdc53b2STrond Myklebust /* 1951837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1952837bb1d7STrond Myklebust * 1953837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1954837bb1d7STrond Myklebust * client. 1955837bb1d7STrond Myklebust */ 1956837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1957837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1958837bb1d7STrond Myklebust { 1959837bb1d7STrond Myklebust int ret; 1960837bb1d7STrond Myklebust 1961837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1962837bb1d7STrond Myklebust if (ret == 0) 1963837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1964837bb1d7STrond Myklebust return ret; 1965837bb1d7STrond Myklebust } 1966837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1967837bb1d7STrond Myklebust 1968837bb1d7STrond Myklebust /* 1969acdc53b2STrond Myklebust * flush the inode to disk. 1970acdc53b2STrond Myklebust */ 1971acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 197234901f70STrond Myklebust { 1973f4ce1299STrond Myklebust int ret; 197434901f70STrond Myklebust 1975f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1976f4ce1299STrond Myklebust 19775bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19786b196875SChuck Lever if (ret) 19796b196875SChuck Lever goto out; 19805bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19816b196875SChuck Lever if (ret < 0) 19826b196875SChuck Lever goto out; 19835bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19846b196875SChuck Lever ret = 0; 1985f4ce1299STrond Myklebust 19866b196875SChuck Lever out: 1987f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1988f4ce1299STrond Myklebust return ret; 19891c75950bSTrond Myklebust } 1990ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19911c75950bSTrond Myklebust 19921b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19931b3b4a1aSTrond Myklebust { 19941b3b4a1aSTrond Myklebust struct nfs_page *req; 19951b3b4a1aSTrond Myklebust int ret = 0; 19961b3b4a1aSTrond Myklebust 1997ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19983e217045SWeston Andros Adamson 19993e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20003e217045SWeston Andros Adamson * request */ 20016d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 20023e217045SWeston Andros Adamson 20033e217045SWeston Andros Adamson if (IS_ERR(req)) { 20043e217045SWeston Andros Adamson ret = PTR_ERR(req); 20053e217045SWeston Andros Adamson } else if (req) { 20063e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 20073e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20083e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20093e217045SWeston Andros Adamson * release it */ 20101b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20111d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20121b3b4a1aSTrond Myklebust } 20133e217045SWeston Andros Adamson 20141b3b4a1aSTrond Myklebust return ret; 20151b3b4a1aSTrond Myklebust } 20161b3b4a1aSTrond Myklebust 20171c75950bSTrond Myklebust /* 20181c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 20191c75950bSTrond Myklebust */ 2020c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 20211c75950bSTrond Myklebust { 202229418aa4SMel Gorman loff_t range_start = page_file_offset(page); 202309cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 20247f2f12d9STrond Myklebust struct writeback_control wbc = { 20257f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20267f2f12d9STrond Myklebust .nr_to_write = 0, 20277f2f12d9STrond Myklebust .range_start = range_start, 20287f2f12d9STrond Myklebust .range_end = range_end, 20297f2f12d9STrond Myklebust }; 20307f2f12d9STrond Myklebust int ret; 20317f2f12d9STrond Myklebust 2032f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2033f4ce1299STrond Myklebust 20340522f6adSTrond Myklebust for (;;) { 2035ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20367f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2037c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20387f2f12d9STrond Myklebust if (ret < 0) 20397f2f12d9STrond Myklebust goto out_error; 20400522f6adSTrond Myklebust continue; 20417f2f12d9STrond Myklebust } 2042f4ce1299STrond Myklebust ret = 0; 20430522f6adSTrond Myklebust if (!PagePrivate(page)) 20440522f6adSTrond Myklebust break; 20450522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20467f2f12d9STrond Myklebust if (ret < 0) 20477f2f12d9STrond Myklebust goto out_error; 20487f2f12d9STrond Myklebust } 20497f2f12d9STrond Myklebust out_error: 2050f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20517f2f12d9STrond Myklebust return ret; 20521c75950bSTrond Myklebust } 20531c75950bSTrond Myklebust 2054074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2055074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2056a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2057074cc1deSTrond Myklebust { 20582da95652SJeff Layton /* 20592da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20602da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20612da95652SJeff Layton * 20622da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20632da95652SJeff Layton * that we can safely release the inode reference while holding 20642da95652SJeff Layton * the page lock. 20652da95652SJeff Layton */ 20662da95652SJeff Layton if (PagePrivate(page)) 20672da95652SJeff Layton return -EBUSY; 2068074cc1deSTrond Myklebust 20698c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20708c209ce7SDavid Howells return -EBUSY; 2071074cc1deSTrond Myklebust 2072a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2073074cc1deSTrond Myklebust } 2074074cc1deSTrond Myklebust #endif 2075074cc1deSTrond Myklebust 2076f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20771da177e4SLinus Torvalds { 20781da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20791e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20801da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 208120c2df83SPaul Mundt NULL); 20821da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20831da177e4SLinus Torvalds return -ENOMEM; 20841da177e4SLinus Torvalds 208593d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20861da177e4SLinus Torvalds nfs_wdata_cachep); 20871da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20883dd4765fSJeff Layton goto out_destroy_write_cache; 20891da177e4SLinus Torvalds 20900b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20910b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20920b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20930b7c0153SFred Isaman NULL); 20940b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20953dd4765fSJeff Layton goto out_destroy_write_mempool; 20960b7c0153SFred Isaman 209793d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20984c100210SYanchuan Nian nfs_cdata_cachep); 20991da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21003dd4765fSJeff Layton goto out_destroy_commit_cache; 21011da177e4SLinus Torvalds 210289a09141SPeter Zijlstra /* 210389a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 210489a09141SPeter Zijlstra * 210589a09141SPeter Zijlstra * 64MB: 8192k 210689a09141SPeter Zijlstra * 128MB: 11585k 210789a09141SPeter Zijlstra * 256MB: 16384k 210889a09141SPeter Zijlstra * 512MB: 23170k 210989a09141SPeter Zijlstra * 1GB: 32768k 211089a09141SPeter Zijlstra * 2GB: 46340k 211189a09141SPeter Zijlstra * 4GB: 65536k 211289a09141SPeter Zijlstra * 8GB: 92681k 211389a09141SPeter Zijlstra * 16GB: 131072k 211489a09141SPeter Zijlstra * 211589a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 211689a09141SPeter Zijlstra * Limit the default to 256M 211789a09141SPeter Zijlstra */ 211889a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 211989a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 212089a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 212189a09141SPeter Zijlstra 21221da177e4SLinus Torvalds return 0; 21233dd4765fSJeff Layton 21243dd4765fSJeff Layton out_destroy_commit_cache: 21253dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21263dd4765fSJeff Layton out_destroy_write_mempool: 21273dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21283dd4765fSJeff Layton out_destroy_write_cache: 21293dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21303dd4765fSJeff Layton return -ENOMEM; 21311da177e4SLinus Torvalds } 21321da177e4SLinus Torvalds 2133266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21341da177e4SLinus Torvalds { 21351da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21363dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21371da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21381a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21391da177e4SLinus Torvalds } 21401da177e4SLinus Torvalds 21414a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21424a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21434a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21440eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21450eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21461ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21474a0de55cSAnna Schumaker }; 2148