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 105fbe77c30SBenjamin Coddington if (p) { 1063feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 107fbe77c30SBenjamin Coddington p->rw_mode = FMODE_WRITE; 108fbe77c30SBenjamin Coddington } 1093feb2d49STrond Myklebust return p; 1103feb2d49STrond Myklebust } 1113feb2d49STrond Myklebust 1121e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 1136c75dc0dSFred Isaman { 1141e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 1153feb2d49STrond Myklebust } 1161da177e4SLinus Torvalds 117919e3bd9STrond Myklebust static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) 118919e3bd9STrond Myklebust { 119919e3bd9STrond Myklebust return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); 120919e3bd9STrond Myklebust } 121919e3bd9STrond Myklebust 122919e3bd9STrond Myklebust static void nfs_io_completion_init(struct nfs_io_completion *ioc, 123919e3bd9STrond Myklebust void (*complete)(void *), void *data) 124919e3bd9STrond Myklebust { 125919e3bd9STrond Myklebust ioc->complete = complete; 126919e3bd9STrond Myklebust ioc->data = data; 127919e3bd9STrond Myklebust kref_init(&ioc->refcount); 128919e3bd9STrond Myklebust } 129919e3bd9STrond Myklebust 130919e3bd9STrond Myklebust static void nfs_io_completion_release(struct kref *kref) 131919e3bd9STrond Myklebust { 132919e3bd9STrond Myklebust struct nfs_io_completion *ioc = container_of(kref, 133919e3bd9STrond Myklebust struct nfs_io_completion, refcount); 134919e3bd9STrond Myklebust ioc->complete(ioc->data); 135919e3bd9STrond Myklebust kfree(ioc); 136919e3bd9STrond Myklebust } 137919e3bd9STrond Myklebust 138919e3bd9STrond Myklebust static void nfs_io_completion_get(struct nfs_io_completion *ioc) 139919e3bd9STrond Myklebust { 140919e3bd9STrond Myklebust if (ioc != NULL) 141919e3bd9STrond Myklebust kref_get(&ioc->refcount); 142919e3bd9STrond Myklebust } 143919e3bd9STrond Myklebust 144919e3bd9STrond Myklebust static void nfs_io_completion_put(struct nfs_io_completion *ioc) 145919e3bd9STrond Myklebust { 146919e3bd9STrond Myklebust if (ioc != NULL) 147919e3bd9STrond Myklebust kref_put(&ioc->refcount, nfs_io_completion_release); 148919e3bd9STrond Myklebust } 149919e3bd9STrond Myklebust 1507b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1517b159fc1STrond Myklebust { 1527b159fc1STrond Myklebust ctx->error = error; 1537b159fc1STrond Myklebust smp_wmb(); 1547b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1557b159fc1STrond Myklebust } 1567b159fc1STrond Myklebust 157bd37d6fcSTrond Myklebust static struct nfs_page * 158bd37d6fcSTrond Myklebust nfs_page_private_request(struct page *page) 159bd37d6fcSTrond Myklebust { 160bd37d6fcSTrond Myklebust if (!PagePrivate(page)) 161bd37d6fcSTrond Myklebust return NULL; 162bd37d6fcSTrond Myklebust return (struct nfs_page *)page_private(page); 163bd37d6fcSTrond Myklebust } 164bd37d6fcSTrond Myklebust 16584d3a9a9SWeston Andros Adamson /* 16684d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 16784d3a9a9SWeston Andros Adamson * 16884d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 16984d3a9a9SWeston Andros Adamson * 17084d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 17184d3a9a9SWeston Andros Adamson */ 17229418aa4SMel Gorman static struct nfs_page * 173b30d2f04STrond Myklebust nfs_page_find_private_request(struct page *page) 174277459d2STrond Myklebust { 175*4b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(page); 176bd37d6fcSTrond Myklebust struct nfs_page *req; 177277459d2STrond Myklebust 178b30d2f04STrond Myklebust if (!PagePrivate(page)) 179b30d2f04STrond Myklebust return NULL; 180*4b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 181bd37d6fcSTrond Myklebust req = nfs_page_private_request(page); 18284d3a9a9SWeston Andros Adamson if (req) { 18384d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 18429418aa4SMel Gorman kref_get(&req->wb_kref); 18584d3a9a9SWeston Andros Adamson } 186*4b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 187b30d2f04STrond Myklebust return req; 188b30d2f04STrond Myklebust } 18929418aa4SMel Gorman 190b30d2f04STrond Myklebust static struct nfs_page * 191b30d2f04STrond Myklebust nfs_page_find_swap_request(struct page *page) 192b30d2f04STrond Myklebust { 193b30d2f04STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 194b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 195b30d2f04STrond Myklebust struct nfs_page *req = NULL; 196b30d2f04STrond Myklebust if (!PageSwapCache(page)) 197b30d2f04STrond Myklebust return NULL; 198e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 199b30d2f04STrond Myklebust if (PageSwapCache(page)) { 200b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 201b30d2f04STrond Myklebust page); 202b30d2f04STrond Myklebust if (req) { 203b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 204b30d2f04STrond Myklebust kref_get(&req->wb_kref); 205b30d2f04STrond Myklebust } 206b30d2f04STrond Myklebust } 207e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 208277459d2STrond Myklebust return req; 209277459d2STrond Myklebust } 210277459d2STrond Myklebust 21184d3a9a9SWeston Andros Adamson /* 21284d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 21384d3a9a9SWeston Andros Adamson * 21484d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 21584d3a9a9SWeston Andros Adamson */ 21684d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 217277459d2STrond Myklebust { 218b30d2f04STrond Myklebust struct nfs_page *req; 219277459d2STrond Myklebust 220b30d2f04STrond Myklebust req = nfs_page_find_private_request(page); 221b30d2f04STrond Myklebust if (!req) 222b30d2f04STrond Myklebust req = nfs_page_find_swap_request(page); 223277459d2STrond Myklebust return req; 224277459d2STrond Myklebust } 225277459d2STrond Myklebust 2261da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2271da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2281da177e4SLinus Torvalds { 229d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 230a3d01454STrond Myklebust loff_t end, i_size; 231a3d01454STrond Myklebust pgoff_t end_index; 2321da177e4SLinus Torvalds 233a3d01454STrond Myklebust spin_lock(&inode->i_lock); 234a3d01454STrond Myklebust i_size = i_size_read(inode); 23509cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2368cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 237a3d01454STrond Myklebust goto out; 238d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2391da177e4SLinus Torvalds if (i_size >= end) 240a3d01454STrond Myklebust goto out; 2411da177e4SLinus Torvalds i_size_write(inode, end); 242a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 243a3d01454STrond Myklebust out: 244a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 247a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 248a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 249a301b777STrond Myklebust { 250d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 251a301b777STrond Myklebust } 252a301b777STrond Myklebust 253d72ddcbaSWeston Andros Adamson /* 254d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 255d72ddcbaSWeston Andros Adamson * @head - head request of page group 256d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 257d72ddcbaSWeston Andros Adamson * 258d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 259d72ddcbaSWeston Andros Adamson * page offset @page_offset. 260d72ddcbaSWeston Andros Adamson * 261d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 262d72ddcbaSWeston Andros Adamson * match is found. 263d72ddcbaSWeston Andros Adamson * 264d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 265d72ddcbaSWeston Andros Adamson */ 266d72ddcbaSWeston Andros Adamson static struct nfs_page * 267d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 268d72ddcbaSWeston Andros Adamson { 269d72ddcbaSWeston Andros Adamson struct nfs_page *req; 270d72ddcbaSWeston Andros Adamson 271d72ddcbaSWeston Andros Adamson req = head; 272d72ddcbaSWeston Andros Adamson do { 273d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 274d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 275d72ddcbaSWeston Andros Adamson return req; 276d72ddcbaSWeston Andros Adamson 277d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 278d72ddcbaSWeston Andros Adamson } while (req != head); 279d72ddcbaSWeston Andros Adamson 280d72ddcbaSWeston Andros Adamson return NULL; 281d72ddcbaSWeston Andros Adamson } 282d72ddcbaSWeston Andros Adamson 283d72ddcbaSWeston Andros Adamson /* 284d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 285d72ddcbaSWeston Andros Adamson * @head - head request of page group 286d72ddcbaSWeston Andros Adamson * 287d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 288d72ddcbaSWeston Andros Adamson * returns false otherwise 289d72ddcbaSWeston Andros Adamson */ 290d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 291d72ddcbaSWeston Andros Adamson { 292d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 293d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 294d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 295d72ddcbaSWeston Andros Adamson 2961344b7eaSTrond Myklebust nfs_page_group_lock(req); 297d72ddcbaSWeston Andros Adamson 2987e8a30f8STrond Myklebust for (;;) { 299d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 3007e8a30f8STrond Myklebust if (!tmp) 3017e8a30f8STrond Myklebust break; 3027e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 303d72ddcbaSWeston Andros Adamson } 304d72ddcbaSWeston Andros Adamson 305d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3067e8a30f8STrond Myklebust return pos >= len; 307d72ddcbaSWeston Andros Adamson } 308d72ddcbaSWeston Andros Adamson 3091da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3101da177e4SLinus Torvalds * covers the full page. 3111da177e4SLinus Torvalds */ 312d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3131da177e4SLinus Torvalds { 314d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 3151da177e4SLinus Torvalds return; 316d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3171da177e4SLinus Torvalds return; 318d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds 3211da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3221da177e4SLinus Torvalds { 323e87b4c7aSNeilBrown int ret = 0; 324cca588d6STrond Myklebust 325e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 326e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 327e87b4c7aSNeilBrown return ret; 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds 3301da177e4SLinus Torvalds /* 33189a09141SPeter Zijlstra * NFS congestion control 33289a09141SPeter Zijlstra */ 33389a09141SPeter Zijlstra 33489a09141SPeter Zijlstra int nfs_congestion_kb; 33589a09141SPeter Zijlstra 33689a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 33789a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 33889a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 33989a09141SPeter Zijlstra 340deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 34189a09141SPeter Zijlstra { 3420db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3430db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3445a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3455a6d41b3STrond Myklebust 346deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 34789a09141SPeter Zijlstra 348277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3490db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3500db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 35189a09141SPeter Zijlstra } 35289a09141SPeter Zijlstra 35320633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 35489a09141SPeter Zijlstra { 35520633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 35689a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 35731a01f09STrond Myklebust bool is_done; 35889a09141SPeter Zijlstra 35931a01f09STrond Myklebust is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); 36031a01f09STrond Myklebust nfs_unlock_request(req); 36131a01f09STrond Myklebust if (!is_done) 36220633f04SWeston Andros Adamson return; 36320633f04SWeston Andros Adamson 36420633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 365c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3660db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 36789a09141SPeter Zijlstra } 36889a09141SPeter Zijlstra 369d4581383SWeston Andros Adamson /* 370d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 371d4581383SWeston Andros Adamson * 372d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 373d4581383SWeston Andros Adamson * 374d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 375d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 376d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 377d4581383SWeston Andros Adamson * 378b5bab9bfSTrond Myklebust * NOTE: this must be called holding page_group bit lock 379b5bab9bfSTrond Myklebust * which will be released before returning. 380d4581383SWeston Andros Adamson * 381d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 382d4581383SWeston Andros Adamson */ 38374a6d4b5STrond Myklebust static void 38474a6d4b5STrond Myklebust nfs_unroll_locks(struct inode *inode, struct nfs_page *head, 3856d17d653STrond Myklebust struct nfs_page *req) 386d4581383SWeston Andros Adamson { 387d4581383SWeston Andros Adamson struct nfs_page *tmp; 388e261f51fSTrond Myklebust 389d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 3905b2b5187STrond Myklebust for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { 3915b2b5187STrond Myklebust if (!kref_read(&tmp->wb_kref)) 3925b2b5187STrond Myklebust continue; 3935b2b5187STrond Myklebust nfs_unlock_and_release_request(tmp); 3945b2b5187STrond Myklebust } 395e261f51fSTrond Myklebust } 396d4581383SWeston Andros Adamson 397d4581383SWeston Andros Adamson /* 398d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 399d4581383SWeston Andros Adamson * 400d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 401d4581383SWeston Andros Adamson * @old_head - the old head of the list 402d4581383SWeston Andros Adamson * 403d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 404d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 405d4581383SWeston Andros Adamson * with a reference held by some other context. 406d4581383SWeston Andros Adamson */ 407d4581383SWeston Andros Adamson static void 408d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 409b66aaa8dSTrond Myklebust struct nfs_page *old_head, 410b66aaa8dSTrond Myklebust struct inode *inode) 411d4581383SWeston Andros Adamson { 412d4581383SWeston Andros Adamson while (destroy_list) { 413d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 414d4581383SWeston Andros Adamson 415d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 416d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 417d4581383SWeston Andros Adamson 418d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 419d4581383SWeston Andros Adamson 420d4581383SWeston Andros Adamson /* make sure old group is not used */ 421d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 422d4581383SWeston Andros Adamson 423902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 424902a4c00STrond Myklebust 4255b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4265b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4275b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 428902a4c00STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) 4295b2b5187STrond Myklebust nfs_free_request(subreq); 4305b2b5187STrond Myklebust continue; 4315b2b5187STrond Myklebust } 432d4581383SWeston Andros Adamson 4335b2b5187STrond Myklebust subreq->wb_head = subreq; 4345b2b5187STrond Myklebust 435b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 436d4581383SWeston Andros Adamson nfs_release_request(subreq); 437a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 438d4581383SWeston Andros Adamson } 4395b2b5187STrond Myklebust 4405b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4415b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4425b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 443d4581383SWeston Andros Adamson } 444d4581383SWeston Andros Adamson } 445d4581383SWeston Andros Adamson 446d4581383SWeston Andros Adamson /* 447d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 448d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 449d4581383SWeston Andros Adamson * operations for this page. 450d4581383SWeston Andros Adamson * 451d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 452d4581383SWeston Andros Adamson * 453d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 454d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 455d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 456d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 457d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 458d4581383SWeston Andros Adamson * 459d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 460d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 461d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 462d4581383SWeston Andros Adamson * error was encountered. 463d4581383SWeston Andros Adamson */ 464d4581383SWeston Andros Adamson static struct nfs_page * 4656d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 466d4581383SWeston Andros Adamson { 467d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 468d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 469d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 470d4581383SWeston Andros Adamson unsigned int total_bytes; 471d4581383SWeston Andros Adamson int ret; 472d4581383SWeston Andros Adamson 473d4581383SWeston Andros Adamson try_again: 474d4581383SWeston Andros Adamson /* 475d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 476d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 477d4581383SWeston Andros Adamson * until the head reference is released. 478d4581383SWeston Andros Adamson */ 479bd37d6fcSTrond Myklebust head = nfs_page_find_head_request(page); 480bd37d6fcSTrond Myklebust if (!head) 481d4581383SWeston Andros Adamson return NULL; 482d4581383SWeston Andros Adamson 483a0e265bcSTrond Myklebust /* lock the page head first in order to avoid an ABBA inefficiency */ 484a0e265bcSTrond Myklebust if (!nfs_lock_request(head)) { 485a0e265bcSTrond Myklebust ret = nfs_wait_on_request(head); 486a0e265bcSTrond Myklebust nfs_release_request(head); 487a0e265bcSTrond Myklebust if (ret < 0) 488a0e265bcSTrond Myklebust return ERR_PTR(ret); 489a0e265bcSTrond Myklebust goto try_again; 490a0e265bcSTrond Myklebust } 491bd37d6fcSTrond Myklebust 492bd37d6fcSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 493bd37d6fcSTrond Myklebust if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { 494bd37d6fcSTrond Myklebust nfs_unlock_and_release_request(head); 495bd37d6fcSTrond Myklebust goto try_again; 496bd37d6fcSTrond Myklebust } 4977c3af975SWeston Andros Adamson 4981344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 499b5bab9bfSTrond Myklebust if (ret < 0) { 500a0e265bcSTrond Myklebust nfs_unlock_and_release_request(head); 501b5bab9bfSTrond Myklebust return ERR_PTR(ret); 5027c3af975SWeston Andros Adamson } 5037c3af975SWeston Andros Adamson 5047c3af975SWeston Andros Adamson /* lock each request in the page group */ 505a0e265bcSTrond Myklebust total_bytes = head->wb_bytes; 506a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 507a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 50874a6d4b5STrond Myklebust 5095b2b5187STrond Myklebust if (!kref_get_unless_zero(&subreq->wb_kref)) 5105b2b5187STrond Myklebust continue; 51174a6d4b5STrond Myklebust while (!nfs_lock_request(subreq)) { 512b5bab9bfSTrond Myklebust /* 51374a6d4b5STrond Myklebust * Unlock page to allow nfs_page_group_sync_on_bit() 51474a6d4b5STrond Myklebust * to succeed 515b5bab9bfSTrond Myklebust */ 51674a6d4b5STrond Myklebust nfs_page_group_unlock(head); 51774a6d4b5STrond Myklebust ret = nfs_wait_on_request(subreq); 51874a6d4b5STrond Myklebust if (!ret) 5191344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 52074a6d4b5STrond Myklebust if (ret < 0) { 52174a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5225b2b5187STrond Myklebust nfs_release_request(subreq); 52374a6d4b5STrond Myklebust nfs_unlock_and_release_request(head); 524d4581383SWeston Andros Adamson return ERR_PTR(ret); 525d4581383SWeston Andros Adamson } 52674a6d4b5STrond Myklebust } 527e14bebf6STrond Myklebust /* 528e14bebf6STrond Myklebust * Subrequests are always contiguous, non overlapping 529e14bebf6STrond Myklebust * and in order - but may be repeated (mirrored writes). 530e14bebf6STrond Myklebust */ 531e14bebf6STrond Myklebust if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 532e14bebf6STrond Myklebust /* keep track of how many bytes this group covers */ 533e14bebf6STrond Myklebust total_bytes += subreq->wb_bytes; 534e14bebf6STrond Myklebust } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 535e14bebf6STrond Myklebust ((subreq->wb_offset + subreq->wb_bytes) > 536e14bebf6STrond Myklebust (head->wb_offset + total_bytes)))) { 53774a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5385b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 53974a6d4b5STrond Myklebust nfs_page_group_unlock(head); 54074a6d4b5STrond Myklebust nfs_unlock_and_release_request(head); 541e14bebf6STrond Myklebust return ERR_PTR(-EIO); 542e14bebf6STrond Myklebust } 543a0e265bcSTrond Myklebust } 544d4581383SWeston Andros Adamson 545d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 546d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 547d4581383SWeston Andros Adamson subreq = head; 548d4581383SWeston Andros Adamson do { 549411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 550d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 551d4581383SWeston Andros Adamson } while (subreq != head); 552d4581383SWeston Andros Adamson 553d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 554d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 555d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 556d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 557d4581383SWeston Andros Adamson head->wb_this_page = head; 558d4581383SWeston Andros Adamson 559d4581383SWeston Andros Adamson /* change head request to cover whole range that 560d4581383SWeston Andros Adamson * the former page group covered */ 561d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 562d4581383SWeston Andros Adamson } 563d4581383SWeston Andros Adamson 564b66aaa8dSTrond Myklebust /* Postpone destruction of this request */ 565b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { 566b66aaa8dSTrond Myklebust set_bit(PG_INODE_REF, &head->wb_flags); 567b66aaa8dSTrond Myklebust kref_get(&head->wb_kref); 568a6b6d5b8STrond Myklebust atomic_long_inc(&NFS_I(inode)->nrequests); 569b66aaa8dSTrond Myklebust } 570b66aaa8dSTrond Myklebust 571d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 572d4581383SWeston Andros Adamson 573b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 574d4581383SWeston Andros Adamson 575b5bab9bfSTrond Myklebust /* Did we lose a race with nfs_inode_remove_request()? */ 576b5bab9bfSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) { 577b5bab9bfSTrond Myklebust nfs_unlock_and_release_request(head); 578b5bab9bfSTrond Myklebust return NULL; 579b5bab9bfSTrond Myklebust } 580b5bab9bfSTrond Myklebust 581bd37d6fcSTrond Myklebust /* still holds ref on head from nfs_page_find_head_request 582d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 583d4581383SWeston Andros Adamson return head; 584612c9384STrond Myklebust } 585074cc1deSTrond Myklebust 5860bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5870bcbf039SPeng Tao { 5880bcbf039SPeng Tao nfs_end_page_writeback(req); 5890bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5900bcbf039SPeng Tao req->wb_page); 5911f84ccdfSFred Isaman nfs_release_request(req); 5920bcbf039SPeng Tao } 5930bcbf039SPeng Tao 594a6598813STrond Myklebust static bool 595a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 596a6598813STrond Myklebust { 597a6598813STrond Myklebust switch (err) { 598a6598813STrond Myklebust case 0: 599a6598813STrond Myklebust case -ERESTARTSYS: 600a6598813STrond Myklebust case -EINTR: 601a6598813STrond Myklebust return false; 602a6598813STrond Myklebust } 603a6598813STrond Myklebust return nfs_error_is_fatal(err); 604e261f51fSTrond Myklebust } 605074cc1deSTrond Myklebust 606074cc1deSTrond Myklebust /* 607074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 608074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 609074cc1deSTrond Myklebust */ 610074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 6116d17d653STrond Myklebust struct page *page) 612074cc1deSTrond Myklebust { 613074cc1deSTrond Myklebust struct nfs_page *req; 614074cc1deSTrond Myklebust int ret = 0; 615074cc1deSTrond Myklebust 6166d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 617074cc1deSTrond Myklebust if (!req) 618074cc1deSTrond Myklebust goto out; 619074cc1deSTrond Myklebust ret = PTR_ERR(req); 620074cc1deSTrond Myklebust if (IS_ERR(req)) 621074cc1deSTrond Myklebust goto out; 622074cc1deSTrond Myklebust 623deed85e7STrond Myklebust nfs_set_page_writeback(page); 624deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 625074cc1deSTrond Myklebust 626deed85e7STrond Myklebust ret = 0; 627a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 628a6598813STrond Myklebust if (nfs_error_is_fatal_on_server(req->wb_context->error)) 629a6598813STrond Myklebust goto out_launder; 630a6598813STrond Myklebust 631f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 632074cc1deSTrond Myklebust ret = pgio->pg_error; 6330bcbf039SPeng Tao /* 634c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6350bcbf039SPeng Tao */ 6360bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 6370bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 638c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 639a6598813STrond Myklebust goto out_launder; 640d6c843b9SPeng Tao } 6410bcbf039SPeng Tao nfs_redirty_request(req); 6420bcbf039SPeng Tao ret = -EAGAIN; 64340f90271STrond Myklebust } else 64440f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 64540f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 646074cc1deSTrond Myklebust out: 647074cc1deSTrond Myklebust return ret; 648a6598813STrond Myklebust out_launder: 649a6598813STrond Myklebust nfs_write_error_remove_page(req); 650a6598813STrond Myklebust return ret; 651e261f51fSTrond Myklebust } 652e261f51fSTrond Myklebust 653d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 654c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 655f758c885STrond Myklebust { 656cfb506e1STrond Myklebust int ret; 657f758c885STrond Myklebust 6588cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 6596d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 660cfb506e1STrond Myklebust if (ret == -EAGAIN) { 661cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 662cfb506e1STrond Myklebust ret = 0; 663cfb506e1STrond Myklebust } 664cfb506e1STrond Myklebust return ret; 665f758c885STrond Myklebust } 666f758c885STrond Myklebust 667e261f51fSTrond Myklebust /* 6681da177e4SLinus Torvalds * Write an mmapped page to the server. 6691da177e4SLinus Torvalds */ 670d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 671c373fff7STrond Myklebust struct writeback_control *wbc) 6721da177e4SLinus Torvalds { 673f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 67440f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 675e261f51fSTrond Myklebust int err; 6761da177e4SLinus Torvalds 67740f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 678811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 679a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 680c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 681f758c885STrond Myklebust nfs_pageio_complete(&pgio); 682f758c885STrond Myklebust if (err < 0) 6834d770ccfSTrond Myklebust return err; 684f758c885STrond Myklebust if (pgio.pg_error < 0) 685f758c885STrond Myklebust return pgio.pg_error; 686f758c885STrond Myklebust return 0; 6874d770ccfSTrond Myklebust } 6884d770ccfSTrond Myklebust 6894d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6904d770ccfSTrond Myklebust { 691f758c885STrond Myklebust int ret; 6924d770ccfSTrond Myklebust 693c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6941da177e4SLinus Torvalds unlock_page(page); 695f758c885STrond Myklebust return ret; 696f758c885STrond Myklebust } 697f758c885STrond Myklebust 698f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 699f758c885STrond Myklebust { 700f758c885STrond Myklebust int ret; 701f758c885STrond Myklebust 702c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 703f758c885STrond Myklebust unlock_page(page); 704f758c885STrond Myklebust return ret; 7051da177e4SLinus Torvalds } 7061da177e4SLinus Torvalds 707919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 708919e3bd9STrond Myklebust { 709919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 710919e3bd9STrond Myklebust } 711919e3bd9STrond Myklebust 7121da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7131da177e4SLinus Torvalds { 7141da177e4SLinus Torvalds struct inode *inode = mapping->host; 715c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 716919e3bd9STrond Myklebust struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); 7171da177e4SLinus Torvalds int err; 7181da177e4SLinus Torvalds 71991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 72091d5b470SChuck Lever 721919e3bd9STrond Myklebust if (ioc) 722919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 723919e3bd9STrond Myklebust 724a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 725a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 726919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 727f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 728c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 729919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 73072cb77f4STrond Myklebust 731f758c885STrond Myklebust if (err < 0) 73272cb77f4STrond Myklebust goto out_err; 73372cb77f4STrond Myklebust err = pgio.pg_error; 73472cb77f4STrond Myklebust if (err < 0) 73572cb77f4STrond Myklebust goto out_err; 736c63c7b05STrond Myklebust return 0; 73772cb77f4STrond Myklebust out_err: 73872cb77f4STrond Myklebust return err; 7391da177e4SLinus Torvalds } 7401da177e4SLinus Torvalds 7411da177e4SLinus Torvalds /* 7421da177e4SLinus Torvalds * Insert a write request into an inode 7431da177e4SLinus Torvalds */ 744d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7451da177e4SLinus Torvalds { 746*4b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7471da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 748e7d39069STrond Myklebust 7492bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7502bfc6e56SWeston Andros Adamson 751e7d39069STrond Myklebust /* Lock the request! */ 7527ad84aa9STrond Myklebust nfs_lock_request(req); 753e7d39069STrond Myklebust 75429418aa4SMel Gorman /* 75529418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 75629418aa4SMel Gorman * with invalidate/truncate. 75729418aa4SMel Gorman */ 758*4b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 759*4b9bb25bSTrond Myklebust if (!nfs_have_writebacks(inode) && 760*4b9bb25bSTrond Myklebust NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) { 761*4b9bb25bSTrond Myklebust spin_lock(&inode->i_lock); 762*4b9bb25bSTrond Myklebust inode->i_version++; 763*4b9bb25bSTrond Myklebust spin_unlock(&inode->i_lock); 764*4b9bb25bSTrond Myklebust } 76529418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7662df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 767deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 768277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 76929418aa4SMel Gorman } 770*4b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 771a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 77217089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 773cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 774cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 775cb1410c7SWeston Andros Adamson * adding subrequests. */ 77617089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 777c03b4024STrond Myklebust kref_get(&req->wb_kref); 7781da177e4SLinus Torvalds } 7791da177e4SLinus Torvalds 7801da177e4SLinus Torvalds /* 78189a09141SPeter Zijlstra * Remove a write request from an inode 7821da177e4SLinus Torvalds */ 7831da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7841da177e4SLinus Torvalds { 785*4b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 786*4b9bb25bSTrond Myklebust struct inode *inode = mapping->host; 7871da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 78820633f04SWeston Andros Adamson struct nfs_page *head; 78920633f04SWeston Andros Adamson 790a6b6d5b8STrond Myklebust atomic_long_dec(&nfsi->nrequests); 79120633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 79220633f04SWeston Andros Adamson head = req->wb_head; 7931da177e4SLinus Torvalds 794*4b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 79567911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 79620633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 79720633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 79820633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 79929418aa4SMel Gorman } 800*4b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 80120633f04SWeston Andros Adamson } 80217089a29SWeston Andros Adamson 80317089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 8041da177e4SLinus Torvalds nfs_release_request(req); 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 80761822ab5STrond Myklebust static void 8086d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 80961822ab5STrond Myklebust { 81067911c8fSAnna Schumaker if (req->wb_page) 81161822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 81261822ab5STrond Myklebust } 81361822ab5STrond Myklebust 8143a3908c8STrond Myklebust /* 8153a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8163a3908c8STrond Myklebust * 8173a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8183a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8193a3908c8STrond Myklebust * 8203a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8213a3908c8STrond Myklebust */ 8223a3908c8STrond Myklebust static struct nfs_page * 8233a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8243a3908c8STrond Myklebust struct page *page) 8253a3908c8STrond Myklebust { 8263a3908c8STrond Myklebust struct nfs_page *freq, *t; 8273a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8283a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8293a3908c8STrond Myklebust 8303a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8313a3908c8STrond Myklebust 8323a3908c8STrond Myklebust /* search through pnfs commit lists */ 8333a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8343a3908c8STrond Myklebust if (freq) 8353a3908c8STrond Myklebust return freq->wb_head; 8363a3908c8STrond Myklebust 8373a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8383a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8393a3908c8STrond Myklebust if (freq->wb_page == page) 8403a3908c8STrond Myklebust return freq->wb_head; 8413a3908c8STrond Myklebust } 8423a3908c8STrond Myklebust 8433a3908c8STrond Myklebust return NULL; 8443a3908c8STrond Myklebust } 8453a3908c8STrond Myklebust 8468dd37758STrond Myklebust /** 84786d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 84886d80f97STrond Myklebust * @req: pointer to a struct nfs_page 84986d80f97STrond Myklebust * @dst: commit list head 85086d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 85186d80f97STrond Myklebust * 85286d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 85386d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 85486d80f97STrond Myklebust * the MM page stats. 85586d80f97STrond Myklebust * 856e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 857e824f99aSTrond Myklebust * nfs_page lock. 85886d80f97STrond Myklebust */ 85986d80f97STrond Myklebust void 86086d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 86186d80f97STrond Myklebust struct nfs_commit_info *cinfo) 86286d80f97STrond Myklebust { 86386d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 86486d80f97STrond Myklebust nfs_list_add_request(req, dst); 8655cb953d4STrond Myklebust atomic_long_inc(&cinfo->mds->ncommit); 86686d80f97STrond Myklebust } 86786d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 86886d80f97STrond Myklebust 86986d80f97STrond Myklebust /** 8708dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8718dd37758STrond Myklebust * @req: pointer to a struct nfs_page 872ea2cf228SFred Isaman * @dst: commit list head 873ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8748dd37758STrond Myklebust * 875ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8768dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8778dd37758STrond Myklebust * the MM page stats. 8788dd37758STrond Myklebust * 879ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8808dd37758STrond Myklebust * holding the nfs_page lock. 8818dd37758STrond Myklebust */ 8828dd37758STrond Myklebust void 8836272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8848dd37758STrond Myklebust { 885e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8866272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 887e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 88867911c8fSAnna Schumaker if (req->wb_page) 88986d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8908dd37758STrond Myklebust } 8918dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8928dd37758STrond Myklebust 8938dd37758STrond Myklebust /** 8948dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8958dd37758STrond Myklebust * @req: pointer to a nfs_page 896ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8978dd37758STrond Myklebust * 898ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8998dd37758STrond Myklebust * number of outstanding requests requiring a commit 9008dd37758STrond Myklebust * It does not update the MM page stats. 9018dd37758STrond Myklebust * 902ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9038dd37758STrond Myklebust */ 9048dd37758STrond Myklebust void 905ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 906ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9078dd37758STrond Myklebust { 9088dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9098dd37758STrond Myklebust return; 9108dd37758STrond Myklebust nfs_list_remove_request(req); 9115cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9128dd37758STrond Myklebust } 9138dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9148dd37758STrond Myklebust 915ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 916ea2cf228SFred Isaman struct inode *inode) 917ea2cf228SFred Isaman { 918fe238e60SDave Wysochanski cinfo->inode = inode; 919ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 920ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 921b359f9d0SFred Isaman cinfo->dreq = NULL; 922f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 923ea2cf228SFred Isaman } 924ea2cf228SFred Isaman 925ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 926ea2cf228SFred Isaman struct inode *inode, 927ea2cf228SFred Isaman struct nfs_direct_req *dreq) 928ea2cf228SFred Isaman { 9291763da12SFred Isaman if (dreq) 9301763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9311763da12SFred Isaman else 932ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 933ea2cf228SFred Isaman } 934ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9358dd37758STrond Myklebust 9361da177e4SLinus Torvalds /* 9371da177e4SLinus Torvalds * Add a request to the inode's commit list. 9381da177e4SLinus Torvalds */ 9391763da12SFred Isaman void 940ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 941b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9421da177e4SLinus Torvalds { 943b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9448dd37758STrond Myklebust return; 9456272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9461da177e4SLinus Torvalds } 9478e821cadSTrond Myklebust 948d6d6dc7cSFred Isaman static void 949d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 950e468bae9STrond Myklebust { 95111fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 95293f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 95393f78d88STejun Heo WB_RECLAIMABLE); 954e468bae9STrond Myklebust } 955d6d6dc7cSFred Isaman 956b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 9578dd37758STrond Myklebust static void 958d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 959d6d6dc7cSFred Isaman { 9608dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9612b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 962ea2cf228SFred Isaman struct nfs_commit_info cinfo; 963d6d6dc7cSFred Isaman 964ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 965e824f99aSTrond Myklebust mutex_lock(&NFS_I(inode)->commit_mutex); 966ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 967ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 968d6d6dc7cSFred Isaman } 969e824f99aSTrond Myklebust mutex_unlock(&NFS_I(inode)->commit_mutex); 9708dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9718dd37758STrond Myklebust } 972e468bae9STrond Myklebust } 973e468bae9STrond Myklebust 974d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9758e821cadSTrond Myklebust { 976c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 977d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 978c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9798e821cadSTrond Myklebust } 9808e821cadSTrond Myklebust 981919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 982919e3bd9STrond Myklebust { 983919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 984919e3bd9STrond Myklebust } 985919e3bd9STrond Myklebust 986061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9876c75dc0dSFred Isaman { 988ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9896c75dc0dSFred Isaman unsigned long bytes = 0; 9906c75dc0dSFred Isaman 9916c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9926c75dc0dSFred Isaman goto out; 993ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9946c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9956c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9966c75dc0dSFred Isaman 9976c75dc0dSFred Isaman bytes += req->wb_bytes; 9986c75dc0dSFred Isaman nfs_list_remove_request(req); 9996c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 10006c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 1001d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 10026c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 10036c75dc0dSFred Isaman goto remove_req; 10046c75dc0dSFred Isaman } 1005c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 1006f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1007b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1008a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10096c75dc0dSFred Isaman goto next; 10106c75dc0dSFred Isaman } 10116c75dc0dSFred Isaman remove_req: 10126c75dc0dSFred Isaman nfs_inode_remove_request(req); 10136c75dc0dSFred Isaman next: 101420633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10153aff4ebbSTrond Myklebust nfs_release_request(req); 10166c75dc0dSFred Isaman } 10176c75dc0dSFred Isaman out: 1018919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10196c75dc0dSFred Isaman hdr->release(hdr); 10206c75dc0dSFred Isaman } 10216c75dc0dSFred Isaman 1022ce59515cSAnna Schumaker unsigned long 1023ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1024fb8a1f11STrond Myklebust { 10255cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1026fb8a1f11STrond Myklebust } 1027fb8a1f11STrond Myklebust 1028e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10291763da12SFred Isaman int 1030ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1031ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1032d6d6dc7cSFred Isaman { 1033d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 1034d6d6dc7cSFred Isaman int ret = 0; 1035d6d6dc7cSFred Isaman 1036d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 10378dd37758STrond Myklebust if (!nfs_lock_request(req)) 10388dd37758STrond Myklebust continue; 10397ad84aa9STrond Myklebust kref_get(&req->wb_kref); 1040ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10418dd37758STrond Myklebust nfs_list_add_request(req, dst); 1042d6d6dc7cSFred Isaman ret++; 10431763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1044d6d6dc7cSFred Isaman break; 1045e824f99aSTrond Myklebust cond_resched(); 1046d6d6dc7cSFred Isaman } 1047d6d6dc7cSFred Isaman return ret; 1048d6d6dc7cSFred Isaman } 1049d6d6dc7cSFred Isaman 10501da177e4SLinus Torvalds /* 10511da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10521da177e4SLinus Torvalds * @inode: NFS inode to scan 1053ea2cf228SFred Isaman * @dst: mds destination list 1054ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10551da177e4SLinus Torvalds * 10561da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10571da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10581da177e4SLinus Torvalds */ 10591763da12SFred Isaman int 1060ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1061ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10621da177e4SLinus Torvalds { 1063d6d6dc7cSFred Isaman int ret = 0; 1064fb8a1f11STrond Myklebust 10655cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10665cb953d4STrond Myklebust return 0; 1067e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10685cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10698dd37758STrond Myklebust const int max = INT_MAX; 1070d6d6dc7cSFred Isaman 1071ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1072ea2cf228SFred Isaman cinfo, max); 1073ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1074d6d6dc7cSFred Isaman } 1075e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1076ff778d02STrond Myklebust return ret; 10771da177e4SLinus Torvalds } 1078d6d6dc7cSFred Isaman 10791da177e4SLinus Torvalds /* 1080e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1081e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10821da177e4SLinus Torvalds * 1083e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1084e7d39069STrond Myklebust * to disk. 10851da177e4SLinus Torvalds */ 1086e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1087e7d39069STrond Myklebust struct page *page, 1088e7d39069STrond Myklebust unsigned int offset, 1089e7d39069STrond Myklebust unsigned int bytes) 10901da177e4SLinus Torvalds { 1091e7d39069STrond Myklebust struct nfs_page *req; 1092e7d39069STrond Myklebust unsigned int rqend; 1093e7d39069STrond Myklebust unsigned int end; 10941da177e4SLinus Torvalds int error; 1095277459d2STrond Myklebust 1096e7d39069STrond Myklebust end = offset + bytes; 1097e7d39069STrond Myklebust 1098f6032f21STrond Myklebust req = nfs_lock_and_join_requests(page); 1099f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1100f6032f21STrond Myklebust return req; 11012bfc6e56SWeston Andros Adamson 1102e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1103e7d39069STrond Myklebust /* 1104e7d39069STrond Myklebust * Tell the caller to flush out the request if 1105e7d39069STrond Myklebust * the offsets are non-contiguous. 1106e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1107e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1108e7d39069STrond Myklebust */ 1109f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1110e7d39069STrond Myklebust goto out_flushme; 1111e7d39069STrond Myklebust 11121da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11131da177e4SLinus Torvalds if (offset < req->wb_offset) { 11141da177e4SLinus Torvalds req->wb_offset = offset; 11151da177e4SLinus Torvalds req->wb_pgbase = offset; 11161da177e4SLinus Torvalds } 11171da177e4SLinus Torvalds if (end > rqend) 11181da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1119e7d39069STrond Myklebust else 1120e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1121e7d39069STrond Myklebust return req; 1122e7d39069STrond Myklebust out_flushme: 1123f6032f21STrond Myklebust /* 1124f6032f21STrond Myklebust * Note: we mark the request dirty here because 1125f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1126f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1127f6032f21STrond Myklebust */ 1128f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1129f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 1130e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1131f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1132e7d39069STrond Myklebust } 11331da177e4SLinus Torvalds 1134e7d39069STrond Myklebust /* 1135e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1136e7d39069STrond Myklebust * 1137e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1138e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1139e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1140e7d39069STrond Myklebust */ 1141e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1142e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1143e7d39069STrond Myklebust { 1144d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1145e7d39069STrond Myklebust struct nfs_page *req; 1146e7d39069STrond Myklebust 1147e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1148e7d39069STrond Myklebust if (req != NULL) 1149e7d39069STrond Myklebust goto out; 11502bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1151e7d39069STrond Myklebust if (IS_ERR(req)) 1152e7d39069STrond Myklebust goto out; 1153d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1154efc91ed0STrond Myklebust out: 115561e930a9STrond Myklebust return req; 11561da177e4SLinus Torvalds } 11571da177e4SLinus Torvalds 1158e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1159e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1160e7d39069STrond Myklebust { 1161e7d39069STrond Myklebust struct nfs_page *req; 1162e7d39069STrond Myklebust 1163e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1164e7d39069STrond Myklebust if (IS_ERR(req)) 1165e7d39069STrond Myklebust return PTR_ERR(req); 1166e7d39069STrond Myklebust /* Update file length */ 1167e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1168d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1169a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11701d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1171e7d39069STrond Myklebust return 0; 1172e7d39069STrond Myklebust } 1173e7d39069STrond Myklebust 11741da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11751da177e4SLinus Torvalds { 1176cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11772a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1178bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11791da177e4SLinus Torvalds struct nfs_page *req; 11801a54533eSTrond Myklebust int do_flush, status; 11811da177e4SLinus Torvalds /* 11821da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11831da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11841da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11851da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11861da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11871da177e4SLinus Torvalds * dropped page. 11881da177e4SLinus Torvalds */ 11891a54533eSTrond Myklebust do { 119084d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11911a54533eSTrond Myklebust if (req == NULL) 11921a54533eSTrond Myklebust return 0; 11932a369153STrond Myklebust l_ctx = req->wb_lock_context; 1194138a2935STrond Myklebust do_flush = req->wb_page != page || 1195138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 1196bd61e0a9SJeff Layton if (l_ctx && flctx && 1197bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1198bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1199d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12005263e31eSJeff Layton } 12011da177e4SLinus Torvalds nfs_release_request(req); 12021a54533eSTrond Myklebust if (!do_flush) 12031a54533eSTrond Myklebust return 0; 1204d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12051a54533eSTrond Myklebust } while (status == 0); 12061a54533eSTrond Myklebust return status; 12071da177e4SLinus Torvalds } 12081da177e4SLinus Torvalds 12091da177e4SLinus Torvalds /* 1210dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1211dc24826bSAndy Adamson * expire soon. 1212dc24826bSAndy Adamson * 1213dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1214dc24826bSAndy Adamson * 1215dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1216dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1217dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1218dc24826bSAndy Adamson */ 1219dc24826bSAndy Adamson int 1220dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1221dc24826bSAndy Adamson { 1222dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1223dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1224dc24826bSAndy Adamson 1225dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1226dc24826bSAndy Adamson } 1227dc24826bSAndy Adamson 1228dc24826bSAndy Adamson /* 1229dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1230dc24826bSAndy Adamson */ 1231ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1232dc24826bSAndy Adamson { 1233ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1234ce52914eSScott Mayhew 1235ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1236dc24826bSAndy Adamson } 1237dc24826bSAndy Adamson 1238dc24826bSAndy Adamson /* 12395d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12405d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12415d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12425d47a356STrond Myklebust */ 12438d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12445d47a356STrond Myklebust { 1245d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1246d529ef83SJeff Layton 12478d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12488d197a56STrond Myklebust goto out; 124918dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1250d529ef83SJeff Layton return false; 12514db72b40SJeff Layton smp_rmb(); 1252d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12538d197a56STrond Myklebust return false; 12548d197a56STrond Myklebust out: 125518dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 125618dd78c4SScott Mayhew return false; 12578d197a56STrond Myklebust return PageUptodate(page) != 0; 12585d47a356STrond Myklebust } 12595d47a356STrond Myklebust 12605263e31eSJeff Layton static bool 12615263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12625263e31eSJeff Layton { 12635263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12645263e31eSJeff Layton fl->fl_type == F_WRLCK; 12655263e31eSJeff Layton } 12665263e31eSJeff Layton 1267c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1268c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1269c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1270c7559663SScott Mayhew * inefficiencies. 1271c7559663SScott Mayhew * 1272263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1273263b4509SScott Mayhew * of the checks. 1274c7559663SScott Mayhew */ 1275c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1276c7559663SScott Mayhew { 12775263e31eSJeff Layton int ret; 12785263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12795263e31eSJeff Layton struct file_lock *fl; 12805263e31eSJeff Layton 1281c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1282c7559663SScott Mayhew return 0; 1283263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1284263b4509SScott Mayhew return 0; 1285c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1286c7559663SScott Mayhew return 1; 1287bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1288bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 12898fa4592aSTrond Myklebust return 1; 12905263e31eSJeff Layton 12915263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12925263e31eSJeff Layton ret = 0; 12936109c850SJeff Layton spin_lock(&flctx->flc_lock); 1294bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1295bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1296bd61e0a9SJeff Layton fl_list); 1297bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12985263e31eSJeff Layton ret = 1; 1299bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13005263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13015263e31eSJeff Layton fl_list); 13025263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13035263e31eSJeff Layton ret = 1; 13045263e31eSJeff Layton } 13056109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13065263e31eSJeff Layton return ret; 1307c7559663SScott Mayhew } 1308c7559663SScott Mayhew 13095d47a356STrond Myklebust /* 13101da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13111da177e4SLinus Torvalds * 13121da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13131da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13141da177e4SLinus Torvalds */ 13151da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13161da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13171da177e4SLinus Torvalds { 1318cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1319d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 13201da177e4SLinus Torvalds int status = 0; 13211da177e4SLinus Torvalds 132291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 132391d5b470SChuck Lever 13246de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13256de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13261da177e4SLinus Torvalds 1327149a4fddSBenjamin Coddington if (!count) 1328149a4fddSBenjamin Coddington goto out; 1329149a4fddSBenjamin Coddington 1330c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 133149a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13321da177e4SLinus Torvalds offset = 0; 13331da177e4SLinus Torvalds } 13341da177e4SLinus Torvalds 1335e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 133603fa9e84STrond Myklebust if (status < 0) 133703fa9e84STrond Myklebust nfs_set_pageerror(page); 133859b7c05fSTrond Myklebust else 133959b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1340149a4fddSBenjamin Coddington out: 134148186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13421da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13431da177e4SLinus Torvalds return status; 13441da177e4SLinus Torvalds } 13451da177e4SLinus Torvalds 13463ff7576dSTrond Myklebust static int flush_task_priority(int how) 13471da177e4SLinus Torvalds { 13481da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13491da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13501da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13511da177e4SLinus Torvalds case FLUSH_LOWPRI: 13521da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13531da177e4SLinus Torvalds } 13541da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds 1357d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1358d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1359abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13601ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13611da177e4SLinus Torvalds { 13623ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13631da177e4SLinus Torvalds 13641ed26f33SAnna Schumaker task_setup_data->priority = priority; 1365abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1366d138d5d1SAndy Adamson 1367abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1368d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1369275acaafSTrond Myklebust } 1370275acaafSTrond Myklebust 13716d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13726d884e8fSFred * call this on each, which will prepare them to be retried on next 13736d884e8fSFred * writeback using standard nfs. 13746d884e8fSFred */ 13756d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13766d884e8fSFred { 13776d884e8fSFred nfs_mark_request_dirty(req); 1378c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 137920633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13803aff4ebbSTrond Myklebust nfs_release_request(req); 13816d884e8fSFred } 13826d884e8fSFred 1383061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13846c75dc0dSFred Isaman { 13856c75dc0dSFred Isaman struct nfs_page *req; 13866c75dc0dSFred Isaman 13876c75dc0dSFred Isaman while (!list_empty(head)) { 13886c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13896c75dc0dSFred Isaman nfs_list_remove_request(req); 13906c75dc0dSFred Isaman nfs_redirty_request(req); 13916c75dc0dSFred Isaman } 13926c75dc0dSFred Isaman } 13936c75dc0dSFred Isaman 1394dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1395dc602dd7STrond Myklebust { 1396dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1397dc602dd7STrond Myklebust } 1398dc602dd7STrond Myklebust 1399061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1400919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1401061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1402061ae2edSFred Isaman .completion = nfs_write_completion, 1403dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1404061ae2edSFred Isaman }; 1405061ae2edSFred Isaman 140657208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1407a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1408061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14091751c363STrond Myklebust { 1410a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 141141d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1412a20c93e3SChristoph Hellwig 1413a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1414a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1415a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1416a20c93e3SChristoph Hellwig #endif 14174a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 1418fbe77c30SBenjamin Coddington server->wsize, ioflags, GFP_NOIO); 14191751c363STrond Myklebust } 1420ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14211751c363STrond Myklebust 1422dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1423dce81290STrond Myklebust { 1424a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1425a7d42ddbSWeston Andros Adamson 14266f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14276f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14286f29b9bbSKinglong Mee 142941d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1430a7d42ddbSWeston Andros Adamson 1431a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1432a7d42ddbSWeston Andros Adamson 1433a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1434a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1435dce81290STrond Myklebust } 14361f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1437dce81290STrond Myklebust 14381da177e4SLinus Torvalds 14390b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14400b7c0153SFred Isaman { 14410b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14420b7c0153SFred Isaman 14430b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14440b7c0153SFred Isaman } 14450b7c0153SFred Isaman 14461f2edbe3STrond Myklebust /* 14471f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14481f2edbe3STrond Myklebust */ 14491f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14501f2edbe3STrond Myklebust { 14511f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14521f2edbe3STrond Myklebust int kill = 0; 1453788e7a89STrond Myklebust 14541f2edbe3STrond Myklebust /* suid always must be killed */ 14551f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14561f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14571f2edbe3STrond Myklebust 14581f2edbe3STrond Myklebust /* 14591f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14601f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14611f2edbe3STrond Myklebust */ 14621f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14631f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14641f2edbe3STrond Myklebust 14651f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14661f2edbe3STrond Myklebust return kill; 14671f2edbe3STrond Myklebust 14681f2edbe3STrond Myklebust return 0; 14691f2edbe3STrond Myklebust } 1470788e7a89STrond Myklebust 1471a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1472a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1473a08a8cd3STrond Myklebust { 1474a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1475a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14762b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1477a08a8cd3STrond Myklebust 1478a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14792b83d3deSTrond Myklebust fattr->size = size; 14802b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 14812b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1482a08a8cd3STrond Myklebust return; 14832b83d3deSTrond Myklebust } 14842b83d3deSTrond Myklebust if (size != fattr->size) 1485a08a8cd3STrond Myklebust return; 1486a08a8cd3STrond Myklebust /* Set attribute barrier */ 1487a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 14882b83d3deSTrond Myklebust /* ...and update size */ 14892b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1490a08a8cd3STrond Myklebust } 1491a08a8cd3STrond Myklebust 1492a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1493a08a8cd3STrond Myklebust { 14942b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1495a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1496a08a8cd3STrond Myklebust 1497a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1498a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1499a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1500a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1501a08a8cd3STrond Myklebust } 1502a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1503a08a8cd3STrond Myklebust 15041da177e4SLinus Torvalds /* 15051da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15061da177e4SLinus Torvalds */ 1507d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1508d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15090eecb214SAnna Schumaker struct inode *inode) 15101da177e4SLinus Torvalds { 1511788e7a89STrond Myklebust int status; 15121da177e4SLinus Torvalds 1513f551e44fSChuck Lever /* 1514f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1515f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1516f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1517f551e44fSChuck Lever * another writer had changed the file, but some applications 1518f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1519f551e44fSChuck Lever */ 1520d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1521788e7a89STrond Myklebust if (status != 0) 15220eecb214SAnna Schumaker return status; 1523d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 152491d5b470SChuck Lever 1525d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1526d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15271da177e4SLinus Torvalds /* We tried a write call, but the server did not 15281da177e4SLinus Torvalds * commit data to stable storage even though we 15291da177e4SLinus Torvalds * requested it. 15301da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15311da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15321da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15331da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15341da177e4SLinus Torvalds */ 15351da177e4SLinus Torvalds static unsigned long complain; 15361da177e4SLinus Torvalds 1537a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15381da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15391da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15401da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1541cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1542d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15431da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15441da177e4SLinus Torvalds } 15451da177e4SLinus Torvalds } 15461f2edbe3STrond Myklebust 15471f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15481f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15491f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15500eecb214SAnna Schumaker return 0; 15510eecb214SAnna Schumaker } 15520eecb214SAnna Schumaker 15530eecb214SAnna Schumaker /* 15540eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15550eecb214SAnna Schumaker */ 1556d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1557d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15580eecb214SAnna Schumaker { 1559d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1560d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15611f2edbe3STrond Myklebust 15621f2edbe3STrond Myklebust if (resp->count < argp->count) { 15631da177e4SLinus Torvalds static unsigned long complain; 15641da177e4SLinus Torvalds 15656c75dc0dSFred Isaman /* This a short write! */ 1566d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 156791d5b470SChuck Lever 15681da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15696c75dc0dSFred Isaman if (resp->count == 0) { 15706c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15716c75dc0dSFred Isaman printk(KERN_WARNING 15726c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 15736c75dc0dSFred Isaman argp->count); 15746c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15756c75dc0dSFred Isaman } 1576d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15776c75dc0dSFred Isaman task->tk_status = -EIO; 15786c75dc0dSFred Isaman return; 15796c75dc0dSFred Isaman } 1580f8417b48SKinglong Mee 1581f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1582f8417b48SKinglong Mee if (!task->tk_ops) { 1583f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1584f8417b48SKinglong Mee return; 1585f8417b48SKinglong Mee } 1586f8417b48SKinglong Mee 15871da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 15881da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 15891da177e4SLinus Torvalds /* Resend from where the server left off */ 1590d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 15911da177e4SLinus Torvalds argp->offset += resp->count; 15921da177e4SLinus Torvalds argp->pgbase += resp->count; 15931da177e4SLinus Torvalds argp->count -= resp->count; 15941da177e4SLinus Torvalds } else { 15951da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 15961da177e4SLinus Torvalds * headaches in the case of a server crash. 15971da177e4SLinus Torvalds */ 15981da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 15991da177e4SLinus Torvalds } 1600d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16011da177e4SLinus Torvalds } 16021da177e4SLinus Torvalds } 16031da177e4SLinus Torvalds 1604af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 160571d0a611STrond Myklebust { 1606af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1607af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1608af7cf057STrond Myklebust } 1609af7cf057STrond Myklebust 1610af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1611af7cf057STrond Myklebust { 1612af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1613af7cf057STrond Myklebust } 1614af7cf057STrond Myklebust 1615af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1616af7cf057STrond Myklebust { 1617af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1618af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 161971d0a611STrond Myklebust } 162071d0a611STrond Myklebust 16210b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16221da177e4SLinus Torvalds { 16230b7c0153SFred Isaman put_nfs_open_context(data->context); 16240b7c0153SFred Isaman nfs_commit_free(data); 16251da177e4SLinus Torvalds } 1626e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16271da177e4SLinus Torvalds 16280b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1629c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16309ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16319f0ec176SAndy Adamson int how, int flags) 16321da177e4SLinus Torvalds { 163307737691STrond Myklebust struct rpc_task *task; 16349ace33cdSFred Isaman int priority = flush_task_priority(how); 1635bdc7f021STrond Myklebust struct rpc_message msg = { 1636bdc7f021STrond Myklebust .rpc_argp = &data->args, 1637bdc7f021STrond Myklebust .rpc_resp = &data->res, 16389ace33cdSFred Isaman .rpc_cred = data->cred, 1639bdc7f021STrond Myklebust }; 164084115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 164107737691STrond Myklebust .task = &data->task, 16429ace33cdSFred Isaman .rpc_client = clnt, 1643bdc7f021STrond Myklebust .rpc_message = &msg, 16449ace33cdSFred Isaman .callback_ops = call_ops, 164584115e1cSTrond Myklebust .callback_data = data, 1646101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16479f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16483ff7576dSTrond Myklebust .priority = priority, 164984115e1cSTrond Myklebust }; 1650788e7a89STrond Myklebust /* Set up the initial task struct. */ 1651c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16521da177e4SLinus Torvalds 1653b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1654bdc7f021STrond Myklebust 16558c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16568c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16578c21c62cSWeston Andros Adamson 165807737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1659dbae4c73STrond Myklebust if (IS_ERR(task)) 1660dbae4c73STrond Myklebust return PTR_ERR(task); 1661d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1662d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 166307737691STrond Myklebust rpc_put_task(task); 1664dbae4c73STrond Myklebust return 0; 16651da177e4SLinus Torvalds } 1666e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16671da177e4SLinus Torvalds 1668378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1669378520b8SPeng Tao { 1670378520b8SPeng Tao loff_t lwb = 0; 1671378520b8SPeng Tao struct nfs_page *req; 1672378520b8SPeng Tao 1673378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1674378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1675378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1676378520b8SPeng Tao 1677378520b8SPeng Tao return lwb; 1678378520b8SPeng Tao } 1679378520b8SPeng Tao 16801da177e4SLinus Torvalds /* 16819ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16829ace33cdSFred Isaman */ 16830b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1684988b6dceSFred Isaman struct list_head *head, 1685f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1686f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16879ace33cdSFred Isaman { 16889ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16892b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16909ace33cdSFred Isaman 16919ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16929ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 16939ace33cdSFred Isaman 16949ace33cdSFred Isaman list_splice_init(head, &data->pages); 16959ace33cdSFred Isaman 16969ace33cdSFred Isaman data->inode = inode; 16979ace33cdSFred Isaman data->cred = first->wb_context->cred; 1698988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1699378520b8SPeng Tao /* only set lwb for pnfs commit */ 1700378520b8SPeng Tao if (lseg) 1701378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17029ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1703f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1704b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17059ace33cdSFred Isaman 17069ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17079ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17089ace33cdSFred Isaman data->args.offset = 0; 17099ace33cdSFred Isaman data->args.count = 0; 17100b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 17119ace33cdSFred Isaman data->res.fattr = &data->fattr; 17129ace33cdSFred Isaman data->res.verf = &data->verf; 17139ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17149ace33cdSFred Isaman } 1715e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17169ace33cdSFred Isaman 1717e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1718ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1719b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1720b57ff130SWeston Andros Adamson u32 ds_commit_idx) 172164bfeb49SFred Isaman { 172264bfeb49SFred Isaman struct nfs_page *req; 172364bfeb49SFred Isaman 172464bfeb49SFred Isaman while (!list_empty(page_list)) { 172564bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 172664bfeb49SFred Isaman nfs_list_remove_request(req); 1727b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1728487b9b8aSTom Haynes if (!cinfo->dreq) 1729487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17301d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 173164bfeb49SFred Isaman } 173264bfeb49SFred Isaman } 1733e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 173464bfeb49SFred Isaman 1735b20135d0STrond Myklebust static void 1736b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1737b20135d0STrond Myklebust struct nfs_page *req) 1738b20135d0STrond Myklebust { 1739b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1740b20135d0STrond Myklebust } 1741b20135d0STrond Myklebust 17429ace33cdSFred Isaman /* 17431da177e4SLinus Torvalds * Commit dirty pages 17441da177e4SLinus Torvalds */ 17451da177e4SLinus Torvalds static int 1746ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1747ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17481da177e4SLinus Torvalds { 17490b7c0153SFred Isaman struct nfs_commit_data *data; 17501da177e4SLinus Torvalds 1751ade8febdSWeston Andros Adamson /* another commit raced with us */ 1752ade8febdSWeston Andros Adamson if (list_empty(head)) 1753ade8febdSWeston Andros Adamson return 0; 1754ade8febdSWeston Andros Adamson 1755518662e0SNeilBrown data = nfs_commitdata_alloc(true); 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds /* Set up the argument struct */ 1758f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1759f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1760c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1761c36aae9aSPeng Tao data->mds_ops, how, 0); 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds 17641da177e4SLinus Torvalds /* 17651da177e4SLinus Torvalds * COMMIT call returned 17661da177e4SLinus Torvalds */ 1767788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17681da177e4SLinus Torvalds { 17690b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17701da177e4SLinus Torvalds 1771a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17721da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17731da177e4SLinus Torvalds 1774788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1775c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1776c9d8f89dSTrond Myklebust } 1777c9d8f89dSTrond Myklebust 1778f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1779c9d8f89dSTrond Myklebust { 1780c9d8f89dSTrond Myklebust struct nfs_page *req; 1781c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1782f453a54aSFred Isaman struct nfs_commit_info cinfo; 1783353db796SNeilBrown struct nfs_server *nfss; 1784788e7a89STrond Myklebust 17851da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17861da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17871da177e4SLinus Torvalds nfs_list_remove_request(req); 178867911c8fSAnna Schumaker if (req->wb_page) 1789d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17901da177e4SLinus Torvalds 17911e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17923d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 17932b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 17941da177e4SLinus Torvalds req->wb_bytes, 17951da177e4SLinus Torvalds (long long)req_offset(req)); 1796c9d8f89dSTrond Myklebust if (status < 0) { 1797c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 179838a33101SKinglong Mee if (req->wb_page) 17991da177e4SLinus Torvalds nfs_inode_remove_request(req); 1800ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18011da177e4SLinus Torvalds goto next; 18021da177e4SLinus Torvalds } 18031da177e4SLinus Torvalds 18041da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18051da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18068fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18071da177e4SLinus Torvalds /* We have a match */ 180838a33101SKinglong Mee if (req->wb_page) 18091da177e4SLinus Torvalds nfs_inode_remove_request(req); 1810ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18111da177e4SLinus Torvalds goto next; 18121da177e4SLinus Torvalds } 18131da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1814ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18156d884e8fSFred nfs_mark_request_dirty(req); 181605990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18171da177e4SLinus Torvalds next: 18181d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18191da177e4SLinus Torvalds } 1820353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1821353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18220db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1823353db796SNeilBrown 1824f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1825af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18265917ce84SFred Isaman } 18275917ce84SFred Isaman 18285917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18295917ce84SFred Isaman { 18300b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18315917ce84SFred Isaman 1832f453a54aSFred Isaman data->completion_ops->completion(data); 1833c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18341da177e4SLinus Torvalds } 1835788e7a89STrond Myklebust 1836788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18370b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1838788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1839788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1840788e7a89STrond Myklebust }; 18411da177e4SLinus Torvalds 1842f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1843f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1844b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1845f453a54aSFred Isaman }; 1846f453a54aSFred Isaman 18471763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1848ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 184984c53ab5SFred Isaman { 185084c53ab5SFred Isaman int status; 185184c53ab5SFred Isaman 1852ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 185384c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1854ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 185584c53ab5SFred Isaman return status; 185684c53ab5SFred Isaman } 185784c53ab5SFred Isaman 1858b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18591da177e4SLinus Torvalds { 18601da177e4SLinus Torvalds LIST_HEAD(head); 1861ea2cf228SFred Isaman struct nfs_commit_info cinfo; 186271d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1863af7cf057STrond Myklebust int error = 0; 1864b8413f98STrond Myklebust int res; 18651da177e4SLinus Torvalds 1866ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1867af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1868ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1869af7cf057STrond Myklebust if (res) 1870ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1871af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18721da177e4SLinus Torvalds if (error < 0) 1873af7cf057STrond Myklebust goto out_error; 1874b8413f98STrond Myklebust if (!may_wait) 1875b8413f98STrond Myklebust goto out_mark_dirty; 1876af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1877b8413f98STrond Myklebust if (error < 0) 1878b8413f98STrond Myklebust return error; 1879c5efa5fcSTrond Myklebust return res; 1880af7cf057STrond Myklebust out_error: 1881af7cf057STrond Myklebust res = error; 1882c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1883c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1884c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1885c5efa5fcSTrond Myklebust * that the data is on the disk. 1886c5efa5fcSTrond Myklebust */ 1887c5efa5fcSTrond Myklebust out_mark_dirty: 1888c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18891da177e4SLinus Torvalds return res; 18901da177e4SLinus Torvalds } 1891b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 18928fc795f7STrond Myklebust 1893ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18948fc795f7STrond Myklebust { 1895420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1896420e3646STrond Myklebust int flags = FLUSH_SYNC; 1897420e3646STrond Myklebust int ret = 0; 18988fc795f7STrond Myklebust 18993236c3e1SJeff Layton /* no commits means nothing needs to be done */ 19005cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 19013236c3e1SJeff Layton return ret; 19023236c3e1SJeff Layton 1903a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1904a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1905a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1906420e3646STrond Myklebust */ 19071a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1908420e3646STrond Myklebust goto out_mark_dirty; 1909420e3646STrond Myklebust 1910a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1911420e3646STrond Myklebust flags = 0; 1912a00dd6c0SJeff Layton } 1913a00dd6c0SJeff Layton 1914420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1915420e3646STrond Myklebust if (ret >= 0) { 1916420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1917420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1918420e3646STrond Myklebust wbc->nr_to_write -= ret; 1919420e3646STrond Myklebust else 1920420e3646STrond Myklebust wbc->nr_to_write = 0; 1921420e3646STrond Myklebust } 19228fc795f7STrond Myklebust return 0; 1923420e3646STrond Myklebust } 1924420e3646STrond Myklebust out_mark_dirty: 19258fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19268fc795f7STrond Myklebust return ret; 19278fc795f7STrond Myklebust } 192889d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1929863a3c6cSAndy Adamson 1930acdc53b2STrond Myklebust /* 1931837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1932837bb1d7STrond Myklebust * 1933837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1934837bb1d7STrond Myklebust * client. 1935837bb1d7STrond Myklebust */ 1936837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1937837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1938837bb1d7STrond Myklebust { 1939837bb1d7STrond Myklebust int ret; 1940837bb1d7STrond Myklebust 1941837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1942837bb1d7STrond Myklebust if (ret == 0) 1943837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1944837bb1d7STrond Myklebust return ret; 1945837bb1d7STrond Myklebust } 1946837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1947837bb1d7STrond Myklebust 1948837bb1d7STrond Myklebust /* 1949acdc53b2STrond Myklebust * flush the inode to disk. 1950acdc53b2STrond Myklebust */ 1951acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 195234901f70STrond Myklebust { 1953f4ce1299STrond Myklebust int ret; 195434901f70STrond Myklebust 1955f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1956f4ce1299STrond Myklebust 19575bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19586b196875SChuck Lever if (ret) 19596b196875SChuck Lever goto out; 19605bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19616b196875SChuck Lever if (ret < 0) 19626b196875SChuck Lever goto out; 19635bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19646b196875SChuck Lever ret = 0; 1965f4ce1299STrond Myklebust 19666b196875SChuck Lever out: 1967f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1968f4ce1299STrond Myklebust return ret; 19691c75950bSTrond Myklebust } 1970ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19711c75950bSTrond Myklebust 19721b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19731b3b4a1aSTrond Myklebust { 19741b3b4a1aSTrond Myklebust struct nfs_page *req; 19751b3b4a1aSTrond Myklebust int ret = 0; 19761b3b4a1aSTrond Myklebust 1977ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19783e217045SWeston Andros Adamson 19793e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 19803e217045SWeston Andros Adamson * request */ 19816d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 19823e217045SWeston Andros Adamson 19833e217045SWeston Andros Adamson if (IS_ERR(req)) { 19843e217045SWeston Andros Adamson ret = PTR_ERR(req); 19853e217045SWeston Andros Adamson } else if (req) { 19863e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 19873e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 19883e217045SWeston Andros Adamson * request from the inode / page_private pointer and 19893e217045SWeston Andros Adamson * release it */ 19901b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 19911d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19921b3b4a1aSTrond Myklebust } 19933e217045SWeston Andros Adamson 19941b3b4a1aSTrond Myklebust return ret; 19951b3b4a1aSTrond Myklebust } 19961b3b4a1aSTrond Myklebust 19971c75950bSTrond Myklebust /* 19981c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 19991c75950bSTrond Myklebust */ 2000c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 20011c75950bSTrond Myklebust { 200229418aa4SMel Gorman loff_t range_start = page_file_offset(page); 200309cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 20047f2f12d9STrond Myklebust struct writeback_control wbc = { 20057f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20067f2f12d9STrond Myklebust .nr_to_write = 0, 20077f2f12d9STrond Myklebust .range_start = range_start, 20087f2f12d9STrond Myklebust .range_end = range_end, 20097f2f12d9STrond Myklebust }; 20107f2f12d9STrond Myklebust int ret; 20117f2f12d9STrond Myklebust 2012f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2013f4ce1299STrond Myklebust 20140522f6adSTrond Myklebust for (;;) { 2015ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20167f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2017c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20187f2f12d9STrond Myklebust if (ret < 0) 20197f2f12d9STrond Myklebust goto out_error; 20200522f6adSTrond Myklebust continue; 20217f2f12d9STrond Myklebust } 2022f4ce1299STrond Myklebust ret = 0; 20230522f6adSTrond Myklebust if (!PagePrivate(page)) 20240522f6adSTrond Myklebust break; 20250522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20267f2f12d9STrond Myklebust if (ret < 0) 20277f2f12d9STrond Myklebust goto out_error; 20287f2f12d9STrond Myklebust } 20297f2f12d9STrond Myklebust out_error: 2030f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20317f2f12d9STrond Myklebust return ret; 20321c75950bSTrond Myklebust } 20331c75950bSTrond Myklebust 2034074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2035074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2036a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2037074cc1deSTrond Myklebust { 20382da95652SJeff Layton /* 20392da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20402da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20412da95652SJeff Layton * 20422da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20432da95652SJeff Layton * that we can safely release the inode reference while holding 20442da95652SJeff Layton * the page lock. 20452da95652SJeff Layton */ 20462da95652SJeff Layton if (PagePrivate(page)) 20472da95652SJeff Layton return -EBUSY; 2048074cc1deSTrond Myklebust 20498c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20508c209ce7SDavid Howells return -EBUSY; 2051074cc1deSTrond Myklebust 2052a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2053074cc1deSTrond Myklebust } 2054074cc1deSTrond Myklebust #endif 2055074cc1deSTrond Myklebust 2056f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20571da177e4SLinus Torvalds { 20581da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20591e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20601da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 206120c2df83SPaul Mundt NULL); 20621da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20631da177e4SLinus Torvalds return -ENOMEM; 20641da177e4SLinus Torvalds 206593d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20661da177e4SLinus Torvalds nfs_wdata_cachep); 20671da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20683dd4765fSJeff Layton goto out_destroy_write_cache; 20691da177e4SLinus Torvalds 20700b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20710b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20720b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20730b7c0153SFred Isaman NULL); 20740b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20753dd4765fSJeff Layton goto out_destroy_write_mempool; 20760b7c0153SFred Isaman 207793d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20784c100210SYanchuan Nian nfs_cdata_cachep); 20791da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 20803dd4765fSJeff Layton goto out_destroy_commit_cache; 20811da177e4SLinus Torvalds 208289a09141SPeter Zijlstra /* 208389a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 208489a09141SPeter Zijlstra * 208589a09141SPeter Zijlstra * 64MB: 8192k 208689a09141SPeter Zijlstra * 128MB: 11585k 208789a09141SPeter Zijlstra * 256MB: 16384k 208889a09141SPeter Zijlstra * 512MB: 23170k 208989a09141SPeter Zijlstra * 1GB: 32768k 209089a09141SPeter Zijlstra * 2GB: 46340k 209189a09141SPeter Zijlstra * 4GB: 65536k 209289a09141SPeter Zijlstra * 8GB: 92681k 209389a09141SPeter Zijlstra * 16GB: 131072k 209489a09141SPeter Zijlstra * 209589a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 209689a09141SPeter Zijlstra * Limit the default to 256M 209789a09141SPeter Zijlstra */ 209889a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 209989a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 210089a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 210189a09141SPeter Zijlstra 21021da177e4SLinus Torvalds return 0; 21033dd4765fSJeff Layton 21043dd4765fSJeff Layton out_destroy_commit_cache: 21053dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21063dd4765fSJeff Layton out_destroy_write_mempool: 21073dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21083dd4765fSJeff Layton out_destroy_write_cache: 21093dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21103dd4765fSJeff Layton return -ENOMEM; 21111da177e4SLinus Torvalds } 21121da177e4SLinus Torvalds 2113266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21141da177e4SLinus Torvalds { 21151da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21163dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21171da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21181a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21191da177e4SLinus Torvalds } 21201da177e4SLinus Torvalds 21214a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21224a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21234a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21240eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21250eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21261ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21274a0de55cSAnna Schumaker }; 2128