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 15784d3a9a9SWeston Andros Adamson /* 15884d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 15984d3a9a9SWeston Andros Adamson * 16084d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 16184d3a9a9SWeston Andros Adamson * 16284d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 16384d3a9a9SWeston Andros Adamson */ 16429418aa4SMel Gorman static struct nfs_page * 16584d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 166277459d2STrond Myklebust { 167277459d2STrond Myklebust struct nfs_page *req = NULL; 168277459d2STrond Myklebust 16929418aa4SMel Gorman if (PagePrivate(page)) 170277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 17102d1426cSWeston Andros Adamson else if (unlikely(PageSwapCache(page))) 17202d1426cSWeston Andros Adamson req = nfs_page_search_commits_for_head_request_locked(nfsi, 17302d1426cSWeston Andros Adamson page); 17429418aa4SMel Gorman 17584d3a9a9SWeston Andros Adamson if (req) { 17684d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 17729418aa4SMel Gorman kref_get(&req->wb_kref); 17884d3a9a9SWeston Andros Adamson } 17929418aa4SMel Gorman 180277459d2STrond Myklebust return req; 181277459d2STrond Myklebust } 182277459d2STrond Myklebust 18384d3a9a9SWeston Andros Adamson /* 18484d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 18584d3a9a9SWeston Andros Adamson * 18684d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 18784d3a9a9SWeston Andros Adamson */ 18884d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 189277459d2STrond Myklebust { 190d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 191277459d2STrond Myklebust struct nfs_page *req = NULL; 192277459d2STrond Myklebust 19382749dd4STrond Myklebust if (PagePrivate(page) || PageSwapCache(page)) { 194587142f8STrond Myklebust spin_lock(&inode->i_lock); 19584d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 196587142f8STrond Myklebust spin_unlock(&inode->i_lock); 19782749dd4STrond Myklebust } 198277459d2STrond Myklebust return req; 199277459d2STrond Myklebust } 200277459d2STrond Myklebust 2011da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2021da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2031da177e4SLinus Torvalds { 204d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 205a3d01454STrond Myklebust loff_t end, i_size; 206a3d01454STrond Myklebust pgoff_t end_index; 2071da177e4SLinus Torvalds 208a3d01454STrond Myklebust spin_lock(&inode->i_lock); 209a3d01454STrond Myklebust i_size = i_size_read(inode); 21009cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2118cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 212a3d01454STrond Myklebust goto out; 213d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2141da177e4SLinus Torvalds if (i_size >= end) 215a3d01454STrond Myklebust goto out; 2161da177e4SLinus Torvalds i_size_write(inode, end); 217a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 218a3d01454STrond Myklebust out: 219a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2201da177e4SLinus Torvalds } 2211da177e4SLinus Torvalds 222a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 223a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 224a301b777STrond Myklebust { 225d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 226a301b777STrond Myklebust } 227a301b777STrond Myklebust 228d72ddcbaSWeston Andros Adamson /* 229d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 230d72ddcbaSWeston Andros Adamson * @head - head request of page group 231d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 232d72ddcbaSWeston Andros Adamson * 233d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 234d72ddcbaSWeston Andros Adamson * page offset @page_offset. 235d72ddcbaSWeston Andros Adamson * 236d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 237d72ddcbaSWeston Andros Adamson * match is found. 238d72ddcbaSWeston Andros Adamson * 239d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 240d72ddcbaSWeston Andros Adamson */ 241d72ddcbaSWeston Andros Adamson static struct nfs_page * 242d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 243d72ddcbaSWeston Andros Adamson { 244d72ddcbaSWeston Andros Adamson struct nfs_page *req; 245d72ddcbaSWeston Andros Adamson 246d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 247d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 248d72ddcbaSWeston Andros Adamson 249d72ddcbaSWeston Andros Adamson req = head; 250d72ddcbaSWeston Andros Adamson do { 251d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 252d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 253d72ddcbaSWeston Andros Adamson return req; 254d72ddcbaSWeston Andros Adamson 255d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 256d72ddcbaSWeston Andros Adamson } while (req != head); 257d72ddcbaSWeston Andros Adamson 258d72ddcbaSWeston Andros Adamson return NULL; 259d72ddcbaSWeston Andros Adamson } 260d72ddcbaSWeston Andros Adamson 261d72ddcbaSWeston Andros Adamson /* 262d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 263d72ddcbaSWeston Andros Adamson * @head - head request of page group 264d72ddcbaSWeston Andros Adamson * 265d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 266d72ddcbaSWeston Andros Adamson * returns false otherwise 267d72ddcbaSWeston Andros Adamson */ 268d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 269d72ddcbaSWeston Andros Adamson { 270d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 271d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 272d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 273d72ddcbaSWeston Andros Adamson 274fd2f3a06SWeston Andros Adamson nfs_page_group_lock(req, false); 275d72ddcbaSWeston Andros Adamson 276d72ddcbaSWeston Andros Adamson do { 277d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 278d72ddcbaSWeston Andros Adamson if (tmp) { 279d72ddcbaSWeston Andros Adamson /* no way this should happen */ 280d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 281d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 282d72ddcbaSWeston Andros Adamson } 283d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 284d72ddcbaSWeston Andros Adamson 285d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 286d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 287d72ddcbaSWeston Andros Adamson return pos == len; 288d72ddcbaSWeston Andros Adamson } 289d72ddcbaSWeston Andros Adamson 2901da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2911da177e4SLinus Torvalds * covers the full page. 2921da177e4SLinus Torvalds */ 293d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2941da177e4SLinus Torvalds { 295d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2961da177e4SLinus Torvalds return; 297d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2981da177e4SLinus Torvalds return; 299d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 3001da177e4SLinus Torvalds } 3011da177e4SLinus Torvalds 3021da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3031da177e4SLinus Torvalds { 304e87b4c7aSNeilBrown int ret = 0; 305cca588d6STrond Myklebust 306e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 307e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 308e87b4c7aSNeilBrown return ret; 3091da177e4SLinus Torvalds } 3101da177e4SLinus Torvalds 3111da177e4SLinus Torvalds /* 31289a09141SPeter Zijlstra * NFS congestion control 31389a09141SPeter Zijlstra */ 31489a09141SPeter Zijlstra 31589a09141SPeter Zijlstra int nfs_congestion_kb; 31689a09141SPeter Zijlstra 31789a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 31889a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 31989a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 32089a09141SPeter Zijlstra 321deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 32289a09141SPeter Zijlstra { 3230db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3240db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3255a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3265a6d41b3STrond Myklebust 327deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 32889a09141SPeter Zijlstra 329277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3300db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3310db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 33289a09141SPeter Zijlstra } 33389a09141SPeter Zijlstra 33420633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 33589a09141SPeter Zijlstra { 33620633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 33789a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 33831a01f09STrond Myklebust bool is_done; 33989a09141SPeter Zijlstra 34031a01f09STrond Myklebust is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); 34131a01f09STrond Myklebust nfs_unlock_request(req); 34231a01f09STrond Myklebust if (!is_done) 34320633f04SWeston Andros Adamson return; 34420633f04SWeston Andros Adamson 34520633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 346c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3470db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 34889a09141SPeter Zijlstra } 34989a09141SPeter Zijlstra 350d4581383SWeston Andros Adamson 351d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 352d4581383SWeston Andros Adamson * @req - an nfs request 353d4581383SWeston Andros Adamson * clears all page group related bits from @req 354d4581383SWeston Andros Adamson */ 355d4581383SWeston Andros Adamson static void 356d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 357e261f51fSTrond Myklebust { 358d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 359d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 360d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 361d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 362d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 363d4581383SWeston Andros Adamson } 364d4581383SWeston Andros Adamson 365d4581383SWeston Andros Adamson 366d4581383SWeston Andros Adamson /* 367d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 368d4581383SWeston Andros Adamson * 369d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 370d4581383SWeston Andros Adamson * 371d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 372d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 373d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 374d4581383SWeston Andros Adamson * 375b5bab9bfSTrond Myklebust * NOTE: this must be called holding page_group bit lock 376b5bab9bfSTrond Myklebust * which will be released before returning. 377d4581383SWeston Andros Adamson * 378d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 379d4581383SWeston Andros Adamson */ 380*74a6d4b5STrond Myklebust static void 381*74a6d4b5STrond Myklebust nfs_unroll_locks(struct inode *inode, struct nfs_page *head, 3826d17d653STrond Myklebust struct nfs_page *req) 383d4581383SWeston Andros Adamson { 384d4581383SWeston Andros Adamson struct nfs_page *tmp; 385e261f51fSTrond Myklebust 386d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 387a0e265bcSTrond Myklebust for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) 388d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 389d4581383SWeston Andros Adamson 390d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 391e261f51fSTrond Myklebust } 392d4581383SWeston Andros Adamson 393d4581383SWeston Andros Adamson /* 394d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 395d4581383SWeston Andros Adamson * 396d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 397d4581383SWeston Andros Adamson * @old_head - the old head of the list 398d4581383SWeston Andros Adamson * 399d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 400d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 401d4581383SWeston Andros Adamson * with a reference held by some other context. 402d4581383SWeston Andros Adamson */ 403d4581383SWeston Andros Adamson static void 404d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 405b66aaa8dSTrond Myklebust struct nfs_page *old_head, 406b66aaa8dSTrond Myklebust struct inode *inode) 407d4581383SWeston Andros Adamson { 408d4581383SWeston Andros Adamson while (destroy_list) { 409d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 410d4581383SWeston Andros Adamson 411d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 412d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 413d4581383SWeston Andros Adamson 414d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 415d4581383SWeston Andros Adamson 416d4581383SWeston Andros Adamson /* make sure old group is not used */ 417d4581383SWeston Andros Adamson subreq->wb_head = subreq; 418d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 419d4581383SWeston Andros Adamson 420d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 421d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 422d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 423d4581383SWeston Andros Adamson 424d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 425d4581383SWeston Andros Adamson /* release ref on old head request */ 426d4581383SWeston Andros Adamson nfs_release_request(old_head); 427d4581383SWeston Andros Adamson 428d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 429d4581383SWeston Andros Adamson 430d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 431b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 432d4581383SWeston Andros Adamson nfs_release_request(subreq); 433b66aaa8dSTrond Myklebust spin_lock(&inode->i_lock); 434b66aaa8dSTrond Myklebust NFS_I(inode)->nrequests--; 435b66aaa8dSTrond Myklebust spin_unlock(&inode->i_lock); 436b66aaa8dSTrond Myklebust } else 437d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 438d4581383SWeston Andros Adamson } else { 439d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 440d4581383SWeston Andros Adamson /* zombie requests have already released the last 441d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 442d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 443d4581383SWeston Andros Adamson * group, simply free the request */ 444d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 445d4581383SWeston Andros Adamson nfs_free_request(subreq); 446d4581383SWeston Andros Adamson } 447d4581383SWeston Andros Adamson } 448d4581383SWeston Andros Adamson } 449d4581383SWeston Andros Adamson 450d4581383SWeston Andros Adamson /* 451d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 452d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 453d4581383SWeston Andros Adamson * operations for this page. 454d4581383SWeston Andros Adamson * 455d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 456d4581383SWeston Andros Adamson * 457d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 458d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 459d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 460d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 461d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 462d4581383SWeston Andros Adamson * 463d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 464d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 465d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 466d4581383SWeston Andros Adamson * error was encountered. 467d4581383SWeston Andros Adamson */ 468d4581383SWeston Andros Adamson static struct nfs_page * 4696d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 470d4581383SWeston Andros Adamson { 471d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 472d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 473d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 474d4581383SWeston Andros Adamson unsigned int total_bytes; 475d4581383SWeston Andros Adamson int ret; 476d4581383SWeston Andros Adamson 477d4581383SWeston Andros Adamson try_again: 478b5bab9bfSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) 479b5bab9bfSTrond Myklebust return NULL; 480d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 481d4581383SWeston Andros Adamson /* 482d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 483d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 484d4581383SWeston Andros Adamson * until the head reference is released. 485d4581383SWeston Andros Adamson */ 486d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 487d4581383SWeston Andros Adamson 488d4581383SWeston Andros Adamson if (!head) { 489587142f8STrond Myklebust spin_unlock(&inode->i_lock); 490d4581383SWeston Andros Adamson return NULL; 491d4581383SWeston Andros Adamson } 492d4581383SWeston Andros Adamson 493a0e265bcSTrond Myklebust /* lock the page head first in order to avoid an ABBA inefficiency */ 494a0e265bcSTrond Myklebust if (!nfs_lock_request(head)) { 495a0e265bcSTrond Myklebust spin_unlock(&inode->i_lock); 496a0e265bcSTrond Myklebust ret = nfs_wait_on_request(head); 497a0e265bcSTrond Myklebust nfs_release_request(head); 498a0e265bcSTrond Myklebust if (ret < 0) 499a0e265bcSTrond Myklebust return ERR_PTR(ret); 500a0e265bcSTrond Myklebust goto try_again; 501a0e265bcSTrond Myklebust } 50294970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 5037c3af975SWeston Andros Adamson 504b5bab9bfSTrond Myklebust ret = nfs_page_group_lock(head, false); 505b5bab9bfSTrond Myklebust if (ret < 0) { 506a0e265bcSTrond Myklebust nfs_unlock_and_release_request(head); 507b5bab9bfSTrond Myklebust return ERR_PTR(ret); 5087c3af975SWeston Andros Adamson } 5097c3af975SWeston Andros Adamson 5107c3af975SWeston Andros Adamson /* lock each request in the page group */ 511a0e265bcSTrond Myklebust total_bytes = head->wb_bytes; 512a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 513a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 514*74a6d4b5STrond Myklebust 515*74a6d4b5STrond Myklebust while (!nfs_lock_request(subreq)) { 516b5bab9bfSTrond Myklebust /* 517*74a6d4b5STrond Myklebust * Unlock page to allow nfs_page_group_sync_on_bit() 518*74a6d4b5STrond Myklebust * to succeed 519b5bab9bfSTrond Myklebust */ 520*74a6d4b5STrond Myklebust nfs_page_group_unlock(head); 521*74a6d4b5STrond Myklebust ret = nfs_wait_on_request(subreq); 522*74a6d4b5STrond Myklebust if (!ret) 523*74a6d4b5STrond Myklebust ret = nfs_page_group_lock(head, false); 524*74a6d4b5STrond Myklebust if (ret < 0) { 525*74a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 526*74a6d4b5STrond Myklebust nfs_unlock_and_release_request(head); 527d4581383SWeston Andros Adamson return ERR_PTR(ret); 528d4581383SWeston Andros Adamson } 529*74a6d4b5STrond Myklebust } 530e14bebf6STrond Myklebust /* 531e14bebf6STrond Myklebust * Subrequests are always contiguous, non overlapping 532e14bebf6STrond Myklebust * and in order - but may be repeated (mirrored writes). 533e14bebf6STrond Myklebust */ 534e14bebf6STrond Myklebust if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 535e14bebf6STrond Myklebust /* keep track of how many bytes this group covers */ 536e14bebf6STrond Myklebust total_bytes += subreq->wb_bytes; 537e14bebf6STrond Myklebust } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 538e14bebf6STrond Myklebust ((subreq->wb_offset + subreq->wb_bytes) > 539e14bebf6STrond Myklebust (head->wb_offset + total_bytes)))) { 540e14bebf6STrond Myklebust nfs_unlock_request(subreq); 541*74a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 542*74a6d4b5STrond Myklebust nfs_page_group_unlock(head); 543*74a6d4b5STrond Myklebust nfs_unlock_and_release_request(head); 544e14bebf6STrond Myklebust return ERR_PTR(-EIO); 545e14bebf6STrond Myklebust } 546a0e265bcSTrond Myklebust } 547d4581383SWeston Andros Adamson 548d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 549d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 550d4581383SWeston Andros Adamson subreq = head; 551d4581383SWeston Andros Adamson do { 552411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 553d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 554d4581383SWeston Andros Adamson } while (subreq != head); 555d4581383SWeston Andros Adamson 556d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 557d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 558d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 559d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 560d4581383SWeston Andros Adamson head->wb_this_page = head; 561d4581383SWeston Andros Adamson 562d4581383SWeston Andros Adamson /* change head request to cover whole range that 563d4581383SWeston Andros Adamson * the former page group covered */ 564d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 565d4581383SWeston Andros Adamson } 566d4581383SWeston Andros Adamson 567b66aaa8dSTrond Myklebust /* Postpone destruction of this request */ 568b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { 569b66aaa8dSTrond Myklebust set_bit(PG_INODE_REF, &head->wb_flags); 570b66aaa8dSTrond Myklebust kref_get(&head->wb_kref); 571b5bab9bfSTrond Myklebust spin_lock(&inode->i_lock); 572b66aaa8dSTrond Myklebust NFS_I(inode)->nrequests++; 573b5bab9bfSTrond Myklebust spin_unlock(&inode->i_lock); 574b66aaa8dSTrond Myklebust } 575b66aaa8dSTrond Myklebust 576d4581383SWeston Andros Adamson /* 577d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 578d4581383SWeston Andros Adamson */ 579d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 580d4581383SWeston Andros Adamson 581d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 582d4581383SWeston Andros Adamson 583b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 584d4581383SWeston Andros Adamson 585b5bab9bfSTrond Myklebust /* Did we lose a race with nfs_inode_remove_request()? */ 586b5bab9bfSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) { 587b5bab9bfSTrond Myklebust nfs_unlock_and_release_request(head); 588b5bab9bfSTrond Myklebust return NULL; 589b5bab9bfSTrond Myklebust } 590b5bab9bfSTrond Myklebust 591d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 592d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 593d4581383SWeston Andros Adamson return head; 594612c9384STrond Myklebust } 595074cc1deSTrond Myklebust 5960bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5970bcbf039SPeng Tao { 5980bcbf039SPeng Tao nfs_end_page_writeback(req); 5990bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 6000bcbf039SPeng Tao req->wb_page); 6011f84ccdfSFred Isaman nfs_release_request(req); 6020bcbf039SPeng Tao } 6030bcbf039SPeng Tao 604a6598813STrond Myklebust static bool 605a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 606a6598813STrond Myklebust { 607a6598813STrond Myklebust switch (err) { 608a6598813STrond Myklebust case 0: 609a6598813STrond Myklebust case -ERESTARTSYS: 610a6598813STrond Myklebust case -EINTR: 611a6598813STrond Myklebust return false; 612a6598813STrond Myklebust } 613a6598813STrond Myklebust return nfs_error_is_fatal(err); 614e261f51fSTrond Myklebust } 615074cc1deSTrond Myklebust 616074cc1deSTrond Myklebust /* 617074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 618074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 619074cc1deSTrond Myklebust */ 620074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 6216d17d653STrond Myklebust struct page *page) 622074cc1deSTrond Myklebust { 623074cc1deSTrond Myklebust struct nfs_page *req; 624074cc1deSTrond Myklebust int ret = 0; 625074cc1deSTrond Myklebust 6266d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 627074cc1deSTrond Myklebust if (!req) 628074cc1deSTrond Myklebust goto out; 629074cc1deSTrond Myklebust ret = PTR_ERR(req); 630074cc1deSTrond Myklebust if (IS_ERR(req)) 631074cc1deSTrond Myklebust goto out; 632074cc1deSTrond Myklebust 633deed85e7STrond Myklebust nfs_set_page_writeback(page); 634deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 635074cc1deSTrond Myklebust 636deed85e7STrond Myklebust ret = 0; 637a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 638a6598813STrond Myklebust if (nfs_error_is_fatal_on_server(req->wb_context->error)) 639a6598813STrond Myklebust goto out_launder; 640a6598813STrond Myklebust 641f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 642074cc1deSTrond Myklebust ret = pgio->pg_error; 6430bcbf039SPeng Tao /* 644c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6450bcbf039SPeng Tao */ 6460bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 6470bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 648c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 649a6598813STrond Myklebust goto out_launder; 650d6c843b9SPeng Tao } 6510bcbf039SPeng Tao nfs_redirty_request(req); 6520bcbf039SPeng Tao ret = -EAGAIN; 65340f90271STrond Myklebust } else 65440f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 65540f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 656074cc1deSTrond Myklebust out: 657074cc1deSTrond Myklebust return ret; 658a6598813STrond Myklebust out_launder: 659a6598813STrond Myklebust nfs_write_error_remove_page(req); 660a6598813STrond Myklebust return ret; 661e261f51fSTrond Myklebust } 662e261f51fSTrond Myklebust 663d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 664c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 665f758c885STrond Myklebust { 666cfb506e1STrond Myklebust int ret; 667f758c885STrond Myklebust 6688cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 6696d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 670cfb506e1STrond Myklebust if (ret == -EAGAIN) { 671cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 672cfb506e1STrond Myklebust ret = 0; 673cfb506e1STrond Myklebust } 674cfb506e1STrond Myklebust return ret; 675f758c885STrond Myklebust } 676f758c885STrond Myklebust 677e261f51fSTrond Myklebust /* 6781da177e4SLinus Torvalds * Write an mmapped page to the server. 6791da177e4SLinus Torvalds */ 680d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 681c373fff7STrond Myklebust struct writeback_control *wbc) 6821da177e4SLinus Torvalds { 683f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 68440f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 685e261f51fSTrond Myklebust int err; 6861da177e4SLinus Torvalds 68740f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 688811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 689a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 690c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 691f758c885STrond Myklebust nfs_pageio_complete(&pgio); 692f758c885STrond Myklebust if (err < 0) 6934d770ccfSTrond Myklebust return err; 694f758c885STrond Myklebust if (pgio.pg_error < 0) 695f758c885STrond Myklebust return pgio.pg_error; 696f758c885STrond Myklebust return 0; 6974d770ccfSTrond Myklebust } 6984d770ccfSTrond Myklebust 6994d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 7004d770ccfSTrond Myklebust { 701f758c885STrond Myklebust int ret; 7024d770ccfSTrond Myklebust 703c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 7041da177e4SLinus Torvalds unlock_page(page); 705f758c885STrond Myklebust return ret; 706f758c885STrond Myklebust } 707f758c885STrond Myklebust 708f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 709f758c885STrond Myklebust { 710f758c885STrond Myklebust int ret; 711f758c885STrond Myklebust 712c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 713f758c885STrond Myklebust unlock_page(page); 714f758c885STrond Myklebust return ret; 7151da177e4SLinus Torvalds } 7161da177e4SLinus Torvalds 717919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 718919e3bd9STrond Myklebust { 719919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 720919e3bd9STrond Myklebust } 721919e3bd9STrond Myklebust 7221da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7231da177e4SLinus Torvalds { 7241da177e4SLinus Torvalds struct inode *inode = mapping->host; 725c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 726919e3bd9STrond Myklebust struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); 7271da177e4SLinus Torvalds int err; 7281da177e4SLinus Torvalds 72991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 73091d5b470SChuck Lever 731919e3bd9STrond Myklebust if (ioc) 732919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 733919e3bd9STrond Myklebust 734a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 735a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 736919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 737f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 738c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 739919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 74072cb77f4STrond Myklebust 741f758c885STrond Myklebust if (err < 0) 74272cb77f4STrond Myklebust goto out_err; 74372cb77f4STrond Myklebust err = pgio.pg_error; 74472cb77f4STrond Myklebust if (err < 0) 74572cb77f4STrond Myklebust goto out_err; 746c63c7b05STrond Myklebust return 0; 74772cb77f4STrond Myklebust out_err: 74872cb77f4STrond Myklebust return err; 7491da177e4SLinus Torvalds } 7501da177e4SLinus Torvalds 7511da177e4SLinus Torvalds /* 7521da177e4SLinus Torvalds * Insert a write request into an inode 7531da177e4SLinus Torvalds */ 754d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7551da177e4SLinus Torvalds { 7561da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 757e7d39069STrond Myklebust 7582bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7592bfc6e56SWeston Andros Adamson 760e7d39069STrond Myklebust /* Lock the request! */ 7617ad84aa9STrond Myklebust nfs_lock_request(req); 762e7d39069STrond Myklebust 763e7d39069STrond Myklebust spin_lock(&inode->i_lock); 764cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 765cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 766a9a4a87aSTrond Myklebust inode->i_version++; 76729418aa4SMel Gorman /* 76829418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 76929418aa4SMel Gorman * with invalidate/truncate. 77029418aa4SMel Gorman */ 77129418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7722df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 773deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 774277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 77529418aa4SMel Gorman } 776cb1410c7SWeston Andros Adamson nfsi->nrequests++; 77717089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 778cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 779cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 780cb1410c7SWeston Andros Adamson * adding subrequests. */ 78117089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 782c03b4024STrond Myklebust kref_get(&req->wb_kref); 783e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7841da177e4SLinus Torvalds } 7851da177e4SLinus Torvalds 7861da177e4SLinus Torvalds /* 78789a09141SPeter Zijlstra * Remove a write request from an inode 7881da177e4SLinus Torvalds */ 7891da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7901da177e4SLinus Torvalds { 7912b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7921da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 79320633f04SWeston Andros Adamson struct nfs_page *head; 79420633f04SWeston Andros Adamson 79520633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 79620633f04SWeston Andros Adamson head = req->wb_head; 7971da177e4SLinus Torvalds 798587142f8STrond Myklebust spin_lock(&inode->i_lock); 79967911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 80020633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 80120633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 80220633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 80329418aa4SMel Gorman } 804cb1410c7SWeston Andros Adamson nfsi->nrequests--; 805cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 806cb1410c7SWeston Andros Adamson } else { 807cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 808cb1410c7SWeston Andros Adamson nfsi->nrequests--; 809587142f8STrond Myklebust spin_unlock(&inode->i_lock); 81020633f04SWeston Andros Adamson } 81117089a29SWeston Andros Adamson 81217089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 8131da177e4SLinus Torvalds nfs_release_request(req); 8141da177e4SLinus Torvalds } 8151da177e4SLinus Torvalds 81661822ab5STrond Myklebust static void 8176d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 81861822ab5STrond Myklebust { 81967911c8fSAnna Schumaker if (req->wb_page) 82061822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 82161822ab5STrond Myklebust } 82261822ab5STrond Myklebust 8233a3908c8STrond Myklebust /* 8243a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8253a3908c8STrond Myklebust * 8263a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8273a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8283a3908c8STrond Myklebust * 8293a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8303a3908c8STrond Myklebust */ 8313a3908c8STrond Myklebust static struct nfs_page * 8323a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8333a3908c8STrond Myklebust struct page *page) 8343a3908c8STrond Myklebust { 8353a3908c8STrond Myklebust struct nfs_page *freq, *t; 8363a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8373a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8383a3908c8STrond Myklebust 8393a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8403a3908c8STrond Myklebust 8413a3908c8STrond Myklebust /* search through pnfs commit lists */ 8423a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8433a3908c8STrond Myklebust if (freq) 8443a3908c8STrond Myklebust return freq->wb_head; 8453a3908c8STrond Myklebust 8463a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8473a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8483a3908c8STrond Myklebust if (freq->wb_page == page) 8493a3908c8STrond Myklebust return freq->wb_head; 8503a3908c8STrond Myklebust } 8513a3908c8STrond Myklebust 8523a3908c8STrond Myklebust return NULL; 8533a3908c8STrond Myklebust } 8543a3908c8STrond Myklebust 8558dd37758STrond Myklebust /** 85686d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 85786d80f97STrond Myklebust * @req: pointer to a struct nfs_page 85886d80f97STrond Myklebust * @dst: commit list head 85986d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 86086d80f97STrond Myklebust * 86186d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 86286d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 86386d80f97STrond Myklebust * the MM page stats. 86486d80f97STrond Myklebust * 865fe238e60SDave Wysochanski * The caller must hold cinfo->inode->i_lock, and the nfs_page lock. 86686d80f97STrond Myklebust */ 86786d80f97STrond Myklebust void 86886d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 86986d80f97STrond Myklebust struct nfs_commit_info *cinfo) 87086d80f97STrond Myklebust { 87186d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 87286d80f97STrond Myklebust nfs_list_add_request(req, dst); 87386d80f97STrond Myklebust cinfo->mds->ncommit++; 87486d80f97STrond Myklebust } 87586d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 87686d80f97STrond Myklebust 87786d80f97STrond Myklebust /** 8788dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8798dd37758STrond Myklebust * @req: pointer to a struct nfs_page 880ea2cf228SFred Isaman * @dst: commit list head 881ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8828dd37758STrond Myklebust * 883ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8848dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8858dd37758STrond Myklebust * the MM page stats. 8868dd37758STrond Myklebust * 887ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8888dd37758STrond Myklebust * holding the nfs_page lock. 8898dd37758STrond Myklebust */ 8908dd37758STrond Myklebust void 8916272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8928dd37758STrond Myklebust { 893fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 8946272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 895fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 89667911c8fSAnna Schumaker if (req->wb_page) 89786d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8988dd37758STrond Myklebust } 8998dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 9008dd37758STrond Myklebust 9018dd37758STrond Myklebust /** 9028dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 9038dd37758STrond Myklebust * @req: pointer to a nfs_page 904ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 9058dd37758STrond Myklebust * 906ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 9078dd37758STrond Myklebust * number of outstanding requests requiring a commit 9088dd37758STrond Myklebust * It does not update the MM page stats. 9098dd37758STrond Myklebust * 910ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9118dd37758STrond Myklebust */ 9128dd37758STrond Myklebust void 913ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 914ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9158dd37758STrond Myklebust { 9168dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9178dd37758STrond Myklebust return; 9188dd37758STrond Myklebust nfs_list_remove_request(req); 919ea2cf228SFred Isaman cinfo->mds->ncommit--; 9208dd37758STrond Myklebust } 9218dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9228dd37758STrond Myklebust 923ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 924ea2cf228SFred Isaman struct inode *inode) 925ea2cf228SFred Isaman { 926fe238e60SDave Wysochanski cinfo->inode = inode; 927ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 928ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 929b359f9d0SFred Isaman cinfo->dreq = NULL; 930f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 931ea2cf228SFred Isaman } 932ea2cf228SFred Isaman 933ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 934ea2cf228SFred Isaman struct inode *inode, 935ea2cf228SFred Isaman struct nfs_direct_req *dreq) 936ea2cf228SFred Isaman { 9371763da12SFred Isaman if (dreq) 9381763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9391763da12SFred Isaman else 940ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 941ea2cf228SFred Isaman } 942ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9438dd37758STrond Myklebust 9441da177e4SLinus Torvalds /* 9451da177e4SLinus Torvalds * Add a request to the inode's commit list. 9461da177e4SLinus Torvalds */ 9471763da12SFred Isaman void 948ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 949b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9501da177e4SLinus Torvalds { 951b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9528dd37758STrond Myklebust return; 9536272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9541da177e4SLinus Torvalds } 9558e821cadSTrond Myklebust 956d6d6dc7cSFred Isaman static void 957d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 958e468bae9STrond Myklebust { 95911fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 96093f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 96193f78d88STejun Heo WB_RECLAIMABLE); 962e468bae9STrond Myklebust } 963d6d6dc7cSFred Isaman 964b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 9658dd37758STrond Myklebust static void 966d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 967d6d6dc7cSFred Isaman { 9688dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9692b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 970ea2cf228SFred Isaman struct nfs_commit_info cinfo; 971d6d6dc7cSFred Isaman 972ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 973b5bab9bfSTrond Myklebust spin_lock(&inode->i_lock); 974ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 975ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 976d6d6dc7cSFred Isaman } 977b5bab9bfSTrond Myklebust spin_unlock(&inode->i_lock); 9788dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9798dd37758STrond Myklebust } 980e468bae9STrond Myklebust } 981e468bae9STrond Myklebust 982d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9838e821cadSTrond Myklebust { 984c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 985d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 986c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9878e821cadSTrond Myklebust } 9888e821cadSTrond Myklebust 989919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 990919e3bd9STrond Myklebust { 991919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 992919e3bd9STrond Myklebust } 993919e3bd9STrond Myklebust 994061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9956c75dc0dSFred Isaman { 996ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9976c75dc0dSFred Isaman unsigned long bytes = 0; 9986c75dc0dSFred Isaman 9996c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 10006c75dc0dSFred Isaman goto out; 1001ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 10026c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 10036c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 10046c75dc0dSFred Isaman 10056c75dc0dSFred Isaman bytes += req->wb_bytes; 10066c75dc0dSFred Isaman nfs_list_remove_request(req); 10076c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 10086c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 1009d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 10106c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 10116c75dc0dSFred Isaman goto remove_req; 10126c75dc0dSFred Isaman } 1013c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 1014f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1015b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1016a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10176c75dc0dSFred Isaman goto next; 10186c75dc0dSFred Isaman } 10196c75dc0dSFred Isaman remove_req: 10206c75dc0dSFred Isaman nfs_inode_remove_request(req); 10216c75dc0dSFred Isaman next: 102220633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10233aff4ebbSTrond Myklebust nfs_release_request(req); 10246c75dc0dSFred Isaman } 10256c75dc0dSFred Isaman out: 1026919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10276c75dc0dSFred Isaman hdr->release(hdr); 10286c75dc0dSFred Isaman } 10296c75dc0dSFred Isaman 1030ce59515cSAnna Schumaker unsigned long 1031ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1032fb8a1f11STrond Myklebust { 1033ea2cf228SFred Isaman return cinfo->mds->ncommit; 1034fb8a1f11STrond Myklebust } 1035fb8a1f11STrond Myklebust 1036fe238e60SDave Wysochanski /* cinfo->inode->i_lock held by caller */ 10371763da12SFred Isaman int 1038ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1039ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1040d6d6dc7cSFred Isaman { 1041d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 1042d6d6dc7cSFred Isaman int ret = 0; 1043d6d6dc7cSFred Isaman 1044d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 10458dd37758STrond Myklebust if (!nfs_lock_request(req)) 10468dd37758STrond Myklebust continue; 10477ad84aa9STrond Myklebust kref_get(&req->wb_kref); 1048fe238e60SDave Wysochanski if (cond_resched_lock(&cinfo->inode->i_lock)) 10493b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 1050ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10518dd37758STrond Myklebust nfs_list_add_request(req, dst); 1052d6d6dc7cSFred Isaman ret++; 10531763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1054d6d6dc7cSFred Isaman break; 1055d6d6dc7cSFred Isaman } 1056d6d6dc7cSFred Isaman return ret; 1057d6d6dc7cSFred Isaman } 1058d6d6dc7cSFred Isaman 10591da177e4SLinus Torvalds /* 10601da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10611da177e4SLinus Torvalds * @inode: NFS inode to scan 1062ea2cf228SFred Isaman * @dst: mds destination list 1063ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10641da177e4SLinus Torvalds * 10651da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10661da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10671da177e4SLinus Torvalds */ 10681763da12SFred Isaman int 1069ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1070ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10711da177e4SLinus Torvalds { 1072d6d6dc7cSFred Isaman int ret = 0; 1073fb8a1f11STrond Myklebust 1074fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 1075ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 10768dd37758STrond Myklebust const int max = INT_MAX; 1077d6d6dc7cSFred Isaman 1078ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1079ea2cf228SFred Isaman cinfo, max); 1080ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1081d6d6dc7cSFred Isaman } 1082fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 1083ff778d02STrond Myklebust return ret; 10841da177e4SLinus Torvalds } 1085d6d6dc7cSFred Isaman 10861da177e4SLinus Torvalds /* 1087e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1088e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10891da177e4SLinus Torvalds * 1090e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1091e7d39069STrond Myklebust * to disk. 10921da177e4SLinus Torvalds */ 1093e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1094e7d39069STrond Myklebust struct page *page, 1095e7d39069STrond Myklebust unsigned int offset, 1096e7d39069STrond Myklebust unsigned int bytes) 10971da177e4SLinus Torvalds { 1098e7d39069STrond Myklebust struct nfs_page *req; 1099e7d39069STrond Myklebust unsigned int rqend; 1100e7d39069STrond Myklebust unsigned int end; 11011da177e4SLinus Torvalds int error; 1102277459d2STrond Myklebust 1103e7d39069STrond Myklebust end = offset + bytes; 1104e7d39069STrond Myklebust 1105f6032f21STrond Myklebust req = nfs_lock_and_join_requests(page); 1106f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1107f6032f21STrond Myklebust return req; 11082bfc6e56SWeston Andros Adamson 1109e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1110e7d39069STrond Myklebust /* 1111e7d39069STrond Myklebust * Tell the caller to flush out the request if 1112e7d39069STrond Myklebust * the offsets are non-contiguous. 1113e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1114e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1115e7d39069STrond Myklebust */ 1116f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1117e7d39069STrond Myklebust goto out_flushme; 1118e7d39069STrond Myklebust 11191da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11201da177e4SLinus Torvalds if (offset < req->wb_offset) { 11211da177e4SLinus Torvalds req->wb_offset = offset; 11221da177e4SLinus Torvalds req->wb_pgbase = offset; 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds if (end > rqend) 11251da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1126e7d39069STrond Myklebust else 1127e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1128e7d39069STrond Myklebust return req; 1129e7d39069STrond Myklebust out_flushme: 1130f6032f21STrond Myklebust /* 1131f6032f21STrond Myklebust * Note: we mark the request dirty here because 1132f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1133f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1134f6032f21STrond Myklebust */ 1135f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1136f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 1137e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1138f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1139e7d39069STrond Myklebust } 11401da177e4SLinus Torvalds 1141e7d39069STrond Myklebust /* 1142e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1143e7d39069STrond Myklebust * 1144e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1145e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1146e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1147e7d39069STrond Myklebust */ 1148e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1149e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1150e7d39069STrond Myklebust { 1151d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1152e7d39069STrond Myklebust struct nfs_page *req; 1153e7d39069STrond Myklebust 1154e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1155e7d39069STrond Myklebust if (req != NULL) 1156e7d39069STrond Myklebust goto out; 11572bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1158e7d39069STrond Myklebust if (IS_ERR(req)) 1159e7d39069STrond Myklebust goto out; 1160d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1161efc91ed0STrond Myklebust out: 116261e930a9STrond Myklebust return req; 11631da177e4SLinus Torvalds } 11641da177e4SLinus Torvalds 1165e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1166e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1167e7d39069STrond Myklebust { 1168e7d39069STrond Myklebust struct nfs_page *req; 1169e7d39069STrond Myklebust 1170e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1171e7d39069STrond Myklebust if (IS_ERR(req)) 1172e7d39069STrond Myklebust return PTR_ERR(req); 1173e7d39069STrond Myklebust /* Update file length */ 1174e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1175d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1176a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11771d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1178e7d39069STrond Myklebust return 0; 1179e7d39069STrond Myklebust } 1180e7d39069STrond Myklebust 11811da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11821da177e4SLinus Torvalds { 1183cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11842a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1185bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11861da177e4SLinus Torvalds struct nfs_page *req; 11871a54533eSTrond Myklebust int do_flush, status; 11881da177e4SLinus Torvalds /* 11891da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11901da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11911da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11921da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11931da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11941da177e4SLinus Torvalds * dropped page. 11951da177e4SLinus Torvalds */ 11961a54533eSTrond Myklebust do { 119784d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11981a54533eSTrond Myklebust if (req == NULL) 11991a54533eSTrond Myklebust return 0; 12002a369153STrond Myklebust l_ctx = req->wb_lock_context; 1201138a2935STrond Myklebust do_flush = req->wb_page != page || 1202138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 1203bd61e0a9SJeff Layton if (l_ctx && flctx && 1204bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1205bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1206d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12075263e31eSJeff Layton } 12081da177e4SLinus Torvalds nfs_release_request(req); 12091a54533eSTrond Myklebust if (!do_flush) 12101a54533eSTrond Myklebust return 0; 1211d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12121a54533eSTrond Myklebust } while (status == 0); 12131a54533eSTrond Myklebust return status; 12141da177e4SLinus Torvalds } 12151da177e4SLinus Torvalds 12161da177e4SLinus Torvalds /* 1217dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1218dc24826bSAndy Adamson * expire soon. 1219dc24826bSAndy Adamson * 1220dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1221dc24826bSAndy Adamson * 1222dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1223dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1224dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1225dc24826bSAndy Adamson */ 1226dc24826bSAndy Adamson int 1227dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1228dc24826bSAndy Adamson { 1229dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1230dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1231dc24826bSAndy Adamson 1232dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1233dc24826bSAndy Adamson } 1234dc24826bSAndy Adamson 1235dc24826bSAndy Adamson /* 1236dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1237dc24826bSAndy Adamson */ 1238ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1239dc24826bSAndy Adamson { 1240ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1241ce52914eSScott Mayhew 1242ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1243dc24826bSAndy Adamson } 1244dc24826bSAndy Adamson 1245dc24826bSAndy Adamson /* 12465d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12475d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12485d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12495d47a356STrond Myklebust */ 12508d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12515d47a356STrond Myklebust { 1252d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1253d529ef83SJeff Layton 12548d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12558d197a56STrond Myklebust goto out; 125618dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1257d529ef83SJeff Layton return false; 12584db72b40SJeff Layton smp_rmb(); 1259d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12608d197a56STrond Myklebust return false; 12618d197a56STrond Myklebust out: 126218dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 126318dd78c4SScott Mayhew return false; 12648d197a56STrond Myklebust return PageUptodate(page) != 0; 12655d47a356STrond Myklebust } 12665d47a356STrond Myklebust 12675263e31eSJeff Layton static bool 12685263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12695263e31eSJeff Layton { 12705263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12715263e31eSJeff Layton fl->fl_type == F_WRLCK; 12725263e31eSJeff Layton } 12735263e31eSJeff Layton 1274c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1275c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1276c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1277c7559663SScott Mayhew * inefficiencies. 1278c7559663SScott Mayhew * 1279263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1280263b4509SScott Mayhew * of the checks. 1281c7559663SScott Mayhew */ 1282c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1283c7559663SScott Mayhew { 12845263e31eSJeff Layton int ret; 12855263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12865263e31eSJeff Layton struct file_lock *fl; 12875263e31eSJeff Layton 1288c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1289c7559663SScott Mayhew return 0; 1290263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1291263b4509SScott Mayhew return 0; 1292c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1293c7559663SScott Mayhew return 1; 1294bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1295bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 12968fa4592aSTrond Myklebust return 1; 12975263e31eSJeff Layton 12985263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12995263e31eSJeff Layton ret = 0; 13006109c850SJeff Layton spin_lock(&flctx->flc_lock); 1301bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1302bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1303bd61e0a9SJeff Layton fl_list); 1304bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13055263e31eSJeff Layton ret = 1; 1306bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13075263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13085263e31eSJeff Layton fl_list); 13095263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13105263e31eSJeff Layton ret = 1; 13115263e31eSJeff Layton } 13126109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13135263e31eSJeff Layton return ret; 1314c7559663SScott Mayhew } 1315c7559663SScott Mayhew 13165d47a356STrond Myklebust /* 13171da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13181da177e4SLinus Torvalds * 13191da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13201da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13211da177e4SLinus Torvalds */ 13221da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13231da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13241da177e4SLinus Torvalds { 1325cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1326d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 13271da177e4SLinus Torvalds int status = 0; 13281da177e4SLinus Torvalds 132991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 133091d5b470SChuck Lever 13316de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13326de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13331da177e4SLinus Torvalds 1334149a4fddSBenjamin Coddington if (!count) 1335149a4fddSBenjamin Coddington goto out; 1336149a4fddSBenjamin Coddington 1337c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 133849a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13391da177e4SLinus Torvalds offset = 0; 13401da177e4SLinus Torvalds } 13411da177e4SLinus Torvalds 1342e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 134303fa9e84STrond Myklebust if (status < 0) 134403fa9e84STrond Myklebust nfs_set_pageerror(page); 134559b7c05fSTrond Myklebust else 134659b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1347149a4fddSBenjamin Coddington out: 134848186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13491da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13501da177e4SLinus Torvalds return status; 13511da177e4SLinus Torvalds } 13521da177e4SLinus Torvalds 13533ff7576dSTrond Myklebust static int flush_task_priority(int how) 13541da177e4SLinus Torvalds { 13551da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13561da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13571da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13581da177e4SLinus Torvalds case FLUSH_LOWPRI: 13591da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13601da177e4SLinus Torvalds } 13611da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13621da177e4SLinus Torvalds } 13631da177e4SLinus Torvalds 1364d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1365d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1366abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13671ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13681da177e4SLinus Torvalds { 13693ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13701da177e4SLinus Torvalds 13711ed26f33SAnna Schumaker task_setup_data->priority = priority; 1372abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1373d138d5d1SAndy Adamson 1374abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1375d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1376275acaafSTrond Myklebust } 1377275acaafSTrond Myklebust 13786d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13796d884e8fSFred * call this on each, which will prepare them to be retried on next 13806d884e8fSFred * writeback using standard nfs. 13816d884e8fSFred */ 13826d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13836d884e8fSFred { 13846d884e8fSFred nfs_mark_request_dirty(req); 1385c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 138620633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13873aff4ebbSTrond Myklebust nfs_release_request(req); 13886d884e8fSFred } 13896d884e8fSFred 1390061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13916c75dc0dSFred Isaman { 13926c75dc0dSFred Isaman struct nfs_page *req; 13936c75dc0dSFred Isaman 13946c75dc0dSFred Isaman while (!list_empty(head)) { 13956c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13966c75dc0dSFred Isaman nfs_list_remove_request(req); 13976c75dc0dSFred Isaman nfs_redirty_request(req); 13986c75dc0dSFred Isaman } 13996c75dc0dSFred Isaman } 14006c75dc0dSFred Isaman 1401dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1402dc602dd7STrond Myklebust { 1403dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1404dc602dd7STrond Myklebust } 1405dc602dd7STrond Myklebust 1406061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1407919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1408061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1409061ae2edSFred Isaman .completion = nfs_write_completion, 1410dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1411061ae2edSFred Isaman }; 1412061ae2edSFred Isaman 141357208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1414a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1415061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14161751c363STrond Myklebust { 1417a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 141841d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1419a20c93e3SChristoph Hellwig 1420a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1421a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1422a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1423a20c93e3SChristoph Hellwig #endif 14244a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 1425fbe77c30SBenjamin Coddington server->wsize, ioflags, GFP_NOIO); 14261751c363STrond Myklebust } 1427ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14281751c363STrond Myklebust 1429dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1430dce81290STrond Myklebust { 1431a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1432a7d42ddbSWeston Andros Adamson 14336f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14346f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14356f29b9bbSKinglong Mee 143641d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1437a7d42ddbSWeston Andros Adamson 1438a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1439a7d42ddbSWeston Andros Adamson 1440a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1441a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1442dce81290STrond Myklebust } 14431f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1444dce81290STrond Myklebust 14451da177e4SLinus Torvalds 14460b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14470b7c0153SFred Isaman { 14480b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14490b7c0153SFred Isaman 14500b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14510b7c0153SFred Isaman } 14520b7c0153SFred Isaman 14531f2edbe3STrond Myklebust /* 14541f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14551f2edbe3STrond Myklebust */ 14561f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14571f2edbe3STrond Myklebust { 14581f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14591f2edbe3STrond Myklebust int kill = 0; 1460788e7a89STrond Myklebust 14611f2edbe3STrond Myklebust /* suid always must be killed */ 14621f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14631f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14641f2edbe3STrond Myklebust 14651f2edbe3STrond Myklebust /* 14661f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14671f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14681f2edbe3STrond Myklebust */ 14691f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14701f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14711f2edbe3STrond Myklebust 14721f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14731f2edbe3STrond Myklebust return kill; 14741f2edbe3STrond Myklebust 14751f2edbe3STrond Myklebust return 0; 14761f2edbe3STrond Myklebust } 1477788e7a89STrond Myklebust 1478a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1479a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1480a08a8cd3STrond Myklebust { 1481a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1482a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14832b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1484a08a8cd3STrond Myklebust 1485a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14862b83d3deSTrond Myklebust fattr->size = size; 14872b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 14882b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1489a08a8cd3STrond Myklebust return; 14902b83d3deSTrond Myklebust } 14912b83d3deSTrond Myklebust if (size != fattr->size) 1492a08a8cd3STrond Myklebust return; 1493a08a8cd3STrond Myklebust /* Set attribute barrier */ 1494a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 14952b83d3deSTrond Myklebust /* ...and update size */ 14962b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1497a08a8cd3STrond Myklebust } 1498a08a8cd3STrond Myklebust 1499a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1500a08a8cd3STrond Myklebust { 15012b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1502a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1503a08a8cd3STrond Myklebust 1504a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1505a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1506a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1507a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1508a08a8cd3STrond Myklebust } 1509a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1510a08a8cd3STrond Myklebust 15111da177e4SLinus Torvalds /* 15121da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15131da177e4SLinus Torvalds */ 1514d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1515d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15160eecb214SAnna Schumaker struct inode *inode) 15171da177e4SLinus Torvalds { 1518788e7a89STrond Myklebust int status; 15191da177e4SLinus Torvalds 1520f551e44fSChuck Lever /* 1521f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1522f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1523f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1524f551e44fSChuck Lever * another writer had changed the file, but some applications 1525f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1526f551e44fSChuck Lever */ 1527d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1528788e7a89STrond Myklebust if (status != 0) 15290eecb214SAnna Schumaker return status; 1530d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 153191d5b470SChuck Lever 1532d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1533d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15341da177e4SLinus Torvalds /* We tried a write call, but the server did not 15351da177e4SLinus Torvalds * commit data to stable storage even though we 15361da177e4SLinus Torvalds * requested it. 15371da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15381da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15391da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15401da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15411da177e4SLinus Torvalds */ 15421da177e4SLinus Torvalds static unsigned long complain; 15431da177e4SLinus Torvalds 1544a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15451da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15461da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15471da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1548cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1549d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15501da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15511da177e4SLinus Torvalds } 15521da177e4SLinus Torvalds } 15531f2edbe3STrond Myklebust 15541f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15551f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15561f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15570eecb214SAnna Schumaker return 0; 15580eecb214SAnna Schumaker } 15590eecb214SAnna Schumaker 15600eecb214SAnna Schumaker /* 15610eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15620eecb214SAnna Schumaker */ 1563d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1564d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15650eecb214SAnna Schumaker { 1566d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1567d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15681f2edbe3STrond Myklebust 15691f2edbe3STrond Myklebust if (resp->count < argp->count) { 15701da177e4SLinus Torvalds static unsigned long complain; 15711da177e4SLinus Torvalds 15726c75dc0dSFred Isaman /* This a short write! */ 1573d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 157491d5b470SChuck Lever 15751da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15766c75dc0dSFred Isaman if (resp->count == 0) { 15776c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15786c75dc0dSFred Isaman printk(KERN_WARNING 15796c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 15806c75dc0dSFred Isaman argp->count); 15816c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15826c75dc0dSFred Isaman } 1583d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15846c75dc0dSFred Isaman task->tk_status = -EIO; 15856c75dc0dSFred Isaman return; 15866c75dc0dSFred Isaman } 1587f8417b48SKinglong Mee 1588f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1589f8417b48SKinglong Mee if (!task->tk_ops) { 1590f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1591f8417b48SKinglong Mee return; 1592f8417b48SKinglong Mee } 1593f8417b48SKinglong Mee 15941da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 15951da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 15961da177e4SLinus Torvalds /* Resend from where the server left off */ 1597d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 15981da177e4SLinus Torvalds argp->offset += resp->count; 15991da177e4SLinus Torvalds argp->pgbase += resp->count; 16001da177e4SLinus Torvalds argp->count -= resp->count; 16011da177e4SLinus Torvalds } else { 16021da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16031da177e4SLinus Torvalds * headaches in the case of a server crash. 16041da177e4SLinus Torvalds */ 16051da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16061da177e4SLinus Torvalds } 1607d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16081da177e4SLinus Torvalds } 16091da177e4SLinus Torvalds } 16101da177e4SLinus Torvalds 1611af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 161271d0a611STrond Myklebust { 1613af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1614af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1615af7cf057STrond Myklebust } 1616af7cf057STrond Myklebust 1617af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1618af7cf057STrond Myklebust { 1619af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1620af7cf057STrond Myklebust } 1621af7cf057STrond Myklebust 1622af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1623af7cf057STrond Myklebust { 1624af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1625af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 162671d0a611STrond Myklebust } 162771d0a611STrond Myklebust 16280b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16291da177e4SLinus Torvalds { 16300b7c0153SFred Isaman put_nfs_open_context(data->context); 16310b7c0153SFred Isaman nfs_commit_free(data); 16321da177e4SLinus Torvalds } 1633e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16341da177e4SLinus Torvalds 16350b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1636c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16379ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16389f0ec176SAndy Adamson int how, int flags) 16391da177e4SLinus Torvalds { 164007737691STrond Myklebust struct rpc_task *task; 16419ace33cdSFred Isaman int priority = flush_task_priority(how); 1642bdc7f021STrond Myklebust struct rpc_message msg = { 1643bdc7f021STrond Myklebust .rpc_argp = &data->args, 1644bdc7f021STrond Myklebust .rpc_resp = &data->res, 16459ace33cdSFred Isaman .rpc_cred = data->cred, 1646bdc7f021STrond Myklebust }; 164784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 164807737691STrond Myklebust .task = &data->task, 16499ace33cdSFred Isaman .rpc_client = clnt, 1650bdc7f021STrond Myklebust .rpc_message = &msg, 16519ace33cdSFred Isaman .callback_ops = call_ops, 165284115e1cSTrond Myklebust .callback_data = data, 1653101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16549f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16553ff7576dSTrond Myklebust .priority = priority, 165684115e1cSTrond Myklebust }; 1657788e7a89STrond Myklebust /* Set up the initial task struct. */ 1658c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16591da177e4SLinus Torvalds 1660b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1661bdc7f021STrond Myklebust 16628c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16638c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16648c21c62cSWeston Andros Adamson 166507737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1666dbae4c73STrond Myklebust if (IS_ERR(task)) 1667dbae4c73STrond Myklebust return PTR_ERR(task); 1668d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1669d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 167007737691STrond Myklebust rpc_put_task(task); 1671dbae4c73STrond Myklebust return 0; 16721da177e4SLinus Torvalds } 1673e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16741da177e4SLinus Torvalds 1675378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1676378520b8SPeng Tao { 1677378520b8SPeng Tao loff_t lwb = 0; 1678378520b8SPeng Tao struct nfs_page *req; 1679378520b8SPeng Tao 1680378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1681378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1682378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1683378520b8SPeng Tao 1684378520b8SPeng Tao return lwb; 1685378520b8SPeng Tao } 1686378520b8SPeng Tao 16871da177e4SLinus Torvalds /* 16889ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16899ace33cdSFred Isaman */ 16900b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1691988b6dceSFred Isaman struct list_head *head, 1692f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1693f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16949ace33cdSFred Isaman { 16959ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16962b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16979ace33cdSFred Isaman 16989ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16999ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17009ace33cdSFred Isaman 17019ace33cdSFred Isaman list_splice_init(head, &data->pages); 17029ace33cdSFred Isaman 17039ace33cdSFred Isaman data->inode = inode; 17049ace33cdSFred Isaman data->cred = first->wb_context->cred; 1705988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1706378520b8SPeng Tao /* only set lwb for pnfs commit */ 1707378520b8SPeng Tao if (lseg) 1708378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17099ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1710f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1711b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17129ace33cdSFred Isaman 17139ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17149ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17159ace33cdSFred Isaman data->args.offset = 0; 17169ace33cdSFred Isaman data->args.count = 0; 17170b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 17189ace33cdSFred Isaman data->res.fattr = &data->fattr; 17199ace33cdSFred Isaman data->res.verf = &data->verf; 17209ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17219ace33cdSFred Isaman } 1722e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17239ace33cdSFred Isaman 1724e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1725ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1726b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1727b57ff130SWeston Andros Adamson u32 ds_commit_idx) 172864bfeb49SFred Isaman { 172964bfeb49SFred Isaman struct nfs_page *req; 173064bfeb49SFred Isaman 173164bfeb49SFred Isaman while (!list_empty(page_list)) { 173264bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 173364bfeb49SFred Isaman nfs_list_remove_request(req); 1734b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1735487b9b8aSTom Haynes if (!cinfo->dreq) 1736487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17371d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 173864bfeb49SFred Isaman } 173964bfeb49SFred Isaman } 1740e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 174164bfeb49SFred Isaman 1742b20135d0STrond Myklebust static void 1743b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1744b20135d0STrond Myklebust struct nfs_page *req) 1745b20135d0STrond Myklebust { 1746b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1747b20135d0STrond Myklebust } 1748b20135d0STrond Myklebust 17499ace33cdSFred Isaman /* 17501da177e4SLinus Torvalds * Commit dirty pages 17511da177e4SLinus Torvalds */ 17521da177e4SLinus Torvalds static int 1753ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1754ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17551da177e4SLinus Torvalds { 17560b7c0153SFred Isaman struct nfs_commit_data *data; 17571da177e4SLinus Torvalds 1758ade8febdSWeston Andros Adamson /* another commit raced with us */ 1759ade8febdSWeston Andros Adamson if (list_empty(head)) 1760ade8febdSWeston Andros Adamson return 0; 1761ade8febdSWeston Andros Adamson 1762518662e0SNeilBrown data = nfs_commitdata_alloc(true); 17631da177e4SLinus Torvalds 17641da177e4SLinus Torvalds /* Set up the argument struct */ 1765f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1766f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1767c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1768c36aae9aSPeng Tao data->mds_ops, how, 0); 17691da177e4SLinus Torvalds } 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds /* 17721da177e4SLinus Torvalds * COMMIT call returned 17731da177e4SLinus Torvalds */ 1774788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17751da177e4SLinus Torvalds { 17760b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17771da177e4SLinus Torvalds 1778a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17791da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17801da177e4SLinus Torvalds 1781788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1782c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1783c9d8f89dSTrond Myklebust } 1784c9d8f89dSTrond Myklebust 1785f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1786c9d8f89dSTrond Myklebust { 1787c9d8f89dSTrond Myklebust struct nfs_page *req; 1788c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1789f453a54aSFred Isaman struct nfs_commit_info cinfo; 1790353db796SNeilBrown struct nfs_server *nfss; 1791788e7a89STrond Myklebust 17921da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17931da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17941da177e4SLinus Torvalds nfs_list_remove_request(req); 179567911c8fSAnna Schumaker if (req->wb_page) 1796d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17971da177e4SLinus Torvalds 17981e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17993d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 18002b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 18011da177e4SLinus Torvalds req->wb_bytes, 18021da177e4SLinus Torvalds (long long)req_offset(req)); 1803c9d8f89dSTrond Myklebust if (status < 0) { 1804c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 180538a33101SKinglong Mee if (req->wb_page) 18061da177e4SLinus Torvalds nfs_inode_remove_request(req); 1807ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18081da177e4SLinus Torvalds goto next; 18091da177e4SLinus Torvalds } 18101da177e4SLinus Torvalds 18111da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18121da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18138fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18141da177e4SLinus Torvalds /* We have a match */ 181538a33101SKinglong Mee if (req->wb_page) 18161da177e4SLinus Torvalds nfs_inode_remove_request(req); 1817ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18181da177e4SLinus Torvalds goto next; 18191da177e4SLinus Torvalds } 18201da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1821ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18226d884e8fSFred nfs_mark_request_dirty(req); 182305990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18241da177e4SLinus Torvalds next: 18251d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18261da177e4SLinus Torvalds } 1827353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1828353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18290db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1830353db796SNeilBrown 1831f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1832af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18335917ce84SFred Isaman } 18345917ce84SFred Isaman 18355917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18365917ce84SFred Isaman { 18370b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18385917ce84SFred Isaman 1839f453a54aSFred Isaman data->completion_ops->completion(data); 1840c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18411da177e4SLinus Torvalds } 1842788e7a89STrond Myklebust 1843788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18440b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1845788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1846788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1847788e7a89STrond Myklebust }; 18481da177e4SLinus Torvalds 1849f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1850f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1851b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1852f453a54aSFred Isaman }; 1853f453a54aSFred Isaman 18541763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1855ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 185684c53ab5SFred Isaman { 185784c53ab5SFred Isaman int status; 185884c53ab5SFred Isaman 1859ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 186084c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1861ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 186284c53ab5SFred Isaman return status; 186384c53ab5SFred Isaman } 186484c53ab5SFred Isaman 1865b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18661da177e4SLinus Torvalds { 18671da177e4SLinus Torvalds LIST_HEAD(head); 1868ea2cf228SFred Isaman struct nfs_commit_info cinfo; 186971d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1870af7cf057STrond Myklebust int error = 0; 1871b8413f98STrond Myklebust int res; 18721da177e4SLinus Torvalds 1873ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1874af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1875ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1876af7cf057STrond Myklebust if (res) 1877ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1878af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18791da177e4SLinus Torvalds if (error < 0) 1880af7cf057STrond Myklebust goto out_error; 1881b8413f98STrond Myklebust if (!may_wait) 1882b8413f98STrond Myklebust goto out_mark_dirty; 1883af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1884b8413f98STrond Myklebust if (error < 0) 1885b8413f98STrond Myklebust return error; 1886c5efa5fcSTrond Myklebust return res; 1887af7cf057STrond Myklebust out_error: 1888af7cf057STrond Myklebust res = error; 1889c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1890c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1891c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1892c5efa5fcSTrond Myklebust * that the data is on the disk. 1893c5efa5fcSTrond Myklebust */ 1894c5efa5fcSTrond Myklebust out_mark_dirty: 1895c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18961da177e4SLinus Torvalds return res; 18971da177e4SLinus Torvalds } 1898b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 18998fc795f7STrond Myklebust 1900ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19018fc795f7STrond Myklebust { 1902420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1903420e3646STrond Myklebust int flags = FLUSH_SYNC; 1904420e3646STrond Myklebust int ret = 0; 19058fc795f7STrond Myklebust 19063236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1907ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 19083236c3e1SJeff Layton return ret; 19093236c3e1SJeff Layton 1910a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1911a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1912a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1913420e3646STrond Myklebust */ 19141a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1915420e3646STrond Myklebust goto out_mark_dirty; 1916420e3646STrond Myklebust 1917a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1918420e3646STrond Myklebust flags = 0; 1919a00dd6c0SJeff Layton } 1920a00dd6c0SJeff Layton 1921420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1922420e3646STrond Myklebust if (ret >= 0) { 1923420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1924420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1925420e3646STrond Myklebust wbc->nr_to_write -= ret; 1926420e3646STrond Myklebust else 1927420e3646STrond Myklebust wbc->nr_to_write = 0; 1928420e3646STrond Myklebust } 19298fc795f7STrond Myklebust return 0; 1930420e3646STrond Myklebust } 1931420e3646STrond Myklebust out_mark_dirty: 19328fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19338fc795f7STrond Myklebust return ret; 19348fc795f7STrond Myklebust } 193589d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1936863a3c6cSAndy Adamson 1937acdc53b2STrond Myklebust /* 1938837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1939837bb1d7STrond Myklebust * 1940837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1941837bb1d7STrond Myklebust * client. 1942837bb1d7STrond Myklebust */ 1943837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1944837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1945837bb1d7STrond Myklebust { 1946837bb1d7STrond Myklebust int ret; 1947837bb1d7STrond Myklebust 1948837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1949837bb1d7STrond Myklebust if (ret == 0) 1950837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1951837bb1d7STrond Myklebust return ret; 1952837bb1d7STrond Myklebust } 1953837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1954837bb1d7STrond Myklebust 1955837bb1d7STrond Myklebust /* 1956acdc53b2STrond Myklebust * flush the inode to disk. 1957acdc53b2STrond Myklebust */ 1958acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 195934901f70STrond Myklebust { 1960f4ce1299STrond Myklebust int ret; 196134901f70STrond Myklebust 1962f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1963f4ce1299STrond Myklebust 19645bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19656b196875SChuck Lever if (ret) 19666b196875SChuck Lever goto out; 19675bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19686b196875SChuck Lever if (ret < 0) 19696b196875SChuck Lever goto out; 19705bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19716b196875SChuck Lever ret = 0; 1972f4ce1299STrond Myklebust 19736b196875SChuck Lever out: 1974f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1975f4ce1299STrond Myklebust return ret; 19761c75950bSTrond Myklebust } 1977ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19781c75950bSTrond Myklebust 19791b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19801b3b4a1aSTrond Myklebust { 19811b3b4a1aSTrond Myklebust struct nfs_page *req; 19821b3b4a1aSTrond Myklebust int ret = 0; 19831b3b4a1aSTrond Myklebust 1984ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19853e217045SWeston Andros Adamson 19863e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 19873e217045SWeston Andros Adamson * request */ 19886d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 19893e217045SWeston Andros Adamson 19903e217045SWeston Andros Adamson if (IS_ERR(req)) { 19913e217045SWeston Andros Adamson ret = PTR_ERR(req); 19923e217045SWeston Andros Adamson } else if (req) { 19933e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 19943e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 19953e217045SWeston Andros Adamson * request from the inode / page_private pointer and 19963e217045SWeston Andros Adamson * release it */ 19971b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 19981d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19991b3b4a1aSTrond Myklebust } 20003e217045SWeston Andros Adamson 20011b3b4a1aSTrond Myklebust return ret; 20021b3b4a1aSTrond Myklebust } 20031b3b4a1aSTrond Myklebust 20041c75950bSTrond Myklebust /* 20051c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 20061c75950bSTrond Myklebust */ 2007c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 20081c75950bSTrond Myklebust { 200929418aa4SMel Gorman loff_t range_start = page_file_offset(page); 201009cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 20117f2f12d9STrond Myklebust struct writeback_control wbc = { 20127f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20137f2f12d9STrond Myklebust .nr_to_write = 0, 20147f2f12d9STrond Myklebust .range_start = range_start, 20157f2f12d9STrond Myklebust .range_end = range_end, 20167f2f12d9STrond Myklebust }; 20177f2f12d9STrond Myklebust int ret; 20187f2f12d9STrond Myklebust 2019f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2020f4ce1299STrond Myklebust 20210522f6adSTrond Myklebust for (;;) { 2022ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20237f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2024c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20257f2f12d9STrond Myklebust if (ret < 0) 20267f2f12d9STrond Myklebust goto out_error; 20270522f6adSTrond Myklebust continue; 20287f2f12d9STrond Myklebust } 2029f4ce1299STrond Myklebust ret = 0; 20300522f6adSTrond Myklebust if (!PagePrivate(page)) 20310522f6adSTrond Myklebust break; 20320522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20337f2f12d9STrond Myklebust if (ret < 0) 20347f2f12d9STrond Myklebust goto out_error; 20357f2f12d9STrond Myklebust } 20367f2f12d9STrond Myklebust out_error: 2037f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20387f2f12d9STrond Myklebust return ret; 20391c75950bSTrond Myklebust } 20401c75950bSTrond Myklebust 2041074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2042074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2043a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2044074cc1deSTrond Myklebust { 20452da95652SJeff Layton /* 20462da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20472da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20482da95652SJeff Layton * 20492da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20502da95652SJeff Layton * that we can safely release the inode reference while holding 20512da95652SJeff Layton * the page lock. 20522da95652SJeff Layton */ 20532da95652SJeff Layton if (PagePrivate(page)) 20542da95652SJeff Layton return -EBUSY; 2055074cc1deSTrond Myklebust 20568c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20578c209ce7SDavid Howells return -EBUSY; 2058074cc1deSTrond Myklebust 2059a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2060074cc1deSTrond Myklebust } 2061074cc1deSTrond Myklebust #endif 2062074cc1deSTrond Myklebust 2063f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20641da177e4SLinus Torvalds { 20651da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20661e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20671da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 206820c2df83SPaul Mundt NULL); 20691da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20701da177e4SLinus Torvalds return -ENOMEM; 20711da177e4SLinus Torvalds 207293d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20731da177e4SLinus Torvalds nfs_wdata_cachep); 20741da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20753dd4765fSJeff Layton goto out_destroy_write_cache; 20761da177e4SLinus Torvalds 20770b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20780b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20790b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20800b7c0153SFred Isaman NULL); 20810b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20823dd4765fSJeff Layton goto out_destroy_write_mempool; 20830b7c0153SFred Isaman 208493d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20854c100210SYanchuan Nian nfs_cdata_cachep); 20861da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 20873dd4765fSJeff Layton goto out_destroy_commit_cache; 20881da177e4SLinus Torvalds 208989a09141SPeter Zijlstra /* 209089a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 209189a09141SPeter Zijlstra * 209289a09141SPeter Zijlstra * 64MB: 8192k 209389a09141SPeter Zijlstra * 128MB: 11585k 209489a09141SPeter Zijlstra * 256MB: 16384k 209589a09141SPeter Zijlstra * 512MB: 23170k 209689a09141SPeter Zijlstra * 1GB: 32768k 209789a09141SPeter Zijlstra * 2GB: 46340k 209889a09141SPeter Zijlstra * 4GB: 65536k 209989a09141SPeter Zijlstra * 8GB: 92681k 210089a09141SPeter Zijlstra * 16GB: 131072k 210189a09141SPeter Zijlstra * 210289a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 210389a09141SPeter Zijlstra * Limit the default to 256M 210489a09141SPeter Zijlstra */ 210589a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 210689a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 210789a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 210889a09141SPeter Zijlstra 21091da177e4SLinus Torvalds return 0; 21103dd4765fSJeff Layton 21113dd4765fSJeff Layton out_destroy_commit_cache: 21123dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21133dd4765fSJeff Layton out_destroy_write_mempool: 21143dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21153dd4765fSJeff Layton out_destroy_write_cache: 21163dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21173dd4765fSJeff Layton return -ENOMEM; 21181da177e4SLinus Torvalds } 21191da177e4SLinus Torvalds 2120266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21211da177e4SLinus Torvalds { 21221da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21233dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21241da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21251a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21261da177e4SLinus Torvalds } 21271da177e4SLinus Torvalds 21284a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21294a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21304a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21310eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21320eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21331ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21344a0de55cSAnna Schumaker }; 2135