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> 261eb5d98fSJeff Layton #include <linux/iversion.h> 273fcfab16SAndrew Morton 287c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 29875bc3fbSTrond Myklebust #include <linux/sched/mm.h> 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds #include "delegation.h" 3249a70f27STrond Myklebust #include "internal.h" 3391d5b470SChuck Lever #include "iostat.h" 34def6ed7eSAndy Adamson #include "nfs4_fs.h" 35074cc1deSTrond Myklebust #include "fscache.h" 3694ad1c80SFred Isaman #include "pnfs.h" 371da177e4SLinus Torvalds 38f4ce1299STrond Myklebust #include "nfstrace.h" 39f4ce1299STrond Myklebust 401da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 411da177e4SLinus Torvalds 421da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 431da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 441da177e4SLinus Torvalds 45919e3bd9STrond Myklebust struct nfs_io_completion { 46919e3bd9STrond Myklebust void (*complete)(void *data); 47919e3bd9STrond Myklebust void *data; 48919e3bd9STrond Myklebust struct kref refcount; 49919e3bd9STrond Myklebust }; 50919e3bd9STrond Myklebust 511da177e4SLinus Torvalds /* 521da177e4SLinus Torvalds * Local function declarations 531da177e4SLinus Torvalds */ 54f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 55788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 56061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 57f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 584a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 59d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 6002d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 6102d1426cSWeston Andros Adamson struct inode *inode); 623a3908c8STrond Myklebust static struct nfs_page * 633a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 643a3908c8STrond Myklebust struct page *page); 651da177e4SLinus Torvalds 66e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 673feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 680b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 691da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 701da177e4SLinus Torvalds 71518662e0SNeilBrown struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) 721da177e4SLinus Torvalds { 73518662e0SNeilBrown struct nfs_commit_data *p; 7440859d7eSChuck Lever 75518662e0SNeilBrown if (never_fail) 76518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 77518662e0SNeilBrown else { 78518662e0SNeilBrown /* It is OK to do some reclaim, not no safe to wait 79518662e0SNeilBrown * for anything to be returned to the pool. 80518662e0SNeilBrown * mempool_alloc() cannot handle that particular combination, 81518662e0SNeilBrown * so we need two separate attempts. 82518662e0SNeilBrown */ 83518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); 84518662e0SNeilBrown if (!p) 85518662e0SNeilBrown p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | 86518662e0SNeilBrown __GFP_NOWARN | __GFP_NORETRY); 87518662e0SNeilBrown if (!p) 88518662e0SNeilBrown return NULL; 89518662e0SNeilBrown } 90518662e0SNeilBrown 911da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 921da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 931da177e4SLinus Torvalds return p; 941da177e4SLinus Torvalds } 95e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 961da177e4SLinus Torvalds 970b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 981da177e4SLinus Torvalds { 991da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 1001da177e4SLinus Torvalds } 101e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 1021da177e4SLinus Torvalds 1031e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 1043feb2d49STrond Myklebust { 1051e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 1063feb2d49STrond Myklebust 1073feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 108fbe77c30SBenjamin Coddington p->rw_mode = FMODE_WRITE; 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 150bd37d6fcSTrond Myklebust static struct nfs_page * 151bd37d6fcSTrond Myklebust nfs_page_private_request(struct page *page) 152bd37d6fcSTrond Myklebust { 153bd37d6fcSTrond Myklebust if (!PagePrivate(page)) 154bd37d6fcSTrond Myklebust return NULL; 155bd37d6fcSTrond Myklebust return (struct nfs_page *)page_private(page); 156bd37d6fcSTrond Myklebust } 157bd37d6fcSTrond Myklebust 15884d3a9a9SWeston Andros Adamson /* 15984d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 16084d3a9a9SWeston Andros Adamson * 16184d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 16284d3a9a9SWeston Andros Adamson * 16384d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 16484d3a9a9SWeston Andros Adamson */ 16529418aa4SMel Gorman static struct nfs_page * 166b30d2f04STrond Myklebust nfs_page_find_private_request(struct page *page) 167277459d2STrond Myklebust { 1684b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(page); 169bd37d6fcSTrond Myklebust struct nfs_page *req; 170277459d2STrond Myklebust 171b30d2f04STrond Myklebust if (!PagePrivate(page)) 172b30d2f04STrond Myklebust return NULL; 1734b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 174bd37d6fcSTrond Myklebust req = nfs_page_private_request(page); 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 } 1794b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 180b30d2f04STrond Myklebust return req; 181b30d2f04STrond Myklebust } 18229418aa4SMel Gorman 183b30d2f04STrond Myklebust static struct nfs_page * 184b30d2f04STrond Myklebust nfs_page_find_swap_request(struct page *page) 185b30d2f04STrond Myklebust { 186b30d2f04STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 187b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 188b30d2f04STrond Myklebust struct nfs_page *req = NULL; 189b30d2f04STrond Myklebust if (!PageSwapCache(page)) 190b30d2f04STrond Myklebust return NULL; 191e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 192b30d2f04STrond Myklebust if (PageSwapCache(page)) { 193b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 194b30d2f04STrond Myklebust page); 195b30d2f04STrond Myklebust if (req) { 196b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 197b30d2f04STrond Myklebust kref_get(&req->wb_kref); 198b30d2f04STrond Myklebust } 199b30d2f04STrond Myklebust } 200e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 201277459d2STrond Myklebust return req; 202277459d2STrond Myklebust } 203277459d2STrond Myklebust 20484d3a9a9SWeston Andros Adamson /* 20584d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 20684d3a9a9SWeston Andros Adamson * 20784d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 20884d3a9a9SWeston Andros Adamson */ 20984d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 210277459d2STrond Myklebust { 211b30d2f04STrond Myklebust struct nfs_page *req; 212277459d2STrond Myklebust 213b30d2f04STrond Myklebust req = nfs_page_find_private_request(page); 214b30d2f04STrond Myklebust if (!req) 215b30d2f04STrond Myklebust req = nfs_page_find_swap_request(page); 216277459d2STrond Myklebust return req; 217277459d2STrond Myklebust } 218277459d2STrond Myklebust 2191da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2201da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2211da177e4SLinus Torvalds { 222d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 223a3d01454STrond Myklebust loff_t end, i_size; 224a3d01454STrond Myklebust pgoff_t end_index; 2251da177e4SLinus Torvalds 226a3d01454STrond Myklebust spin_lock(&inode->i_lock); 227a3d01454STrond Myklebust i_size = i_size_read(inode); 22809cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2298cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 230a3d01454STrond Myklebust goto out; 231d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2321da177e4SLinus Torvalds if (i_size >= end) 233a3d01454STrond Myklebust goto out; 2341da177e4SLinus Torvalds i_size_write(inode, end); 235f6cdfa6dSTrond Myklebust NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 236a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 237a3d01454STrond Myklebust out: 238a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2391da177e4SLinus Torvalds } 2401da177e4SLinus Torvalds 241a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 242d2ceb7e5SBenjamin Coddington static void nfs_set_pageerror(struct address_space *mapping) 243a301b777STrond Myklebust { 244d2ceb7e5SBenjamin Coddington nfs_zap_mapping(mapping->host, mapping); 245a301b777STrond Myklebust } 246a301b777STrond Myklebust 2476fbda89bSTrond Myklebust static void nfs_mapping_set_error(struct page *page, int error) 2486fbda89bSTrond Myklebust { 2496fbda89bSTrond Myklebust SetPageError(page); 2506fbda89bSTrond Myklebust mapping_set_error(page_file_mapping(page), error); 2516fbda89bSTrond Myklebust } 2526fbda89bSTrond Myklebust 253d72ddcbaSWeston Andros Adamson /* 254d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 255d72ddcbaSWeston Andros Adamson * @head - head request of page group 256d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 257d72ddcbaSWeston Andros Adamson * 258d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 259d72ddcbaSWeston Andros Adamson * page offset @page_offset. 260d72ddcbaSWeston Andros Adamson * 261d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 262d72ddcbaSWeston Andros Adamson * match is found. 263d72ddcbaSWeston Andros Adamson * 264d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 265d72ddcbaSWeston Andros Adamson */ 266d72ddcbaSWeston Andros Adamson static struct nfs_page * 267d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 268d72ddcbaSWeston Andros Adamson { 269d72ddcbaSWeston Andros Adamson struct nfs_page *req; 270d72ddcbaSWeston Andros Adamson 271d72ddcbaSWeston Andros Adamson req = head; 272d72ddcbaSWeston Andros Adamson do { 273d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 274d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 275d72ddcbaSWeston Andros Adamson return req; 276d72ddcbaSWeston Andros Adamson 277d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 278d72ddcbaSWeston Andros Adamson } while (req != head); 279d72ddcbaSWeston Andros Adamson 280d72ddcbaSWeston Andros Adamson return NULL; 281d72ddcbaSWeston Andros Adamson } 282d72ddcbaSWeston Andros Adamson 283d72ddcbaSWeston Andros Adamson /* 284d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 285d72ddcbaSWeston Andros Adamson * @head - head request of page group 286d72ddcbaSWeston Andros Adamson * 287d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 288d72ddcbaSWeston Andros Adamson * returns false otherwise 289d72ddcbaSWeston Andros Adamson */ 290d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 291d72ddcbaSWeston Andros Adamson { 292d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 293d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 294d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 295d72ddcbaSWeston Andros Adamson 2961344b7eaSTrond Myklebust nfs_page_group_lock(req); 297d72ddcbaSWeston Andros Adamson 2987e8a30f8STrond Myklebust for (;;) { 299d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 3007e8a30f8STrond Myklebust if (!tmp) 3017e8a30f8STrond Myklebust break; 3027e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 303d72ddcbaSWeston Andros Adamson } 304d72ddcbaSWeston Andros Adamson 305d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3067e8a30f8STrond Myklebust return pos >= len; 307d72ddcbaSWeston Andros Adamson } 308d72ddcbaSWeston Andros Adamson 3091da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3101da177e4SLinus Torvalds * covers the full page. 3111da177e4SLinus Torvalds */ 312d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3131da177e4SLinus Torvalds { 314d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 3151da177e4SLinus Torvalds return; 316d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3171da177e4SLinus Torvalds return; 318d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds 3211da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3221da177e4SLinus Torvalds { 323e87b4c7aSNeilBrown int ret = 0; 324cca588d6STrond Myklebust 325e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 326e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 327e87b4c7aSNeilBrown return ret; 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds 3301da177e4SLinus Torvalds /* 33189a09141SPeter Zijlstra * NFS congestion control 33289a09141SPeter Zijlstra */ 33389a09141SPeter Zijlstra 33489a09141SPeter Zijlstra int nfs_congestion_kb; 33589a09141SPeter Zijlstra 33689a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 33789a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 33889a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 33989a09141SPeter Zijlstra 340deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 34189a09141SPeter Zijlstra { 3420db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3430db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3445a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3455a6d41b3STrond Myklebust 346deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 34789a09141SPeter Zijlstra 348277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3490db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3500db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 35189a09141SPeter Zijlstra } 35289a09141SPeter Zijlstra 35320633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 35489a09141SPeter Zijlstra { 35520633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 35689a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 35731a01f09STrond Myklebust bool is_done; 35889a09141SPeter Zijlstra 35931a01f09STrond Myklebust is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); 36031a01f09STrond Myklebust nfs_unlock_request(req); 36131a01f09STrond Myklebust if (!is_done) 36220633f04SWeston Andros Adamson return; 36320633f04SWeston Andros Adamson 36420633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 365c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3660db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 36789a09141SPeter Zijlstra } 36889a09141SPeter Zijlstra 369d4581383SWeston Andros Adamson /* 370d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 371d4581383SWeston Andros Adamson * 372d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 373d4581383SWeston Andros Adamson * 374d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 375d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 376d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 377d4581383SWeston Andros Adamson * 378b5bab9bfSTrond Myklebust * NOTE: this must be called holding page_group bit lock 379b5bab9bfSTrond Myklebust * which will be released before returning. 380d4581383SWeston Andros Adamson * 381d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 382d4581383SWeston Andros Adamson */ 38374a6d4b5STrond Myklebust static void 38474a6d4b5STrond Myklebust nfs_unroll_locks(struct inode *inode, struct nfs_page *head, 3856d17d653STrond Myklebust struct nfs_page *req) 386d4581383SWeston Andros Adamson { 387d4581383SWeston Andros Adamson struct nfs_page *tmp; 388e261f51fSTrond Myklebust 389d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 3905b2b5187STrond Myklebust for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { 3915b2b5187STrond Myklebust if (!kref_read(&tmp->wb_kref)) 3925b2b5187STrond Myklebust continue; 3935b2b5187STrond Myklebust nfs_unlock_and_release_request(tmp); 3945b2b5187STrond Myklebust } 395e261f51fSTrond Myklebust } 396d4581383SWeston Andros Adamson 397d4581383SWeston Andros Adamson /* 398d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 399d4581383SWeston Andros Adamson * 400d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 401d4581383SWeston Andros Adamson * @old_head - the old head of the list 402d4581383SWeston Andros Adamson * 403d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 404d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 405d4581383SWeston Andros Adamson * with a reference held by some other context. 406d4581383SWeston Andros Adamson */ 407d4581383SWeston Andros Adamson static void 408d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 409b66aaa8dSTrond Myklebust struct nfs_page *old_head, 410b66aaa8dSTrond Myklebust struct inode *inode) 411d4581383SWeston Andros Adamson { 412d4581383SWeston Andros Adamson while (destroy_list) { 413d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 414d4581383SWeston Andros Adamson 415d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 416d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 417d4581383SWeston Andros Adamson 418d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 419d4581383SWeston Andros Adamson 420d4581383SWeston Andros Adamson /* make sure old group is not used */ 421d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 422d4581383SWeston Andros Adamson 423902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 424902a4c00STrond Myklebust 4255b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4265b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4275b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 428902a4c00STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) 4295b2b5187STrond Myklebust nfs_free_request(subreq); 4305b2b5187STrond Myklebust continue; 4315b2b5187STrond Myklebust } 432d4581383SWeston Andros Adamson 4335b2b5187STrond Myklebust subreq->wb_head = subreq; 4345b2b5187STrond Myklebust 435b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 436d4581383SWeston Andros Adamson nfs_release_request(subreq); 437a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 438d4581383SWeston Andros Adamson } 4395b2b5187STrond Myklebust 4405b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4415b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4425b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 443d4581383SWeston Andros Adamson } 444d4581383SWeston Andros Adamson } 445d4581383SWeston Andros Adamson 446d4581383SWeston Andros Adamson /* 447d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 448d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 449d4581383SWeston Andros Adamson * operations for this page. 450d4581383SWeston Andros Adamson * 451d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 452d4581383SWeston Andros Adamson * 453d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 454d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 455d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 456d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 457d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 458d4581383SWeston Andros Adamson * 459d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 460d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 461d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 462d4581383SWeston Andros Adamson * error was encountered. 463d4581383SWeston Andros Adamson */ 464d4581383SWeston Andros Adamson static struct nfs_page * 4656d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 466d4581383SWeston Andros Adamson { 467d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 468d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 469d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 470d4581383SWeston Andros Adamson unsigned int total_bytes; 471d4581383SWeston Andros Adamson int ret; 472d4581383SWeston Andros Adamson 473d4581383SWeston Andros Adamson try_again: 474d4581383SWeston Andros Adamson /* 475d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 476d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 477d4581383SWeston Andros Adamson * until the head reference is released. 478d4581383SWeston Andros Adamson */ 479bd37d6fcSTrond Myklebust head = nfs_page_find_head_request(page); 480bd37d6fcSTrond Myklebust if (!head) 481d4581383SWeston Andros Adamson return NULL; 482d4581383SWeston Andros Adamson 483a0e265bcSTrond Myklebust /* lock the page head first in order to avoid an ABBA inefficiency */ 484a0e265bcSTrond Myklebust if (!nfs_lock_request(head)) { 485a0e265bcSTrond Myklebust ret = nfs_wait_on_request(head); 486a0e265bcSTrond Myklebust nfs_release_request(head); 487a0e265bcSTrond Myklebust if (ret < 0) 488a0e265bcSTrond Myklebust return ERR_PTR(ret); 489a0e265bcSTrond Myklebust goto try_again; 490a0e265bcSTrond Myklebust } 491bd37d6fcSTrond Myklebust 492bd37d6fcSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 493bd37d6fcSTrond Myklebust if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { 494bd37d6fcSTrond Myklebust nfs_unlock_and_release_request(head); 495bd37d6fcSTrond Myklebust goto try_again; 496bd37d6fcSTrond Myklebust } 4977c3af975SWeston Andros Adamson 4981344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 4990671d8f1SMarkus Elfring if (ret < 0) 5000671d8f1SMarkus Elfring goto release_request; 5017c3af975SWeston Andros Adamson 5027c3af975SWeston Andros Adamson /* lock each request in the page group */ 503a0e265bcSTrond Myklebust total_bytes = head->wb_bytes; 504a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 505a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 50674a6d4b5STrond Myklebust 5071bd5d6d0STrond Myklebust if (!kref_get_unless_zero(&subreq->wb_kref)) { 5081bd5d6d0STrond Myklebust if (subreq->wb_offset == head->wb_offset + total_bytes) 5091bd5d6d0STrond Myklebust total_bytes += subreq->wb_bytes; 5105b2b5187STrond Myklebust continue; 5111bd5d6d0STrond Myklebust } 5121bd5d6d0STrond Myklebust 51374a6d4b5STrond Myklebust while (!nfs_lock_request(subreq)) { 514b5bab9bfSTrond Myklebust /* 51574a6d4b5STrond Myklebust * Unlock page to allow nfs_page_group_sync_on_bit() 51674a6d4b5STrond Myklebust * to succeed 517b5bab9bfSTrond Myklebust */ 51874a6d4b5STrond Myklebust nfs_page_group_unlock(head); 51974a6d4b5STrond Myklebust ret = nfs_wait_on_request(subreq); 52074a6d4b5STrond Myklebust if (!ret) 5211344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 52274a6d4b5STrond Myklebust if (ret < 0) { 52374a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5245b2b5187STrond Myklebust nfs_release_request(subreq); 5250671d8f1SMarkus Elfring goto release_request; 526d4581383SWeston Andros Adamson } 52774a6d4b5STrond Myklebust } 528e14bebf6STrond Myklebust /* 529e14bebf6STrond Myklebust * Subrequests are always contiguous, non overlapping 530e14bebf6STrond Myklebust * and in order - but may be repeated (mirrored writes). 531e14bebf6STrond Myklebust */ 532e14bebf6STrond Myklebust if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 533e14bebf6STrond Myklebust /* keep track of how many bytes this group covers */ 534e14bebf6STrond Myklebust total_bytes += subreq->wb_bytes; 535e14bebf6STrond Myklebust } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 536e14bebf6STrond Myklebust ((subreq->wb_offset + subreq->wb_bytes) > 537e14bebf6STrond Myklebust (head->wb_offset + total_bytes)))) { 5388b77484fSTrond Myklebust nfs_page_group_unlock(head); 53974a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5405b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 5410671d8f1SMarkus Elfring ret = -EIO; 5420671d8f1SMarkus Elfring goto release_request; 543e14bebf6STrond Myklebust } 544a0e265bcSTrond Myklebust } 545d4581383SWeston Andros Adamson 546d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 547d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 548d4581383SWeston Andros Adamson subreq = head; 549d4581383SWeston Andros Adamson do { 550411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 551d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 552d4581383SWeston Andros Adamson } while (subreq != head); 553d4581383SWeston Andros Adamson 554d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 555d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 556d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 557d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 558d4581383SWeston Andros Adamson head->wb_this_page = head; 559d4581383SWeston Andros Adamson 560d4581383SWeston Andros Adamson /* change head request to cover whole range that 561d4581383SWeston Andros Adamson * the former page group covered */ 562d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 563d4581383SWeston Andros Adamson } 564d4581383SWeston Andros Adamson 565b66aaa8dSTrond Myklebust /* Postpone destruction of this request */ 566b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { 567b66aaa8dSTrond Myklebust set_bit(PG_INODE_REF, &head->wb_flags); 568b66aaa8dSTrond Myklebust kref_get(&head->wb_kref); 569a6b6d5b8STrond Myklebust atomic_long_inc(&NFS_I(inode)->nrequests); 570b66aaa8dSTrond Myklebust } 571b66aaa8dSTrond Myklebust 572d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 573d4581383SWeston Andros Adamson 574b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 575d4581383SWeston Andros Adamson 576b5bab9bfSTrond Myklebust /* Did we lose a race with nfs_inode_remove_request()? */ 577b5bab9bfSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) { 578b5bab9bfSTrond Myklebust nfs_unlock_and_release_request(head); 579b5bab9bfSTrond Myklebust return NULL; 580b5bab9bfSTrond Myklebust } 581b5bab9bfSTrond Myklebust 582bd37d6fcSTrond Myklebust /* still holds ref on head from nfs_page_find_head_request 583d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 584d4581383SWeston Andros Adamson return head; 5850671d8f1SMarkus Elfring 5860671d8f1SMarkus Elfring release_request: 5870671d8f1SMarkus Elfring nfs_unlock_and_release_request(head); 5880671d8f1SMarkus Elfring return ERR_PTR(ret); 589612c9384STrond Myklebust } 590074cc1deSTrond Myklebust 5916fbda89bSTrond Myklebust static void nfs_write_error(struct nfs_page *req, int error) 5920bcbf039SPeng Tao { 5936fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, error); 5940bcbf039SPeng Tao nfs_end_page_writeback(req); 5951f84ccdfSFred Isaman nfs_release_request(req); 5960bcbf039SPeng Tao } 5970bcbf039SPeng Tao 598a6598813STrond Myklebust static bool 599a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 600a6598813STrond Myklebust { 601a6598813STrond Myklebust switch (err) { 602a6598813STrond Myklebust case 0: 603a6598813STrond Myklebust case -ERESTARTSYS: 604a6598813STrond Myklebust case -EINTR: 605a6598813STrond Myklebust return false; 606a6598813STrond Myklebust } 607a6598813STrond Myklebust return nfs_error_is_fatal(err); 608e261f51fSTrond Myklebust } 609074cc1deSTrond Myklebust 610074cc1deSTrond Myklebust /* 611074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 612074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 613074cc1deSTrond Myklebust */ 614074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 6156d17d653STrond Myklebust struct page *page) 616074cc1deSTrond Myklebust { 6176fbda89bSTrond Myklebust struct address_space *mapping; 618074cc1deSTrond Myklebust struct nfs_page *req; 619074cc1deSTrond Myklebust int ret = 0; 620074cc1deSTrond Myklebust 6216d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 622074cc1deSTrond Myklebust if (!req) 623074cc1deSTrond Myklebust goto out; 624074cc1deSTrond Myklebust ret = PTR_ERR(req); 625074cc1deSTrond Myklebust if (IS_ERR(req)) 626074cc1deSTrond Myklebust goto out; 627074cc1deSTrond Myklebust 628deed85e7STrond Myklebust nfs_set_page_writeback(page); 629deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 630074cc1deSTrond Myklebust 631a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 6326fbda89bSTrond Myklebust ret = 0; 6336fbda89bSTrond Myklebust mapping = page_file_mapping(page); 6346fbda89bSTrond Myklebust if (test_bit(AS_ENOSPC, &mapping->flags) || 6356fbda89bSTrond Myklebust test_bit(AS_EIO, &mapping->flags)) 636a6598813STrond Myklebust goto out_launder; 637a6598813STrond Myklebust 638f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 639074cc1deSTrond Myklebust ret = pgio->pg_error; 6400bcbf039SPeng Tao /* 641c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6420bcbf039SPeng Tao */ 6430bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 644c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 645a6598813STrond Myklebust goto out_launder; 6468fc75bedSTrond Myklebust } else 6470bcbf039SPeng Tao ret = -EAGAIN; 6488fc75bedSTrond Myklebust nfs_redirty_request(req); 64940f90271STrond Myklebust } else 65040f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 65140f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 652074cc1deSTrond Myklebust out: 653074cc1deSTrond Myklebust return ret; 654a6598813STrond Myklebust out_launder: 6556fbda89bSTrond Myklebust nfs_write_error(req, ret); 65614bebe3cSTrond Myklebust return 0; 657e261f51fSTrond Myklebust } 658e261f51fSTrond Myklebust 659d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 660c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 661f758c885STrond Myklebust { 662cfb506e1STrond Myklebust int ret; 663f758c885STrond Myklebust 6648cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 6656d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 666cfb506e1STrond Myklebust if (ret == -EAGAIN) { 667cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 668cfb506e1STrond Myklebust ret = 0; 669cfb506e1STrond Myklebust } 670cfb506e1STrond Myklebust return ret; 671f758c885STrond Myklebust } 672f758c885STrond Myklebust 673e261f51fSTrond Myklebust /* 6741da177e4SLinus Torvalds * Write an mmapped page to the server. 6751da177e4SLinus Torvalds */ 676d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 677c373fff7STrond Myklebust struct writeback_control *wbc) 6781da177e4SLinus Torvalds { 679f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 68040f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 681e261f51fSTrond Myklebust int err; 6821da177e4SLinus Torvalds 68340f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 684811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 685a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 686c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 687f758c885STrond Myklebust nfs_pageio_complete(&pgio); 688f758c885STrond Myklebust if (err < 0) 6894d770ccfSTrond Myklebust return err; 690f758c885STrond Myklebust if (pgio.pg_error < 0) 691f758c885STrond Myklebust return pgio.pg_error; 692f758c885STrond Myklebust return 0; 6934d770ccfSTrond Myklebust } 6944d770ccfSTrond Myklebust 6954d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6964d770ccfSTrond Myklebust { 697f758c885STrond Myklebust int ret; 6984d770ccfSTrond Myklebust 699c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 7001da177e4SLinus Torvalds unlock_page(page); 701f758c885STrond Myklebust return ret; 702f758c885STrond Myklebust } 703f758c885STrond Myklebust 704f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 705f758c885STrond Myklebust { 706f758c885STrond Myklebust int ret; 707f758c885STrond Myklebust 708c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 709f758c885STrond Myklebust unlock_page(page); 710f758c885STrond Myklebust return ret; 7111da177e4SLinus Torvalds } 7121da177e4SLinus Torvalds 713919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 714919e3bd9STrond Myklebust { 715919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 716919e3bd9STrond Myklebust } 717919e3bd9STrond Myklebust 7181da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7191da177e4SLinus Torvalds { 7201da177e4SLinus Torvalds struct inode *inode = mapping->host; 721c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 722875bc3fbSTrond Myklebust struct nfs_io_completion *ioc; 723875bc3fbSTrond Myklebust unsigned int pflags = memalloc_nofs_save(); 7241da177e4SLinus Torvalds int err; 7251da177e4SLinus Torvalds 72691d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 72791d5b470SChuck Lever 728875bc3fbSTrond Myklebust ioc = nfs_io_completion_alloc(GFP_NOFS); 729919e3bd9STrond Myklebust if (ioc) 730919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 731919e3bd9STrond Myklebust 732a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 733a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 734919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 735f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 736c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 737919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 73872cb77f4STrond Myklebust 739875bc3fbSTrond Myklebust memalloc_nofs_restore(pflags); 740875bc3fbSTrond 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 { 7564b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7571da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 758e7d39069STrond Myklebust 7592bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7602bfc6e56SWeston Andros Adamson 761e7d39069STrond Myklebust /* Lock the request! */ 7627ad84aa9STrond Myklebust nfs_lock_request(req); 763e7d39069STrond Myklebust 76429418aa4SMel Gorman /* 76529418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 76629418aa4SMel Gorman * with invalidate/truncate. 76729418aa4SMel Gorman */ 7684b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 7694b9bb25bSTrond Myklebust if (!nfs_have_writebacks(inode) && 7701eb5d98fSJeff Layton NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 7711eb5d98fSJeff Layton inode_inc_iversion_raw(inode); 77229418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7732df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 774deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 775277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 77629418aa4SMel Gorman } 7774b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 778a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 77917089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 780cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 781cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 782cb1410c7SWeston Andros Adamson * adding subrequests. */ 78317089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 784c03b4024STrond Myklebust kref_get(&req->wb_kref); 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds 7871da177e4SLinus Torvalds /* 78889a09141SPeter Zijlstra * Remove a write request from an inode 7891da177e4SLinus Torvalds */ 7901da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7911da177e4SLinus Torvalds { 7924b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7934b9bb25bSTrond Myklebust struct inode *inode = mapping->host; 7941da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 79520633f04SWeston Andros Adamson struct nfs_page *head; 79620633f04SWeston Andros Adamson 797a6b6d5b8STrond Myklebust atomic_long_dec(&nfsi->nrequests); 79820633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 79920633f04SWeston Andros Adamson head = req->wb_head; 8001da177e4SLinus Torvalds 8014b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 80267911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 80320633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 80420633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 80520633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 80629418aa4SMel Gorman } 8074b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 80820633f04SWeston Andros Adamson } 80917089a29SWeston Andros Adamson 81017089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 8111da177e4SLinus Torvalds nfs_release_request(req); 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds 81461822ab5STrond Myklebust static void 8156d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 81661822ab5STrond Myklebust { 81767911c8fSAnna Schumaker if (req->wb_page) 81861822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 81961822ab5STrond Myklebust } 82061822ab5STrond Myklebust 8213a3908c8STrond Myklebust /* 8223a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8233a3908c8STrond Myklebust * 8243a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8253a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8263a3908c8STrond Myklebust * 8273a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8283a3908c8STrond Myklebust */ 8293a3908c8STrond Myklebust static struct nfs_page * 8303a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8313a3908c8STrond Myklebust struct page *page) 8323a3908c8STrond Myklebust { 8333a3908c8STrond Myklebust struct nfs_page *freq, *t; 8343a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8353a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8363a3908c8STrond Myklebust 8373a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8383a3908c8STrond Myklebust 8393a3908c8STrond Myklebust /* search through pnfs commit lists */ 8403a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8413a3908c8STrond Myklebust if (freq) 8423a3908c8STrond Myklebust return freq->wb_head; 8433a3908c8STrond Myklebust 8443a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8453a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8463a3908c8STrond Myklebust if (freq->wb_page == page) 8473a3908c8STrond Myklebust return freq->wb_head; 8483a3908c8STrond Myklebust } 8493a3908c8STrond Myklebust 8503a3908c8STrond Myklebust return NULL; 8513a3908c8STrond Myklebust } 8523a3908c8STrond Myklebust 8538dd37758STrond Myklebust /** 85486d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 85586d80f97STrond Myklebust * @req: pointer to a struct nfs_page 85686d80f97STrond Myklebust * @dst: commit list head 85786d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 85886d80f97STrond Myklebust * 85986d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 86086d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 86186d80f97STrond Myklebust * the MM page stats. 86286d80f97STrond Myklebust * 863e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 864e824f99aSTrond Myklebust * nfs_page lock. 86586d80f97STrond Myklebust */ 86686d80f97STrond Myklebust void 86786d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 86886d80f97STrond Myklebust struct nfs_commit_info *cinfo) 86986d80f97STrond Myklebust { 87086d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 87186d80f97STrond Myklebust nfs_list_add_request(req, dst); 8725cb953d4STrond Myklebust atomic_long_inc(&cinfo->mds->ncommit); 87386d80f97STrond Myklebust } 87486d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 87586d80f97STrond Myklebust 87686d80f97STrond Myklebust /** 8778dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8788dd37758STrond Myklebust * @req: pointer to a struct nfs_page 879ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8808dd37758STrond Myklebust * 881ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8828dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8838dd37758STrond Myklebust * the MM page stats. 8848dd37758STrond Myklebust * 885ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8868dd37758STrond Myklebust * holding the nfs_page lock. 8878dd37758STrond Myklebust */ 8888dd37758STrond Myklebust void 8896272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8908dd37758STrond Myklebust { 891e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8926272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 893e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 89467911c8fSAnna Schumaker if (req->wb_page) 89586d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8968dd37758STrond Myklebust } 8978dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8988dd37758STrond Myklebust 8998dd37758STrond Myklebust /** 9008dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 9018dd37758STrond Myklebust * @req: pointer to a nfs_page 902ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 9038dd37758STrond Myklebust * 904ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 9058dd37758STrond Myklebust * number of outstanding requests requiring a commit 9068dd37758STrond Myklebust * It does not update the MM page stats. 9078dd37758STrond Myklebust * 908ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9098dd37758STrond Myklebust */ 9108dd37758STrond Myklebust void 911ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 912ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9138dd37758STrond Myklebust { 9148dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9158dd37758STrond Myklebust return; 9168dd37758STrond Myklebust nfs_list_remove_request(req); 9175cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9188dd37758STrond Myklebust } 9198dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9208dd37758STrond Myklebust 921ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 922ea2cf228SFred Isaman struct inode *inode) 923ea2cf228SFred Isaman { 924fe238e60SDave Wysochanski cinfo->inode = inode; 925ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 926ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 927b359f9d0SFred Isaman cinfo->dreq = NULL; 928f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 929ea2cf228SFred Isaman } 930ea2cf228SFred Isaman 931ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 932ea2cf228SFred Isaman struct inode *inode, 933ea2cf228SFred Isaman struct nfs_direct_req *dreq) 934ea2cf228SFred Isaman { 9351763da12SFred Isaman if (dreq) 9361763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9371763da12SFred Isaman else 938ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 939ea2cf228SFred Isaman } 940ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9418dd37758STrond Myklebust 9421da177e4SLinus Torvalds /* 9431da177e4SLinus Torvalds * Add a request to the inode's commit list. 9441da177e4SLinus Torvalds */ 9451763da12SFred Isaman void 946ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 947b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9481da177e4SLinus Torvalds { 949b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9508dd37758STrond Myklebust return; 9516272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9521da177e4SLinus Torvalds } 9538e821cadSTrond Myklebust 954d6d6dc7cSFred Isaman static void 955d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 956e468bae9STrond Myklebust { 95711fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 95893f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 95993f78d88STejun Heo WB_RECLAIMABLE); 960e468bae9STrond Myklebust } 961d6d6dc7cSFred Isaman 962b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 9638dd37758STrond Myklebust static void 964d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 965d6d6dc7cSFred Isaman { 9668dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 967*9fcd5960STrond Myklebust struct nfs_open_context *ctx = nfs_req_openctx(req); 968*9fcd5960STrond Myklebust struct inode *inode = d_inode(ctx->dentry); 969ea2cf228SFred Isaman struct nfs_commit_info cinfo; 970d6d6dc7cSFred Isaman 971ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 972e824f99aSTrond Myklebust mutex_lock(&NFS_I(inode)->commit_mutex); 973ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 974ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 975d6d6dc7cSFred Isaman } 976e824f99aSTrond Myklebust mutex_unlock(&NFS_I(inode)->commit_mutex); 9778dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9788dd37758STrond Myklebust } 979e468bae9STrond Myklebust } 980e468bae9STrond Myklebust 981d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9828e821cadSTrond Myklebust { 983c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 984d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 985c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9868e821cadSTrond Myklebust } 9878e821cadSTrond Myklebust 988919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 989919e3bd9STrond Myklebust { 990919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 991919e3bd9STrond Myklebust } 992919e3bd9STrond Myklebust 993061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9946c75dc0dSFred Isaman { 995ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9966c75dc0dSFred Isaman unsigned long bytes = 0; 9976c75dc0dSFred Isaman 9986c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9996c75dc0dSFred Isaman goto out; 1000ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 10016c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 10026c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 10036c75dc0dSFred Isaman 10046c75dc0dSFred Isaman bytes += req->wb_bytes; 10056c75dc0dSFred Isaman nfs_list_remove_request(req); 10066c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 10076c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 1008d2ceb7e5SBenjamin Coddington nfs_set_pageerror(page_file_mapping(req->wb_page)); 10096fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, hdr->error); 10106c75dc0dSFred Isaman goto remove_req; 10116c75dc0dSFred Isaman } 1012c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 101333344e0fSTrond Myklebust /* Reset wb_nio, since the write was successful. */ 101433344e0fSTrond Myklebust req->wb_nio = 0; 1015f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1016b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1017a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10186c75dc0dSFred Isaman goto next; 10196c75dc0dSFred Isaman } 10206c75dc0dSFred Isaman remove_req: 10216c75dc0dSFred Isaman nfs_inode_remove_request(req); 10226c75dc0dSFred Isaman next: 102320633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10243aff4ebbSTrond Myklebust nfs_release_request(req); 10256c75dc0dSFred Isaman } 10266c75dc0dSFred Isaman out: 1027919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10286c75dc0dSFred Isaman hdr->release(hdr); 10296c75dc0dSFred Isaman } 10306c75dc0dSFred Isaman 1031ce59515cSAnna Schumaker unsigned long 1032ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1033fb8a1f11STrond Myklebust { 10345cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1035fb8a1f11STrond Myklebust } 1036fb8a1f11STrond Myklebust 1037e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10381763da12SFred Isaman int 1039ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1040ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1041d6d6dc7cSFred Isaman { 1042137da553STrond Myklebust struct nfs_page *req, *tmp; 1043d6d6dc7cSFred Isaman int ret = 0; 1044d6d6dc7cSFred Isaman 1045137da553STrond Myklebust restart: 1046137da553STrond Myklebust list_for_each_entry_safe(req, tmp, src, wb_list) { 10477ad84aa9STrond Myklebust kref_get(&req->wb_kref); 10482ce209c4STrond Myklebust if (!nfs_lock_request(req)) { 10492ce209c4STrond Myklebust int status; 1050137da553STrond Myklebust 1051137da553STrond Myklebust /* Prevent deadlock with nfs_lock_and_join_requests */ 1052137da553STrond Myklebust if (!list_empty(dst)) { 1053137da553STrond Myklebust nfs_release_request(req); 1054137da553STrond Myklebust continue; 1055137da553STrond Myklebust } 1056137da553STrond Myklebust /* Ensure we make progress to prevent livelock */ 10572ce209c4STrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 10582ce209c4STrond Myklebust status = nfs_wait_on_request(req); 10592ce209c4STrond Myklebust nfs_release_request(req); 10602ce209c4STrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10612ce209c4STrond Myklebust if (status < 0) 10622ce209c4STrond Myklebust break; 1063137da553STrond Myklebust goto restart; 10642ce209c4STrond Myklebust } 1065ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10665d2a9d9dSTrond Myklebust clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); 10678dd37758STrond Myklebust nfs_list_add_request(req, dst); 1068d6d6dc7cSFred Isaman ret++; 10691763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1070d6d6dc7cSFred Isaman break; 1071e824f99aSTrond Myklebust cond_resched(); 1072d6d6dc7cSFred Isaman } 1073d6d6dc7cSFred Isaman return ret; 1074d6d6dc7cSFred Isaman } 10755d2a9d9dSTrond Myklebust EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 1076d6d6dc7cSFred Isaman 10771da177e4SLinus Torvalds /* 10781da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10791da177e4SLinus Torvalds * @inode: NFS inode to scan 1080ea2cf228SFred Isaman * @dst: mds destination list 1081ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10821da177e4SLinus Torvalds * 10831da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10841da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10851da177e4SLinus Torvalds */ 10861763da12SFred Isaman int 1087ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1088ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10891da177e4SLinus Torvalds { 1090d6d6dc7cSFred Isaman int ret = 0; 1091fb8a1f11STrond Myklebust 10925cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10935cb953d4STrond Myklebust return 0; 1094e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10955cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10968dd37758STrond Myklebust const int max = INT_MAX; 1097d6d6dc7cSFred Isaman 1098ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1099ea2cf228SFred Isaman cinfo, max); 1100ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1101d6d6dc7cSFred Isaman } 1102e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1103ff778d02STrond Myklebust return ret; 11041da177e4SLinus Torvalds } 1105d6d6dc7cSFred Isaman 11061da177e4SLinus Torvalds /* 1107e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1108e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 11091da177e4SLinus Torvalds * 1110e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1111e7d39069STrond Myklebust * to disk. 11121da177e4SLinus Torvalds */ 1113e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1114e7d39069STrond Myklebust struct page *page, 1115e7d39069STrond Myklebust unsigned int offset, 1116e7d39069STrond Myklebust unsigned int bytes) 11171da177e4SLinus Torvalds { 1118e7d39069STrond Myklebust struct nfs_page *req; 1119e7d39069STrond Myklebust unsigned int rqend; 1120e7d39069STrond Myklebust unsigned int end; 11211da177e4SLinus Torvalds int error; 1122277459d2STrond Myklebust 1123e7d39069STrond Myklebust end = offset + bytes; 1124e7d39069STrond Myklebust 1125f6032f21STrond Myklebust req = nfs_lock_and_join_requests(page); 1126f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1127f6032f21STrond Myklebust return req; 11282bfc6e56SWeston Andros Adamson 1129e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1130e7d39069STrond Myklebust /* 1131e7d39069STrond Myklebust * Tell the caller to flush out the request if 1132e7d39069STrond Myklebust * the offsets are non-contiguous. 1133e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1134e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1135e7d39069STrond Myklebust */ 1136f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1137e7d39069STrond Myklebust goto out_flushme; 1138e7d39069STrond Myklebust 11391da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11401da177e4SLinus Torvalds if (offset < req->wb_offset) { 11411da177e4SLinus Torvalds req->wb_offset = offset; 11421da177e4SLinus Torvalds req->wb_pgbase = offset; 11431da177e4SLinus Torvalds } 11441da177e4SLinus Torvalds if (end > rqend) 11451da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1146e7d39069STrond Myklebust else 1147e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 114833344e0fSTrond Myklebust req->wb_nio = 0; 1149e7d39069STrond Myklebust return req; 1150e7d39069STrond Myklebust out_flushme: 1151f6032f21STrond Myklebust /* 1152f6032f21STrond Myklebust * Note: we mark the request dirty here because 1153f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1154f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1155f6032f21STrond Myklebust */ 1156f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1157f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 1158e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1159f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1160e7d39069STrond Myklebust } 11611da177e4SLinus Torvalds 1162e7d39069STrond Myklebust /* 1163e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1164e7d39069STrond Myklebust * 1165e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1166e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1167e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1168e7d39069STrond Myklebust */ 1169e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1170e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1171e7d39069STrond Myklebust { 1172d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1173e7d39069STrond Myklebust struct nfs_page *req; 1174e7d39069STrond Myklebust 1175e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1176e7d39069STrond Myklebust if (req != NULL) 1177e7d39069STrond Myklebust goto out; 117828b1d3f5STrond Myklebust req = nfs_create_request(ctx, page, offset, bytes); 1179e7d39069STrond Myklebust if (IS_ERR(req)) 1180e7d39069STrond Myklebust goto out; 1181d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1182efc91ed0STrond Myklebust out: 118361e930a9STrond Myklebust return req; 11841da177e4SLinus Torvalds } 11851da177e4SLinus Torvalds 1186e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1187e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1188e7d39069STrond Myklebust { 1189e7d39069STrond Myklebust struct nfs_page *req; 1190e7d39069STrond Myklebust 1191e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1192e7d39069STrond Myklebust if (IS_ERR(req)) 1193e7d39069STrond Myklebust return PTR_ERR(req); 1194e7d39069STrond Myklebust /* Update file length */ 1195e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1196d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1197a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11981d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1199e7d39069STrond Myklebust return 0; 1200e7d39069STrond Myklebust } 1201e7d39069STrond Myklebust 12021da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 12031da177e4SLinus Torvalds { 1204cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 12052a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1206bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 12071da177e4SLinus Torvalds struct nfs_page *req; 12081a54533eSTrond Myklebust int do_flush, status; 12091da177e4SLinus Torvalds /* 12101da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 12111da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 12121da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 12131da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 12141da177e4SLinus Torvalds * Also do the same if we find a request from an existing 12151da177e4SLinus Torvalds * dropped page. 12161da177e4SLinus Torvalds */ 12171a54533eSTrond Myklebust do { 121884d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 12191a54533eSTrond Myklebust if (req == NULL) 12201a54533eSTrond Myklebust return 0; 12212a369153STrond Myklebust l_ctx = req->wb_lock_context; 1222138a2935STrond Myklebust do_flush = req->wb_page != page || 1223*9fcd5960STrond Myklebust !nfs_match_open_context(nfs_req_openctx(req), ctx); 1224bd61e0a9SJeff Layton if (l_ctx && flctx && 1225bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1226bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1227d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12285263e31eSJeff Layton } 12291da177e4SLinus Torvalds nfs_release_request(req); 12301a54533eSTrond Myklebust if (!do_flush) 12311a54533eSTrond Myklebust return 0; 1232d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12331a54533eSTrond Myklebust } while (status == 0); 12341a54533eSTrond Myklebust return status; 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds /* 1238dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1239dc24826bSAndy Adamson * expire soon. 1240dc24826bSAndy Adamson * 1241dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1242dc24826bSAndy Adamson * 1243dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1244dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1245dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1246dc24826bSAndy Adamson */ 1247dc24826bSAndy Adamson int 1248dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1249dc24826bSAndy Adamson { 1250dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1251dc24826bSAndy Adamson 1252ddf529eeSNeilBrown if (nfs_ctx_key_to_expire(ctx, inode) && 1253ddf529eeSNeilBrown !ctx->ll_cred) 1254ddf529eeSNeilBrown /* Already expired! */ 1255ddf529eeSNeilBrown return -EACCES; 1256ddf529eeSNeilBrown return 0; 1257dc24826bSAndy Adamson } 1258dc24826bSAndy Adamson 1259dc24826bSAndy Adamson /* 1260dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1261dc24826bSAndy Adamson */ 1262ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1263dc24826bSAndy Adamson { 1264ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1265ddf529eeSNeilBrown struct rpc_cred *cred = ctx->ll_cred; 1266ddf529eeSNeilBrown struct auth_cred acred = { 1267a52458b4SNeilBrown .cred = ctx->cred, 1268ddf529eeSNeilBrown }; 1269ce52914eSScott Mayhew 1270ddf529eeSNeilBrown if (cred && !cred->cr_ops->crmatch(&acred, cred, 0)) { 1271ddf529eeSNeilBrown put_rpccred(cred); 1272ddf529eeSNeilBrown ctx->ll_cred = NULL; 1273ddf529eeSNeilBrown cred = NULL; 1274ddf529eeSNeilBrown } 1275ddf529eeSNeilBrown if (!cred) 1276ddf529eeSNeilBrown cred = auth->au_ops->lookup_cred(auth, &acred, 0); 1277ddf529eeSNeilBrown if (!cred || IS_ERR(cred)) 1278ddf529eeSNeilBrown return true; 1279ddf529eeSNeilBrown ctx->ll_cred = cred; 1280ddf529eeSNeilBrown return !!(cred->cr_ops->crkey_timeout && 1281ddf529eeSNeilBrown cred->cr_ops->crkey_timeout(cred)); 1282dc24826bSAndy Adamson } 1283dc24826bSAndy Adamson 1284dc24826bSAndy Adamson /* 12855d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12865d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12875d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12885d47a356STrond Myklebust */ 12898d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12905d47a356STrond Myklebust { 1291d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1292d529ef83SJeff Layton 12938d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12948d197a56STrond Myklebust goto out; 129518dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1296d529ef83SJeff Layton return false; 12974db72b40SJeff Layton smp_rmb(); 1298d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12998d197a56STrond Myklebust return false; 13008d197a56STrond Myklebust out: 130118dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 130218dd78c4SScott Mayhew return false; 13038d197a56STrond Myklebust return PageUptodate(page) != 0; 13045d47a356STrond Myklebust } 13055d47a356STrond Myklebust 13065263e31eSJeff Layton static bool 13075263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 13085263e31eSJeff Layton { 13095263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 13105263e31eSJeff Layton fl->fl_type == F_WRLCK; 13115263e31eSJeff Layton } 13125263e31eSJeff Layton 1313c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1314c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1315c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1316c7559663SScott Mayhew * inefficiencies. 1317c7559663SScott Mayhew * 1318263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1319263b4509SScott Mayhew * of the checks. 1320c7559663SScott Mayhew */ 1321c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1322c7559663SScott Mayhew { 13235263e31eSJeff Layton int ret; 13245263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 13255263e31eSJeff Layton struct file_lock *fl; 13265263e31eSJeff Layton 1327c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1328c7559663SScott Mayhew return 0; 1329263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1330263b4509SScott Mayhew return 0; 1331c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1332c7559663SScott Mayhew return 1; 1333bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1334bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13358fa4592aSTrond Myklebust return 1; 13365263e31eSJeff Layton 13375263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13385263e31eSJeff Layton ret = 0; 13396109c850SJeff Layton spin_lock(&flctx->flc_lock); 1340bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1341bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1342bd61e0a9SJeff Layton fl_list); 1343bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13445263e31eSJeff Layton ret = 1; 1345bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13465263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13475263e31eSJeff Layton fl_list); 13485263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13495263e31eSJeff Layton ret = 1; 13505263e31eSJeff Layton } 13516109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13525263e31eSJeff Layton return ret; 1353c7559663SScott Mayhew } 1354c7559663SScott Mayhew 13555d47a356STrond Myklebust /* 13561da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13571da177e4SLinus Torvalds * 13581da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13591da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13601da177e4SLinus Torvalds */ 13611da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13621da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13631da177e4SLinus Torvalds { 1364cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1365d2ceb7e5SBenjamin Coddington struct address_space *mapping = page_file_mapping(page); 1366d2ceb7e5SBenjamin Coddington struct inode *inode = mapping->host; 13671da177e4SLinus Torvalds int status = 0; 13681da177e4SLinus Torvalds 136991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 137091d5b470SChuck Lever 13716de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13726de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13731da177e4SLinus Torvalds 1374149a4fddSBenjamin Coddington if (!count) 1375149a4fddSBenjamin Coddington goto out; 1376149a4fddSBenjamin Coddington 1377c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 137849a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13791da177e4SLinus Torvalds offset = 0; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds 1382e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 138303fa9e84STrond Myklebust if (status < 0) 1384d2ceb7e5SBenjamin Coddington nfs_set_pageerror(mapping); 138559b7c05fSTrond Myklebust else 138659b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1387149a4fddSBenjamin Coddington out: 138848186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13891da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13901da177e4SLinus Torvalds return status; 13911da177e4SLinus Torvalds } 13921da177e4SLinus Torvalds 13933ff7576dSTrond Myklebust static int flush_task_priority(int how) 13941da177e4SLinus Torvalds { 13951da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13961da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13971da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13981da177e4SLinus Torvalds case FLUSH_LOWPRI: 13991da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 14001da177e4SLinus Torvalds } 14011da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 14021da177e4SLinus Torvalds } 14031da177e4SLinus Torvalds 1404d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1405d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1406abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 14071ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 14081da177e4SLinus Torvalds { 14093ff7576dSTrond Myklebust int priority = flush_task_priority(how); 14101da177e4SLinus Torvalds 14111ed26f33SAnna Schumaker task_setup_data->priority = priority; 1412fb91fb0eSAnna Schumaker rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); 14138224b273SChuck Lever trace_nfs_initiate_write(hdr->inode, hdr->io_start, hdr->good_bytes, 14148224b273SChuck Lever hdr->args.stable); 1415275acaafSTrond Myklebust } 1416275acaafSTrond Myklebust 14176d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 14186d884e8fSFred * call this on each, which will prepare them to be retried on next 14196d884e8fSFred * writeback using standard nfs. 14206d884e8fSFred */ 14216d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14226d884e8fSFred { 142333344e0fSTrond Myklebust /* Bump the transmission count */ 142433344e0fSTrond Myklebust req->wb_nio++; 14256d884e8fSFred nfs_mark_request_dirty(req); 1426*9fcd5960STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); 142720633f04SWeston Andros Adamson nfs_end_page_writeback(req); 14283aff4ebbSTrond Myklebust nfs_release_request(req); 14296d884e8fSFred } 14306d884e8fSFred 1431df3accb8STrond Myklebust static void nfs_async_write_error(struct list_head *head, int error) 14326c75dc0dSFred Isaman { 14336c75dc0dSFred Isaman struct nfs_page *req; 14346c75dc0dSFred Isaman 14356c75dc0dSFred Isaman while (!list_empty(head)) { 14366c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14376c75dc0dSFred Isaman nfs_list_remove_request(req); 14386fbda89bSTrond Myklebust if (nfs_error_is_fatal(error)) 14396fbda89bSTrond Myklebust nfs_write_error(req, error); 14406fbda89bSTrond Myklebust else 14416c75dc0dSFred Isaman nfs_redirty_request(req); 14426c75dc0dSFred Isaman } 14436c75dc0dSFred Isaman } 14446c75dc0dSFred Isaman 1445dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1446dc602dd7STrond Myklebust { 1447df3accb8STrond Myklebust nfs_async_write_error(&hdr->pages, 0); 14487be7b3caSTrond Myklebust filemap_fdatawrite_range(hdr->inode->i_mapping, hdr->args.offset, 14497be7b3caSTrond Myklebust hdr->args.offset + hdr->args.count - 1); 1450dc602dd7STrond Myklebust } 1451dc602dd7STrond Myklebust 1452061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1453919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1454061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1455061ae2edSFred Isaman .completion = nfs_write_completion, 1456dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1457061ae2edSFred Isaman }; 1458061ae2edSFred Isaman 145957208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1460a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1461061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14621751c363STrond Myklebust { 1463a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 146441d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1465a20c93e3SChristoph Hellwig 1466a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1467a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1468a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1469a20c93e3SChristoph Hellwig #endif 14704a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 14713bde7afdSTrond Myklebust server->wsize, ioflags); 14721751c363STrond Myklebust } 1473ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14741751c363STrond Myklebust 1475dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1476dce81290STrond Myklebust { 1477a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1478a7d42ddbSWeston Andros Adamson 14796f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14806f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14816f29b9bbSKinglong Mee 148241d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1483a7d42ddbSWeston Andros Adamson 1484a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1485a7d42ddbSWeston Andros Adamson 1486a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1487a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1488dce81290STrond Myklebust } 14891f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1490dce81290STrond Myklebust 14911da177e4SLinus Torvalds 14920b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14930b7c0153SFred Isaman { 14940b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14950b7c0153SFred Isaman 14960b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14970b7c0153SFred Isaman } 14980b7c0153SFred Isaman 14991f2edbe3STrond Myklebust /* 15001f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 15011f2edbe3STrond Myklebust */ 15021f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 15031f2edbe3STrond Myklebust { 15041f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 15051f2edbe3STrond Myklebust int kill = 0; 1506788e7a89STrond Myklebust 15071f2edbe3STrond Myklebust /* suid always must be killed */ 15081f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 15091f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 15101f2edbe3STrond Myklebust 15111f2edbe3STrond Myklebust /* 15121f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 15131f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 15141f2edbe3STrond Myklebust */ 15151f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 15161f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 15171f2edbe3STrond Myklebust 15181f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 15191f2edbe3STrond Myklebust return kill; 15201f2edbe3STrond Myklebust 15211f2edbe3STrond Myklebust return 0; 15221f2edbe3STrond Myklebust } 1523788e7a89STrond Myklebust 1524a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1525a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1526a08a8cd3STrond Myklebust { 1527a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1528a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 15292b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1530a08a8cd3STrond Myklebust 1531a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15322b83d3deSTrond Myklebust fattr->size = size; 15332b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15342b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1535a08a8cd3STrond Myklebust return; 15362b83d3deSTrond Myklebust } 15372b83d3deSTrond Myklebust if (size != fattr->size) 1538a08a8cd3STrond Myklebust return; 1539a08a8cd3STrond Myklebust /* Set attribute barrier */ 1540a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15412b83d3deSTrond Myklebust /* ...and update size */ 15422b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1543a08a8cd3STrond Myklebust } 1544a08a8cd3STrond Myklebust 1545a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1546a08a8cd3STrond Myklebust { 15472b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1548a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1549a08a8cd3STrond Myklebust 1550a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1551a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1552a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1553a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1554a08a8cd3STrond Myklebust } 1555a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1556a08a8cd3STrond Myklebust 15571da177e4SLinus Torvalds /* 15581da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15591da177e4SLinus Torvalds */ 1560d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1561d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15620eecb214SAnna Schumaker struct inode *inode) 15631da177e4SLinus Torvalds { 1564788e7a89STrond Myklebust int status; 15651da177e4SLinus Torvalds 1566f551e44fSChuck Lever /* 1567f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1568f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1569f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1570f551e44fSChuck Lever * another writer had changed the file, but some applications 1571f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1572f551e44fSChuck Lever */ 1573d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1574788e7a89STrond Myklebust if (status != 0) 15750eecb214SAnna Schumaker return status; 15768224b273SChuck Lever 1577d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 15788224b273SChuck Lever trace_nfs_writeback_done(inode, task->tk_status, 15798224b273SChuck Lever hdr->args.offset, hdr->res.verf); 158091d5b470SChuck Lever 1581d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1582d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15831da177e4SLinus Torvalds /* We tried a write call, but the server did not 15841da177e4SLinus Torvalds * commit data to stable storage even though we 15851da177e4SLinus Torvalds * requested it. 15861da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15871da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15881da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15891da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15901da177e4SLinus Torvalds */ 15911da177e4SLinus Torvalds static unsigned long complain; 15921da177e4SLinus Torvalds 1593a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15941da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15951da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15961da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1597cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1598d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15991da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 16001da177e4SLinus Torvalds } 16011da177e4SLinus Torvalds } 16021f2edbe3STrond Myklebust 16031f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 160416e14375STrond Myklebust if (nfs_should_remove_suid(inode)) { 160516e14375STrond Myklebust spin_lock(&inode->i_lock); 160616e14375STrond Myklebust NFS_I(inode)->cache_validity |= NFS_INO_INVALID_OTHER; 160716e14375STrond Myklebust spin_unlock(&inode->i_lock); 160816e14375STrond Myklebust } 16090eecb214SAnna Schumaker return 0; 16100eecb214SAnna Schumaker } 16110eecb214SAnna Schumaker 16120eecb214SAnna Schumaker /* 16130eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 16140eecb214SAnna Schumaker */ 1615d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1616d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 16170eecb214SAnna Schumaker { 1618d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1619d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 16201f2edbe3STrond Myklebust 16211f2edbe3STrond Myklebust if (resp->count < argp->count) { 16221da177e4SLinus Torvalds static unsigned long complain; 16231da177e4SLinus Torvalds 16246c75dc0dSFred Isaman /* This a short write! */ 1625d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 162691d5b470SChuck Lever 16271da177e4SLinus Torvalds /* Has the server at least made some progress? */ 16286c75dc0dSFred Isaman if (resp->count == 0) { 16296c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 16306c75dc0dSFred Isaman printk(KERN_WARNING 16316c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16326c75dc0dSFred Isaman argp->count); 16336c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16346c75dc0dSFred Isaman } 1635d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16366c75dc0dSFred Isaman task->tk_status = -EIO; 16376c75dc0dSFred Isaman return; 16386c75dc0dSFred Isaman } 1639f8417b48SKinglong Mee 1640f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1641f8417b48SKinglong Mee if (!task->tk_ops) { 1642f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1643f8417b48SKinglong Mee return; 1644f8417b48SKinglong Mee } 1645f8417b48SKinglong Mee 16461da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16471da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16481da177e4SLinus Torvalds /* Resend from where the server left off */ 1649d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16501da177e4SLinus Torvalds argp->offset += resp->count; 16511da177e4SLinus Torvalds argp->pgbase += resp->count; 16521da177e4SLinus Torvalds argp->count -= resp->count; 16531da177e4SLinus Torvalds } else { 16541da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16551da177e4SLinus Torvalds * headaches in the case of a server crash. 16561da177e4SLinus Torvalds */ 16571da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16581da177e4SLinus Torvalds } 1659d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16601da177e4SLinus Torvalds } 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 1663af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 166471d0a611STrond Myklebust { 1665723c921eSPeter Zijlstra return wait_var_event_killable(&cinfo->rpcs_out, 1666723c921eSPeter Zijlstra !atomic_read(&cinfo->rpcs_out)); 1667af7cf057STrond Myklebust } 1668af7cf057STrond Myklebust 1669af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1670af7cf057STrond Myklebust { 1671af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1672af7cf057STrond Myklebust } 1673af7cf057STrond Myklebust 1674af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1675af7cf057STrond Myklebust { 1676af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1677723c921eSPeter Zijlstra wake_up_var(&cinfo->rpcs_out); 167871d0a611STrond Myklebust } 167971d0a611STrond Myklebust 16800b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16811da177e4SLinus Torvalds { 16820b7c0153SFred Isaman put_nfs_open_context(data->context); 16830b7c0153SFred Isaman nfs_commit_free(data); 16841da177e4SLinus Torvalds } 1685e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16861da177e4SLinus Torvalds 16870b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1688c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16899ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16909f0ec176SAndy Adamson int how, int flags) 16911da177e4SLinus Torvalds { 169207737691STrond Myklebust struct rpc_task *task; 16939ace33cdSFred Isaman int priority = flush_task_priority(how); 1694bdc7f021STrond Myklebust struct rpc_message msg = { 1695bdc7f021STrond Myklebust .rpc_argp = &data->args, 1696bdc7f021STrond Myklebust .rpc_resp = &data->res, 16979ace33cdSFred Isaman .rpc_cred = data->cred, 1698bdc7f021STrond Myklebust }; 169984115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 170007737691STrond Myklebust .task = &data->task, 17019ace33cdSFred Isaman .rpc_client = clnt, 1702bdc7f021STrond Myklebust .rpc_message = &msg, 17039ace33cdSFred Isaman .callback_ops = call_ops, 170484115e1cSTrond Myklebust .callback_data = data, 1705101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 17069f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 17073ff7576dSTrond Myklebust .priority = priority, 170884115e1cSTrond Myklebust }; 1709788e7a89STrond Myklebust /* Set up the initial task struct. */ 1710e9ae1ee2SAnna Schumaker nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); 17118224b273SChuck Lever trace_nfs_initiate_commit(data); 17121da177e4SLinus Torvalds 1713b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1714bdc7f021STrond Myklebust 171507737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1716dbae4c73STrond Myklebust if (IS_ERR(task)) 1717dbae4c73STrond Myklebust return PTR_ERR(task); 1718d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1719d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 172007737691STrond Myklebust rpc_put_task(task); 1721dbae4c73STrond Myklebust return 0; 17221da177e4SLinus Torvalds } 1723e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 17241da177e4SLinus Torvalds 1725378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1726378520b8SPeng Tao { 1727378520b8SPeng Tao loff_t lwb = 0; 1728378520b8SPeng Tao struct nfs_page *req; 1729378520b8SPeng Tao 1730378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1731378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1732378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1733378520b8SPeng Tao 1734378520b8SPeng Tao return lwb; 1735378520b8SPeng Tao } 1736378520b8SPeng Tao 17371da177e4SLinus Torvalds /* 17389ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17399ace33cdSFred Isaman */ 17400b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1741988b6dceSFred Isaman struct list_head *head, 1742f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1743f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17449ace33cdSFred Isaman { 17459ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 1746*9fcd5960STrond Myklebust struct nfs_open_context *ctx = nfs_req_openctx(first); 1747*9fcd5960STrond Myklebust struct inode *inode = d_inode(ctx->dentry); 17489ace33cdSFred Isaman 17499ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17509ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17519ace33cdSFred Isaman 17529ace33cdSFred Isaman list_splice_init(head, &data->pages); 17539ace33cdSFred Isaman 17549ace33cdSFred Isaman data->inode = inode; 1755*9fcd5960STrond Myklebust data->cred = ctx->cred; 1756988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1757378520b8SPeng Tao /* only set lwb for pnfs commit */ 1758378520b8SPeng Tao if (lseg) 1759378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17609ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1761f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1762b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17639ace33cdSFred Isaman 17649ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17659ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17669ace33cdSFred Isaman data->args.offset = 0; 17679ace33cdSFred Isaman data->args.count = 0; 1768*9fcd5960STrond Myklebust data->context = get_nfs_open_context(ctx); 17699ace33cdSFred Isaman data->res.fattr = &data->fattr; 17709ace33cdSFred Isaman data->res.verf = &data->verf; 17719ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17729ace33cdSFred Isaman } 1773e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17749ace33cdSFred Isaman 1775e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1776ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1777b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1778b57ff130SWeston Andros Adamson u32 ds_commit_idx) 177964bfeb49SFred Isaman { 178064bfeb49SFred Isaman struct nfs_page *req; 178164bfeb49SFred Isaman 178264bfeb49SFred Isaman while (!list_empty(page_list)) { 178364bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 178464bfeb49SFred Isaman nfs_list_remove_request(req); 1785b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1786487b9b8aSTom Haynes if (!cinfo->dreq) 1787487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17881d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 178964bfeb49SFred Isaman } 179064bfeb49SFred Isaman } 1791e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 179264bfeb49SFred Isaman 1793b20135d0STrond Myklebust static void 1794b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1795b20135d0STrond Myklebust struct nfs_page *req) 1796b20135d0STrond Myklebust { 1797b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1798b20135d0STrond Myklebust } 1799b20135d0STrond Myklebust 18009ace33cdSFred Isaman /* 18011da177e4SLinus Torvalds * Commit dirty pages 18021da177e4SLinus Torvalds */ 18031da177e4SLinus Torvalds static int 1804ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1805ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 18061da177e4SLinus Torvalds { 18070b7c0153SFred Isaman struct nfs_commit_data *data; 18081da177e4SLinus Torvalds 1809ade8febdSWeston Andros Adamson /* another commit raced with us */ 1810ade8febdSWeston Andros Adamson if (list_empty(head)) 1811ade8febdSWeston Andros Adamson return 0; 1812ade8febdSWeston Andros Adamson 1813518662e0SNeilBrown data = nfs_commitdata_alloc(true); 18141da177e4SLinus Torvalds 18151da177e4SLinus Torvalds /* Set up the argument struct */ 1816f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1817f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1818c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1819c36aae9aSPeng Tao data->mds_ops, how, 0); 18201da177e4SLinus Torvalds } 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds /* 18231da177e4SLinus Torvalds * COMMIT call returned 18241da177e4SLinus Torvalds */ 1825788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 18261da177e4SLinus Torvalds { 18270b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18281da177e4SLinus Torvalds 1829a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 18301da177e4SLinus Torvalds task->tk_pid, task->tk_status); 18311da177e4SLinus Torvalds 1832788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1833c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 18348224b273SChuck Lever trace_nfs_commit_done(data); 1835c9d8f89dSTrond Myklebust } 1836c9d8f89dSTrond Myklebust 1837f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1838c9d8f89dSTrond Myklebust { 1839c9d8f89dSTrond Myklebust struct nfs_page *req; 1840c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1841f453a54aSFred Isaman struct nfs_commit_info cinfo; 1842353db796SNeilBrown struct nfs_server *nfss; 1843788e7a89STrond Myklebust 18441da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18451da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18461da177e4SLinus Torvalds nfs_list_remove_request(req); 184767911c8fSAnna Schumaker if (req->wb_page) 1848d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 18491da177e4SLinus Torvalds 18501e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 1851*9fcd5960STrond Myklebust nfs_req_openctx(req)->dentry->d_sb->s_id, 1852*9fcd5960STrond Myklebust (unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)), 18531da177e4SLinus Torvalds req->wb_bytes, 18541da177e4SLinus Torvalds (long long)req_offset(req)); 1855c9d8f89dSTrond Myklebust if (status < 0) { 18566fbda89bSTrond Myklebust if (req->wb_page) { 18576fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, status); 18581da177e4SLinus Torvalds nfs_inode_remove_request(req); 18596fbda89bSTrond Myklebust } 1860ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18611da177e4SLinus Torvalds goto next; 18621da177e4SLinus Torvalds } 18631da177e4SLinus Torvalds 18641da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18651da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18668fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18671da177e4SLinus Torvalds /* We have a match */ 186838a33101SKinglong Mee if (req->wb_page) 18691da177e4SLinus Torvalds nfs_inode_remove_request(req); 1870ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18711da177e4SLinus Torvalds goto next; 18721da177e4SLinus Torvalds } 18731da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1874ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18756d884e8fSFred nfs_mark_request_dirty(req); 1876*9fcd5960STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); 18771da177e4SLinus Torvalds next: 18781d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18797f1bda44STrond Myklebust /* Latency breaker */ 18807f1bda44STrond Myklebust cond_resched(); 18811da177e4SLinus Torvalds } 1882353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1883353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18840db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1885353db796SNeilBrown 1886f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1887af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18885917ce84SFred Isaman } 18895917ce84SFred Isaman 18905917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18915917ce84SFred Isaman { 18920b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18935917ce84SFred Isaman 1894f453a54aSFred Isaman data->completion_ops->completion(data); 1895c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18961da177e4SLinus Torvalds } 1897788e7a89STrond Myklebust 1898788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18990b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1900788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1901788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1902788e7a89STrond Myklebust }; 19031da177e4SLinus Torvalds 1904f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1905f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1906b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1907f453a54aSFred Isaman }; 1908f453a54aSFred Isaman 19091763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1910ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 191184c53ab5SFred Isaman { 191284c53ab5SFred Isaman int status; 191384c53ab5SFred Isaman 1914ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 191584c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1916ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 191784c53ab5SFred Isaman return status; 191884c53ab5SFred Isaman } 191984c53ab5SFred Isaman 1920c4f24df9STrond Myklebust static int __nfs_commit_inode(struct inode *inode, int how, 1921c4f24df9STrond Myklebust struct writeback_control *wbc) 19221da177e4SLinus Torvalds { 19231da177e4SLinus Torvalds LIST_HEAD(head); 1924ea2cf228SFred Isaman struct nfs_commit_info cinfo; 192571d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1926c4f24df9STrond Myklebust int ret, nscan; 19271da177e4SLinus Torvalds 1928ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1929af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1930c4f24df9STrond Myklebust for (;;) { 1931c4f24df9STrond Myklebust ret = nscan = nfs_scan_commit(inode, &head, &cinfo); 1932c4f24df9STrond Myklebust if (ret <= 0) 1933c4f24df9STrond Myklebust break; 1934c4f24df9STrond Myklebust ret = nfs_generic_commit_list(inode, &head, how, &cinfo); 1935c4f24df9STrond Myklebust if (ret < 0) 1936c4f24df9STrond Myklebust break; 1937c4f24df9STrond Myklebust ret = 0; 1938c4f24df9STrond Myklebust if (wbc && wbc->sync_mode == WB_SYNC_NONE) { 1939c4f24df9STrond Myklebust if (nscan < wbc->nr_to_write) 1940c4f24df9STrond Myklebust wbc->nr_to_write -= nscan; 1941c4f24df9STrond Myklebust else 1942c4f24df9STrond Myklebust wbc->nr_to_write = 0; 1943c4f24df9STrond Myklebust } 1944c4f24df9STrond Myklebust if (nscan < INT_MAX) 1945c4f24df9STrond Myklebust break; 1946c4f24df9STrond Myklebust cond_resched(); 1947c4f24df9STrond Myklebust } 1948af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 1949c4f24df9STrond Myklebust if (ret || !may_wait) 1950c4f24df9STrond Myklebust return ret; 1951c4f24df9STrond Myklebust return wait_on_commit(cinfo.mds); 1952c4f24df9STrond Myklebust } 1953c4f24df9STrond Myklebust 1954c4f24df9STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 1955c4f24df9STrond Myklebust { 1956c4f24df9STrond Myklebust return __nfs_commit_inode(inode, how, NULL); 19571da177e4SLinus Torvalds } 1958b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19598fc795f7STrond Myklebust 1960ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19618fc795f7STrond Myklebust { 1962420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1963420e3646STrond Myklebust int flags = FLUSH_SYNC; 1964420e3646STrond Myklebust int ret = 0; 19658fc795f7STrond Myklebust 1966c4f24df9STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 19673236c3e1SJeff Layton /* no commits means nothing needs to be done */ 19685cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 1969c4f24df9STrond Myklebust goto check_requests_outstanding; 19703236c3e1SJeff Layton 1971a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1972a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1973420e3646STrond Myklebust */ 19741a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1975420e3646STrond Myklebust goto out_mark_dirty; 1976420e3646STrond Myklebust 1977a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1978420e3646STrond Myklebust flags = 0; 1979a00dd6c0SJeff Layton } 1980a00dd6c0SJeff Layton 1981c4f24df9STrond Myklebust ret = __nfs_commit_inode(inode, flags, wbc); 1982c4f24df9STrond Myklebust if (!ret) { 1983c4f24df9STrond Myklebust if (flags & FLUSH_SYNC) 19848fc795f7STrond Myklebust return 0; 1985c4f24df9STrond Myklebust } else if (atomic_long_read(&nfsi->commit_info.ncommit)) 1986c4f24df9STrond Myklebust goto out_mark_dirty; 1987c4f24df9STrond Myklebust 1988c4f24df9STrond Myklebust check_requests_outstanding: 1989c4f24df9STrond Myklebust if (!atomic_read(&nfsi->commit_info.rpcs_out)) 1990c4f24df9STrond Myklebust return ret; 1991420e3646STrond Myklebust out_mark_dirty: 19928fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19938fc795f7STrond Myklebust return ret; 19948fc795f7STrond Myklebust } 199589d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1996863a3c6cSAndy Adamson 1997acdc53b2STrond Myklebust /* 1998837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1999837bb1d7STrond Myklebust * 2000837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 2001837bb1d7STrond Myklebust * client. 2002837bb1d7STrond Myklebust */ 2003837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 2004837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 2005837bb1d7STrond Myklebust { 2006837bb1d7STrond Myklebust int ret; 2007837bb1d7STrond Myklebust 2008837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 2009837bb1d7STrond Myklebust if (ret == 0) 2010837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 2011837bb1d7STrond Myklebust return ret; 2012837bb1d7STrond Myklebust } 2013837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 2014837bb1d7STrond Myklebust 2015837bb1d7STrond Myklebust /* 2016acdc53b2STrond Myklebust * flush the inode to disk. 2017acdc53b2STrond Myklebust */ 2018acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 201934901f70STrond Myklebust { 2020f4ce1299STrond Myklebust int ret; 202134901f70STrond Myklebust 2022f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 2023f4ce1299STrond Myklebust 20245bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 20256b196875SChuck Lever if (ret) 20266b196875SChuck Lever goto out; 20275bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20286b196875SChuck Lever if (ret < 0) 20296b196875SChuck Lever goto out; 20305bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 20316b196875SChuck Lever ret = 0; 2032f4ce1299STrond Myklebust 20336b196875SChuck Lever out: 2034f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 2035f4ce1299STrond Myklebust return ret; 20361c75950bSTrond Myklebust } 2037ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 20381c75950bSTrond Myklebust 20391b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 20401b3b4a1aSTrond Myklebust { 20411b3b4a1aSTrond Myklebust struct nfs_page *req; 20421b3b4a1aSTrond Myklebust int ret = 0; 20431b3b4a1aSTrond Myklebust 2044ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20453e217045SWeston Andros Adamson 20463e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20473e217045SWeston Andros Adamson * request */ 20486d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 20493e217045SWeston Andros Adamson 20503e217045SWeston Andros Adamson if (IS_ERR(req)) { 20513e217045SWeston Andros Adamson ret = PTR_ERR(req); 20523e217045SWeston Andros Adamson } else if (req) { 20533e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 20543e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20553e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20563e217045SWeston Andros Adamson * release it */ 20571b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20581d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20591b3b4a1aSTrond Myklebust } 20603e217045SWeston Andros Adamson 20611b3b4a1aSTrond Myklebust return ret; 20621b3b4a1aSTrond Myklebust } 20631b3b4a1aSTrond Myklebust 20641c75950bSTrond Myklebust /* 20651c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 20661c75950bSTrond Myklebust */ 2067c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 20681c75950bSTrond Myklebust { 206929418aa4SMel Gorman loff_t range_start = page_file_offset(page); 207009cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 20717f2f12d9STrond Myklebust struct writeback_control wbc = { 20727f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20737f2f12d9STrond Myklebust .nr_to_write = 0, 20747f2f12d9STrond Myklebust .range_start = range_start, 20757f2f12d9STrond Myklebust .range_end = range_end, 20767f2f12d9STrond Myklebust }; 20777f2f12d9STrond Myklebust int ret; 20787f2f12d9STrond Myklebust 2079f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2080f4ce1299STrond Myklebust 20810522f6adSTrond Myklebust for (;;) { 2082ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20837f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2084c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20857f2f12d9STrond Myklebust if (ret < 0) 20867f2f12d9STrond Myklebust goto out_error; 20870522f6adSTrond Myklebust continue; 20887f2f12d9STrond Myklebust } 2089f4ce1299STrond Myklebust ret = 0; 20900522f6adSTrond Myklebust if (!PagePrivate(page)) 20910522f6adSTrond Myklebust break; 20920522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20937f2f12d9STrond Myklebust if (ret < 0) 20947f2f12d9STrond Myklebust goto out_error; 20957f2f12d9STrond Myklebust } 20967f2f12d9STrond Myklebust out_error: 2097f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20987f2f12d9STrond Myklebust return ret; 20991c75950bSTrond Myklebust } 21001c75950bSTrond Myklebust 2101074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2102074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2103a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2104074cc1deSTrond Myklebust { 21052da95652SJeff Layton /* 21062da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 21072da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 21082da95652SJeff Layton * 21092da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 21102da95652SJeff Layton * that we can safely release the inode reference while holding 21112da95652SJeff Layton * the page lock. 21122da95652SJeff Layton */ 21132da95652SJeff Layton if (PagePrivate(page)) 21142da95652SJeff Layton return -EBUSY; 2115074cc1deSTrond Myklebust 21168c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 21178c209ce7SDavid Howells return -EBUSY; 2118074cc1deSTrond Myklebust 2119a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2120074cc1deSTrond Myklebust } 2121074cc1deSTrond Myklebust #endif 2122074cc1deSTrond Myklebust 2123f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 21241da177e4SLinus Torvalds { 21251da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 21261e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 21271da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 212820c2df83SPaul Mundt NULL); 21291da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 21301da177e4SLinus Torvalds return -ENOMEM; 21311da177e4SLinus Torvalds 213293d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 21331da177e4SLinus Torvalds nfs_wdata_cachep); 21341da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 21353dd4765fSJeff Layton goto out_destroy_write_cache; 21361da177e4SLinus Torvalds 21370b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 21380b7c0153SFred Isaman sizeof(struct nfs_commit_data), 21390b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21400b7c0153SFred Isaman NULL); 21410b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21423dd4765fSJeff Layton goto out_destroy_write_mempool; 21430b7c0153SFred Isaman 214493d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21454c100210SYanchuan Nian nfs_cdata_cachep); 21461da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21473dd4765fSJeff Layton goto out_destroy_commit_cache; 21481da177e4SLinus Torvalds 214989a09141SPeter Zijlstra /* 215089a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 215189a09141SPeter Zijlstra * 215289a09141SPeter Zijlstra * 64MB: 8192k 215389a09141SPeter Zijlstra * 128MB: 11585k 215489a09141SPeter Zijlstra * 256MB: 16384k 215589a09141SPeter Zijlstra * 512MB: 23170k 215689a09141SPeter Zijlstra * 1GB: 32768k 215789a09141SPeter Zijlstra * 2GB: 46340k 215889a09141SPeter Zijlstra * 4GB: 65536k 215989a09141SPeter Zijlstra * 8GB: 92681k 216089a09141SPeter Zijlstra * 16GB: 131072k 216189a09141SPeter Zijlstra * 216289a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 216389a09141SPeter Zijlstra * Limit the default to 256M 216489a09141SPeter Zijlstra */ 2165ca79b0c2SArun KS nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); 216689a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 216789a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 216889a09141SPeter Zijlstra 21691da177e4SLinus Torvalds return 0; 21703dd4765fSJeff Layton 21713dd4765fSJeff Layton out_destroy_commit_cache: 21723dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21733dd4765fSJeff Layton out_destroy_write_mempool: 21743dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21753dd4765fSJeff Layton out_destroy_write_cache: 21763dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21773dd4765fSJeff Layton return -ENOMEM; 21781da177e4SLinus Torvalds } 21791da177e4SLinus Torvalds 2180266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21811da177e4SLinus Torvalds { 21821da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21833dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21841da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21851a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21861da177e4SLinus Torvalds } 21871da177e4SLinus Torvalds 21884a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21894a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21904a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21910eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21920eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21931ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21944a0de55cSAnna Schumaker }; 2195