11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/file.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1992 Rick Sladkey 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Changes Copyright (C) 1994 by Florian La Roche 71da177e4SLinus Torvalds * - Do not copy data too often around in the kernel. 81da177e4SLinus Torvalds * - In nfs_file_read the return value of kmalloc wasn't checked. 91da177e4SLinus Torvalds * - Put in a better version of read look-ahead buffering. Original idea 101da177e4SLinus Torvalds * and implementation by Wai S Kok elekokws@ee.nus.sg. 111da177e4SLinus Torvalds * 121da177e4SLinus Torvalds * Expire cache on write to a file by Wai S Kok (Oct 1994). 131da177e4SLinus Torvalds * 141da177e4SLinus Torvalds * Total rewrite of read side for new NFS buffer cache.. Linus. 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * nfs regular file handling functions 171da177e4SLinus Torvalds */ 181da177e4SLinus Torvalds 191da177e4SLinus Torvalds #include <linux/time.h> 201da177e4SLinus Torvalds #include <linux/kernel.h> 211da177e4SLinus Torvalds #include <linux/errno.h> 221da177e4SLinus Torvalds #include <linux/fcntl.h> 231da177e4SLinus Torvalds #include <linux/stat.h> 241da177e4SLinus Torvalds #include <linux/nfs_fs.h> 251da177e4SLinus Torvalds #include <linux/nfs_mount.h> 261da177e4SLinus Torvalds #include <linux/mm.h> 271da177e4SLinus Torvalds #include <linux/slab.h> 281da177e4SLinus Torvalds #include <linux/pagemap.h> 291da177e4SLinus Torvalds #include <linux/smp_lock.h> 30e8edc6e0SAlexey Dobriyan #include <linux/aio.h> 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds #include <asm/uaccess.h> 331da177e4SLinus Torvalds #include <asm/system.h> 341da177e4SLinus Torvalds 351da177e4SLinus Torvalds #include "delegation.h" 3694387fb1STrond Myklebust #include "internal.h" 3791d5b470SChuck Lever #include "iostat.h" 381da177e4SLinus Torvalds 391da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_FILE 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds static int nfs_file_open(struct inode *, struct file *); 421da177e4SLinus Torvalds static int nfs_file_release(struct inode *, struct file *); 43980802e3STrond Myklebust static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin); 441da177e4SLinus Torvalds static int nfs_file_mmap(struct file *, struct vm_area_struct *); 45f0930fffSJens Axboe static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos, 46f0930fffSJens Axboe struct pipe_inode_info *pipe, 47f0930fffSJens Axboe size_t count, unsigned int flags); 48027445c3SBadari Pulavarty static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov, 49027445c3SBadari Pulavarty unsigned long nr_segs, loff_t pos); 50027445c3SBadari Pulavarty static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov, 51027445c3SBadari Pulavarty unsigned long nr_segs, loff_t pos); 5275e1fcc0SMiklos Szeredi static int nfs_file_flush(struct file *, fl_owner_t id); 531da177e4SLinus Torvalds static int nfs_fsync(struct file *, struct dentry *dentry, int datasync); 541da177e4SLinus Torvalds static int nfs_check_flags(int flags); 551da177e4SLinus Torvalds static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl); 561da177e4SLinus Torvalds static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl); 57370f6599SJ. Bruce Fields static int nfs_setlease(struct file *file, long arg, struct file_lock **fl); 581da177e4SLinus Torvalds 5994387fb1STrond Myklebust static struct vm_operations_struct nfs_file_vm_ops; 6094387fb1STrond Myklebust 614b6f5d20SArjan van de Ven const struct file_operations nfs_file_operations = { 62980802e3STrond Myklebust .llseek = nfs_file_llseek, 631da177e4SLinus Torvalds .read = do_sync_read, 641da177e4SLinus Torvalds .write = do_sync_write, 651da177e4SLinus Torvalds .aio_read = nfs_file_read, 661da177e4SLinus Torvalds .aio_write = nfs_file_write, 67240ee831SBryan Wu #ifdef CONFIG_MMU 681da177e4SLinus Torvalds .mmap = nfs_file_mmap, 69240ee831SBryan Wu #else 70240ee831SBryan Wu .mmap = generic_file_mmap, 71240ee831SBryan Wu #endif 721da177e4SLinus Torvalds .open = nfs_file_open, 731da177e4SLinus Torvalds .flush = nfs_file_flush, 741da177e4SLinus Torvalds .release = nfs_file_release, 751da177e4SLinus Torvalds .fsync = nfs_fsync, 761da177e4SLinus Torvalds .lock = nfs_lock, 771da177e4SLinus Torvalds .flock = nfs_flock, 78f0930fffSJens Axboe .splice_read = nfs_file_splice_read, 791da177e4SLinus Torvalds .check_flags = nfs_check_flags, 80370f6599SJ. Bruce Fields .setlease = nfs_setlease, 811da177e4SLinus Torvalds }; 821da177e4SLinus Torvalds 8392e1d5beSArjan van de Ven const struct inode_operations nfs_file_inode_operations = { 841da177e4SLinus Torvalds .permission = nfs_permission, 851da177e4SLinus Torvalds .getattr = nfs_getattr, 861da177e4SLinus Torvalds .setattr = nfs_setattr, 871da177e4SLinus Torvalds }; 881da177e4SLinus Torvalds 89b7fa0554SAndreas Gruenbacher #ifdef CONFIG_NFS_V3 9092e1d5beSArjan van de Ven const struct inode_operations nfs3_file_inode_operations = { 91b7fa0554SAndreas Gruenbacher .permission = nfs_permission, 92b7fa0554SAndreas Gruenbacher .getattr = nfs_getattr, 93b7fa0554SAndreas Gruenbacher .setattr = nfs_setattr, 94b7fa0554SAndreas Gruenbacher .listxattr = nfs3_listxattr, 95b7fa0554SAndreas Gruenbacher .getxattr = nfs3_getxattr, 96b7fa0554SAndreas Gruenbacher .setxattr = nfs3_setxattr, 97b7fa0554SAndreas Gruenbacher .removexattr = nfs3_removexattr, 98b7fa0554SAndreas Gruenbacher }; 99b7fa0554SAndreas Gruenbacher #endif /* CONFIG_NFS_v3 */ 100b7fa0554SAndreas Gruenbacher 1011da177e4SLinus Torvalds /* Hack for future NFS swap support */ 1021da177e4SLinus Torvalds #ifndef IS_SWAPFILE 1031da177e4SLinus Torvalds # define IS_SWAPFILE(inode) (0) 1041da177e4SLinus Torvalds #endif 1051da177e4SLinus Torvalds 1061da177e4SLinus Torvalds static int nfs_check_flags(int flags) 1071da177e4SLinus Torvalds { 1081da177e4SLinus Torvalds if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) 1091da177e4SLinus Torvalds return -EINVAL; 1101da177e4SLinus Torvalds 1111da177e4SLinus Torvalds return 0; 1121da177e4SLinus Torvalds } 1131da177e4SLinus Torvalds 1141da177e4SLinus Torvalds /* 1151da177e4SLinus Torvalds * Open file 1161da177e4SLinus Torvalds */ 1171da177e4SLinus Torvalds static int 1181da177e4SLinus Torvalds nfs_file_open(struct inode *inode, struct file *filp) 1191da177e4SLinus Torvalds { 1201da177e4SLinus Torvalds int res; 1211da177e4SLinus Torvalds 1221da177e4SLinus Torvalds res = nfs_check_flags(filp->f_flags); 1231da177e4SLinus Torvalds if (res) 1241da177e4SLinus Torvalds return res; 1251da177e4SLinus Torvalds 12691d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSOPEN); 1271da177e4SLinus Torvalds lock_kernel(); 1281f163415SDavid Howells res = NFS_PROTO(inode)->file_open(inode, filp); 1291da177e4SLinus Torvalds unlock_kernel(); 1301da177e4SLinus Torvalds return res; 1311da177e4SLinus Torvalds } 1321da177e4SLinus Torvalds 1331da177e4SLinus Torvalds static int 1341da177e4SLinus Torvalds nfs_file_release(struct inode *inode, struct file *filp) 1351da177e4SLinus Torvalds { 1361da177e4SLinus Torvalds /* Ensure that dirty pages are flushed out with the right creds */ 1371da177e4SLinus Torvalds if (filp->f_mode & FMODE_WRITE) 138a49c3c77STrond Myklebust nfs_wb_all(filp->f_path.dentry->d_inode); 13991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSRELEASE); 1401da177e4SLinus Torvalds return NFS_PROTO(inode)->file_release(inode, filp); 1411da177e4SLinus Torvalds } 1421da177e4SLinus Torvalds 143980802e3STrond Myklebust /** 144980802e3STrond Myklebust * nfs_revalidate_size - Revalidate the file size 145980802e3STrond Myklebust * @inode - pointer to inode struct 146980802e3STrond Myklebust * @file - pointer to struct file 147980802e3STrond Myklebust * 148980802e3STrond Myklebust * Revalidates the file length. This is basically a wrapper around 149980802e3STrond Myklebust * nfs_revalidate_inode() that takes into account the fact that we may 150980802e3STrond Myklebust * have cached writes (in which case we don't care about the server's 151980802e3STrond Myklebust * idea of what the file length is), or O_DIRECT (in which case we 152980802e3STrond Myklebust * shouldn't trust the cache). 153980802e3STrond Myklebust */ 154980802e3STrond Myklebust static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) 155980802e3STrond Myklebust { 156980802e3STrond Myklebust struct nfs_server *server = NFS_SERVER(inode); 157980802e3STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 158980802e3STrond Myklebust 159980802e3STrond Myklebust if (server->flags & NFS_MOUNT_NOAC) 160980802e3STrond Myklebust goto force_reval; 161980802e3STrond Myklebust if (filp->f_flags & O_DIRECT) 162980802e3STrond Myklebust goto force_reval; 163980802e3STrond Myklebust if (nfsi->npages != 0) 164980802e3STrond Myklebust return 0; 16555296809SChuck Lever if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode)) 166fe51beecSTrond Myklebust return 0; 167980802e3STrond Myklebust force_reval: 168980802e3STrond Myklebust return __nfs_revalidate_inode(server, inode); 169980802e3STrond Myklebust } 170980802e3STrond Myklebust 171980802e3STrond Myklebust static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin) 172980802e3STrond Myklebust { 173980802e3STrond Myklebust /* origin == SEEK_END => we must revalidate the cached file length */ 174aec5e175SJosef 'Jeff' Sipek if (origin == SEEK_END) { 175980802e3STrond Myklebust struct inode *inode = filp->f_mapping->host; 176980802e3STrond Myklebust int retval = nfs_revalidate_file_size(inode, filp); 177980802e3STrond Myklebust if (retval < 0) 178980802e3STrond Myklebust return (loff_t)retval; 179980802e3STrond Myklebust } 180980802e3STrond Myklebust return remote_llseek(filp, offset, origin); 181980802e3STrond Myklebust } 182980802e3STrond Myklebust 1831da177e4SLinus Torvalds /* 1847b159fc1STrond Myklebust * Helper for nfs_file_flush() and nfs_fsync() 1857b159fc1STrond Myklebust * 1867b159fc1STrond Myklebust * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to 1877b159fc1STrond Myklebust * disk, but it retrieves and clears ctx->error after synching, despite 1887b159fc1STrond Myklebust * the two being set at the same time in nfs_context_set_write_error(). 1897b159fc1STrond Myklebust * This is because the former is used to notify the _next_ call to 1907b159fc1STrond Myklebust * nfs_file_write() that a write error occured, and hence cause it to 1917b159fc1STrond Myklebust * fall back to doing a synchronous write. 1927b159fc1STrond Myklebust */ 1937b159fc1STrond Myklebust static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode) 1947b159fc1STrond Myklebust { 1957b159fc1STrond Myklebust int have_error, status; 1967b159fc1STrond Myklebust int ret = 0; 1977b159fc1STrond Myklebust 1987b159fc1STrond Myklebust have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1997b159fc1STrond Myklebust status = nfs_wb_all(inode); 2007b159fc1STrond Myklebust have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 2017b159fc1STrond Myklebust if (have_error) 2027b159fc1STrond Myklebust ret = xchg(&ctx->error, 0); 2037b159fc1STrond Myklebust if (!ret) 2047b159fc1STrond Myklebust ret = status; 2057b159fc1STrond Myklebust return ret; 2067b159fc1STrond Myklebust } 2077b159fc1STrond Myklebust 2087b159fc1STrond Myklebust /* 2091da177e4SLinus Torvalds * Flush all dirty pages, and check for write errors. 2101da177e4SLinus Torvalds * 2111da177e4SLinus Torvalds */ 2121da177e4SLinus Torvalds static int 21375e1fcc0SMiklos Szeredi nfs_file_flush(struct file *file, fl_owner_t id) 2141da177e4SLinus Torvalds { 215cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 21601cce933SJosef "Jeff" Sipek struct inode *inode = file->f_path.dentry->d_inode; 2171da177e4SLinus Torvalds int status; 2181da177e4SLinus Torvalds 2191da177e4SLinus Torvalds dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino); 2201da177e4SLinus Torvalds 2211da177e4SLinus Torvalds if ((file->f_mode & FMODE_WRITE) == 0) 2221da177e4SLinus Torvalds return 0; 22391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSFLUSH); 2247b159fc1STrond Myklebust 2251da177e4SLinus Torvalds /* Ensure that data+attribute caches are up to date after close() */ 2267b159fc1STrond Myklebust status = nfs_do_fsync(ctx, inode); 2273338c143STrond Myklebust if (!status) 2283338c143STrond Myklebust nfs_revalidate_inode(NFS_SERVER(inode), inode); 2291da177e4SLinus Torvalds return status; 2301da177e4SLinus Torvalds } 2311da177e4SLinus Torvalds 2321da177e4SLinus Torvalds static ssize_t 233027445c3SBadari Pulavarty nfs_file_read(struct kiocb *iocb, const struct iovec *iov, 234027445c3SBadari Pulavarty unsigned long nr_segs, loff_t pos) 2351da177e4SLinus Torvalds { 23601cce933SJosef "Jeff" Sipek struct dentry * dentry = iocb->ki_filp->f_path.dentry; 2371da177e4SLinus Torvalds struct inode * inode = dentry->d_inode; 2381da177e4SLinus Torvalds ssize_t result; 239027445c3SBadari Pulavarty size_t count = iov_length(iov, nr_segs); 2401da177e4SLinus Torvalds 2411da177e4SLinus Torvalds if (iocb->ki_filp->f_flags & O_DIRECT) 242027445c3SBadari Pulavarty return nfs_file_direct_read(iocb, iov, nr_segs, pos); 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n", 2451da177e4SLinus Torvalds dentry->d_parent->d_name.name, dentry->d_name.name, 2461da177e4SLinus Torvalds (unsigned long) count, (unsigned long) pos); 2471da177e4SLinus Torvalds 24844b11874STrond Myklebust result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); 24991d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count); 2501da177e4SLinus Torvalds if (!result) 251027445c3SBadari Pulavarty result = generic_file_aio_read(iocb, iov, nr_segs, pos); 2521da177e4SLinus Torvalds return result; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2551da177e4SLinus Torvalds static ssize_t 256f0930fffSJens Axboe nfs_file_splice_read(struct file *filp, loff_t *ppos, 257f0930fffSJens Axboe struct pipe_inode_info *pipe, size_t count, 258f0930fffSJens Axboe unsigned int flags) 2591da177e4SLinus Torvalds { 26001cce933SJosef "Jeff" Sipek struct dentry *dentry = filp->f_path.dentry; 2611da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 2621da177e4SLinus Torvalds ssize_t res; 2631da177e4SLinus Torvalds 264f0930fffSJens Axboe dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n", 2651da177e4SLinus Torvalds dentry->d_parent->d_name.name, dentry->d_name.name, 2661da177e4SLinus Torvalds (unsigned long) count, (unsigned long long) *ppos); 2671da177e4SLinus Torvalds 26844b11874STrond Myklebust res = nfs_revalidate_mapping(inode, filp->f_mapping); 2691da177e4SLinus Torvalds if (!res) 270f0930fffSJens Axboe res = generic_file_splice_read(filp, ppos, pipe, count, flags); 2711da177e4SLinus Torvalds return res; 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 2741da177e4SLinus Torvalds static int 2751da177e4SLinus Torvalds nfs_file_mmap(struct file * file, struct vm_area_struct * vma) 2761da177e4SLinus Torvalds { 27701cce933SJosef "Jeff" Sipek struct dentry *dentry = file->f_path.dentry; 2781da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 2791da177e4SLinus Torvalds int status; 2801da177e4SLinus Torvalds 2811da177e4SLinus Torvalds dfprintk(VFS, "nfs: mmap(%s/%s)\n", 2821da177e4SLinus Torvalds dentry->d_parent->d_name.name, dentry->d_name.name); 2831da177e4SLinus Torvalds 28444b11874STrond Myklebust status = nfs_revalidate_mapping(inode, file->f_mapping); 28594387fb1STrond Myklebust if (!status) { 28694387fb1STrond Myklebust vma->vm_ops = &nfs_file_vm_ops; 28794387fb1STrond Myklebust vma->vm_flags |= VM_CAN_NONLINEAR; 28894387fb1STrond Myklebust file_accessed(file); 28994387fb1STrond Myklebust } 2901da177e4SLinus Torvalds return status; 2911da177e4SLinus Torvalds } 2921da177e4SLinus Torvalds 2931da177e4SLinus Torvalds /* 2941da177e4SLinus Torvalds * Flush any dirty pages for this process, and check for write errors. 2951da177e4SLinus Torvalds * The return status from this call provides a reliable indication of 2961da177e4SLinus Torvalds * whether any write errors occurred for this process. 2971da177e4SLinus Torvalds */ 2981da177e4SLinus Torvalds static int 2991da177e4SLinus Torvalds nfs_fsync(struct file *file, struct dentry *dentry, int datasync) 3001da177e4SLinus Torvalds { 301cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 3021da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 3031da177e4SLinus Torvalds 3041da177e4SLinus Torvalds dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino); 3051da177e4SLinus Torvalds 30691d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 3077b159fc1STrond Myklebust return nfs_do_fsync(ctx, inode); 3081da177e4SLinus Torvalds } 3091da177e4SLinus Torvalds 3101da177e4SLinus Torvalds /* 3114899f9c8SNick Piggin * This does the "real" work of the write. We must allocate and lock the 3124899f9c8SNick Piggin * page to be sent back to the generic routine, which then copies the 3134899f9c8SNick Piggin * data from user space. 3141da177e4SLinus Torvalds * 3151da177e4SLinus Torvalds * If the writer ends up delaying the write, the writer needs to 3161da177e4SLinus Torvalds * increment the page use counts until he is done with the page. 3171da177e4SLinus Torvalds */ 3184899f9c8SNick Piggin static int nfs_write_begin(struct file *file, struct address_space *mapping, 3194899f9c8SNick Piggin loff_t pos, unsigned len, unsigned flags, 3204899f9c8SNick Piggin struct page **pagep, void **fsdata) 3211da177e4SLinus Torvalds { 3224899f9c8SNick Piggin int ret; 3234899f9c8SNick Piggin pgoff_t index; 3244899f9c8SNick Piggin struct page *page; 3254899f9c8SNick Piggin index = pos >> PAGE_CACHE_SHIFT; 3264899f9c8SNick Piggin 3274899f9c8SNick Piggin page = __grab_cache_page(mapping, index); 3284899f9c8SNick Piggin if (!page) 3294899f9c8SNick Piggin return -ENOMEM; 3304899f9c8SNick Piggin *pagep = page; 3314899f9c8SNick Piggin 3324899f9c8SNick Piggin ret = nfs_flush_incompatible(file, page); 3334899f9c8SNick Piggin if (ret) { 3344899f9c8SNick Piggin unlock_page(page); 3354899f9c8SNick Piggin page_cache_release(page); 3364899f9c8SNick Piggin } 3374899f9c8SNick Piggin return ret; 3381da177e4SLinus Torvalds } 3391da177e4SLinus Torvalds 3404899f9c8SNick Piggin static int nfs_write_end(struct file *file, struct address_space *mapping, 3414899f9c8SNick Piggin loff_t pos, unsigned len, unsigned copied, 3424899f9c8SNick Piggin struct page *page, void *fsdata) 3431da177e4SLinus Torvalds { 3444899f9c8SNick Piggin unsigned offset = pos & (PAGE_CACHE_SIZE - 1); 3454899f9c8SNick Piggin int status; 3461da177e4SLinus Torvalds 347efc91ed0STrond Myklebust /* 348efc91ed0STrond Myklebust * Zero any uninitialised parts of the page, and then mark the page 349efc91ed0STrond Myklebust * as up to date if it turns out that we're extending the file. 350efc91ed0STrond Myklebust */ 351efc91ed0STrond Myklebust if (!PageUptodate(page)) { 352efc91ed0STrond Myklebust unsigned pglen = nfs_page_length(page); 353efc91ed0STrond Myklebust unsigned end = offset + len; 354efc91ed0STrond Myklebust 355efc91ed0STrond Myklebust if (pglen == 0) { 356efc91ed0STrond Myklebust zero_user_segments(page, 0, offset, 357efc91ed0STrond Myklebust end, PAGE_CACHE_SIZE); 358efc91ed0STrond Myklebust SetPageUptodate(page); 359efc91ed0STrond Myklebust } else if (end >= pglen) { 360efc91ed0STrond Myklebust zero_user_segment(page, end, PAGE_CACHE_SIZE); 361efc91ed0STrond Myklebust if (offset == 0) 362efc91ed0STrond Myklebust SetPageUptodate(page); 363efc91ed0STrond Myklebust } else 364efc91ed0STrond Myklebust zero_user_segment(page, pglen, PAGE_CACHE_SIZE); 365efc91ed0STrond Myklebust } 366efc91ed0STrond Myklebust 3671da177e4SLinus Torvalds lock_kernel(); 3684899f9c8SNick Piggin status = nfs_updatepage(file, page, offset, copied); 3691da177e4SLinus Torvalds unlock_kernel(); 3704899f9c8SNick Piggin 3714899f9c8SNick Piggin unlock_page(page); 3724899f9c8SNick Piggin page_cache_release(page); 3734899f9c8SNick Piggin 3743d509e54SChuck Lever if (status < 0) 3753d509e54SChuck Lever return status; 3763d509e54SChuck Lever return copied; 3771da177e4SLinus Torvalds } 3781da177e4SLinus Torvalds 3792ff28e22SNeilBrown static void nfs_invalidate_page(struct page *page, unsigned long offset) 380cd52ed35STrond Myklebust { 3811c75950bSTrond Myklebust if (offset != 0) 3821c75950bSTrond Myklebust return; 383d2ccddf0STrond Myklebust /* Cancel any unstarted writes on this page */ 3841b3b4a1aSTrond Myklebust nfs_wb_page_cancel(page->mapping->host, page); 385cd52ed35STrond Myklebust } 386cd52ed35STrond Myklebust 387cd52ed35STrond Myklebust static int nfs_release_page(struct page *page, gfp_t gfp) 388cd52ed35STrond Myklebust { 389e3db7691STrond Myklebust /* If PagePrivate() is set, then the page is not freeable */ 390ddeff520SNikita Danilov return 0; 391e3db7691STrond Myklebust } 392e3db7691STrond Myklebust 393e3db7691STrond Myklebust static int nfs_launder_page(struct page *page) 394e3db7691STrond Myklebust { 395e3db7691STrond Myklebust return nfs_wb_page(page->mapping->host, page); 396cd52ed35STrond Myklebust } 397cd52ed35STrond Myklebust 398f5e54d6eSChristoph Hellwig const struct address_space_operations nfs_file_aops = { 3991da177e4SLinus Torvalds .readpage = nfs_readpage, 4001da177e4SLinus Torvalds .readpages = nfs_readpages, 4019cccef95STrond Myklebust .set_page_dirty = __set_page_dirty_nobuffers, 4021da177e4SLinus Torvalds .writepage = nfs_writepage, 4031da177e4SLinus Torvalds .writepages = nfs_writepages, 4044899f9c8SNick Piggin .write_begin = nfs_write_begin, 4054899f9c8SNick Piggin .write_end = nfs_write_end, 406cd52ed35STrond Myklebust .invalidatepage = nfs_invalidate_page, 407cd52ed35STrond Myklebust .releasepage = nfs_release_page, 4081da177e4SLinus Torvalds .direct_IO = nfs_direct_IO, 409e3db7691STrond Myklebust .launder_page = nfs_launder_page, 4101da177e4SLinus Torvalds }; 4111da177e4SLinus Torvalds 41294387fb1STrond Myklebust static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page) 41394387fb1STrond Myklebust { 41494387fb1STrond Myklebust struct file *filp = vma->vm_file; 41594387fb1STrond Myklebust unsigned pagelen; 41694387fb1STrond Myklebust int ret = -EINVAL; 4174899f9c8SNick Piggin struct address_space *mapping; 41894387fb1STrond Myklebust 41994387fb1STrond Myklebust lock_page(page); 4204899f9c8SNick Piggin mapping = page->mapping; 4218b1f9ee5STrond Myklebust if (mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping) 4228b1f9ee5STrond Myklebust goto out_unlock; 4238b1f9ee5STrond Myklebust 4248b1f9ee5STrond Myklebust ret = 0; 4254899f9c8SNick Piggin pagelen = nfs_page_length(page); 4268b1f9ee5STrond Myklebust if (pagelen == 0) 4278b1f9ee5STrond Myklebust goto out_unlock; 4288b1f9ee5STrond Myklebust 4298b1f9ee5STrond Myklebust ret = nfs_flush_incompatible(filp, page); 4308b1f9ee5STrond Myklebust if (ret != 0) 4318b1f9ee5STrond Myklebust goto out_unlock; 4328b1f9ee5STrond Myklebust 4338b1f9ee5STrond Myklebust ret = nfs_updatepage(filp, page, 0, pagelen); 4348b1f9ee5STrond Myklebust if (ret == 0) 4358b1f9ee5STrond Myklebust ret = pagelen; 4368b1f9ee5STrond Myklebust out_unlock: 4374899f9c8SNick Piggin unlock_page(page); 43894387fb1STrond Myklebust return ret; 43994387fb1STrond Myklebust } 44094387fb1STrond Myklebust 44194387fb1STrond Myklebust static struct vm_operations_struct nfs_file_vm_ops = { 44294387fb1STrond Myklebust .fault = filemap_fault, 44394387fb1STrond Myklebust .page_mkwrite = nfs_vm_page_mkwrite, 44494387fb1STrond Myklebust }; 44594387fb1STrond Myklebust 4467b159fc1STrond Myklebust static int nfs_need_sync_write(struct file *filp, struct inode *inode) 4477b159fc1STrond Myklebust { 4487b159fc1STrond Myklebust struct nfs_open_context *ctx; 4497b159fc1STrond Myklebust 4507b159fc1STrond Myklebust if (IS_SYNC(inode) || (filp->f_flags & O_SYNC)) 4517b159fc1STrond Myklebust return 1; 452cd3758e3STrond Myklebust ctx = nfs_file_open_context(filp); 4537b159fc1STrond Myklebust if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) 4547b159fc1STrond Myklebust return 1; 4557b159fc1STrond Myklebust return 0; 4567b159fc1STrond Myklebust } 4577b159fc1STrond Myklebust 458027445c3SBadari Pulavarty static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, 459027445c3SBadari Pulavarty unsigned long nr_segs, loff_t pos) 4601da177e4SLinus Torvalds { 46101cce933SJosef "Jeff" Sipek struct dentry * dentry = iocb->ki_filp->f_path.dentry; 4621da177e4SLinus Torvalds struct inode * inode = dentry->d_inode; 4631da177e4SLinus Torvalds ssize_t result; 464027445c3SBadari Pulavarty size_t count = iov_length(iov, nr_segs); 4651da177e4SLinus Torvalds 4661da177e4SLinus Torvalds if (iocb->ki_filp->f_flags & O_DIRECT) 467027445c3SBadari Pulavarty return nfs_file_direct_write(iocb, iov, nr_segs, pos); 4681da177e4SLinus Torvalds 469027445c3SBadari Pulavarty dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n", 4701da177e4SLinus Torvalds dentry->d_parent->d_name.name, dentry->d_name.name, 471027445c3SBadari Pulavarty inode->i_ino, (unsigned long) count, (long long) pos); 4721da177e4SLinus Torvalds 4731da177e4SLinus Torvalds result = -EBUSY; 4741da177e4SLinus Torvalds if (IS_SWAPFILE(inode)) 4751da177e4SLinus Torvalds goto out_swapfile; 4767d52e862STrond Myklebust /* 4777d52e862STrond Myklebust * O_APPEND implies that we must revalidate the file length. 4787d52e862STrond Myklebust */ 4797d52e862STrond Myklebust if (iocb->ki_filp->f_flags & O_APPEND) { 4807d52e862STrond Myklebust result = nfs_revalidate_file_size(inode, iocb->ki_filp); 4811da177e4SLinus Torvalds if (result) 4821da177e4SLinus Torvalds goto out; 483fe51beecSTrond Myklebust } 4841da177e4SLinus Torvalds 4851da177e4SLinus Torvalds result = count; 4861da177e4SLinus Torvalds if (!count) 4871da177e4SLinus Torvalds goto out; 4881da177e4SLinus Torvalds 48991d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count); 490027445c3SBadari Pulavarty result = generic_file_aio_write(iocb, iov, nr_segs, pos); 491200baa21STrond Myklebust /* Return error values for O_SYNC and IS_SYNC() */ 4927b159fc1STrond Myklebust if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { 493cd3758e3STrond Myklebust int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode); 494200baa21STrond Myklebust if (err < 0) 495200baa21STrond Myklebust result = err; 496200baa21STrond Myklebust } 4971da177e4SLinus Torvalds out: 4981da177e4SLinus Torvalds return result; 4991da177e4SLinus Torvalds 5001da177e4SLinus Torvalds out_swapfile: 5011da177e4SLinus Torvalds printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); 5021da177e4SLinus Torvalds goto out; 5031da177e4SLinus Torvalds } 5041da177e4SLinus Torvalds 5051da177e4SLinus Torvalds static int do_getlk(struct file *filp, int cmd, struct file_lock *fl) 5061da177e4SLinus Torvalds { 5071da177e4SLinus Torvalds struct inode *inode = filp->f_mapping->host; 5081da177e4SLinus Torvalds int status = 0; 5091da177e4SLinus Torvalds 5101da177e4SLinus Torvalds lock_kernel(); 511039c4d7aSTrond Myklebust /* Try local locking first */ 5126d34ac19SJ. Bruce Fields posix_test_lock(filp, fl); 5136d34ac19SJ. Bruce Fields if (fl->fl_type != F_UNLCK) { 5146d34ac19SJ. Bruce Fields /* found a conflict */ 515039c4d7aSTrond Myklebust goto out; 5161da177e4SLinus Torvalds } 517039c4d7aSTrond Myklebust 518039c4d7aSTrond Myklebust if (nfs_have_delegation(inode, FMODE_READ)) 519039c4d7aSTrond Myklebust goto out_noconflict; 520039c4d7aSTrond Myklebust 521039c4d7aSTrond Myklebust if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM) 522039c4d7aSTrond Myklebust goto out_noconflict; 523039c4d7aSTrond Myklebust 524039c4d7aSTrond Myklebust status = NFS_PROTO(inode)->lock(filp, cmd, fl); 525039c4d7aSTrond Myklebust out: 5261da177e4SLinus Torvalds unlock_kernel(); 5271da177e4SLinus Torvalds return status; 528039c4d7aSTrond Myklebust out_noconflict: 529039c4d7aSTrond Myklebust fl->fl_type = F_UNLCK; 530039c4d7aSTrond Myklebust goto out; 5311da177e4SLinus Torvalds } 5321da177e4SLinus Torvalds 5331da177e4SLinus Torvalds static int do_vfs_lock(struct file *file, struct file_lock *fl) 5341da177e4SLinus Torvalds { 5351da177e4SLinus Torvalds int res = 0; 5361da177e4SLinus Torvalds switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { 5371da177e4SLinus Torvalds case FL_POSIX: 5381da177e4SLinus Torvalds res = posix_lock_file_wait(file, fl); 5391da177e4SLinus Torvalds break; 5401da177e4SLinus Torvalds case FL_FLOCK: 5411da177e4SLinus Torvalds res = flock_lock_file_wait(file, fl); 5421da177e4SLinus Torvalds break; 5431da177e4SLinus Torvalds default: 5441da177e4SLinus Torvalds BUG(); 5451da177e4SLinus Torvalds } 5461da177e4SLinus Torvalds if (res < 0) 54746bae1a9SNeil Brown dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager" 54846bae1a9SNeil Brown " - error %d!\n", 5493110ff80SHarvey Harrison __func__, res); 5501da177e4SLinus Torvalds return res; 5511da177e4SLinus Torvalds } 5521da177e4SLinus Torvalds 5531da177e4SLinus Torvalds static int do_unlk(struct file *filp, int cmd, struct file_lock *fl) 5541da177e4SLinus Torvalds { 5551da177e4SLinus Torvalds struct inode *inode = filp->f_mapping->host; 5561da177e4SLinus Torvalds int status; 5571da177e4SLinus Torvalds 5581da177e4SLinus Torvalds /* 5591da177e4SLinus Torvalds * Flush all pending writes before doing anything 5601da177e4SLinus Torvalds * with locks.. 5611da177e4SLinus Torvalds */ 56229884df0STrond Myklebust nfs_sync_mapping(filp->f_mapping); 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds /* NOTE: special case 5651da177e4SLinus Torvalds * If we're signalled while cleaning up locks on process exit, we 5661da177e4SLinus Torvalds * still need to complete the unlock. 5671da177e4SLinus Torvalds */ 5681da177e4SLinus Torvalds lock_kernel(); 5691da177e4SLinus Torvalds /* Use local locking if mounted with "-onolock" */ 5701da177e4SLinus Torvalds if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) 5711da177e4SLinus Torvalds status = NFS_PROTO(inode)->lock(filp, cmd, fl); 5721da177e4SLinus Torvalds else 5731da177e4SLinus Torvalds status = do_vfs_lock(filp, fl); 5741da177e4SLinus Torvalds unlock_kernel(); 5751da177e4SLinus Torvalds return status; 5761da177e4SLinus Torvalds } 5771da177e4SLinus Torvalds 5781da177e4SLinus Torvalds static int do_setlk(struct file *filp, int cmd, struct file_lock *fl) 5791da177e4SLinus Torvalds { 5801da177e4SLinus Torvalds struct inode *inode = filp->f_mapping->host; 5811da177e4SLinus Torvalds int status; 5821da177e4SLinus Torvalds 5831da177e4SLinus Torvalds /* 5841da177e4SLinus Torvalds * Flush all pending writes before doing anything 5851da177e4SLinus Torvalds * with locks.. 5861da177e4SLinus Torvalds */ 58729884df0STrond Myklebust status = nfs_sync_mapping(filp->f_mapping); 58829884df0STrond Myklebust if (status != 0) 5891da177e4SLinus Torvalds goto out; 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds lock_kernel(); 5921da177e4SLinus Torvalds /* Use local locking if mounted with "-onolock" */ 593c4d7c402STrond Myklebust if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) 5941da177e4SLinus Torvalds status = NFS_PROTO(inode)->lock(filp, cmd, fl); 595c4d7c402STrond Myklebust else 5961da177e4SLinus Torvalds status = do_vfs_lock(filp, fl); 5971da177e4SLinus Torvalds unlock_kernel(); 5981da177e4SLinus Torvalds if (status < 0) 5991da177e4SLinus Torvalds goto out; 6001da177e4SLinus Torvalds /* 6011da177e4SLinus Torvalds * Make sure we clear the cache whenever we try to get the lock. 6021da177e4SLinus Torvalds * This makes locking act as a cache coherency point. 6031da177e4SLinus Torvalds */ 60429884df0STrond Myklebust nfs_sync_mapping(filp->f_mapping); 605b5418383STrond Myklebust if (!nfs_have_delegation(inode, FMODE_READ)) 6061da177e4SLinus Torvalds nfs_zap_caches(inode); 6071da177e4SLinus Torvalds out: 6081da177e4SLinus Torvalds return status; 6091da177e4SLinus Torvalds } 6101da177e4SLinus Torvalds 6111da177e4SLinus Torvalds /* 6121da177e4SLinus Torvalds * Lock a (portion of) a file 6131da177e4SLinus Torvalds */ 6141da177e4SLinus Torvalds static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) 6151da177e4SLinus Torvalds { 6161da177e4SLinus Torvalds struct inode * inode = filp->f_mapping->host; 6172116271aSTrond Myklebust int ret = -ENOLCK; 6181da177e4SLinus Torvalds 6191da177e4SLinus Torvalds dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n", 6201da177e4SLinus Torvalds inode->i_sb->s_id, inode->i_ino, 6211da177e4SLinus Torvalds fl->fl_type, fl->fl_flags, 6221da177e4SLinus Torvalds (long long)fl->fl_start, (long long)fl->fl_end); 62391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSLOCK); 6241da177e4SLinus Torvalds 6251da177e4SLinus Torvalds /* No mandatory locks over NFS */ 626dfad9441SPavel Emelyanov if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) 6272116271aSTrond Myklebust goto out_err; 6282116271aSTrond Myklebust 6292116271aSTrond Myklebust if (NFS_PROTO(inode)->lock_check_bounds != NULL) { 6302116271aSTrond Myklebust ret = NFS_PROTO(inode)->lock_check_bounds(fl); 6312116271aSTrond Myklebust if (ret < 0) 6322116271aSTrond Myklebust goto out_err; 6332116271aSTrond Myklebust } 6341da177e4SLinus Torvalds 6351da177e4SLinus Torvalds if (IS_GETLK(cmd)) 6362116271aSTrond Myklebust ret = do_getlk(filp, cmd, fl); 6372116271aSTrond Myklebust else if (fl->fl_type == F_UNLCK) 6382116271aSTrond Myklebust ret = do_unlk(filp, cmd, fl); 6392116271aSTrond Myklebust else 6402116271aSTrond Myklebust ret = do_setlk(filp, cmd, fl); 6412116271aSTrond Myklebust out_err: 6422116271aSTrond Myklebust return ret; 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds 6451da177e4SLinus Torvalds /* 6461da177e4SLinus Torvalds * Lock a (portion of) a file 6471da177e4SLinus Torvalds */ 6481da177e4SLinus Torvalds static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) 6491da177e4SLinus Torvalds { 6501da177e4SLinus Torvalds dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n", 65101cce933SJosef "Jeff" Sipek filp->f_path.dentry->d_inode->i_sb->s_id, 65201cce933SJosef "Jeff" Sipek filp->f_path.dentry->d_inode->i_ino, 6531da177e4SLinus Torvalds fl->fl_type, fl->fl_flags); 6541da177e4SLinus Torvalds 6551da177e4SLinus Torvalds /* 6561da177e4SLinus Torvalds * No BSD flocks over NFS allowed. 6571da177e4SLinus Torvalds * Note: we could try to fake a POSIX lock request here by 6581da177e4SLinus Torvalds * using ((u32) filp | 0x80000000) or some such as the pid. 6591da177e4SLinus Torvalds * Not sure whether that would be unique, though, or whether 6601da177e4SLinus Torvalds * that would break in other places. 6611da177e4SLinus Torvalds */ 6621da177e4SLinus Torvalds if (!(fl->fl_flags & FL_FLOCK)) 6631da177e4SLinus Torvalds return -ENOLCK; 6641da177e4SLinus Torvalds 6651da177e4SLinus Torvalds /* We're simulating flock() locks using posix locks on the server */ 6661da177e4SLinus Torvalds fl->fl_owner = (fl_owner_t)filp; 6671da177e4SLinus Torvalds fl->fl_start = 0; 6681da177e4SLinus Torvalds fl->fl_end = OFFSET_MAX; 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds if (fl->fl_type == F_UNLCK) 6711da177e4SLinus Torvalds return do_unlk(filp, cmd, fl); 6721da177e4SLinus Torvalds return do_setlk(filp, cmd, fl); 6731da177e4SLinus Torvalds } 674370f6599SJ. Bruce Fields 675370f6599SJ. Bruce Fields static int nfs_setlease(struct file *file, long arg, struct file_lock **fl) 676370f6599SJ. Bruce Fields { 677370f6599SJ. Bruce Fields /* 678370f6599SJ. Bruce Fields * There is no protocol support for leases, so we have no way 679370f6599SJ. Bruce Fields * to implement them correctly in the face of opens by other 680370f6599SJ. Bruce Fields * clients. 681370f6599SJ. Bruce Fields */ 682370f6599SJ. Bruce Fields return -EINVAL; 683370f6599SJ. Bruce Fields } 684