Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/fs/nfs/file.c |
| 3 | * |
| 4 | * Copyright (C) 1992 Rick Sladkey |
| 5 | * |
| 6 | * Changes Copyright (C) 1994 by Florian La Roche |
| 7 | * - Do not copy data too often around in the kernel. |
| 8 | * - In nfs_file_read the return value of kmalloc wasn't checked. |
| 9 | * - Put in a better version of read look-ahead buffering. Original idea |
| 10 | * and implementation by Wai S Kok elekokws@ee.nus.sg. |
| 11 | * |
| 12 | * Expire cache on write to a file by Wai S Kok (Oct 1994). |
| 13 | * |
| 14 | * Total rewrite of read side for new NFS buffer cache.. Linus. |
| 15 | * |
| 16 | * nfs regular file handling functions |
| 17 | */ |
| 18 | |
| 19 | #include <linux/time.h> |
| 20 | #include <linux/kernel.h> |
| 21 | #include <linux/errno.h> |
| 22 | #include <linux/fcntl.h> |
| 23 | #include <linux/stat.h> |
| 24 | #include <linux/nfs_fs.h> |
| 25 | #include <linux/nfs_mount.h> |
| 26 | #include <linux/mm.h> |
| 27 | #include <linux/slab.h> |
| 28 | #include <linux/pagemap.h> |
| 29 | #include <linux/smp_lock.h> |
| 30 | |
| 31 | #include <asm/uaccess.h> |
| 32 | #include <asm/system.h> |
| 33 | |
| 34 | #include "delegation.h" |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 35 | #include "iostat.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | |
| 37 | #define NFSDBG_FACILITY NFSDBG_FILE |
| 38 | |
| 39 | static int nfs_file_open(struct inode *, struct file *); |
| 40 | static int nfs_file_release(struct inode *, struct file *); |
Trond Myklebust | 980802e | 2005-06-13 11:14:01 -0400 | [diff] [blame] | 41 | static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | static int nfs_file_mmap(struct file *, struct vm_area_struct *); |
| 43 | static ssize_t nfs_file_sendfile(struct file *, loff_t *, size_t, read_actor_t, void *); |
| 44 | static ssize_t nfs_file_read(struct kiocb *, char __user *, size_t, loff_t); |
| 45 | static ssize_t nfs_file_write(struct kiocb *, const char __user *, size_t, loff_t); |
| 46 | static int nfs_file_flush(struct file *); |
| 47 | static int nfs_fsync(struct file *, struct dentry *dentry, int datasync); |
| 48 | static int nfs_check_flags(int flags); |
| 49 | static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl); |
| 50 | static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl); |
| 51 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 52 | const struct file_operations nfs_file_operations = { |
Trond Myklebust | 980802e | 2005-06-13 11:14:01 -0400 | [diff] [blame] | 53 | .llseek = nfs_file_llseek, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | .read = do_sync_read, |
| 55 | .write = do_sync_write, |
| 56 | .aio_read = nfs_file_read, |
| 57 | .aio_write = nfs_file_write, |
| 58 | .mmap = nfs_file_mmap, |
| 59 | .open = nfs_file_open, |
| 60 | .flush = nfs_file_flush, |
| 61 | .release = nfs_file_release, |
| 62 | .fsync = nfs_fsync, |
| 63 | .lock = nfs_lock, |
| 64 | .flock = nfs_flock, |
| 65 | .sendfile = nfs_file_sendfile, |
| 66 | .check_flags = nfs_check_flags, |
| 67 | }; |
| 68 | |
| 69 | struct inode_operations nfs_file_inode_operations = { |
| 70 | .permission = nfs_permission, |
| 71 | .getattr = nfs_getattr, |
| 72 | .setattr = nfs_setattr, |
| 73 | }; |
| 74 | |
Andreas Gruenbacher | b7fa055 | 2005-06-22 17:16:27 +0000 | [diff] [blame] | 75 | #ifdef CONFIG_NFS_V3 |
| 76 | struct inode_operations nfs3_file_inode_operations = { |
| 77 | .permission = nfs_permission, |
| 78 | .getattr = nfs_getattr, |
| 79 | .setattr = nfs_setattr, |
| 80 | .listxattr = nfs3_listxattr, |
| 81 | .getxattr = nfs3_getxattr, |
| 82 | .setxattr = nfs3_setxattr, |
| 83 | .removexattr = nfs3_removexattr, |
| 84 | }; |
| 85 | #endif /* CONFIG_NFS_v3 */ |
| 86 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 87 | /* Hack for future NFS swap support */ |
| 88 | #ifndef IS_SWAPFILE |
| 89 | # define IS_SWAPFILE(inode) (0) |
| 90 | #endif |
| 91 | |
| 92 | static int nfs_check_flags(int flags) |
| 93 | { |
| 94 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) |
| 95 | return -EINVAL; |
| 96 | |
| 97 | return 0; |
| 98 | } |
| 99 | |
| 100 | /* |
| 101 | * Open file |
| 102 | */ |
| 103 | static int |
| 104 | nfs_file_open(struct inode *inode, struct file *filp) |
| 105 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | int res; |
| 107 | |
| 108 | res = nfs_check_flags(filp->f_flags); |
| 109 | if (res) |
| 110 | return res; |
| 111 | |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 112 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | lock_kernel(); |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 114 | res = NFS_SERVER(inode)->rpc_ops->file_open(inode, filp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | unlock_kernel(); |
| 116 | return res; |
| 117 | } |
| 118 | |
| 119 | static int |
| 120 | nfs_file_release(struct inode *inode, struct file *filp) |
| 121 | { |
| 122 | /* Ensure that dirty pages are flushed out with the right creds */ |
| 123 | if (filp->f_mode & FMODE_WRITE) |
| 124 | filemap_fdatawrite(filp->f_mapping); |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 125 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | return NFS_PROTO(inode)->file_release(inode, filp); |
| 127 | } |
| 128 | |
Trond Myklebust | 980802e | 2005-06-13 11:14:01 -0400 | [diff] [blame] | 129 | /** |
Trond Myklebust | fe51beec | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 130 | * nfs_revalidate_file - Revalidate the page cache & related metadata |
| 131 | * @inode - pointer to inode struct |
| 132 | * @file - pointer to file |
| 133 | */ |
| 134 | static int nfs_revalidate_file(struct inode *inode, struct file *filp) |
| 135 | { |
Chuck Lever | 5529680 | 2005-08-18 11:24:09 -0700 | [diff] [blame] | 136 | struct nfs_inode *nfsi = NFS_I(inode); |
Trond Myklebust | fe51beec | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 137 | int retval = 0; |
| 138 | |
Trond Myklebust | b3c52da | 2005-10-17 06:02:00 -0400 | [diff] [blame] | 139 | if ((nfsi->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATTR)) |
| 140 | || nfs_attribute_timeout(inode)) |
Trond Myklebust | fe51beec | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 141 | retval = __nfs_revalidate_inode(NFS_SERVER(inode), inode); |
| 142 | nfs_revalidate_mapping(inode, filp->f_mapping); |
| 143 | return 0; |
| 144 | } |
| 145 | |
| 146 | /** |
Trond Myklebust | 980802e | 2005-06-13 11:14:01 -0400 | [diff] [blame] | 147 | * nfs_revalidate_size - Revalidate the file size |
| 148 | * @inode - pointer to inode struct |
| 149 | * @file - pointer to struct file |
| 150 | * |
| 151 | * Revalidates the file length. This is basically a wrapper around |
| 152 | * nfs_revalidate_inode() that takes into account the fact that we may |
| 153 | * have cached writes (in which case we don't care about the server's |
| 154 | * idea of what the file length is), or O_DIRECT (in which case we |
| 155 | * shouldn't trust the cache). |
| 156 | */ |
| 157 | static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) |
| 158 | { |
| 159 | struct nfs_server *server = NFS_SERVER(inode); |
| 160 | struct nfs_inode *nfsi = NFS_I(inode); |
| 161 | |
| 162 | if (server->flags & NFS_MOUNT_NOAC) |
| 163 | goto force_reval; |
| 164 | if (filp->f_flags & O_DIRECT) |
| 165 | goto force_reval; |
| 166 | if (nfsi->npages != 0) |
| 167 | return 0; |
Chuck Lever | 5529680 | 2005-08-18 11:24:09 -0700 | [diff] [blame] | 168 | if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode)) |
Trond Myklebust | fe51beec | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 169 | return 0; |
Trond Myklebust | 980802e | 2005-06-13 11:14:01 -0400 | [diff] [blame] | 170 | force_reval: |
| 171 | return __nfs_revalidate_inode(server, inode); |
| 172 | } |
| 173 | |
| 174 | static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin) |
| 175 | { |
| 176 | /* origin == SEEK_END => we must revalidate the cached file length */ |
| 177 | if (origin == 2) { |
| 178 | struct inode *inode = filp->f_mapping->host; |
| 179 | int retval = nfs_revalidate_file_size(inode, filp); |
| 180 | if (retval < 0) |
| 181 | return (loff_t)retval; |
| 182 | } |
| 183 | return remote_llseek(filp, offset, origin); |
| 184 | } |
| 185 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | /* |
| 187 | * Flush all dirty pages, and check for write errors. |
| 188 | * |
| 189 | */ |
| 190 | static int |
| 191 | nfs_file_flush(struct file *file) |
| 192 | { |
| 193 | struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; |
| 194 | struct inode *inode = file->f_dentry->d_inode; |
| 195 | int status; |
| 196 | |
| 197 | dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino); |
| 198 | |
| 199 | if ((file->f_mode & FMODE_WRITE) == 0) |
| 200 | return 0; |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 201 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | lock_kernel(); |
| 203 | /* Ensure that data+attribute caches are up to date after close() */ |
| 204 | status = nfs_wb_all(inode); |
| 205 | if (!status) { |
| 206 | status = ctx->error; |
| 207 | ctx->error = 0; |
Trond Myklebust | 3338c14 | 2005-10-27 22:12:41 -0400 | [diff] [blame] | 208 | if (!status) |
| 209 | nfs_revalidate_inode(NFS_SERVER(inode), inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | } |
| 211 | unlock_kernel(); |
| 212 | return status; |
| 213 | } |
| 214 | |
| 215 | static ssize_t |
| 216 | nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos) |
| 217 | { |
| 218 | struct dentry * dentry = iocb->ki_filp->f_dentry; |
| 219 | struct inode * inode = dentry->d_inode; |
| 220 | ssize_t result; |
| 221 | |
| 222 | #ifdef CONFIG_NFS_DIRECTIO |
| 223 | if (iocb->ki_filp->f_flags & O_DIRECT) |
| 224 | return nfs_file_direct_read(iocb, buf, count, pos); |
| 225 | #endif |
| 226 | |
| 227 | dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n", |
| 228 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 229 | (unsigned long) count, (unsigned long) pos); |
| 230 | |
Trond Myklebust | fe51beec | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 231 | result = nfs_revalidate_file(inode, iocb->ki_filp); |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 232 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | if (!result) |
| 234 | result = generic_file_aio_read(iocb, buf, count, pos); |
| 235 | return result; |
| 236 | } |
| 237 | |
| 238 | static ssize_t |
| 239 | nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count, |
| 240 | read_actor_t actor, void *target) |
| 241 | { |
| 242 | struct dentry *dentry = filp->f_dentry; |
| 243 | struct inode *inode = dentry->d_inode; |
| 244 | ssize_t res; |
| 245 | |
| 246 | dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n", |
| 247 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 248 | (unsigned long) count, (unsigned long long) *ppos); |
| 249 | |
Trond Myklebust | fe51beec | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 250 | res = nfs_revalidate_file(inode, filp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | if (!res) |
| 252 | res = generic_file_sendfile(filp, ppos, count, actor, target); |
| 253 | return res; |
| 254 | } |
| 255 | |
| 256 | static int |
| 257 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
| 258 | { |
| 259 | struct dentry *dentry = file->f_dentry; |
| 260 | struct inode *inode = dentry->d_inode; |
| 261 | int status; |
| 262 | |
| 263 | dfprintk(VFS, "nfs: mmap(%s/%s)\n", |
| 264 | dentry->d_parent->d_name.name, dentry->d_name.name); |
| 265 | |
Trond Myklebust | fe51beec | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 266 | status = nfs_revalidate_file(inode, file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | if (!status) |
| 268 | status = generic_file_mmap(file, vma); |
| 269 | return status; |
| 270 | } |
| 271 | |
| 272 | /* |
| 273 | * Flush any dirty pages for this process, and check for write errors. |
| 274 | * The return status from this call provides a reliable indication of |
| 275 | * whether any write errors occurred for this process. |
| 276 | */ |
| 277 | static int |
| 278 | nfs_fsync(struct file *file, struct dentry *dentry, int datasync) |
| 279 | { |
| 280 | struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; |
| 281 | struct inode *inode = dentry->d_inode; |
| 282 | int status; |
| 283 | |
| 284 | dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino); |
| 285 | |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 286 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | lock_kernel(); |
| 288 | status = nfs_wb_all(inode); |
| 289 | if (!status) { |
| 290 | status = ctx->error; |
| 291 | ctx->error = 0; |
| 292 | } |
| 293 | unlock_kernel(); |
| 294 | return status; |
| 295 | } |
| 296 | |
| 297 | /* |
| 298 | * This does the "real" work of the write. The generic routine has |
| 299 | * allocated the page, locked it, done all the page alignment stuff |
| 300 | * calculations etc. Now we should just copy the data from user |
| 301 | * space and write it back to the real medium.. |
| 302 | * |
| 303 | * If the writer ends up delaying the write, the writer needs to |
| 304 | * increment the page use counts until he is done with the page. |
| 305 | */ |
| 306 | static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to) |
| 307 | { |
| 308 | return nfs_flush_incompatible(file, page); |
| 309 | } |
| 310 | |
| 311 | static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to) |
| 312 | { |
| 313 | long status; |
| 314 | |
| 315 | lock_kernel(); |
| 316 | status = nfs_updatepage(file, page, offset, to-offset); |
| 317 | unlock_kernel(); |
| 318 | return status; |
| 319 | } |
| 320 | |
NeilBrown | 2ff28e2 | 2006-03-26 01:37:18 -0800 | [diff] [blame] | 321 | static void nfs_invalidate_page(struct page *page, unsigned long offset) |
Trond Myklebust | cd52ed3 | 2006-03-20 13:44:04 -0500 | [diff] [blame] | 322 | { |
| 323 | /* FIXME: we really should cancel any unstarted writes on this page */ |
Trond Myklebust | cd52ed3 | 2006-03-20 13:44:04 -0500 | [diff] [blame] | 324 | } |
| 325 | |
| 326 | static int nfs_release_page(struct page *page, gfp_t gfp) |
| 327 | { |
| 328 | return !nfs_wb_page(page->mapping->host, page); |
| 329 | } |
| 330 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | struct address_space_operations nfs_file_aops = { |
| 332 | .readpage = nfs_readpage, |
| 333 | .readpages = nfs_readpages, |
| 334 | .set_page_dirty = __set_page_dirty_nobuffers, |
| 335 | .writepage = nfs_writepage, |
| 336 | .writepages = nfs_writepages, |
| 337 | .prepare_write = nfs_prepare_write, |
| 338 | .commit_write = nfs_commit_write, |
Trond Myklebust | cd52ed3 | 2006-03-20 13:44:04 -0500 | [diff] [blame] | 339 | .invalidatepage = nfs_invalidate_page, |
| 340 | .releasepage = nfs_release_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | #ifdef CONFIG_NFS_DIRECTIO |
| 342 | .direct_IO = nfs_direct_IO, |
| 343 | #endif |
| 344 | }; |
| 345 | |
| 346 | /* |
| 347 | * Write to a file (through the page cache). |
| 348 | */ |
| 349 | static ssize_t |
| 350 | nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos) |
| 351 | { |
| 352 | struct dentry * dentry = iocb->ki_filp->f_dentry; |
| 353 | struct inode * inode = dentry->d_inode; |
| 354 | ssize_t result; |
| 355 | |
| 356 | #ifdef CONFIG_NFS_DIRECTIO |
| 357 | if (iocb->ki_filp->f_flags & O_DIRECT) |
| 358 | return nfs_file_direct_write(iocb, buf, count, pos); |
| 359 | #endif |
| 360 | |
| 361 | dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n", |
| 362 | dentry->d_parent->d_name.name, dentry->d_name.name, |
| 363 | inode->i_ino, (unsigned long) count, (unsigned long) pos); |
| 364 | |
| 365 | result = -EBUSY; |
| 366 | if (IS_SWAPFILE(inode)) |
| 367 | goto out_swapfile; |
Trond Myklebust | 7d52e86 | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 368 | /* |
| 369 | * O_APPEND implies that we must revalidate the file length. |
| 370 | */ |
| 371 | if (iocb->ki_filp->f_flags & O_APPEND) { |
| 372 | result = nfs_revalidate_file_size(inode, iocb->ki_filp); |
| 373 | if (result) |
| 374 | goto out; |
Trond Myklebust | fe51beec | 2005-06-22 17:16:30 +0000 | [diff] [blame] | 375 | } |
| 376 | nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | |
| 378 | result = count; |
| 379 | if (!count) |
| 380 | goto out; |
| 381 | |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 382 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | result = generic_file_aio_write(iocb, buf, count, pos); |
| 384 | out: |
| 385 | return result; |
| 386 | |
| 387 | out_swapfile: |
| 388 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); |
| 389 | goto out; |
| 390 | } |
| 391 | |
| 392 | static int do_getlk(struct file *filp, int cmd, struct file_lock *fl) |
| 393 | { |
Andy Adamson | 8dc7c31 | 2006-03-20 13:44:26 -0500 | [diff] [blame] | 394 | struct file_lock cfl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | struct inode *inode = filp->f_mapping->host; |
| 396 | int status = 0; |
| 397 | |
| 398 | lock_kernel(); |
Trond Myklebust | 039c4d7 | 2005-10-18 14:20:16 -0700 | [diff] [blame] | 399 | /* Try local locking first */ |
Andy Adamson | 8dc7c31 | 2006-03-20 13:44:26 -0500 | [diff] [blame] | 400 | if (posix_test_lock(filp, fl, &cfl)) { |
Trond Myklebust | e4cd038 | 2006-03-20 13:44:44 -0500 | [diff] [blame] | 401 | fl->fl_start = cfl.fl_start; |
| 402 | fl->fl_end = cfl.fl_end; |
| 403 | fl->fl_type = cfl.fl_type; |
| 404 | fl->fl_pid = cfl.fl_pid; |
Trond Myklebust | 039c4d7 | 2005-10-18 14:20:16 -0700 | [diff] [blame] | 405 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | } |
Trond Myklebust | 039c4d7 | 2005-10-18 14:20:16 -0700 | [diff] [blame] | 407 | |
| 408 | if (nfs_have_delegation(inode, FMODE_READ)) |
| 409 | goto out_noconflict; |
| 410 | |
| 411 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM) |
| 412 | goto out_noconflict; |
| 413 | |
| 414 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
| 415 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | unlock_kernel(); |
| 417 | return status; |
Trond Myklebust | 039c4d7 | 2005-10-18 14:20:16 -0700 | [diff] [blame] | 418 | out_noconflict: |
| 419 | fl->fl_type = F_UNLCK; |
| 420 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | } |
| 422 | |
| 423 | static int do_vfs_lock(struct file *file, struct file_lock *fl) |
| 424 | { |
| 425 | int res = 0; |
| 426 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { |
| 427 | case FL_POSIX: |
| 428 | res = posix_lock_file_wait(file, fl); |
| 429 | break; |
| 430 | case FL_FLOCK: |
| 431 | res = flock_lock_file_wait(file, fl); |
| 432 | break; |
| 433 | default: |
| 434 | BUG(); |
| 435 | } |
| 436 | if (res < 0) |
| 437 | printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n", |
| 438 | __FUNCTION__); |
| 439 | return res; |
| 440 | } |
| 441 | |
| 442 | static int do_unlk(struct file *filp, int cmd, struct file_lock *fl) |
| 443 | { |
| 444 | struct inode *inode = filp->f_mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | int status; |
| 446 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 447 | /* |
| 448 | * Flush all pending writes before doing anything |
| 449 | * with locks.. |
| 450 | */ |
Trond Myklebust | 29884df | 2005-12-13 16:13:54 -0500 | [diff] [blame] | 451 | nfs_sync_mapping(filp->f_mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 452 | |
| 453 | /* NOTE: special case |
| 454 | * If we're signalled while cleaning up locks on process exit, we |
| 455 | * still need to complete the unlock. |
| 456 | */ |
| 457 | lock_kernel(); |
| 458 | /* Use local locking if mounted with "-onolock" */ |
| 459 | if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) |
| 460 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
| 461 | else |
| 462 | status = do_vfs_lock(filp, fl); |
| 463 | unlock_kernel(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | return status; |
| 465 | } |
| 466 | |
| 467 | static int do_setlk(struct file *filp, int cmd, struct file_lock *fl) |
| 468 | { |
| 469 | struct inode *inode = filp->f_mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | int status; |
| 471 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | /* |
| 473 | * Flush all pending writes before doing anything |
| 474 | * with locks.. |
| 475 | */ |
Trond Myklebust | 29884df | 2005-12-13 16:13:54 -0500 | [diff] [blame] | 476 | status = nfs_sync_mapping(filp->f_mapping); |
| 477 | if (status != 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | goto out; |
| 479 | |
| 480 | lock_kernel(); |
| 481 | /* Use local locking if mounted with "-onolock" */ |
| 482 | if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) { |
| 483 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
| 484 | /* If we were signalled we still need to ensure that |
| 485 | * we clean up any state on the server. We therefore |
| 486 | * record the lock call as having succeeded in order to |
| 487 | * ensure that locks_remove_posix() cleans it out when |
| 488 | * the process exits. |
| 489 | */ |
| 490 | if (status == -EINTR || status == -ERESTARTSYS) |
| 491 | do_vfs_lock(filp, fl); |
| 492 | } else |
| 493 | status = do_vfs_lock(filp, fl); |
| 494 | unlock_kernel(); |
| 495 | if (status < 0) |
| 496 | goto out; |
| 497 | /* |
| 498 | * Make sure we clear the cache whenever we try to get the lock. |
| 499 | * This makes locking act as a cache coherency point. |
| 500 | */ |
Trond Myklebust | 29884df | 2005-12-13 16:13:54 -0500 | [diff] [blame] | 501 | nfs_sync_mapping(filp->f_mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 502 | nfs_zap_caches(inode); |
| 503 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 504 | return status; |
| 505 | } |
| 506 | |
| 507 | /* |
| 508 | * Lock a (portion of) a file |
| 509 | */ |
| 510 | static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
| 511 | { |
| 512 | struct inode * inode = filp->f_mapping->host; |
| 513 | |
| 514 | dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n", |
| 515 | inode->i_sb->s_id, inode->i_ino, |
| 516 | fl->fl_type, fl->fl_flags, |
| 517 | (long long)fl->fl_start, (long long)fl->fl_end); |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 518 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | |
| 520 | /* No mandatory locks over NFS */ |
ASANO Masahiro | 0800c5f | 2005-12-22 13:24:54 +0900 | [diff] [blame] | 521 | if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID && |
| 522 | fl->fl_type != F_UNLCK) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 523 | return -ENOLCK; |
| 524 | |
| 525 | if (IS_GETLK(cmd)) |
| 526 | return do_getlk(filp, cmd, fl); |
| 527 | if (fl->fl_type == F_UNLCK) |
| 528 | return do_unlk(filp, cmd, fl); |
| 529 | return do_setlk(filp, cmd, fl); |
| 530 | } |
| 531 | |
| 532 | /* |
| 533 | * Lock a (portion of) a file |
| 534 | */ |
| 535 | static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
| 536 | { |
| 537 | struct inode * inode = filp->f_mapping->host; |
| 538 | |
| 539 | dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n", |
| 540 | inode->i_sb->s_id, inode->i_ino, |
| 541 | fl->fl_type, fl->fl_flags); |
| 542 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 543 | /* |
| 544 | * No BSD flocks over NFS allowed. |
| 545 | * Note: we could try to fake a POSIX lock request here by |
| 546 | * using ((u32) filp | 0x80000000) or some such as the pid. |
| 547 | * Not sure whether that would be unique, though, or whether |
| 548 | * that would break in other places. |
| 549 | */ |
| 550 | if (!(fl->fl_flags & FL_FLOCK)) |
| 551 | return -ENOLCK; |
| 552 | |
| 553 | /* We're simulating flock() locks using posix locks on the server */ |
| 554 | fl->fl_owner = (fl_owner_t)filp; |
| 555 | fl->fl_start = 0; |
| 556 | fl->fl_end = OFFSET_MAX; |
| 557 | |
| 558 | if (fl->fl_type == F_UNLCK) |
| 559 | return do_unlk(filp, cmd, fl); |
| 560 | return do_setlk(filp, cmd, fl); |
| 561 | } |