1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/nfs/file.c 4 * 5 * Copyright (C) 1992 Rick Sladkey 6 */ 7 #include <linux/fs.h> 8 #include <linux/file.h> 9 #include <linux/falloc.h> 10 #include <linux/mount.h> 11 #include <linux/nfs_fs.h> 12 #include "delegation.h" 13 #include "internal.h" 14 #include "iostat.h" 15 #include "fscache.h" 16 #include "pnfs.h" 17 18 #include "nfstrace.h" 19 20 #ifdef CONFIG_NFS_V4_2 21 #include "nfs42.h" 22 #endif 23 24 #define NFSDBG_FACILITY NFSDBG_FILE 25 26 static int 27 nfs4_file_open(struct inode *inode, struct file *filp) 28 { 29 struct nfs_open_context *ctx; 30 struct dentry *dentry = file_dentry(filp); 31 struct dentry *parent = NULL; 32 struct inode *dir; 33 unsigned openflags = filp->f_flags; 34 struct iattr attr; 35 int err; 36 37 /* 38 * If no cached dentry exists or if it's negative, NFSv4 handled the 39 * opens in ->lookup() or ->create(). 40 * 41 * We only get this far for a cached positive dentry. We skipped 42 * revalidation, so handle it here by dropping the dentry and returning 43 * -EOPENSTALE. The VFS will retry the lookup/create/open. 44 */ 45 46 dprintk("NFS: open file(%pd2)\n", dentry); 47 48 err = nfs_check_flags(openflags); 49 if (err) 50 return err; 51 52 if ((openflags & O_ACCMODE) == 3) 53 return nfs_open(inode, filp); 54 55 /* We can't create new files here */ 56 openflags &= ~(O_CREAT|O_EXCL); 57 58 parent = dget_parent(dentry); 59 dir = d_inode(parent); 60 61 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp); 62 err = PTR_ERR(ctx); 63 if (IS_ERR(ctx)) 64 goto out; 65 66 attr.ia_valid = ATTR_OPEN; 67 if (openflags & O_TRUNC) { 68 attr.ia_valid |= ATTR_SIZE; 69 attr.ia_size = 0; 70 filemap_write_and_wait(inode->i_mapping); 71 } 72 73 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL); 74 if (IS_ERR(inode)) { 75 err = PTR_ERR(inode); 76 switch (err) { 77 default: 78 goto out_put_ctx; 79 case -ENOENT: 80 case -ESTALE: 81 case -EISDIR: 82 case -ENOTDIR: 83 case -ELOOP: 84 goto out_drop; 85 } 86 } 87 if (inode != d_inode(dentry)) 88 goto out_drop; 89 90 nfs_file_set_open_context(filp, ctx); 91 nfs_fscache_open_file(inode, filp); 92 err = 0; 93 94 out_put_ctx: 95 put_nfs_open_context(ctx); 96 out: 97 dput(parent); 98 return err; 99 100 out_drop: 101 d_drop(dentry); 102 err = -EOPENSTALE; 103 goto out_put_ctx; 104 } 105 106 /* 107 * Flush all dirty pages, and check for write errors. 108 */ 109 static int 110 nfs4_file_flush(struct file *file, fl_owner_t id) 111 { 112 struct inode *inode = file_inode(file); 113 114 dprintk("NFS: flush(%pD2)\n", file); 115 116 nfs_inc_stats(inode, NFSIOS_VFSFLUSH); 117 if ((file->f_mode & FMODE_WRITE) == 0) 118 return 0; 119 120 /* 121 * If we're holding a write delegation, then check if we're required 122 * to flush the i/o on close. If not, then just start the i/o now. 123 */ 124 if (!nfs4_delegation_flush_on_close(inode)) 125 return filemap_fdatawrite(file->f_mapping); 126 127 /* Flush writes to the server and return any errors */ 128 return nfs_wb_all(inode); 129 } 130 131 #ifdef CONFIG_NFS_V4_2 132 static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in, 133 struct file *file_out, loff_t pos_out, 134 size_t count, unsigned int flags) 135 { 136 struct nfs42_copy_notify_res *cn_resp = NULL; 137 struct nl4_server *nss = NULL; 138 nfs4_stateid *cnrs = NULL; 139 ssize_t ret; 140 bool sync = false; 141 142 /* Only offload copy if superblock is the same */ 143 if (file_in->f_op != &nfs4_file_operations) 144 return -EXDEV; 145 if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY)) 146 return -EOPNOTSUPP; 147 if (file_inode(file_in) == file_inode(file_out)) 148 return -EOPNOTSUPP; 149 /* if the copy size if smaller than 2 RPC payloads, make it 150 * synchronous 151 */ 152 if (count <= 2 * NFS_SERVER(file_inode(file_in))->rsize) 153 sync = true; 154 retry: 155 if (!nfs42_files_from_same_server(file_in, file_out)) { 156 /* for inter copy, if copy size if smaller than 12 RPC 157 * payloads, fallback to traditional copy. There are 158 * 14 RPCs during an NFSv4.x mount between source/dest 159 * servers. 160 */ 161 if (sync || 162 count <= 14 * NFS_SERVER(file_inode(file_in))->rsize) 163 return -EOPNOTSUPP; 164 cn_resp = kzalloc(sizeof(struct nfs42_copy_notify_res), 165 GFP_NOFS); 166 if (unlikely(cn_resp == NULL)) 167 return -ENOMEM; 168 169 ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp); 170 if (ret) { 171 ret = -EOPNOTSUPP; 172 goto out; 173 } 174 nss = &cn_resp->cnr_src; 175 cnrs = &cn_resp->cnr_stateid; 176 } 177 ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count, 178 nss, cnrs, sync); 179 out: 180 if (!nfs42_files_from_same_server(file_in, file_out)) 181 kfree(cn_resp); 182 if (ret == -EAGAIN) 183 goto retry; 184 return ret; 185 } 186 187 static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in, 188 struct file *file_out, loff_t pos_out, 189 size_t count, unsigned int flags) 190 { 191 ssize_t ret; 192 193 ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count, 194 flags); 195 if (ret == -EOPNOTSUPP || ret == -EXDEV) 196 ret = generic_copy_file_range(file_in, pos_in, file_out, 197 pos_out, count, flags); 198 return ret; 199 } 200 201 static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence) 202 { 203 loff_t ret; 204 205 switch (whence) { 206 case SEEK_HOLE: 207 case SEEK_DATA: 208 ret = nfs42_proc_llseek(filep, offset, whence); 209 if (ret != -ENOTSUPP) 210 return ret; 211 /* Fall through */ 212 default: 213 return nfs_file_llseek(filep, offset, whence); 214 } 215 } 216 217 static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len) 218 { 219 struct inode *inode = file_inode(filep); 220 long ret; 221 222 if (!S_ISREG(inode->i_mode)) 223 return -EOPNOTSUPP; 224 225 if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))) 226 return -EOPNOTSUPP; 227 228 ret = inode_newsize_ok(inode, offset + len); 229 if (ret < 0) 230 return ret; 231 232 if (mode & FALLOC_FL_PUNCH_HOLE) 233 return nfs42_proc_deallocate(filep, offset, len); 234 return nfs42_proc_allocate(filep, offset, len); 235 } 236 237 static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off, 238 struct file *dst_file, loff_t dst_off, loff_t count, 239 unsigned int remap_flags) 240 { 241 struct inode *dst_inode = file_inode(dst_file); 242 struct nfs_server *server = NFS_SERVER(dst_inode); 243 struct inode *src_inode = file_inode(src_file); 244 unsigned int bs = server->clone_blksize; 245 bool same_inode = false; 246 int ret; 247 248 /* NFS does not support deduplication. */ 249 if (remap_flags & REMAP_FILE_DEDUP) 250 return -EOPNOTSUPP; 251 252 if (remap_flags & ~REMAP_FILE_ADVISORY) 253 return -EINVAL; 254 255 if (IS_SWAPFILE(dst_inode) || IS_SWAPFILE(src_inode)) 256 return -ETXTBSY; 257 258 /* check alignment w.r.t. clone_blksize */ 259 ret = -EINVAL; 260 if (bs) { 261 if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs)) 262 goto out; 263 if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count)) 264 goto out; 265 } 266 267 if (src_inode == dst_inode) 268 same_inode = true; 269 270 /* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */ 271 if (same_inode) { 272 inode_lock(src_inode); 273 } else if (dst_inode < src_inode) { 274 inode_lock_nested(dst_inode, I_MUTEX_PARENT); 275 inode_lock_nested(src_inode, I_MUTEX_CHILD); 276 } else { 277 inode_lock_nested(src_inode, I_MUTEX_PARENT); 278 inode_lock_nested(dst_inode, I_MUTEX_CHILD); 279 } 280 281 /* flush all pending writes on both src and dst so that server 282 * has the latest data */ 283 ret = nfs_sync_inode(src_inode); 284 if (ret) 285 goto out_unlock; 286 ret = nfs_sync_inode(dst_inode); 287 if (ret) 288 goto out_unlock; 289 290 ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count); 291 292 /* truncate inode page cache of the dst range so that future reads can fetch 293 * new data from server */ 294 if (!ret) 295 truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1); 296 297 out_unlock: 298 if (same_inode) { 299 inode_unlock(src_inode); 300 } else if (dst_inode < src_inode) { 301 inode_unlock(src_inode); 302 inode_unlock(dst_inode); 303 } else { 304 inode_unlock(dst_inode); 305 inode_unlock(src_inode); 306 } 307 out: 308 return ret < 0 ? ret : count; 309 } 310 311 static int read_name_gen = 1; 312 #define SSC_READ_NAME_BODY "ssc_read_%d" 313 314 struct file * 315 nfs42_ssc_open(struct vfsmount *ss_mnt, struct nfs_fh *src_fh, 316 nfs4_stateid *stateid) 317 { 318 struct nfs_fattr fattr; 319 struct file *filep, *res; 320 struct nfs_server *server; 321 struct inode *r_ino = NULL; 322 struct nfs_open_context *ctx; 323 struct nfs4_state_owner *sp; 324 char *read_name = NULL; 325 int len, status = 0; 326 327 server = NFS_SERVER(ss_mnt->mnt_root->d_inode); 328 329 nfs_fattr_init(&fattr); 330 331 status = nfs4_proc_getattr(server, src_fh, &fattr, NULL, NULL); 332 if (status < 0) { 333 res = ERR_PTR(status); 334 goto out; 335 } 336 337 res = ERR_PTR(-ENOMEM); 338 len = strlen(SSC_READ_NAME_BODY) + 16; 339 read_name = kzalloc(len, GFP_NOFS); 340 if (read_name == NULL) 341 goto out; 342 snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++); 343 344 r_ino = nfs_fhget(ss_mnt->mnt_root->d_inode->i_sb, src_fh, &fattr, 345 NULL); 346 if (IS_ERR(r_ino)) { 347 res = ERR_CAST(r_ino); 348 goto out_free_name; 349 } 350 351 filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, FMODE_READ, 352 r_ino->i_fop); 353 if (IS_ERR(filep)) { 354 res = ERR_CAST(filep); 355 goto out_free_name; 356 } 357 filep->f_mode |= FMODE_READ; 358 359 ctx = alloc_nfs_open_context(filep->f_path.dentry, filep->f_mode, 360 filep); 361 if (IS_ERR(ctx)) { 362 res = ERR_CAST(ctx); 363 goto out_filep; 364 } 365 366 res = ERR_PTR(-EINVAL); 367 sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL); 368 if (sp == NULL) 369 goto out_ctx; 370 371 ctx->state = nfs4_get_open_state(r_ino, sp); 372 if (ctx->state == NULL) 373 goto out_stateowner; 374 375 set_bit(NFS_SRV_SSC_COPY_STATE, &ctx->state->flags); 376 set_bit(NFS_OPEN_STATE, &ctx->state->flags); 377 memcpy(&ctx->state->open_stateid.other, &stateid->other, 378 NFS4_STATEID_OTHER_SIZE); 379 update_open_stateid(ctx->state, stateid, NULL, filep->f_mode); 380 381 nfs_file_set_open_context(filep, ctx); 382 put_nfs_open_context(ctx); 383 384 file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping); 385 res = filep; 386 out_free_name: 387 kfree(read_name); 388 out: 389 return res; 390 out_stateowner: 391 nfs4_put_state_owner(sp); 392 out_ctx: 393 put_nfs_open_context(ctx); 394 out_filep: 395 fput(filep); 396 goto out_free_name; 397 } 398 EXPORT_SYMBOL_GPL(nfs42_ssc_open); 399 void nfs42_ssc_close(struct file *filep) 400 { 401 struct nfs_open_context *ctx = nfs_file_open_context(filep); 402 403 ctx->state->flags = 0; 404 } 405 EXPORT_SYMBOL_GPL(nfs42_ssc_close); 406 #endif /* CONFIG_NFS_V4_2 */ 407 408 const struct file_operations nfs4_file_operations = { 409 .read_iter = nfs_file_read, 410 .write_iter = nfs_file_write, 411 .mmap = nfs_file_mmap, 412 .open = nfs4_file_open, 413 .flush = nfs4_file_flush, 414 .release = nfs_file_release, 415 .fsync = nfs_file_fsync, 416 .lock = nfs_lock, 417 .flock = nfs_flock, 418 .splice_read = generic_file_splice_read, 419 .splice_write = iter_file_splice_write, 420 .check_flags = nfs_check_flags, 421 .setlease = simple_nosetlease, 422 #ifdef CONFIG_NFS_V4_2 423 .copy_file_range = nfs4_copy_file_range, 424 .llseek = nfs4_file_llseek, 425 .fallocate = nfs42_fallocate, 426 .remap_file_range = nfs42_remap_file_range, 427 #else 428 .llseek = nfs_file_llseek, 429 #endif 430 }; 431