| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * linux/fs/nfs/unlink.c |
| 4 | * |
| 5 | * nfs sillydelete handling |
| 6 | * |
| 7 | */ |
| 8 | |
| 9 | #include <linux/slab.h> |
| 10 | #include <linux/string.h> |
| 11 | #include <linux/dcache.h> |
| 12 | #include <linux/sunrpc/sched.h> |
| 13 | #include <linux/sunrpc/clnt.h> |
| 14 | #include <linux/nfs_fs.h> |
| 15 | #include <linux/sched.h> |
| 16 | #include <linux/wait.h> |
| 17 | #include <linux/namei.h> |
| 18 | #include <linux/fsnotify.h> |
| 19 | |
| 20 | #include "internal.h" |
| 21 | #include "nfs4_fs.h" |
| 22 | #include "iostat.h" |
| 23 | #include "delegation.h" |
| 24 | |
| 25 | #include "nfstrace.h" |
| 26 | |
| 27 | /** |
| 28 | * nfs_free_unlinkdata - release data from a sillydelete operation. |
| 29 | * @data: pointer to unlink structure. |
| 30 | */ |
| 31 | static void |
| 32 | nfs_free_unlinkdata(struct nfs_unlinkdata *data) |
| 33 | { |
| 34 | put_cred(data->cred); |
| 35 | kfree(data->args.name.name); |
| 36 | kfree(data); |
| 37 | } |
| 38 | |
| 39 | /** |
| 40 | * nfs_async_unlink_done - Sillydelete post-processing |
| 41 | * @task: rpc_task of the sillydelete |
| 42 | * @calldata: pointer to nfs_unlinkdata |
| 43 | * |
| 44 | * Do the directory attribute update. |
| 45 | */ |
| 46 | static void nfs_async_unlink_done(struct rpc_task *task, void *calldata) |
| 47 | { |
| 48 | struct nfs_unlinkdata *data = calldata; |
| 49 | struct inode *dir = d_inode(data->dentry->d_parent); |
| 50 | |
| 51 | trace_nfs_sillyrename_unlink(data, task->tk_status); |
| 52 | if (!NFS_PROTO(dir)->unlink_done(task, dir)) |
| 53 | rpc_restart_call_prepare(task); |
| 54 | } |
| 55 | |
| 56 | /** |
| 57 | * nfs_async_unlink_release - Release the sillydelete data. |
| 58 | * @calldata: struct nfs_unlinkdata to release |
| 59 | * |
| 60 | * We need to call nfs_put_unlinkdata as a 'tk_release' task since the |
| 61 | * rpc_task would be freed too. |
| 62 | */ |
| 63 | static void nfs_async_unlink_release(void *calldata) |
| 64 | { |
| 65 | struct nfs_unlinkdata *data = calldata; |
| 66 | struct dentry *dentry = data->dentry; |
| 67 | struct super_block *sb = dentry->d_sb; |
| 68 | |
| 69 | up_read_non_owner(&NFS_I(d_inode(dentry->d_parent))->rmdir_sem); |
| 70 | d_lookup_done(dentry); |
| 71 | nfs_free_unlinkdata(data); |
| 72 | dput(dentry); |
| 73 | nfs_sb_deactive(sb); |
| 74 | } |
| 75 | |
| 76 | static void nfs_unlink_prepare(struct rpc_task *task, void *calldata) |
| 77 | { |
| 78 | struct nfs_unlinkdata *data = calldata; |
| 79 | struct inode *dir = d_inode(data->dentry->d_parent); |
| 80 | NFS_PROTO(dir)->unlink_rpc_prepare(task, data); |
| 81 | } |
| 82 | |
| 83 | static const struct rpc_call_ops nfs_unlink_ops = { |
| 84 | .rpc_call_done = nfs_async_unlink_done, |
| 85 | .rpc_release = nfs_async_unlink_release, |
| 86 | .rpc_call_prepare = nfs_unlink_prepare, |
| 87 | }; |
| 88 | |
| 89 | static void nfs_do_call_unlink(struct inode *inode, struct nfs_unlinkdata *data) |
| 90 | { |
| 91 | struct rpc_message msg = { |
| 92 | .rpc_argp = &data->args, |
| 93 | .rpc_resp = &data->res, |
| 94 | .rpc_cred = data->cred, |
| 95 | }; |
| 96 | struct rpc_task_setup task_setup_data = { |
| 97 | .rpc_message = &msg, |
| 98 | .callback_ops = &nfs_unlink_ops, |
| 99 | .callback_data = data, |
| 100 | .workqueue = nfsiod_workqueue, |
| 101 | .flags = RPC_TASK_ASYNC | RPC_TASK_CRED_NOREF, |
| 102 | }; |
| 103 | struct rpc_task *task; |
| 104 | struct inode *dir = d_inode(data->dentry->d_parent); |
| 105 | nfs_sb_active(dir->i_sb); |
| 106 | data->args.fh = NFS_FH(dir); |
| 107 | nfs_fattr_init(data->res.dir_attr); |
| 108 | |
| 109 | NFS_PROTO(dir)->unlink_setup(&msg, data->dentry, inode); |
| 110 | |
| 111 | task_setup_data.rpc_client = NFS_CLIENT(dir); |
| 112 | task = rpc_run_task(&task_setup_data); |
| 113 | if (!IS_ERR(task)) |
| 114 | rpc_put_task_async(task); |
| 115 | } |
| 116 | |
| 117 | static int nfs_call_unlink(struct dentry *dentry, struct inode *inode, struct nfs_unlinkdata *data) |
| 118 | { |
| 119 | struct inode *dir = d_inode(dentry->d_parent); |
| 120 | struct dentry *alias; |
| 121 | |
| 122 | down_read_non_owner(&NFS_I(dir)->rmdir_sem); |
| 123 | alias = d_alloc_parallel(dentry->d_parent, &data->args.name, &data->wq); |
| 124 | if (IS_ERR(alias)) { |
| 125 | up_read_non_owner(&NFS_I(dir)->rmdir_sem); |
| 126 | return 0; |
| 127 | } |
| 128 | if (!d_in_lookup(alias)) { |
| 129 | int ret; |
| 130 | void *devname_garbage = NULL; |
| 131 | |
| 132 | /* |
| 133 | * Hey, we raced with lookup... See if we need to transfer |
| 134 | * the sillyrename information to the aliased dentry. |
| 135 | */ |
| 136 | spin_lock(&alias->d_lock); |
| 137 | if (d_really_is_positive(alias) && |
| 138 | !(alias->d_flags & DCACHE_NFSFS_RENAMED)) { |
| 139 | devname_garbage = alias->d_fsdata; |
| 140 | alias->d_fsdata = data; |
| 141 | alias->d_flags |= DCACHE_NFSFS_RENAMED; |
| 142 | ret = 1; |
| 143 | } else |
| 144 | ret = 0; |
| 145 | spin_unlock(&alias->d_lock); |
| 146 | dput(alias); |
| 147 | up_read_non_owner(&NFS_I(dir)->rmdir_sem); |
| 148 | /* |
| 149 | * If we'd displaced old cached devname, free it. At that |
| 150 | * point dentry is definitely not a root, so we won't need |
| 151 | * that anymore. |
| 152 | */ |
| 153 | kfree(devname_garbage); |
| 154 | return ret; |
| 155 | } |
| 156 | data->dentry = alias; |
| 157 | nfs_do_call_unlink(inode, data); |
| 158 | return 1; |
| 159 | } |
| 160 | |
| 161 | /** |
| 162 | * nfs_async_unlink - asynchronous unlinking of a file |
| 163 | * @dentry: parent directory of dentry |
| 164 | * @name: name of dentry to unlink |
| 165 | */ |
| 166 | static int |
| 167 | nfs_async_unlink(struct dentry *dentry, const struct qstr *name) |
| 168 | { |
| 169 | struct nfs_unlinkdata *data; |
| 170 | int status = -ENOMEM; |
| 171 | void *devname_garbage = NULL; |
| 172 | |
| 173 | data = kzalloc(sizeof(*data), GFP_KERNEL); |
| 174 | if (data == NULL) |
| 175 | goto out; |
| 176 | data->args.name.name = kstrdup(name->name, GFP_KERNEL); |
| 177 | if (!data->args.name.name) |
| 178 | goto out_free; |
| 179 | data->args.name.len = name->len; |
| 180 | |
| 181 | data->cred = get_current_cred(); |
| 182 | data->res.dir_attr = &data->dir_attr; |
| 183 | init_waitqueue_head(&data->wq); |
| 184 | |
| 185 | status = -EBUSY; |
| 186 | spin_lock(&dentry->d_lock); |
| 187 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) |
| 188 | goto out_unlock; |
| 189 | dentry->d_flags |= DCACHE_NFSFS_RENAMED; |
| 190 | devname_garbage = dentry->d_fsdata; |
| 191 | dentry->d_fsdata = data; |
| 192 | spin_unlock(&dentry->d_lock); |
| 193 | /* |
| 194 | * If we'd displaced old cached devname, free it. At that |
| 195 | * point dentry is definitely not a root, so we won't need |
| 196 | * that anymore. |
| 197 | */ |
| 198 | kfree(devname_garbage); |
| 199 | return 0; |
| 200 | out_unlock: |
| 201 | spin_unlock(&dentry->d_lock); |
| 202 | put_cred(data->cred); |
| 203 | kfree(data->args.name.name); |
| 204 | out_free: |
| 205 | kfree(data); |
| 206 | out: |
| 207 | return status; |
| 208 | } |
| 209 | |
| 210 | /** |
| 211 | * nfs_complete_unlink - Initialize completion of the sillydelete |
| 212 | * @dentry: dentry to delete |
| 213 | * @inode: inode |
| 214 | * |
| 215 | * Since we're most likely to be called by dentry_iput(), we |
| 216 | * only use the dentry to find the sillydelete. We then copy the name |
| 217 | * into the qstr. |
| 218 | */ |
| 219 | void |
| 220 | nfs_complete_unlink(struct dentry *dentry, struct inode *inode) |
| 221 | { |
| 222 | struct nfs_unlinkdata *data; |
| 223 | |
| 224 | spin_lock(&dentry->d_lock); |
| 225 | dentry->d_flags &= ~DCACHE_NFSFS_RENAMED; |
| 226 | data = dentry->d_fsdata; |
| 227 | dentry->d_fsdata = NULL; |
| 228 | spin_unlock(&dentry->d_lock); |
| 229 | |
| 230 | if (NFS_STALE(inode) || !nfs_call_unlink(dentry, inode, data)) |
| 231 | nfs_free_unlinkdata(data); |
| 232 | } |
| 233 | |
| 234 | /* Cancel a queued async unlink. Called when a sillyrename run fails. */ |
| 235 | static void |
| 236 | nfs_cancel_async_unlink(struct dentry *dentry) |
| 237 | { |
| 238 | spin_lock(&dentry->d_lock); |
| 239 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
| 240 | struct nfs_unlinkdata *data = dentry->d_fsdata; |
| 241 | |
| 242 | dentry->d_flags &= ~DCACHE_NFSFS_RENAMED; |
| 243 | dentry->d_fsdata = NULL; |
| 244 | spin_unlock(&dentry->d_lock); |
| 245 | nfs_free_unlinkdata(data); |
| 246 | return; |
| 247 | } |
| 248 | spin_unlock(&dentry->d_lock); |
| 249 | } |
| 250 | |
| 251 | /** |
| 252 | * nfs_async_rename_done - Sillyrename post-processing |
| 253 | * @task: rpc_task of the sillyrename |
| 254 | * @calldata: nfs_renamedata for the sillyrename |
| 255 | * |
| 256 | * Do the directory attribute updates and the d_move |
| 257 | */ |
| 258 | static void nfs_async_rename_done(struct rpc_task *task, void *calldata) |
| 259 | { |
| 260 | struct nfs_renamedata *data = calldata; |
| 261 | struct inode *old_dir = data->old_dir; |
| 262 | struct inode *new_dir = data->new_dir; |
| 263 | struct dentry *old_dentry = data->old_dentry; |
| 264 | |
| 265 | trace_nfs_sillyrename_rename(old_dir, old_dentry, |
| 266 | new_dir, data->new_dentry, task->tk_status); |
| 267 | if (!NFS_PROTO(old_dir)->rename_done(task, old_dir, new_dir)) { |
| 268 | rpc_restart_call_prepare(task); |
| 269 | return; |
| 270 | } |
| 271 | |
| 272 | if (data->complete) |
| 273 | data->complete(task, data); |
| 274 | } |
| 275 | |
| 276 | /** |
| 277 | * nfs_async_rename_release - Release the sillyrename data. |
| 278 | * @calldata: the struct nfs_renamedata to be released |
| 279 | */ |
| 280 | static void nfs_async_rename_release(void *calldata) |
| 281 | { |
| 282 | struct nfs_renamedata *data = calldata; |
| 283 | struct super_block *sb = data->old_dir->i_sb; |
| 284 | |
| 285 | if (d_really_is_positive(data->old_dentry)) |
| 286 | nfs_mark_for_revalidate(d_inode(data->old_dentry)); |
| 287 | |
| 288 | /* The result of the rename is unknown. Play it safe by |
| 289 | * forcing a new lookup */ |
| 290 | if (data->cancelled) { |
| 291 | spin_lock(&data->old_dir->i_lock); |
| 292 | nfs_force_lookup_revalidate(data->old_dir); |
| 293 | spin_unlock(&data->old_dir->i_lock); |
| 294 | if (data->new_dir != data->old_dir) { |
| 295 | spin_lock(&data->new_dir->i_lock); |
| 296 | nfs_force_lookup_revalidate(data->new_dir); |
| 297 | spin_unlock(&data->new_dir->i_lock); |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | dput(data->old_dentry); |
| 302 | dput(data->new_dentry); |
| 303 | iput(data->old_dir); |
| 304 | iput(data->new_dir); |
| 305 | nfs_sb_deactive(sb); |
| 306 | put_cred(data->cred); |
| 307 | kfree(data); |
| 308 | } |
| 309 | |
| 310 | static void nfs_rename_prepare(struct rpc_task *task, void *calldata) |
| 311 | { |
| 312 | struct nfs_renamedata *data = calldata; |
| 313 | NFS_PROTO(data->old_dir)->rename_rpc_prepare(task, data); |
| 314 | } |
| 315 | |
| 316 | static const struct rpc_call_ops nfs_rename_ops = { |
| 317 | .rpc_call_done = nfs_async_rename_done, |
| 318 | .rpc_release = nfs_async_rename_release, |
| 319 | .rpc_call_prepare = nfs_rename_prepare, |
| 320 | }; |
| 321 | |
| 322 | /** |
| 323 | * nfs_async_rename - perform an asynchronous rename operation |
| 324 | * @old_dir: directory that currently holds the dentry to be renamed |
| 325 | * @new_dir: target directory for the rename |
| 326 | * @old_dentry: original dentry to be renamed |
| 327 | * @new_dentry: dentry to which the old_dentry should be renamed |
| 328 | * @complete: Function to run on successful completion |
| 329 | * |
| 330 | * It's expected that valid references to the dentries and inodes are held |
| 331 | */ |
| 332 | struct rpc_task * |
| 333 | nfs_async_rename(struct inode *old_dir, struct inode *new_dir, |
| 334 | struct dentry *old_dentry, struct dentry *new_dentry, |
| 335 | void (*complete)(struct rpc_task *, struct nfs_renamedata *)) |
| 336 | { |
| 337 | struct nfs_renamedata *data; |
| 338 | struct rpc_message msg = { }; |
| 339 | struct rpc_task_setup task_setup_data = { |
| 340 | .rpc_message = &msg, |
| 341 | .callback_ops = &nfs_rename_ops, |
| 342 | .workqueue = nfsiod_workqueue, |
| 343 | .rpc_client = NFS_CLIENT(old_dir), |
| 344 | .flags = RPC_TASK_ASYNC | RPC_TASK_CRED_NOREF, |
| 345 | }; |
| 346 | |
| 347 | data = kzalloc(sizeof(*data), GFP_KERNEL); |
| 348 | if (data == NULL) |
| 349 | return ERR_PTR(-ENOMEM); |
| 350 | task_setup_data.callback_data = data; |
| 351 | |
| 352 | data->cred = get_current_cred(); |
| 353 | |
| 354 | msg.rpc_argp = &data->args; |
| 355 | msg.rpc_resp = &data->res; |
| 356 | msg.rpc_cred = data->cred; |
| 357 | |
| 358 | /* set up nfs_renamedata */ |
| 359 | data->old_dir = old_dir; |
| 360 | ihold(old_dir); |
| 361 | data->new_dir = new_dir; |
| 362 | ihold(new_dir); |
| 363 | data->old_dentry = dget(old_dentry); |
| 364 | data->new_dentry = dget(new_dentry); |
| 365 | nfs_fattr_init(&data->old_fattr); |
| 366 | nfs_fattr_init(&data->new_fattr); |
| 367 | data->complete = complete; |
| 368 | |
| 369 | /* set up nfs_renameargs */ |
| 370 | data->args.old_dir = NFS_FH(old_dir); |
| 371 | data->args.old_name = &old_dentry->d_name; |
| 372 | data->args.new_dir = NFS_FH(new_dir); |
| 373 | data->args.new_name = &new_dentry->d_name; |
| 374 | |
| 375 | /* set up nfs_renameres */ |
| 376 | data->res.old_fattr = &data->old_fattr; |
| 377 | data->res.new_fattr = &data->new_fattr; |
| 378 | |
| 379 | nfs_sb_active(old_dir->i_sb); |
| 380 | |
| 381 | NFS_PROTO(data->old_dir)->rename_setup(&msg, old_dentry, new_dentry); |
| 382 | |
| 383 | return rpc_run_task(&task_setup_data); |
| 384 | } |
| 385 | |
| 386 | /* |
| 387 | * Perform tasks needed when a sillyrename is done such as cancelling the |
| 388 | * queued async unlink if it failed. |
| 389 | */ |
| 390 | static void |
| 391 | nfs_complete_sillyrename(struct rpc_task *task, struct nfs_renamedata *data) |
| 392 | { |
| 393 | struct dentry *dentry = data->old_dentry; |
| 394 | |
| 395 | if (task->tk_status != 0) { |
| 396 | nfs_cancel_async_unlink(dentry); |
| 397 | return; |
| 398 | } |
| 399 | } |
| 400 | |
| 401 | #define SILLYNAME_PREFIX ".nfs" |
| 402 | #define SILLYNAME_PREFIX_LEN ((unsigned)sizeof(SILLYNAME_PREFIX) - 1) |
| 403 | #define SILLYNAME_FILEID_LEN ((unsigned)sizeof(u64) << 1) |
| 404 | #define SILLYNAME_COUNTER_LEN ((unsigned)sizeof(unsigned int) << 1) |
| 405 | #define SILLYNAME_LEN (SILLYNAME_PREFIX_LEN + \ |
| 406 | SILLYNAME_FILEID_LEN + \ |
| 407 | SILLYNAME_COUNTER_LEN) |
| 408 | |
| 409 | /** |
| 410 | * nfs_sillyrename - Perform a silly-rename of a dentry |
| 411 | * @dir: inode of directory that contains dentry |
| 412 | * @dentry: dentry to be sillyrenamed |
| 413 | * |
| 414 | * NFSv2/3 is stateless and the server doesn't know when the client is |
| 415 | * holding a file open. To prevent application problems when a file is |
| 416 | * unlinked while it's still open, the client performs a "silly-rename". |
| 417 | * That is, it renames the file to a hidden file in the same directory, |
| 418 | * and only performs the unlink once the last reference to it is put. |
| 419 | * |
| 420 | * The final cleanup is done during dentry_iput. |
| 421 | * |
| 422 | * (Note: NFSv4 is stateful, and has opens, so in theory an NFSv4 server |
| 423 | * could take responsibility for keeping open files referenced. The server |
| 424 | * would also need to ensure that opened-but-deleted files were kept over |
| 425 | * reboots. However, we may not assume a server does so. (RFC 5661 |
| 426 | * does provide an OPEN4_RESULT_PRESERVE_UNLINKED flag that a server can |
| 427 | * use to advertise that it does this; some day we may take advantage of |
| 428 | * it.)) |
| 429 | */ |
| 430 | int |
| 431 | nfs_sillyrename(struct inode *dir, struct dentry *dentry) |
| 432 | { |
| 433 | static unsigned int sillycounter; |
| 434 | unsigned char silly[SILLYNAME_LEN + 1]; |
| 435 | unsigned long long fileid; |
| 436 | struct dentry *sdentry; |
| 437 | struct inode *inode = d_inode(dentry); |
| 438 | struct rpc_task *task; |
| 439 | int error = -EBUSY; |
| 440 | |
| 441 | dfprintk(VFS, "NFS: silly-rename(%pd2, ct=%d)\n", |
| 442 | dentry, d_count(dentry)); |
| 443 | nfs_inc_stats(dir, NFSIOS_SILLYRENAME); |
| 444 | |
| 445 | /* |
| 446 | * We don't allow a dentry to be silly-renamed twice. |
| 447 | */ |
| 448 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) |
| 449 | goto out; |
| 450 | |
| 451 | fileid = NFS_FILEID(d_inode(dentry)); |
| 452 | |
| 453 | sdentry = NULL; |
| 454 | do { |
| 455 | int slen; |
| 456 | dput(sdentry); |
| 457 | sillycounter++; |
| 458 | slen = scnprintf(silly, sizeof(silly), |
| 459 | SILLYNAME_PREFIX "%0*llx%0*x", |
| 460 | SILLYNAME_FILEID_LEN, fileid, |
| 461 | SILLYNAME_COUNTER_LEN, sillycounter); |
| 462 | |
| 463 | dfprintk(VFS, "NFS: trying to rename %pd to %s\n", |
| 464 | dentry, silly); |
| 465 | |
| 466 | sdentry = lookup_one_len(silly, dentry->d_parent, slen); |
| 467 | /* |
| 468 | * N.B. Better to return EBUSY here ... it could be |
| 469 | * dangerous to delete the file while it's in use. |
| 470 | */ |
| 471 | if (IS_ERR(sdentry)) |
| 472 | goto out; |
| 473 | } while (d_inode(sdentry) != NULL); /* need negative lookup */ |
| 474 | |
| 475 | ihold(inode); |
| 476 | |
| 477 | /* queue unlink first. Can't do this from rpc_release as it |
| 478 | * has to allocate memory |
| 479 | */ |
| 480 | error = nfs_async_unlink(dentry, &sdentry->d_name); |
| 481 | if (error) |
| 482 | goto out_dput; |
| 483 | |
| 484 | /* run the rename task, undo unlink if it fails */ |
| 485 | task = nfs_async_rename(dir, dir, dentry, sdentry, |
| 486 | nfs_complete_sillyrename); |
| 487 | if (IS_ERR(task)) { |
| 488 | error = -EBUSY; |
| 489 | nfs_cancel_async_unlink(dentry); |
| 490 | goto out_dput; |
| 491 | } |
| 492 | |
| 493 | /* wait for the RPC task to complete, unless a SIGKILL intervenes */ |
| 494 | error = rpc_wait_for_completion_task(task); |
| 495 | if (error == 0) |
| 496 | error = task->tk_status; |
| 497 | switch (error) { |
| 498 | case 0: |
| 499 | /* The rename succeeded */ |
| 500 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
| 501 | spin_lock(&inode->i_lock); |
| 502 | NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); |
| 503 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE |
| 504 | | NFS_INO_INVALID_CTIME |
| 505 | | NFS_INO_REVAL_FORCED; |
| 506 | spin_unlock(&inode->i_lock); |
| 507 | d_move(dentry, sdentry); |
| 508 | break; |
| 509 | case -ERESTARTSYS: |
| 510 | /* The result of the rename is unknown. Play it safe by |
| 511 | * forcing a new lookup */ |
| 512 | d_drop(dentry); |
| 513 | d_drop(sdentry); |
| 514 | } |
| 515 | rpc_put_task(task); |
| 516 | out_dput: |
| 517 | iput(inode); |
| 518 | dput(sdentry); |
| 519 | out: |
| 520 | return error; |
| 521 | } |