| 1 | /* |
| 2 | * GPL HEADER START |
| 3 | * |
| 4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License version 2 only, |
| 8 | * as published by the Free Software Foundation. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, but |
| 11 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | * General Public License version 2 for more details (a copy is included |
| 14 | * in the LICENSE file that accompanied this code). |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * version 2 along with this program; If not, see |
| 18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf |
| 19 | * |
| 20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
| 21 | * CA 95054 USA or visit www.sun.com if you need additional information or |
| 22 | * have any questions. |
| 23 | * |
| 24 | * GPL HEADER END |
| 25 | */ |
| 26 | /* |
| 27 | * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. |
| 28 | * Use is subject to license terms. |
| 29 | * |
| 30 | * Copyright (c) 2011, 2012, Intel Corporation. |
| 31 | */ |
| 32 | /* |
| 33 | * This file is part of Lustre, http://www.lustre.org/ |
| 34 | * Lustre is a trademark of Sun Microsystems, Inc. |
| 35 | */ |
| 36 | |
| 37 | #include <linux/fs.h> |
| 38 | #include <linux/sched.h> |
| 39 | #include <linux/mm.h> |
| 40 | #include <linux/quotaops.h> |
| 41 | #include <linux/highmem.h> |
| 42 | #include <linux/pagemap.h> |
| 43 | #include <linux/security.h> |
| 44 | |
| 45 | #define DEBUG_SUBSYSTEM S_LLITE |
| 46 | |
| 47 | #include <obd_support.h> |
| 48 | #include <lustre_fid.h> |
| 49 | #include <lustre_lite.h> |
| 50 | #include <lustre_dlm.h> |
| 51 | #include <lustre_ver.h> |
| 52 | #include "llite_internal.h" |
| 53 | |
| 54 | static int ll_create_it(struct inode *, struct dentry *, |
| 55 | int, struct lookup_intent *); |
| 56 | |
| 57 | /* |
| 58 | * Check if we have something mounted at the named dchild. |
| 59 | * In such a case there would always be dentry present. |
| 60 | */ |
| 61 | static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild, |
| 62 | struct qstr *name) |
| 63 | { |
| 64 | int mounted = 0; |
| 65 | |
| 66 | if (unlikely(dchild)) { |
| 67 | mounted = d_mountpoint(dchild); |
| 68 | } else if (dparent) { |
| 69 | dchild = d_lookup(dparent, name); |
| 70 | if (dchild) { |
| 71 | mounted = d_mountpoint(dchild); |
| 72 | dput(dchild); |
| 73 | } |
| 74 | } |
| 75 | return mounted; |
| 76 | } |
| 77 | |
| 78 | int ll_unlock(__u32 mode, struct lustre_handle *lockh) |
| 79 | { |
| 80 | ldlm_lock_decref(lockh, mode); |
| 81 | |
| 82 | return 0; |
| 83 | } |
| 84 | |
| 85 | |
| 86 | /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */ |
| 87 | static int ll_test_inode(struct inode *inode, void *opaque) |
| 88 | { |
| 89 | struct ll_inode_info *lli = ll_i2info(inode); |
| 90 | struct lustre_md *md = opaque; |
| 91 | |
| 92 | if (unlikely(!(md->body->valid & OBD_MD_FLID))) { |
| 93 | CERROR("MDS body missing FID\n"); |
| 94 | return 0; |
| 95 | } |
| 96 | |
| 97 | if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1)) |
| 98 | return 0; |
| 99 | |
| 100 | return 1; |
| 101 | } |
| 102 | |
| 103 | static int ll_set_inode(struct inode *inode, void *opaque) |
| 104 | { |
| 105 | struct ll_inode_info *lli = ll_i2info(inode); |
| 106 | struct mdt_body *body = ((struct lustre_md *)opaque)->body; |
| 107 | |
| 108 | if (unlikely(!(body->valid & OBD_MD_FLID))) { |
| 109 | CERROR("MDS body missing FID\n"); |
| 110 | return -EINVAL; |
| 111 | } |
| 112 | |
| 113 | lli->lli_fid = body->fid1; |
| 114 | if (unlikely(!(body->valid & OBD_MD_FLTYPE))) { |
| 115 | CERROR("Can not initialize inode "DFID" without object type: " |
| 116 | "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid); |
| 117 | return -EINVAL; |
| 118 | } |
| 119 | |
| 120 | inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT); |
| 121 | if (unlikely(inode->i_mode == 0)) { |
| 122 | CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid)); |
| 123 | return -EINVAL; |
| 124 | } |
| 125 | |
| 126 | ll_lli_init(lli); |
| 127 | |
| 128 | return 0; |
| 129 | } |
| 130 | |
| 131 | |
| 132 | /* |
| 133 | * Get an inode by inode number (already instantiated by the intent lookup). |
| 134 | * Returns inode or NULL |
| 135 | */ |
| 136 | struct inode *ll_iget(struct super_block *sb, ino_t hash, |
| 137 | struct lustre_md *md) |
| 138 | { |
| 139 | struct inode *inode; |
| 140 | |
| 141 | LASSERT(hash != 0); |
| 142 | inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md); |
| 143 | |
| 144 | if (inode) { |
| 145 | if (inode->i_state & I_NEW) { |
| 146 | int rc = 0; |
| 147 | |
| 148 | ll_read_inode2(inode, md); |
| 149 | if (S_ISREG(inode->i_mode) && |
| 150 | ll_i2info(inode)->lli_clob == NULL) { |
| 151 | CDEBUG(D_INODE, |
| 152 | "%s: apply lsm %p to inode "DFID".\n", |
| 153 | ll_get_fsname(sb, NULL, 0), md->lsm, |
| 154 | PFID(ll_inode2fid(inode))); |
| 155 | rc = cl_file_inode_init(inode, md); |
| 156 | } |
| 157 | if (rc != 0) { |
| 158 | make_bad_inode(inode); |
| 159 | unlock_new_inode(inode); |
| 160 | iput(inode); |
| 161 | inode = ERR_PTR(rc); |
| 162 | } else |
| 163 | unlock_new_inode(inode); |
| 164 | } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) |
| 165 | ll_update_inode(inode, md); |
| 166 | CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n", |
| 167 | inode, PFID(&md->body->fid1)); |
| 168 | } |
| 169 | return inode; |
| 170 | } |
| 171 | |
| 172 | static void ll_invalidate_negative_children(struct inode *dir) |
| 173 | { |
| 174 | struct dentry *dentry, *tmp_subdir; |
| 175 | struct ll_d_hlist_node *p; |
| 176 | |
| 177 | ll_lock_dcache(dir); |
| 178 | ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) { |
| 179 | spin_lock(&dentry->d_lock); |
| 180 | if (!list_empty(&dentry->d_subdirs)) { |
| 181 | struct dentry *child; |
| 182 | |
| 183 | list_for_each_entry_safe(child, tmp_subdir, |
| 184 | &dentry->d_subdirs, |
| 185 | d_u.d_child) { |
| 186 | if (child->d_inode == NULL) |
| 187 | d_lustre_invalidate(child, 1); |
| 188 | } |
| 189 | } |
| 190 | spin_unlock(&dentry->d_lock); |
| 191 | } |
| 192 | ll_unlock_dcache(dir); |
| 193 | } |
| 194 | |
| 195 | int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, |
| 196 | void *data, int flag) |
| 197 | { |
| 198 | int rc; |
| 199 | struct lustre_handle lockh; |
| 200 | |
| 201 | switch (flag) { |
| 202 | case LDLM_CB_BLOCKING: |
| 203 | ldlm_lock2handle(lock, &lockh); |
| 204 | rc = ldlm_cli_cancel(&lockh, LCF_ASYNC); |
| 205 | if (rc < 0) { |
| 206 | CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc); |
| 207 | return rc; |
| 208 | } |
| 209 | break; |
| 210 | case LDLM_CB_CANCELING: { |
| 211 | struct inode *inode = ll_inode_from_resource_lock(lock); |
| 212 | struct ll_inode_info *lli; |
| 213 | __u64 bits = lock->l_policy_data.l_inodebits.bits; |
| 214 | struct lu_fid *fid; |
| 215 | ldlm_mode_t mode = lock->l_req_mode; |
| 216 | |
| 217 | /* Inode is set to lock->l_resource->lr_lvb_inode |
| 218 | * for mdc - bug 24555 */ |
| 219 | LASSERT(lock->l_ast_data == NULL); |
| 220 | |
| 221 | /* Invalidate all dentries associated with this inode */ |
| 222 | if (inode == NULL) |
| 223 | break; |
| 224 | |
| 225 | LASSERT(lock->l_flags & LDLM_FL_CANCELING); |
| 226 | |
| 227 | if (bits & MDS_INODELOCK_XATTR) |
| 228 | ll_xattr_cache_destroy(inode); |
| 229 | |
| 230 | /* For OPEN locks we differentiate between lock modes |
| 231 | * LCK_CR, LCK_CW, LCK_PR - bug 22891 */ |
| 232 | if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE | |
| 233 | MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM)) |
| 234 | ll_have_md_lock(inode, &bits, LCK_MINMODE); |
| 235 | |
| 236 | if (bits & MDS_INODELOCK_OPEN) |
| 237 | ll_have_md_lock(inode, &bits, mode); |
| 238 | |
| 239 | fid = ll_inode2fid(inode); |
| 240 | if (!fid_res_name_eq(fid, &lock->l_resource->lr_name)) |
| 241 | LDLM_ERROR(lock, "data mismatch with object " |
| 242 | DFID" (%p)", PFID(fid), inode); |
| 243 | |
| 244 | if (bits & MDS_INODELOCK_OPEN) { |
| 245 | int flags = 0; |
| 246 | switch (lock->l_req_mode) { |
| 247 | case LCK_CW: |
| 248 | flags = FMODE_WRITE; |
| 249 | break; |
| 250 | case LCK_PR: |
| 251 | flags = FMODE_EXEC; |
| 252 | break; |
| 253 | case LCK_CR: |
| 254 | flags = FMODE_READ; |
| 255 | break; |
| 256 | default: |
| 257 | CERROR("Unexpected lock mode for OPEN lock " |
| 258 | "%d, inode %ld\n", lock->l_req_mode, |
| 259 | inode->i_ino); |
| 260 | } |
| 261 | ll_md_real_close(inode, flags); |
| 262 | } |
| 263 | |
| 264 | lli = ll_i2info(inode); |
| 265 | if (bits & MDS_INODELOCK_LAYOUT) { |
| 266 | struct cl_object_conf conf = { { 0 } }; |
| 267 | |
| 268 | conf.coc_opc = OBJECT_CONF_INVALIDATE; |
| 269 | conf.coc_inode = inode; |
| 270 | rc = ll_layout_conf(inode, &conf); |
| 271 | if (rc) |
| 272 | CDEBUG(D_INODE, "invaliding layout %d.\n", rc); |
| 273 | } |
| 274 | |
| 275 | if (bits & MDS_INODELOCK_UPDATE) { |
| 276 | spin_lock(&lli->lli_lock); |
| 277 | lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK; |
| 278 | spin_unlock(&lli->lli_lock); |
| 279 | } |
| 280 | |
| 281 | if (S_ISDIR(inode->i_mode) && |
| 282 | (bits & MDS_INODELOCK_UPDATE)) { |
| 283 | CDEBUG(D_INODE, "invalidating inode %lu\n", |
| 284 | inode->i_ino); |
| 285 | truncate_inode_pages(inode->i_mapping, 0); |
| 286 | ll_invalidate_negative_children(inode); |
| 287 | } |
| 288 | |
| 289 | if (inode->i_sb->s_root && |
| 290 | inode != inode->i_sb->s_root->d_inode && |
| 291 | (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))) |
| 292 | ll_invalidate_aliases(inode); |
| 293 | iput(inode); |
| 294 | break; |
| 295 | } |
| 296 | default: |
| 297 | LBUG(); |
| 298 | } |
| 299 | |
| 300 | return 0; |
| 301 | } |
| 302 | |
| 303 | __u32 ll_i2suppgid(struct inode *i) |
| 304 | { |
| 305 | if (in_group_p(i->i_gid)) |
| 306 | return (__u32)from_kgid(&init_user_ns, i->i_gid); |
| 307 | else |
| 308 | return (__u32)(-1); |
| 309 | } |
| 310 | |
| 311 | /* Pack the required supplementary groups into the supplied groups array. |
| 312 | * If we don't need to use the groups from the target inode(s) then we |
| 313 | * instead pack one or more groups from the user's supplementary group |
| 314 | * array in case it might be useful. Not needed if doing an MDS-side upcall. */ |
| 315 | void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2) |
| 316 | { |
| 317 | #if 0 |
| 318 | int i; |
| 319 | #endif |
| 320 | |
| 321 | LASSERT(i1 != NULL); |
| 322 | LASSERT(suppgids != NULL); |
| 323 | |
| 324 | suppgids[0] = ll_i2suppgid(i1); |
| 325 | |
| 326 | if (i2) |
| 327 | suppgids[1] = ll_i2suppgid(i2); |
| 328 | else |
| 329 | suppgids[1] = -1; |
| 330 | |
| 331 | #if 0 |
| 332 | for (i = 0; i < current_ngroups; i++) { |
| 333 | if (suppgids[0] == -1) { |
| 334 | if (current_groups[i] != suppgids[1]) |
| 335 | suppgids[0] = current_groups[i]; |
| 336 | continue; |
| 337 | } |
| 338 | if (suppgids[1] == -1) { |
| 339 | if (current_groups[i] != suppgids[0]) |
| 340 | suppgids[1] = current_groups[i]; |
| 341 | continue; |
| 342 | } |
| 343 | break; |
| 344 | } |
| 345 | #endif |
| 346 | } |
| 347 | |
| 348 | /* |
| 349 | * try to reuse three types of dentry: |
| 350 | * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid |
| 351 | * by concurrent .revalidate). |
| 352 | * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may |
| 353 | * be cleared by others calling d_lustre_revalidate). |
| 354 | * 3. DISCONNECTED alias. |
| 355 | */ |
| 356 | static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry) |
| 357 | { |
| 358 | struct dentry *alias, *discon_alias, *invalid_alias; |
| 359 | struct ll_d_hlist_node *p; |
| 360 | |
| 361 | if (ll_d_hlist_empty(&inode->i_dentry)) |
| 362 | return NULL; |
| 363 | |
| 364 | discon_alias = invalid_alias = NULL; |
| 365 | |
| 366 | ll_lock_dcache(inode); |
| 367 | ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) { |
| 368 | LASSERT(alias != dentry); |
| 369 | |
| 370 | spin_lock(&alias->d_lock); |
| 371 | if (alias->d_flags & DCACHE_DISCONNECTED) |
| 372 | /* LASSERT(last_discon == NULL); LU-405, bz 20055 */ |
| 373 | discon_alias = alias; |
| 374 | else if (alias->d_parent == dentry->d_parent && |
| 375 | alias->d_name.hash == dentry->d_name.hash && |
| 376 | alias->d_name.len == dentry->d_name.len && |
| 377 | memcmp(alias->d_name.name, dentry->d_name.name, |
| 378 | dentry->d_name.len) == 0) |
| 379 | invalid_alias = alias; |
| 380 | spin_unlock(&alias->d_lock); |
| 381 | |
| 382 | if (invalid_alias) |
| 383 | break; |
| 384 | } |
| 385 | alias = invalid_alias ?: discon_alias ?: NULL; |
| 386 | if (alias) { |
| 387 | spin_lock(&alias->d_lock); |
| 388 | dget_dlock(alias); |
| 389 | spin_unlock(&alias->d_lock); |
| 390 | } |
| 391 | ll_unlock_dcache(inode); |
| 392 | |
| 393 | return alias; |
| 394 | } |
| 395 | |
| 396 | /* |
| 397 | * Similar to d_splice_alias(), but lustre treats invalid alias |
| 398 | * similar to DCACHE_DISCONNECTED, and tries to use it anyway. |
| 399 | */ |
| 400 | struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de) |
| 401 | { |
| 402 | struct dentry *new; |
| 403 | int rc; |
| 404 | |
| 405 | if (inode) { |
| 406 | new = ll_find_alias(inode, de); |
| 407 | if (new) { |
| 408 | rc = ll_d_init(new); |
| 409 | if (rc < 0) { |
| 410 | dput(new); |
| 411 | return ERR_PTR(rc); |
| 412 | } |
| 413 | d_move(new, de); |
| 414 | iput(inode); |
| 415 | CDEBUG(D_DENTRY, |
| 416 | "Reuse dentry %p inode %p refc %d flags %#x\n", |
| 417 | new, new->d_inode, d_count(new), new->d_flags); |
| 418 | return new; |
| 419 | } |
| 420 | } |
| 421 | rc = ll_d_init(de); |
| 422 | if (rc < 0) |
| 423 | return ERR_PTR(rc); |
| 424 | d_add(de, inode); |
| 425 | CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n", |
| 426 | de, de->d_inode, d_count(de), de->d_flags); |
| 427 | return de; |
| 428 | } |
| 429 | |
| 430 | int ll_lookup_it_finish(struct ptlrpc_request *request, |
| 431 | struct lookup_intent *it, void *data) |
| 432 | { |
| 433 | struct it_cb_data *icbd = data; |
| 434 | struct dentry **de = icbd->icbd_childp; |
| 435 | struct inode *parent = icbd->icbd_parent; |
| 436 | struct inode *inode = NULL; |
| 437 | __u64 bits = 0; |
| 438 | int rc; |
| 439 | |
| 440 | /* NB 1 request reference will be taken away by ll_intent_lock() |
| 441 | * when I return */ |
| 442 | CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it, |
| 443 | it->d.lustre.it_disposition); |
| 444 | if (!it_disposition(it, DISP_LOOKUP_NEG)) { |
| 445 | rc = ll_prep_inode(&inode, request, (*de)->d_sb, it); |
| 446 | if (rc) |
| 447 | return rc; |
| 448 | |
| 449 | ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits); |
| 450 | |
| 451 | /* We used to query real size from OSTs here, but actually |
| 452 | this is not needed. For stat() calls size would be updated |
| 453 | from subsequent do_revalidate()->ll_inode_revalidate_it() in |
| 454 | 2.4 and |
| 455 | vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6 |
| 456 | Everybody else who needs correct file size would call |
| 457 | ll_glimpse_size or some equivalent themselves anyway. |
| 458 | Also see bug 7198. */ |
| 459 | } |
| 460 | |
| 461 | /* Only hash *de if it is unhashed (new dentry). |
| 462 | * Atoimc_open may passin hashed dentries for open. |
| 463 | */ |
| 464 | if (d_unhashed(*de)) { |
| 465 | *de = ll_splice_alias(inode, *de); |
| 466 | if (IS_ERR(*de)) |
| 467 | return PTR_ERR(*de); |
| 468 | } |
| 469 | |
| 470 | if (!it_disposition(it, DISP_LOOKUP_NEG)) { |
| 471 | /* we have lookup look - unhide dentry */ |
| 472 | if (bits & MDS_INODELOCK_LOOKUP) |
| 473 | d_lustre_revalidate(*de); |
| 474 | } else if (!it_disposition(it, DISP_OPEN_CREATE)) { |
| 475 | /* If file created on server, don't depend on parent UPDATE |
| 476 | * lock to unhide it. It is left hidden and next lookup can |
| 477 | * find it in ll_splice_alias. |
| 478 | */ |
| 479 | /* Check that parent has UPDATE lock. */ |
| 480 | struct lookup_intent parent_it = { |
| 481 | .it_op = IT_GETATTR, |
| 482 | .d.lustre.it_lock_handle = 0 }; |
| 483 | |
| 484 | if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, |
| 485 | &ll_i2info(parent)->lli_fid, NULL)) { |
| 486 | d_lustre_revalidate(*de); |
| 487 | ll_intent_release(&parent_it); |
| 488 | } |
| 489 | } |
| 490 | |
| 491 | return 0; |
| 492 | } |
| 493 | |
| 494 | static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry, |
| 495 | struct lookup_intent *it, int lookup_flags) |
| 496 | { |
| 497 | struct lookup_intent lookup_it = { .it_op = IT_LOOKUP }; |
| 498 | struct dentry *save = dentry, *retval; |
| 499 | struct ptlrpc_request *req = NULL; |
| 500 | struct md_op_data *op_data; |
| 501 | struct it_cb_data icbd; |
| 502 | __u32 opc; |
| 503 | int rc; |
| 504 | |
| 505 | if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen) |
| 506 | return ERR_PTR(-ENAMETOOLONG); |
| 507 | |
| 508 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n", |
| 509 | dentry->d_name.len, dentry->d_name.name, parent->i_ino, |
| 510 | parent->i_generation, parent, LL_IT2STR(it)); |
| 511 | |
| 512 | if (d_mountpoint(dentry)) |
| 513 | CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it)); |
| 514 | |
| 515 | ll_frob_intent(&it, &lookup_it); |
| 516 | |
| 517 | if (it->it_op == IT_GETATTR) { |
| 518 | rc = ll_statahead_enter(parent, &dentry, 0); |
| 519 | if (rc == 1) { |
| 520 | if (dentry == save) |
| 521 | GOTO(out, retval = NULL); |
| 522 | GOTO(out, retval = dentry); |
| 523 | } |
| 524 | } |
| 525 | |
| 526 | icbd.icbd_childp = &dentry; |
| 527 | icbd.icbd_parent = parent; |
| 528 | |
| 529 | if (it->it_op & IT_CREAT) |
| 530 | opc = LUSTRE_OPC_CREATE; |
| 531 | else |
| 532 | opc = LUSTRE_OPC_ANY; |
| 533 | |
| 534 | op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name, |
| 535 | dentry->d_name.len, lookup_flags, opc, |
| 536 | NULL); |
| 537 | if (IS_ERR(op_data)) |
| 538 | return (void *)op_data; |
| 539 | |
| 540 | /* enforce umask if acl disabled or MDS doesn't support umask */ |
| 541 | if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent))) |
| 542 | it->it_create_mode &= ~current_umask(); |
| 543 | |
| 544 | rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it, |
| 545 | lookup_flags, &req, ll_md_blocking_ast, 0); |
| 546 | ll_finish_md_op_data(op_data); |
| 547 | if (rc < 0) |
| 548 | GOTO(out, retval = ERR_PTR(rc)); |
| 549 | |
| 550 | rc = ll_lookup_it_finish(req, it, &icbd); |
| 551 | if (rc != 0) { |
| 552 | ll_intent_release(it); |
| 553 | GOTO(out, retval = ERR_PTR(rc)); |
| 554 | } |
| 555 | |
| 556 | if ((it->it_op & IT_OPEN) && dentry->d_inode && |
| 557 | !S_ISREG(dentry->d_inode->i_mode) && |
| 558 | !S_ISDIR(dentry->d_inode->i_mode)) { |
| 559 | ll_release_openhandle(dentry, it); |
| 560 | } |
| 561 | ll_lookup_finish_locks(it, dentry); |
| 562 | |
| 563 | if (dentry == save) |
| 564 | GOTO(out, retval = NULL); |
| 565 | else |
| 566 | GOTO(out, retval = dentry); |
| 567 | out: |
| 568 | if (req) |
| 569 | ptlrpc_req_finished(req); |
| 570 | if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry)) |
| 571 | ll_statahead_mark(parent, dentry); |
| 572 | return retval; |
| 573 | } |
| 574 | |
| 575 | static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry, |
| 576 | unsigned int flags) |
| 577 | { |
| 578 | struct lookup_intent *itp, it = { .it_op = IT_GETATTR }; |
| 579 | struct dentry *de; |
| 580 | |
| 581 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n", |
| 582 | dentry->d_name.len, dentry->d_name.name, parent->i_ino, |
| 583 | parent->i_generation, parent, flags); |
| 584 | |
| 585 | /* Optimize away (CREATE && !OPEN). Let .create handle the race. */ |
| 586 | if ((flags & LOOKUP_CREATE ) && !(flags & LOOKUP_OPEN)) { |
| 587 | __d_lustre_invalidate(dentry); |
| 588 | d_add(dentry, NULL); |
| 589 | return NULL; |
| 590 | } |
| 591 | |
| 592 | if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE)) |
| 593 | itp = NULL; |
| 594 | else |
| 595 | itp = ⁢ |
| 596 | de = ll_lookup_it(parent, dentry, itp, 0); |
| 597 | |
| 598 | if (itp != NULL) |
| 599 | ll_intent_release(itp); |
| 600 | |
| 601 | return de; |
| 602 | } |
| 603 | |
| 604 | /* |
| 605 | * For cached negative dentry and new dentry, handle lookup/create/open |
| 606 | * together. |
| 607 | */ |
| 608 | static int ll_atomic_open(struct inode *dir, struct dentry *dentry, |
| 609 | struct file *file, unsigned open_flags, |
| 610 | umode_t mode, int *opened) |
| 611 | { |
| 612 | struct lookup_intent *it; |
| 613 | struct dentry *de; |
| 614 | long long lookup_flags = LOOKUP_OPEN; |
| 615 | int rc = 0; |
| 616 | |
| 617 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p," |
| 618 | "open_flags %x,mode %x opened %d\n", |
| 619 | dentry->d_name.len, dentry->d_name.name, dir->i_ino, |
| 620 | dir->i_generation, dir, file, open_flags, mode, *opened); |
| 621 | |
| 622 | OBD_ALLOC(it, sizeof(*it)); |
| 623 | if (!it) |
| 624 | return -ENOMEM; |
| 625 | |
| 626 | it->it_op = IT_OPEN; |
| 627 | if (open_flags & O_CREAT) { |
| 628 | it->it_op |= IT_CREAT; |
| 629 | lookup_flags |= LOOKUP_CREATE; |
| 630 | } |
| 631 | it->it_create_mode = (mode & S_IALLUGO) | S_IFREG; |
| 632 | it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags); |
| 633 | |
| 634 | /* Dentry added to dcache tree in ll_lookup_it */ |
| 635 | de = ll_lookup_it(dir, dentry, it, lookup_flags); |
| 636 | if (IS_ERR(de)) |
| 637 | rc = PTR_ERR(de); |
| 638 | else if (de != NULL) |
| 639 | dentry = de; |
| 640 | |
| 641 | if (!rc) { |
| 642 | if (it_disposition(it, DISP_OPEN_CREATE)) { |
| 643 | /* Dentry instantiated in ll_create_it. */ |
| 644 | rc = ll_create_it(dir, dentry, mode, it); |
| 645 | if (rc) { |
| 646 | /* We dget in ll_splice_alias. */ |
| 647 | if (de != NULL) |
| 648 | dput(de); |
| 649 | goto out_release; |
| 650 | } |
| 651 | |
| 652 | *opened |= FILE_CREATED; |
| 653 | } |
| 654 | if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) { |
| 655 | /* Open dentry. */ |
| 656 | if (S_ISFIFO(dentry->d_inode->i_mode)) { |
| 657 | /* We cannot call open here as it would |
| 658 | * deadlock. |
| 659 | */ |
| 660 | if (it_disposition(it, DISP_ENQ_OPEN_REF)) |
| 661 | ptlrpc_req_finished( |
| 662 | (struct ptlrpc_request *) |
| 663 | it->d.lustre.it_data); |
| 664 | rc = finish_no_open(file, de); |
| 665 | } else { |
| 666 | file->private_data = it; |
| 667 | rc = finish_open(file, dentry, NULL, opened); |
| 668 | /* We dget in ll_splice_alias. finish_open takes |
| 669 | * care of dget for fd open. |
| 670 | */ |
| 671 | if (de != NULL) |
| 672 | dput(de); |
| 673 | } |
| 674 | } else { |
| 675 | rc = finish_no_open(file, de); |
| 676 | } |
| 677 | } |
| 678 | |
| 679 | out_release: |
| 680 | ll_intent_release(it); |
| 681 | OBD_FREE(it, sizeof(*it)); |
| 682 | |
| 683 | return rc; |
| 684 | } |
| 685 | |
| 686 | |
| 687 | /* We depend on "mode" being set with the proper file type/umask by now */ |
| 688 | static struct inode *ll_create_node(struct inode *dir, const char *name, |
| 689 | int namelen, const void *data, int datalen, |
| 690 | int mode, __u64 extra, |
| 691 | struct lookup_intent *it) |
| 692 | { |
| 693 | struct inode *inode = NULL; |
| 694 | struct ptlrpc_request *request = NULL; |
| 695 | struct ll_sb_info *sbi = ll_i2sbi(dir); |
| 696 | int rc; |
| 697 | |
| 698 | LASSERT(it && it->d.lustre.it_disposition); |
| 699 | |
| 700 | LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF)); |
| 701 | request = it->d.lustre.it_data; |
| 702 | it_clear_disposition(it, DISP_ENQ_CREATE_REF); |
| 703 | rc = ll_prep_inode(&inode, request, dir->i_sb, it); |
| 704 | if (rc) |
| 705 | GOTO(out, inode = ERR_PTR(rc)); |
| 706 | |
| 707 | LASSERT(ll_d_hlist_empty(&inode->i_dentry)); |
| 708 | |
| 709 | /* We asked for a lock on the directory, but were granted a |
| 710 | * lock on the inode. Since we finally have an inode pointer, |
| 711 | * stuff it in the lock. */ |
| 712 | CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n", |
| 713 | inode, inode->i_ino, inode->i_generation); |
| 714 | ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL); |
| 715 | out: |
| 716 | ptlrpc_req_finished(request); |
| 717 | return inode; |
| 718 | } |
| 719 | |
| 720 | /* |
| 721 | * By the time this is called, we already have created the directory cache |
| 722 | * entry for the new file, but it is so far negative - it has no inode. |
| 723 | * |
| 724 | * We defer creating the OBD object(s) until open, to keep the intent and |
| 725 | * non-intent code paths similar, and also because we do not have the MDS |
| 726 | * inode number before calling ll_create_node() (which is needed for LOV), |
| 727 | * so we would need to do yet another RPC to the MDS to store the LOV EA |
| 728 | * data on the MDS. If needed, we would pass the PACKED lmm as data and |
| 729 | * lmm_size in datalen (the MDS still has code which will handle that). |
| 730 | * |
| 731 | * If the create succeeds, we fill in the inode information |
| 732 | * with d_instantiate(). |
| 733 | */ |
| 734 | static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode, |
| 735 | struct lookup_intent *it) |
| 736 | { |
| 737 | struct inode *inode; |
| 738 | int rc = 0; |
| 739 | |
| 740 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n", |
| 741 | dentry->d_name.len, dentry->d_name.name, dir->i_ino, |
| 742 | dir->i_generation, dir, LL_IT2STR(it)); |
| 743 | |
| 744 | rc = it_open_error(DISP_OPEN_CREATE, it); |
| 745 | if (rc) |
| 746 | return rc; |
| 747 | |
| 748 | inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len, |
| 749 | NULL, 0, mode, 0, it); |
| 750 | if (IS_ERR(inode)) |
| 751 | return PTR_ERR(inode); |
| 752 | |
| 753 | if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL)) |
| 754 | ll_i2info(inode)->lli_volatile = true; |
| 755 | |
| 756 | d_instantiate(dentry, inode); |
| 757 | return 0; |
| 758 | } |
| 759 | |
| 760 | static void ll_update_times(struct ptlrpc_request *request, |
| 761 | struct inode *inode) |
| 762 | { |
| 763 | struct mdt_body *body = req_capsule_server_get(&request->rq_pill, |
| 764 | &RMF_MDT_BODY); |
| 765 | |
| 766 | LASSERT(body); |
| 767 | if (body->valid & OBD_MD_FLMTIME && |
| 768 | body->mtime > LTIME_S(inode->i_mtime)) { |
| 769 | CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n", |
| 770 | inode->i_ino, LTIME_S(inode->i_mtime), body->mtime); |
| 771 | LTIME_S(inode->i_mtime) = body->mtime; |
| 772 | } |
| 773 | if (body->valid & OBD_MD_FLCTIME && |
| 774 | body->ctime > LTIME_S(inode->i_ctime)) |
| 775 | LTIME_S(inode->i_ctime) = body->ctime; |
| 776 | } |
| 777 | |
| 778 | static int ll_new_node(struct inode *dir, struct qstr *name, |
| 779 | const char *tgt, int mode, int rdev, |
| 780 | struct dentry *dchild, __u32 opc) |
| 781 | { |
| 782 | struct ptlrpc_request *request = NULL; |
| 783 | struct md_op_data *op_data; |
| 784 | struct inode *inode = NULL; |
| 785 | struct ll_sb_info *sbi = ll_i2sbi(dir); |
| 786 | int tgt_len = 0; |
| 787 | int err; |
| 788 | |
| 789 | if (unlikely(tgt != NULL)) |
| 790 | tgt_len = strlen(tgt) + 1; |
| 791 | |
| 792 | op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, |
| 793 | name->len, 0, opc, NULL); |
| 794 | if (IS_ERR(op_data)) |
| 795 | GOTO(err_exit, err = PTR_ERR(op_data)); |
| 796 | |
| 797 | err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode, |
| 798 | from_kuid(&init_user_ns, current_fsuid()), |
| 799 | from_kgid(&init_user_ns, current_fsgid()), |
| 800 | cfs_curproc_cap_pack(), rdev, &request); |
| 801 | ll_finish_md_op_data(op_data); |
| 802 | if (err) |
| 803 | GOTO(err_exit, err); |
| 804 | |
| 805 | ll_update_times(request, dir); |
| 806 | |
| 807 | if (dchild) { |
| 808 | err = ll_prep_inode(&inode, request, dchild->d_sb, NULL); |
| 809 | if (err) |
| 810 | GOTO(err_exit, err); |
| 811 | |
| 812 | d_instantiate(dchild, inode); |
| 813 | } |
| 814 | err_exit: |
| 815 | ptlrpc_req_finished(request); |
| 816 | |
| 817 | return err; |
| 818 | } |
| 819 | |
| 820 | static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode, |
| 821 | unsigned rdev, struct dentry *dchild) |
| 822 | { |
| 823 | int err; |
| 824 | |
| 825 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n", |
| 826 | name->len, name->name, dir->i_ino, dir->i_generation, dir, |
| 827 | mode, rdev); |
| 828 | |
| 829 | if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir))) |
| 830 | mode &= ~current_umask(); |
| 831 | |
| 832 | switch (mode & S_IFMT) { |
| 833 | case 0: |
| 834 | mode |= S_IFREG; /* for mode = 0 case, fallthrough */ |
| 835 | case S_IFREG: |
| 836 | case S_IFCHR: |
| 837 | case S_IFBLK: |
| 838 | case S_IFIFO: |
| 839 | case S_IFSOCK: |
| 840 | err = ll_new_node(dir, name, NULL, mode, rdev, dchild, |
| 841 | LUSTRE_OPC_MKNOD); |
| 842 | break; |
| 843 | case S_IFDIR: |
| 844 | err = -EPERM; |
| 845 | break; |
| 846 | default: |
| 847 | err = -EINVAL; |
| 848 | } |
| 849 | |
| 850 | if (!err) |
| 851 | ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1); |
| 852 | |
| 853 | return err; |
| 854 | } |
| 855 | |
| 856 | /* |
| 857 | * Plain create. Intent create is handled in atomic_open. |
| 858 | */ |
| 859 | static int ll_create_nd(struct inode *dir, struct dentry *dentry, |
| 860 | umode_t mode, bool want_excl) |
| 861 | { |
| 862 | int rc; |
| 863 | |
| 864 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)," |
| 865 | "flags=%u, excl=%d\n", |
| 866 | dentry->d_name.len, dentry->d_name.name, dir->i_ino, |
| 867 | dir->i_generation, dir, mode, want_excl); |
| 868 | |
| 869 | rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry); |
| 870 | |
| 871 | ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1); |
| 872 | |
| 873 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n", |
| 874 | dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry)); |
| 875 | |
| 876 | return rc; |
| 877 | } |
| 878 | |
| 879 | static int ll_symlink_generic(struct inode *dir, struct qstr *name, |
| 880 | const char *tgt, struct dentry *dchild) |
| 881 | { |
| 882 | int err; |
| 883 | |
| 884 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n", |
| 885 | name->len, name->name, dir->i_ino, dir->i_generation, |
| 886 | dir, 3000, tgt); |
| 887 | |
| 888 | err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO, |
| 889 | 0, dchild, LUSTRE_OPC_SYMLINK); |
| 890 | |
| 891 | if (!err) |
| 892 | ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1); |
| 893 | |
| 894 | return err; |
| 895 | } |
| 896 | |
| 897 | static int ll_link_generic(struct inode *src, struct inode *dir, |
| 898 | struct qstr *name, struct dentry *dchild) |
| 899 | { |
| 900 | struct ll_sb_info *sbi = ll_i2sbi(dir); |
| 901 | struct ptlrpc_request *request = NULL; |
| 902 | struct md_op_data *op_data; |
| 903 | int err; |
| 904 | |
| 905 | CDEBUG(D_VFSTRACE, |
| 906 | "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n", |
| 907 | src->i_ino, src->i_generation, src, dir->i_ino, |
| 908 | dir->i_generation, dir, name->len, name->name); |
| 909 | |
| 910 | op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len, |
| 911 | 0, LUSTRE_OPC_ANY, NULL); |
| 912 | if (IS_ERR(op_data)) |
| 913 | return PTR_ERR(op_data); |
| 914 | |
| 915 | err = md_link(sbi->ll_md_exp, op_data, &request); |
| 916 | ll_finish_md_op_data(op_data); |
| 917 | if (err) |
| 918 | GOTO(out, err); |
| 919 | |
| 920 | ll_update_times(request, dir); |
| 921 | ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1); |
| 922 | out: |
| 923 | ptlrpc_req_finished(request); |
| 924 | return err; |
| 925 | } |
| 926 | |
| 927 | static int ll_mkdir_generic(struct inode *dir, struct qstr *name, |
| 928 | int mode, struct dentry *dchild) |
| 929 | |
| 930 | { |
| 931 | int err; |
| 932 | |
| 933 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n", |
| 934 | name->len, name->name, dir->i_ino, dir->i_generation, dir); |
| 935 | |
| 936 | if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir))) |
| 937 | mode &= ~current_umask(); |
| 938 | mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR; |
| 939 | err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR); |
| 940 | |
| 941 | if (!err) |
| 942 | ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1); |
| 943 | |
| 944 | return err; |
| 945 | } |
| 946 | |
| 947 | /* Try to find the child dentry by its name. |
| 948 | If found, put the result fid into @fid. */ |
| 949 | static void ll_get_child_fid(struct inode * dir, struct qstr *name, |
| 950 | struct lu_fid *fid) |
| 951 | { |
| 952 | struct dentry *parent, *child; |
| 953 | |
| 954 | parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias); |
| 955 | child = d_lookup(parent, name); |
| 956 | if (child) { |
| 957 | if (child->d_inode) |
| 958 | *fid = *ll_inode2fid(child->d_inode); |
| 959 | dput(child); |
| 960 | } |
| 961 | } |
| 962 | |
| 963 | static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent, |
| 964 | struct dentry *dchild, struct qstr *name) |
| 965 | { |
| 966 | struct ptlrpc_request *request = NULL; |
| 967 | struct md_op_data *op_data; |
| 968 | int rc; |
| 969 | |
| 970 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n", |
| 971 | name->len, name->name, dir->i_ino, dir->i_generation, dir); |
| 972 | |
| 973 | if (unlikely(ll_d_mountpoint(dparent, dchild, name))) |
| 974 | return -EBUSY; |
| 975 | |
| 976 | op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, |
| 977 | S_IFDIR, LUSTRE_OPC_ANY, NULL); |
| 978 | if (IS_ERR(op_data)) |
| 979 | return PTR_ERR(op_data); |
| 980 | |
| 981 | ll_get_child_fid(dir, name, &op_data->op_fid3); |
| 982 | op_data->op_fid2 = op_data->op_fid3; |
| 983 | rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request); |
| 984 | ll_finish_md_op_data(op_data); |
| 985 | if (rc == 0) { |
| 986 | ll_update_times(request, dir); |
| 987 | ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1); |
| 988 | } |
| 989 | |
| 990 | ptlrpc_req_finished(request); |
| 991 | return rc; |
| 992 | } |
| 993 | |
| 994 | /** |
| 995 | * Remove dir entry |
| 996 | **/ |
| 997 | int ll_rmdir_entry(struct inode *dir, char *name, int namelen) |
| 998 | { |
| 999 | struct ptlrpc_request *request = NULL; |
| 1000 | struct md_op_data *op_data; |
| 1001 | int rc; |
| 1002 | |
| 1003 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n", |
| 1004 | namelen, name, dir->i_ino, dir->i_generation, dir); |
| 1005 | |
| 1006 | op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name), |
| 1007 | S_IFDIR, LUSTRE_OPC_ANY, NULL); |
| 1008 | if (IS_ERR(op_data)) |
| 1009 | return PTR_ERR(op_data); |
| 1010 | op_data->op_cli_flags |= CLI_RM_ENTRY; |
| 1011 | rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request); |
| 1012 | ll_finish_md_op_data(op_data); |
| 1013 | if (rc == 0) { |
| 1014 | ll_update_times(request, dir); |
| 1015 | ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1); |
| 1016 | } |
| 1017 | |
| 1018 | ptlrpc_req_finished(request); |
| 1019 | return rc; |
| 1020 | } |
| 1021 | |
| 1022 | int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir) |
| 1023 | { |
| 1024 | struct mdt_body *body; |
| 1025 | struct lov_mds_md *eadata; |
| 1026 | struct lov_stripe_md *lsm = NULL; |
| 1027 | struct obd_trans_info oti = { 0 }; |
| 1028 | struct obdo *oa; |
| 1029 | struct obd_capa *oc = NULL; |
| 1030 | int rc; |
| 1031 | |
| 1032 | /* req is swabbed so this is safe */ |
| 1033 | body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY); |
| 1034 | if (!(body->valid & OBD_MD_FLEASIZE)) |
| 1035 | return 0; |
| 1036 | |
| 1037 | if (body->eadatasize == 0) { |
| 1038 | CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n"); |
| 1039 | GOTO(out, rc = -EPROTO); |
| 1040 | } |
| 1041 | |
| 1042 | /* The MDS sent back the EA because we unlinked the last reference |
| 1043 | * to this file. Use this EA to unlink the objects on the OST. |
| 1044 | * It's opaque so we don't swab here; we leave it to obd_unpackmd() to |
| 1045 | * check it is complete and sensible. */ |
| 1046 | eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD, |
| 1047 | body->eadatasize); |
| 1048 | LASSERT(eadata != NULL); |
| 1049 | |
| 1050 | rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize); |
| 1051 | if (rc < 0) { |
| 1052 | CERROR("obd_unpackmd: %d\n", rc); |
| 1053 | GOTO(out, rc); |
| 1054 | } |
| 1055 | LASSERT(rc >= sizeof(*lsm)); |
| 1056 | |
| 1057 | OBDO_ALLOC(oa); |
| 1058 | if (oa == NULL) |
| 1059 | GOTO(out_free_memmd, rc = -ENOMEM); |
| 1060 | |
| 1061 | oa->o_oi = lsm->lsm_oi; |
| 1062 | oa->o_mode = body->mode & S_IFMT; |
| 1063 | oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP; |
| 1064 | |
| 1065 | if (body->valid & OBD_MD_FLCOOKIE) { |
| 1066 | oa->o_valid |= OBD_MD_FLCOOKIE; |
| 1067 | oti.oti_logcookies = |
| 1068 | req_capsule_server_sized_get(&request->rq_pill, |
| 1069 | &RMF_LOGCOOKIES, |
| 1070 | sizeof(struct llog_cookie) * |
| 1071 | lsm->lsm_stripe_count); |
| 1072 | if (oti.oti_logcookies == NULL) { |
| 1073 | oa->o_valid &= ~OBD_MD_FLCOOKIE; |
| 1074 | body->valid &= ~OBD_MD_FLCOOKIE; |
| 1075 | } |
| 1076 | } |
| 1077 | |
| 1078 | if (body->valid & OBD_MD_FLOSSCAPA) { |
| 1079 | rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc); |
| 1080 | if (rc) |
| 1081 | GOTO(out_free_memmd, rc); |
| 1082 | } |
| 1083 | |
| 1084 | rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti, |
| 1085 | ll_i2mdexp(dir), oc); |
| 1086 | capa_put(oc); |
| 1087 | if (rc) |
| 1088 | CERROR("obd destroy objid "DOSTID" error %d\n", |
| 1089 | POSTID(&lsm->lsm_oi), rc); |
| 1090 | out_free_memmd: |
| 1091 | obd_free_memmd(ll_i2dtexp(dir), &lsm); |
| 1092 | OBDO_FREE(oa); |
| 1093 | out: |
| 1094 | return rc; |
| 1095 | } |
| 1096 | |
| 1097 | /* ll_unlink_generic() doesn't update the inode with the new link count. |
| 1098 | * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there |
| 1099 | * is any lock existing. They will recycle dentries and inodes based upon locks |
| 1100 | * too. b=20433 */ |
| 1101 | static int ll_unlink_generic(struct inode *dir, struct dentry *dparent, |
| 1102 | struct dentry *dchild, struct qstr *name) |
| 1103 | { |
| 1104 | struct ptlrpc_request *request = NULL; |
| 1105 | struct md_op_data *op_data; |
| 1106 | int rc; |
| 1107 | CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n", |
| 1108 | name->len, name->name, dir->i_ino, dir->i_generation, dir); |
| 1109 | |
| 1110 | /* |
| 1111 | * XXX: unlink bind mountpoint maybe call to here, |
| 1112 | * just check it as vfs_unlink does. |
| 1113 | */ |
| 1114 | if (unlikely(ll_d_mountpoint(dparent, dchild, name))) |
| 1115 | return -EBUSY; |
| 1116 | |
| 1117 | op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, |
| 1118 | name->len, 0, LUSTRE_OPC_ANY, NULL); |
| 1119 | if (IS_ERR(op_data)) |
| 1120 | return PTR_ERR(op_data); |
| 1121 | |
| 1122 | ll_get_child_fid(dir, name, &op_data->op_fid3); |
| 1123 | op_data->op_fid2 = op_data->op_fid3; |
| 1124 | rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request); |
| 1125 | ll_finish_md_op_data(op_data); |
| 1126 | if (rc) |
| 1127 | GOTO(out, rc); |
| 1128 | |
| 1129 | ll_update_times(request, dir); |
| 1130 | ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1); |
| 1131 | |
| 1132 | rc = ll_objects_destroy(request, dir); |
| 1133 | out: |
| 1134 | ptlrpc_req_finished(request); |
| 1135 | return rc; |
| 1136 | } |
| 1137 | |
| 1138 | static int ll_rename_generic(struct inode *src, struct dentry *src_dparent, |
| 1139 | struct dentry *src_dchild, struct qstr *src_name, |
| 1140 | struct inode *tgt, struct dentry *tgt_dparent, |
| 1141 | struct dentry *tgt_dchild, struct qstr *tgt_name) |
| 1142 | { |
| 1143 | struct ptlrpc_request *request = NULL; |
| 1144 | struct ll_sb_info *sbi = ll_i2sbi(src); |
| 1145 | struct md_op_data *op_data; |
| 1146 | int err; |
| 1147 | |
| 1148 | CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s," |
| 1149 | "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name, |
| 1150 | src->i_ino, src->i_generation, src, tgt_name->len, |
| 1151 | tgt_name->name, tgt->i_ino, tgt->i_generation, tgt); |
| 1152 | |
| 1153 | if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) || |
| 1154 | ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name))) |
| 1155 | return -EBUSY; |
| 1156 | |
| 1157 | op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0, |
| 1158 | LUSTRE_OPC_ANY, NULL); |
| 1159 | if (IS_ERR(op_data)) |
| 1160 | return PTR_ERR(op_data); |
| 1161 | |
| 1162 | ll_get_child_fid(src, src_name, &op_data->op_fid3); |
| 1163 | ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4); |
| 1164 | err = md_rename(sbi->ll_md_exp, op_data, |
| 1165 | src_name->name, src_name->len, |
| 1166 | tgt_name->name, tgt_name->len, &request); |
| 1167 | ll_finish_md_op_data(op_data); |
| 1168 | if (!err) { |
| 1169 | ll_update_times(request, src); |
| 1170 | ll_update_times(request, tgt); |
| 1171 | ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1); |
| 1172 | err = ll_objects_destroy(request, src); |
| 1173 | } |
| 1174 | |
| 1175 | ptlrpc_req_finished(request); |
| 1176 | |
| 1177 | return err; |
| 1178 | } |
| 1179 | |
| 1180 | static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode, |
| 1181 | dev_t rdev) |
| 1182 | { |
| 1183 | return ll_mknod_generic(dir, &dchild->d_name, mode, |
| 1184 | old_encode_dev(rdev), dchild); |
| 1185 | } |
| 1186 | |
| 1187 | static int ll_unlink(struct inode * dir, struct dentry *dentry) |
| 1188 | { |
| 1189 | return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name); |
| 1190 | } |
| 1191 | |
| 1192 | static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode) |
| 1193 | { |
| 1194 | return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry); |
| 1195 | } |
| 1196 | |
| 1197 | static int ll_rmdir(struct inode *dir, struct dentry *dentry) |
| 1198 | { |
| 1199 | return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name); |
| 1200 | } |
| 1201 | |
| 1202 | static int ll_symlink(struct inode *dir, struct dentry *dentry, |
| 1203 | const char *oldname) |
| 1204 | { |
| 1205 | return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry); |
| 1206 | } |
| 1207 | |
| 1208 | static int ll_link(struct dentry *old_dentry, struct inode *dir, |
| 1209 | struct dentry *new_dentry) |
| 1210 | { |
| 1211 | return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name, |
| 1212 | new_dentry); |
| 1213 | } |
| 1214 | |
| 1215 | static int ll_rename(struct inode *old_dir, struct dentry *old_dentry, |
| 1216 | struct inode *new_dir, struct dentry *new_dentry) |
| 1217 | { |
| 1218 | int err; |
| 1219 | err = ll_rename_generic(old_dir, NULL, |
| 1220 | old_dentry, &old_dentry->d_name, |
| 1221 | new_dir, NULL, new_dentry, |
| 1222 | &new_dentry->d_name); |
| 1223 | if (!err) { |
| 1224 | d_move(old_dentry, new_dentry); |
| 1225 | } |
| 1226 | return err; |
| 1227 | } |
| 1228 | |
| 1229 | struct inode_operations ll_dir_inode_operations = { |
| 1230 | .mknod = ll_mknod, |
| 1231 | .atomic_open = ll_atomic_open, |
| 1232 | .lookup = ll_lookup_nd, |
| 1233 | .create = ll_create_nd, |
| 1234 | /* We need all these non-raw things for NFSD, to not patch it. */ |
| 1235 | .unlink = ll_unlink, |
| 1236 | .mkdir = ll_mkdir, |
| 1237 | .rmdir = ll_rmdir, |
| 1238 | .symlink = ll_symlink, |
| 1239 | .link = ll_link, |
| 1240 | .rename = ll_rename, |
| 1241 | .setattr = ll_setattr, |
| 1242 | .getattr = ll_getattr, |
| 1243 | .permission = ll_inode_permission, |
| 1244 | .setxattr = ll_setxattr, |
| 1245 | .getxattr = ll_getxattr, |
| 1246 | .listxattr = ll_listxattr, |
| 1247 | .removexattr = ll_removexattr, |
| 1248 | .get_acl = ll_get_acl, |
| 1249 | }; |
| 1250 | |
| 1251 | struct inode_operations ll_special_inode_operations = { |
| 1252 | .setattr = ll_setattr, |
| 1253 | .getattr = ll_getattr, |
| 1254 | .permission = ll_inode_permission, |
| 1255 | .setxattr = ll_setxattr, |
| 1256 | .getxattr = ll_getxattr, |
| 1257 | .listxattr = ll_listxattr, |
| 1258 | .removexattr = ll_removexattr, |
| 1259 | .get_acl = ll_get_acl, |
| 1260 | }; |