| 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 | * lustre/llite/dir.c |
| 37 | * |
| 38 | * Directory code for lustre client. |
| 39 | */ |
| 40 | |
| 41 | #include <linux/fs.h> |
| 42 | #include <linux/pagemap.h> |
| 43 | #include <linux/mm.h> |
| 44 | #include <asm/uaccess.h> |
| 45 | #include <linux/buffer_head.h> /* for wait_on_buffer */ |
| 46 | #include <linux/pagevec.h> |
| 47 | #include <linux/prefetch.h> |
| 48 | |
| 49 | #define DEBUG_SUBSYSTEM S_LLITE |
| 50 | |
| 51 | #include "../include/obd_support.h" |
| 52 | #include "../include/obd_class.h" |
| 53 | #include "../include/lustre_lib.h" |
| 54 | #include "../include/lustre/lustre_idl.h" |
| 55 | #include "../include/lustre_lite.h" |
| 56 | #include "../include/lustre_dlm.h" |
| 57 | #include "../include/lustre_fid.h" |
| 58 | #include "llite_internal.h" |
| 59 | |
| 60 | /* |
| 61 | * (new) readdir implementation overview. |
| 62 | * |
| 63 | * Original lustre readdir implementation cached exact copy of raw directory |
| 64 | * pages on the client. These pages were indexed in client page cache by |
| 65 | * logical offset in the directory file. This design, while very simple and |
| 66 | * intuitive had some inherent problems: |
| 67 | * |
| 68 | * . it implies that byte offset to the directory entry serves as a |
| 69 | * telldir(3)/seekdir(3) cookie, but that offset is not stable: in |
| 70 | * ext3/htree directory entries may move due to splits, and more |
| 71 | * importantly, |
| 72 | * |
| 73 | * . it is incompatible with the design of split directories for cmd3, |
| 74 | * that assumes that names are distributed across nodes based on their |
| 75 | * hash, and so readdir should be done in hash order. |
| 76 | * |
| 77 | * New readdir implementation does readdir in hash order, and uses hash of a |
| 78 | * file name as a telldir/seekdir cookie. This led to number of complications: |
| 79 | * |
| 80 | * . hash is not unique, so it cannot be used to index cached directory |
| 81 | * pages on the client (note, that it requires a whole pageful of hash |
| 82 | * collided entries to cause two pages to have identical hashes); |
| 83 | * |
| 84 | * . hash is not unique, so it cannot, strictly speaking, be used as an |
| 85 | * entry cookie. ext3/htree has the same problem and lustre implementation |
| 86 | * mimics their solution: seekdir(hash) positions directory at the first |
| 87 | * entry with the given hash. |
| 88 | * |
| 89 | * Client side. |
| 90 | * |
| 91 | * 0. caching |
| 92 | * |
| 93 | * Client caches directory pages using hash of the first entry as an index. As |
| 94 | * noted above hash is not unique, so this solution doesn't work as is: |
| 95 | * special processing is needed for "page hash chains" (i.e., sequences of |
| 96 | * pages filled with entries all having the same hash value). |
| 97 | * |
| 98 | * First, such chains have to be detected. To this end, server returns to the |
| 99 | * client the hash of the first entry on the page next to one returned. When |
| 100 | * client detects that this hash is the same as hash of the first entry on the |
| 101 | * returned page, page hash collision has to be handled. Pages in the |
| 102 | * hash chain, except first one, are termed "overflow pages". |
| 103 | * |
| 104 | * Solution to index uniqueness problem is to not cache overflow |
| 105 | * pages. Instead, when page hash collision is detected, all overflow pages |
| 106 | * from emerging chain are immediately requested from the server and placed in |
| 107 | * a special data structure (struct ll_dir_chain). This data structure is used |
| 108 | * by ll_readdir() to process entries from overflow pages. When readdir |
| 109 | * invocation finishes, overflow pages are discarded. If page hash collision |
| 110 | * chain weren't completely processed, next call to readdir will again detect |
| 111 | * page hash collision, again read overflow pages in, process next portion of |
| 112 | * entries and again discard the pages. This is not as wasteful as it looks, |
| 113 | * because, given reasonable hash, page hash collisions are extremely rare. |
| 114 | * |
| 115 | * 1. directory positioning |
| 116 | * |
| 117 | * When seekdir(hash) is called, original |
| 118 | * |
| 119 | * |
| 120 | * |
| 121 | * |
| 122 | * |
| 123 | * |
| 124 | * |
| 125 | * |
| 126 | * Server. |
| 127 | * |
| 128 | * identification of and access to overflow pages |
| 129 | * |
| 130 | * page format |
| 131 | * |
| 132 | * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains |
| 133 | * a header lu_dirpage which describes the start/end hash, and whether this |
| 134 | * page is empty (contains no dir entry) or hash collide with next page. |
| 135 | * After client receives reply, several pages will be integrated into dir page |
| 136 | * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the |
| 137 | * lu_dirpage for this integrated page will be adjusted. See |
| 138 | * lmv_adjust_dirpages(). |
| 139 | * |
| 140 | */ |
| 141 | |
| 142 | /* returns the page unlocked, but with a reference */ |
| 143 | static int ll_dir_filler(void *_hash, struct page *page0) |
| 144 | { |
| 145 | struct inode *inode = page0->mapping->host; |
| 146 | int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH; |
| 147 | struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp; |
| 148 | struct ptlrpc_request *request; |
| 149 | struct mdt_body *body; |
| 150 | struct md_op_data *op_data; |
| 151 | __u64 hash = *((__u64 *)_hash); |
| 152 | struct page **page_pool; |
| 153 | struct page *page; |
| 154 | struct lu_dirpage *dp; |
| 155 | int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT; |
| 156 | int nrdpgs = 0; /* number of pages read actually */ |
| 157 | int npages; |
| 158 | int i; |
| 159 | int rc; |
| 160 | |
| 161 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash %llu\n", |
| 162 | inode->i_ino, inode->i_generation, inode, hash); |
| 163 | |
| 164 | LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES); |
| 165 | |
| 166 | page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS); |
| 167 | if (page_pool) { |
| 168 | page_pool[0] = page0; |
| 169 | } else { |
| 170 | page_pool = &page0; |
| 171 | max_pages = 1; |
| 172 | } |
| 173 | for (npages = 1; npages < max_pages; npages++) { |
| 174 | page = page_cache_alloc_cold(inode->i_mapping); |
| 175 | if (!page) |
| 176 | break; |
| 177 | page_pool[npages] = page; |
| 178 | } |
| 179 | |
| 180 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, |
| 181 | LUSTRE_OPC_ANY, NULL); |
| 182 | op_data->op_npages = npages; |
| 183 | op_data->op_offset = hash; |
| 184 | rc = md_readpage(exp, op_data, page_pool, &request); |
| 185 | ll_finish_md_op_data(op_data); |
| 186 | if (rc == 0) { |
| 187 | body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY); |
| 188 | /* Checked by mdc_readpage() */ |
| 189 | LASSERT(body != NULL); |
| 190 | |
| 191 | if (body->valid & OBD_MD_FLSIZE) |
| 192 | cl_isize_write(inode, body->size); |
| 193 | |
| 194 | nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1) |
| 195 | >> PAGE_CACHE_SHIFT; |
| 196 | SetPageUptodate(page0); |
| 197 | } |
| 198 | unlock_page(page0); |
| 199 | ptlrpc_req_finished(request); |
| 200 | |
| 201 | CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages); |
| 202 | |
| 203 | ll_pagevec_init(&lru_pvec, 0); |
| 204 | for (i = 1; i < npages; i++) { |
| 205 | unsigned long offset; |
| 206 | int ret; |
| 207 | |
| 208 | page = page_pool[i]; |
| 209 | |
| 210 | if (rc < 0 || i >= nrdpgs) { |
| 211 | page_cache_release(page); |
| 212 | continue; |
| 213 | } |
| 214 | |
| 215 | SetPageUptodate(page); |
| 216 | |
| 217 | dp = kmap(page); |
| 218 | hash = le64_to_cpu(dp->ldp_hash_start); |
| 219 | kunmap(page); |
| 220 | |
| 221 | offset = hash_x_index(hash, hash64); |
| 222 | |
| 223 | prefetchw(&page->flags); |
| 224 | ret = add_to_page_cache_lru(page, inode->i_mapping, offset, |
| 225 | GFP_KERNEL); |
| 226 | if (ret == 0) { |
| 227 | unlock_page(page); |
| 228 | if (ll_pagevec_add(&lru_pvec, page) == 0) |
| 229 | ll_pagevec_lru_add_file(&lru_pvec); |
| 230 | } else { |
| 231 | CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:" |
| 232 | " %d\n", offset, ret); |
| 233 | } |
| 234 | page_cache_release(page); |
| 235 | } |
| 236 | ll_pagevec_lru_add_file(&lru_pvec); |
| 237 | |
| 238 | if (page_pool != &page0) |
| 239 | OBD_FREE(page_pool, sizeof(struct page *) * max_pages); |
| 240 | return rc; |
| 241 | } |
| 242 | |
| 243 | static void ll_check_page(struct inode *dir, struct page *page) |
| 244 | { |
| 245 | /* XXX: check page format later */ |
| 246 | SetPageChecked(page); |
| 247 | } |
| 248 | |
| 249 | void ll_release_page(struct page *page, int remove) |
| 250 | { |
| 251 | kunmap(page); |
| 252 | if (remove) { |
| 253 | lock_page(page); |
| 254 | if (likely(page->mapping != NULL)) |
| 255 | truncate_complete_page(page->mapping, page); |
| 256 | unlock_page(page); |
| 257 | } |
| 258 | page_cache_release(page); |
| 259 | } |
| 260 | |
| 261 | /* |
| 262 | * Find, kmap and return page that contains given hash. |
| 263 | */ |
| 264 | static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash, |
| 265 | __u64 *start, __u64 *end) |
| 266 | { |
| 267 | int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH; |
| 268 | struct address_space *mapping = dir->i_mapping; |
| 269 | /* |
| 270 | * Complement of hash is used as an index so that |
| 271 | * radix_tree_gang_lookup() can be used to find a page with starting |
| 272 | * hash _smaller_ than one we are looking for. |
| 273 | */ |
| 274 | unsigned long offset = hash_x_index(*hash, hash64); |
| 275 | struct page *page; |
| 276 | int found; |
| 277 | |
| 278 | spin_lock_irq(&mapping->tree_lock); |
| 279 | found = radix_tree_gang_lookup(&mapping->page_tree, |
| 280 | (void **)&page, offset, 1); |
| 281 | if (found > 0) { |
| 282 | struct lu_dirpage *dp; |
| 283 | |
| 284 | page_cache_get(page); |
| 285 | spin_unlock_irq(&mapping->tree_lock); |
| 286 | /* |
| 287 | * In contrast to find_lock_page() we are sure that directory |
| 288 | * page cannot be truncated (while DLM lock is held) and, |
| 289 | * hence, can avoid restart. |
| 290 | * |
| 291 | * In fact, page cannot be locked here at all, because |
| 292 | * ll_dir_filler() does synchronous io. |
| 293 | */ |
| 294 | wait_on_page_locked(page); |
| 295 | if (PageUptodate(page)) { |
| 296 | dp = kmap(page); |
| 297 | if (BITS_PER_LONG == 32 && hash64) { |
| 298 | *start = le64_to_cpu(dp->ldp_hash_start) >> 32; |
| 299 | *end = le64_to_cpu(dp->ldp_hash_end) >> 32; |
| 300 | *hash = *hash >> 32; |
| 301 | } else { |
| 302 | *start = le64_to_cpu(dp->ldp_hash_start); |
| 303 | *end = le64_to_cpu(dp->ldp_hash_end); |
| 304 | } |
| 305 | LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n", |
| 306 | *start, *end, *hash); |
| 307 | CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n", |
| 308 | offset, *start, *end, *hash); |
| 309 | if (*hash > *end) { |
| 310 | ll_release_page(page, 0); |
| 311 | page = NULL; |
| 312 | } else if (*end != *start && *hash == *end) { |
| 313 | /* |
| 314 | * upon hash collision, remove this page, |
| 315 | * otherwise put page reference, and |
| 316 | * ll_get_dir_page() will issue RPC to fetch |
| 317 | * the page we want. |
| 318 | */ |
| 319 | ll_release_page(page, |
| 320 | le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE); |
| 321 | page = NULL; |
| 322 | } |
| 323 | } else { |
| 324 | page_cache_release(page); |
| 325 | page = ERR_PTR(-EIO); |
| 326 | } |
| 327 | |
| 328 | } else { |
| 329 | spin_unlock_irq(&mapping->tree_lock); |
| 330 | page = NULL; |
| 331 | } |
| 332 | return page; |
| 333 | } |
| 334 | |
| 335 | struct page *ll_get_dir_page(struct inode *dir, __u64 hash, |
| 336 | struct ll_dir_chain *chain) |
| 337 | { |
| 338 | ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} }; |
| 339 | struct address_space *mapping = dir->i_mapping; |
| 340 | struct lustre_handle lockh; |
| 341 | struct lu_dirpage *dp; |
| 342 | struct page *page; |
| 343 | ldlm_mode_t mode; |
| 344 | int rc; |
| 345 | __u64 start = 0; |
| 346 | __u64 end = 0; |
| 347 | __u64 lhash = hash; |
| 348 | struct ll_inode_info *lli = ll_i2info(dir); |
| 349 | int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH; |
| 350 | |
| 351 | mode = LCK_PR; |
| 352 | rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED, |
| 353 | ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh); |
| 354 | if (!rc) { |
| 355 | struct ldlm_enqueue_info einfo = { |
| 356 | .ei_type = LDLM_IBITS, |
| 357 | .ei_mode = mode, |
| 358 | .ei_cb_bl = ll_md_blocking_ast, |
| 359 | .ei_cb_cp = ldlm_completion_ast, |
| 360 | }; |
| 361 | struct lookup_intent it = { .it_op = IT_READDIR }; |
| 362 | struct ptlrpc_request *request; |
| 363 | struct md_op_data *op_data; |
| 364 | |
| 365 | op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0, |
| 366 | LUSTRE_OPC_ANY, NULL); |
| 367 | if (IS_ERR(op_data)) |
| 368 | return (void *)op_data; |
| 369 | |
| 370 | rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it, |
| 371 | op_data, &lockh, NULL, 0, NULL, 0); |
| 372 | |
| 373 | ll_finish_md_op_data(op_data); |
| 374 | |
| 375 | request = (struct ptlrpc_request *)it.d.lustre.it_data; |
| 376 | if (request) |
| 377 | ptlrpc_req_finished(request); |
| 378 | if (rc < 0) { |
| 379 | CERROR("lock enqueue: "DFID" at %llu: rc %d\n", |
| 380 | PFID(ll_inode2fid(dir)), hash, rc); |
| 381 | return ERR_PTR(rc); |
| 382 | } |
| 383 | |
| 384 | CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n", |
| 385 | dir, dir->i_ino, dir->i_generation); |
| 386 | md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, |
| 387 | &it.d.lustre.it_lock_handle, dir, NULL); |
| 388 | } else { |
| 389 | /* for cross-ref object, l_ast_data of the lock may not be set, |
| 390 | * we reset it here */ |
| 391 | md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie, |
| 392 | dir, NULL); |
| 393 | } |
| 394 | ldlm_lock_dump_handle(D_OTHER, &lockh); |
| 395 | |
| 396 | mutex_lock(&lli->lli_readdir_mutex); |
| 397 | page = ll_dir_page_locate(dir, &lhash, &start, &end); |
| 398 | if (IS_ERR(page)) { |
| 399 | CERROR("dir page locate: "DFID" at %llu: rc %ld\n", |
| 400 | PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page)); |
| 401 | goto out_unlock; |
| 402 | } else if (page != NULL) { |
| 403 | /* |
| 404 | * XXX nikita: not entirely correct handling of a corner case: |
| 405 | * suppose hash chain of entries with hash value HASH crosses |
| 406 | * border between pages P0 and P1. First both P0 and P1 are |
| 407 | * cached, seekdir() is called for some entry from the P0 part |
| 408 | * of the chain. Later P0 goes out of cache. telldir(HASH) |
| 409 | * happens and finds P1, as it starts with matching hash |
| 410 | * value. Remaining entries from P0 part of the chain are |
| 411 | * skipped. (Is that really a bug?) |
| 412 | * |
| 413 | * Possible solutions: 0. don't cache P1 is such case, handle |
| 414 | * it as an "overflow" page. 1. invalidate all pages at |
| 415 | * once. 2. use HASH|1 as an index for P1. |
| 416 | */ |
| 417 | goto hash_collision; |
| 418 | } |
| 419 | |
| 420 | page = read_cache_page(mapping, hash_x_index(hash, hash64), |
| 421 | ll_dir_filler, &lhash); |
| 422 | if (IS_ERR(page)) { |
| 423 | CERROR("read cache page: "DFID" at %llu: rc %ld\n", |
| 424 | PFID(ll_inode2fid(dir)), hash, PTR_ERR(page)); |
| 425 | goto out_unlock; |
| 426 | } |
| 427 | |
| 428 | wait_on_page_locked(page); |
| 429 | (void)kmap(page); |
| 430 | if (!PageUptodate(page)) { |
| 431 | CERROR("page not updated: "DFID" at %llu: rc %d\n", |
| 432 | PFID(ll_inode2fid(dir)), hash, -5); |
| 433 | goto fail; |
| 434 | } |
| 435 | if (!PageChecked(page)) |
| 436 | ll_check_page(dir, page); |
| 437 | if (PageError(page)) { |
| 438 | CERROR("page error: "DFID" at %llu: rc %d\n", |
| 439 | PFID(ll_inode2fid(dir)), hash, -5); |
| 440 | goto fail; |
| 441 | } |
| 442 | hash_collision: |
| 443 | dp = page_address(page); |
| 444 | if (BITS_PER_LONG == 32 && hash64) { |
| 445 | start = le64_to_cpu(dp->ldp_hash_start) >> 32; |
| 446 | end = le64_to_cpu(dp->ldp_hash_end) >> 32; |
| 447 | lhash = hash >> 32; |
| 448 | } else { |
| 449 | start = le64_to_cpu(dp->ldp_hash_start); |
| 450 | end = le64_to_cpu(dp->ldp_hash_end); |
| 451 | lhash = hash; |
| 452 | } |
| 453 | if (end == start) { |
| 454 | LASSERT(start == lhash); |
| 455 | CWARN("Page-wide hash collision: %llu\n", end); |
| 456 | if (BITS_PER_LONG == 32 && hash64) |
| 457 | CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n", |
| 458 | le64_to_cpu(dp->ldp_hash_start), |
| 459 | le64_to_cpu(dp->ldp_hash_end), hash); |
| 460 | /* |
| 461 | * Fetch whole overflow chain... |
| 462 | * |
| 463 | * XXX not yet. |
| 464 | */ |
| 465 | goto fail; |
| 466 | } |
| 467 | out_unlock: |
| 468 | mutex_unlock(&lli->lli_readdir_mutex); |
| 469 | ldlm_lock_decref(&lockh, mode); |
| 470 | return page; |
| 471 | |
| 472 | fail: |
| 473 | ll_release_page(page, 1); |
| 474 | page = ERR_PTR(-EIO); |
| 475 | goto out_unlock; |
| 476 | } |
| 477 | |
| 478 | int ll_dir_read(struct inode *inode, struct dir_context *ctx) |
| 479 | { |
| 480 | struct ll_inode_info *info = ll_i2info(inode); |
| 481 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
| 482 | __u64 pos = ctx->pos; |
| 483 | int api32 = ll_need_32bit_api(sbi); |
| 484 | int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH; |
| 485 | struct page *page; |
| 486 | struct ll_dir_chain chain; |
| 487 | int done = 0; |
| 488 | int rc = 0; |
| 489 | |
| 490 | ll_dir_chain_init(&chain); |
| 491 | |
| 492 | page = ll_get_dir_page(inode, pos, &chain); |
| 493 | |
| 494 | while (rc == 0 && !done) { |
| 495 | struct lu_dirpage *dp; |
| 496 | struct lu_dirent *ent; |
| 497 | |
| 498 | if (!IS_ERR(page)) { |
| 499 | /* |
| 500 | * If page is empty (end of directory is reached), |
| 501 | * use this value. |
| 502 | */ |
| 503 | __u64 hash = MDS_DIR_END_OFF; |
| 504 | __u64 next; |
| 505 | |
| 506 | dp = page_address(page); |
| 507 | for (ent = lu_dirent_start(dp); ent != NULL && !done; |
| 508 | ent = lu_dirent_next(ent)) { |
| 509 | __u16 type; |
| 510 | int namelen; |
| 511 | struct lu_fid fid; |
| 512 | __u64 lhash; |
| 513 | __u64 ino; |
| 514 | |
| 515 | /* |
| 516 | * XXX: implement correct swabbing here. |
| 517 | */ |
| 518 | |
| 519 | hash = le64_to_cpu(ent->lde_hash); |
| 520 | if (hash < pos) |
| 521 | /* |
| 522 | * Skip until we find target hash |
| 523 | * value. |
| 524 | */ |
| 525 | continue; |
| 526 | |
| 527 | namelen = le16_to_cpu(ent->lde_namelen); |
| 528 | if (namelen == 0) |
| 529 | /* |
| 530 | * Skip dummy record. |
| 531 | */ |
| 532 | continue; |
| 533 | |
| 534 | if (api32 && hash64) |
| 535 | lhash = hash >> 32; |
| 536 | else |
| 537 | lhash = hash; |
| 538 | fid_le_to_cpu(&fid, &ent->lde_fid); |
| 539 | ino = cl_fid_build_ino(&fid, api32); |
| 540 | type = ll_dirent_type_get(ent); |
| 541 | ctx->pos = lhash; |
| 542 | /* For 'll_nfs_get_name_filldir()', it will try |
| 543 | * to access the 'ent' through its 'lde_name', |
| 544 | * so the parameter 'name' for 'ctx->actor()' |
| 545 | * must be part of the 'ent'. |
| 546 | */ |
| 547 | done = !dir_emit(ctx, ent->lde_name, |
| 548 | namelen, ino, type); |
| 549 | } |
| 550 | next = le64_to_cpu(dp->ldp_hash_end); |
| 551 | if (!done) { |
| 552 | pos = next; |
| 553 | if (pos == MDS_DIR_END_OFF) { |
| 554 | /* |
| 555 | * End of directory reached. |
| 556 | */ |
| 557 | done = 1; |
| 558 | ll_release_page(page, 0); |
| 559 | } else if (1 /* chain is exhausted*/) { |
| 560 | /* |
| 561 | * Normal case: continue to the next |
| 562 | * page. |
| 563 | */ |
| 564 | ll_release_page(page, |
| 565 | le32_to_cpu(dp->ldp_flags) & |
| 566 | LDF_COLLIDE); |
| 567 | next = pos; |
| 568 | page = ll_get_dir_page(inode, pos, |
| 569 | &chain); |
| 570 | } else { |
| 571 | /* |
| 572 | * go into overflow page. |
| 573 | */ |
| 574 | LASSERT(le32_to_cpu(dp->ldp_flags) & |
| 575 | LDF_COLLIDE); |
| 576 | ll_release_page(page, 1); |
| 577 | } |
| 578 | } else { |
| 579 | pos = hash; |
| 580 | ll_release_page(page, 0); |
| 581 | } |
| 582 | } else { |
| 583 | rc = PTR_ERR(page); |
| 584 | CERROR("error reading dir "DFID" at %lu: rc %d\n", |
| 585 | PFID(&info->lli_fid), (unsigned long)pos, rc); |
| 586 | } |
| 587 | } |
| 588 | |
| 589 | ctx->pos = pos; |
| 590 | ll_dir_chain_fini(&chain); |
| 591 | return rc; |
| 592 | } |
| 593 | |
| 594 | static int ll_readdir(struct file *filp, struct dir_context *ctx) |
| 595 | { |
| 596 | struct inode *inode = filp->f_dentry->d_inode; |
| 597 | struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp); |
| 598 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
| 599 | int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH; |
| 600 | int api32 = ll_need_32bit_api(sbi); |
| 601 | int rc; |
| 602 | |
| 603 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu " |
| 604 | " 32bit_api %d\n", inode->i_ino, inode->i_generation, |
| 605 | inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32); |
| 606 | |
| 607 | if (lfd->lfd_pos == MDS_DIR_END_OFF) { |
| 608 | /* |
| 609 | * end-of-file. |
| 610 | */ |
| 611 | rc = 0; |
| 612 | goto out; |
| 613 | } |
| 614 | |
| 615 | ctx->pos = lfd->lfd_pos; |
| 616 | rc = ll_dir_read(inode, ctx); |
| 617 | lfd->lfd_pos = ctx->pos; |
| 618 | if (ctx->pos == MDS_DIR_END_OFF) { |
| 619 | if (api32) |
| 620 | ctx->pos = LL_DIR_END_OFF_32BIT; |
| 621 | else |
| 622 | ctx->pos = LL_DIR_END_OFF; |
| 623 | } else { |
| 624 | if (api32 && hash64) |
| 625 | ctx->pos >>= 32; |
| 626 | } |
| 627 | filp->f_version = inode->i_version; |
| 628 | |
| 629 | out: |
| 630 | if (!rc) |
| 631 | ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1); |
| 632 | |
| 633 | return rc; |
| 634 | } |
| 635 | |
| 636 | static int ll_send_mgc_param(struct obd_export *mgc, char *string) |
| 637 | { |
| 638 | struct mgs_send_param *msp; |
| 639 | int rc = 0; |
| 640 | |
| 641 | msp = kzalloc(sizeof(*msp), GFP_NOFS); |
| 642 | if (!msp) |
| 643 | return -ENOMEM; |
| 644 | |
| 645 | strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN); |
| 646 | rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO, |
| 647 | sizeof(struct mgs_send_param), msp, NULL); |
| 648 | if (rc) |
| 649 | CERROR("Failed to set parameter: %d\n", rc); |
| 650 | OBD_FREE_PTR(msp); |
| 651 | |
| 652 | return rc; |
| 653 | } |
| 654 | |
| 655 | int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump, |
| 656 | char *filename) |
| 657 | { |
| 658 | struct ptlrpc_request *request = NULL; |
| 659 | struct md_op_data *op_data; |
| 660 | struct ll_sb_info *sbi = ll_i2sbi(dir); |
| 661 | int mode; |
| 662 | int err; |
| 663 | |
| 664 | mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR; |
| 665 | op_data = ll_prep_md_op_data(NULL, dir, NULL, filename, |
| 666 | strlen(filename), mode, LUSTRE_OPC_MKDIR, |
| 667 | lump); |
| 668 | if (IS_ERR(op_data)) { |
| 669 | err = PTR_ERR(op_data); |
| 670 | goto err_exit; |
| 671 | } |
| 672 | |
| 673 | op_data->op_cli_flags |= CLI_SET_MEA; |
| 674 | err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode, |
| 675 | from_kuid(&init_user_ns, current_fsuid()), |
| 676 | from_kgid(&init_user_ns, current_fsgid()), |
| 677 | cfs_curproc_cap_pack(), 0, &request); |
| 678 | ll_finish_md_op_data(op_data); |
| 679 | if (err) |
| 680 | goto err_exit; |
| 681 | err_exit: |
| 682 | ptlrpc_req_finished(request); |
| 683 | return err; |
| 684 | } |
| 685 | |
| 686 | int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump, |
| 687 | int set_default) |
| 688 | { |
| 689 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
| 690 | struct md_op_data *op_data; |
| 691 | struct ptlrpc_request *req = NULL; |
| 692 | int rc = 0; |
| 693 | struct lustre_sb_info *lsi = s2lsi(inode->i_sb); |
| 694 | struct obd_device *mgc = lsi->lsi_mgc; |
| 695 | int lum_size; |
| 696 | |
| 697 | if (lump != NULL) { |
| 698 | /* |
| 699 | * This is coming from userspace, so should be in |
| 700 | * local endian. But the MDS would like it in little |
| 701 | * endian, so we swab it before we send it. |
| 702 | */ |
| 703 | switch (lump->lmm_magic) { |
| 704 | case LOV_USER_MAGIC_V1: { |
| 705 | if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1)) |
| 706 | lustre_swab_lov_user_md_v1(lump); |
| 707 | lum_size = sizeof(struct lov_user_md_v1); |
| 708 | break; |
| 709 | } |
| 710 | case LOV_USER_MAGIC_V3: { |
| 711 | if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3)) |
| 712 | lustre_swab_lov_user_md_v3( |
| 713 | (struct lov_user_md_v3 *)lump); |
| 714 | lum_size = sizeof(struct lov_user_md_v3); |
| 715 | break; |
| 716 | } |
| 717 | default: { |
| 718 | CDEBUG(D_IOCTL, "bad userland LOV MAGIC:" |
| 719 | " %#08x != %#08x nor %#08x\n", |
| 720 | lump->lmm_magic, LOV_USER_MAGIC_V1, |
| 721 | LOV_USER_MAGIC_V3); |
| 722 | return -EINVAL; |
| 723 | } |
| 724 | } |
| 725 | } else { |
| 726 | lum_size = sizeof(struct lov_user_md_v1); |
| 727 | } |
| 728 | |
| 729 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, |
| 730 | LUSTRE_OPC_ANY, NULL); |
| 731 | if (IS_ERR(op_data)) |
| 732 | return PTR_ERR(op_data); |
| 733 | |
| 734 | if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC)) |
| 735 | op_data->op_cli_flags |= CLI_SET_MEA; |
| 736 | |
| 737 | /* swabbing is done in lov_setstripe() on server side */ |
| 738 | rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, |
| 739 | NULL, 0, &req, NULL); |
| 740 | ll_finish_md_op_data(op_data); |
| 741 | ptlrpc_req_finished(req); |
| 742 | if (rc) { |
| 743 | if (rc != -EPERM && rc != -EACCES) |
| 744 | CERROR("mdc_setattr fails: rc = %d\n", rc); |
| 745 | } |
| 746 | |
| 747 | /* In the following we use the fact that LOV_USER_MAGIC_V1 and |
| 748 | LOV_USER_MAGIC_V3 have the same initial fields so we do not |
| 749 | need to make the distinction between the 2 versions */ |
| 750 | if (set_default && mgc->u.cli.cl_mgc_mgsexp) { |
| 751 | char *param = NULL; |
| 752 | char *buf; |
| 753 | |
| 754 | param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS); |
| 755 | if (!param) { |
| 756 | rc = -ENOMEM; |
| 757 | goto end; |
| 758 | } |
| 759 | |
| 760 | buf = param; |
| 761 | /* Get fsname and assume devname to be -MDT0000. */ |
| 762 | ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN); |
| 763 | strcat(buf, "-MDT0000.lov"); |
| 764 | buf += strlen(buf); |
| 765 | |
| 766 | /* Set root stripesize */ |
| 767 | sprintf(buf, ".stripesize=%u", |
| 768 | lump ? le32_to_cpu(lump->lmm_stripe_size) : 0); |
| 769 | rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param); |
| 770 | if (rc) |
| 771 | goto end; |
| 772 | |
| 773 | /* Set root stripecount */ |
| 774 | sprintf(buf, ".stripecount=%hd", |
| 775 | lump ? le16_to_cpu(lump->lmm_stripe_count) : 0); |
| 776 | rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param); |
| 777 | if (rc) |
| 778 | goto end; |
| 779 | |
| 780 | /* Set root stripeoffset */ |
| 781 | sprintf(buf, ".stripeoffset=%hd", |
| 782 | lump ? le16_to_cpu(lump->lmm_stripe_offset) : |
| 783 | (typeof(lump->lmm_stripe_offset))(-1)); |
| 784 | rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param); |
| 785 | |
| 786 | end: |
| 787 | if (param != NULL) |
| 788 | OBD_FREE(param, MGS_PARAM_MAXLEN); |
| 789 | } |
| 790 | return rc; |
| 791 | } |
| 792 | |
| 793 | int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp, |
| 794 | int *lmm_size, struct ptlrpc_request **request) |
| 795 | { |
| 796 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
| 797 | struct mdt_body *body; |
| 798 | struct lov_mds_md *lmm = NULL; |
| 799 | struct ptlrpc_request *req = NULL; |
| 800 | int rc, lmmsize; |
| 801 | struct md_op_data *op_data; |
| 802 | |
| 803 | rc = ll_get_default_mdsize(sbi, &lmmsize); |
| 804 | if (rc) |
| 805 | return rc; |
| 806 | |
| 807 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, |
| 808 | 0, lmmsize, LUSTRE_OPC_ANY, |
| 809 | NULL); |
| 810 | if (IS_ERR(op_data)) |
| 811 | return PTR_ERR(op_data); |
| 812 | |
| 813 | op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; |
| 814 | rc = md_getattr(sbi->ll_md_exp, op_data, &req); |
| 815 | ll_finish_md_op_data(op_data); |
| 816 | if (rc < 0) { |
| 817 | CDEBUG(D_INFO, "md_getattr failed on inode " |
| 818 | "%lu/%u: rc %d\n", inode->i_ino, |
| 819 | inode->i_generation, rc); |
| 820 | goto out; |
| 821 | } |
| 822 | |
| 823 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); |
| 824 | LASSERT(body != NULL); |
| 825 | |
| 826 | lmmsize = body->eadatasize; |
| 827 | |
| 828 | if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) || |
| 829 | lmmsize == 0) { |
| 830 | rc = -ENODATA; |
| 831 | goto out; |
| 832 | } |
| 833 | |
| 834 | lmm = req_capsule_server_sized_get(&req->rq_pill, |
| 835 | &RMF_MDT_MD, lmmsize); |
| 836 | LASSERT(lmm != NULL); |
| 837 | |
| 838 | /* |
| 839 | * This is coming from the MDS, so is probably in |
| 840 | * little endian. We convert it to host endian before |
| 841 | * passing it to userspace. |
| 842 | */ |
| 843 | /* We don't swab objects for directories */ |
| 844 | switch (le32_to_cpu(lmm->lmm_magic)) { |
| 845 | case LOV_MAGIC_V1: |
| 846 | if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) |
| 847 | lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); |
| 848 | break; |
| 849 | case LOV_MAGIC_V3: |
| 850 | if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) |
| 851 | lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); |
| 852 | break; |
| 853 | default: |
| 854 | CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic); |
| 855 | rc = -EPROTO; |
| 856 | } |
| 857 | out: |
| 858 | *lmmp = lmm; |
| 859 | *lmm_size = lmmsize; |
| 860 | *request = req; |
| 861 | return rc; |
| 862 | } |
| 863 | |
| 864 | /* |
| 865 | * Get MDT index for the inode. |
| 866 | */ |
| 867 | int ll_get_mdt_idx(struct inode *inode) |
| 868 | { |
| 869 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
| 870 | struct md_op_data *op_data; |
| 871 | int rc, mdtidx; |
| 872 | |
| 873 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, |
| 874 | 0, LUSTRE_OPC_ANY, NULL); |
| 875 | if (IS_ERR(op_data)) |
| 876 | return PTR_ERR(op_data); |
| 877 | |
| 878 | op_data->op_flags |= MF_GET_MDT_IDX; |
| 879 | rc = md_getattr(sbi->ll_md_exp, op_data, NULL); |
| 880 | mdtidx = op_data->op_mds; |
| 881 | ll_finish_md_op_data(op_data); |
| 882 | if (rc < 0) { |
| 883 | CDEBUG(D_INFO, "md_getattr_name: %d\n", rc); |
| 884 | return rc; |
| 885 | } |
| 886 | return mdtidx; |
| 887 | } |
| 888 | |
| 889 | /** |
| 890 | * Generic handler to do any pre-copy work. |
| 891 | * |
| 892 | * It send a first hsm_progress (with extent length == 0) to coordinator as a |
| 893 | * first information for it that real work has started. |
| 894 | * |
| 895 | * Moreover, for a ARCHIVE request, it will sample the file data version and |
| 896 | * store it in \a copy. |
| 897 | * |
| 898 | * \return 0 on success. |
| 899 | */ |
| 900 | static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy) |
| 901 | { |
| 902 | struct ll_sb_info *sbi = ll_s2sbi(sb); |
| 903 | struct hsm_progress_kernel hpk; |
| 904 | int rc; |
| 905 | |
| 906 | /* Forge a hsm_progress based on data from copy. */ |
| 907 | hpk.hpk_fid = copy->hc_hai.hai_fid; |
| 908 | hpk.hpk_cookie = copy->hc_hai.hai_cookie; |
| 909 | hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset; |
| 910 | hpk.hpk_extent.length = 0; |
| 911 | hpk.hpk_flags = 0; |
| 912 | hpk.hpk_errval = 0; |
| 913 | hpk.hpk_data_version = 0; |
| 914 | |
| 915 | |
| 916 | /* For archive request, we need to read the current file version. */ |
| 917 | if (copy->hc_hai.hai_action == HSMA_ARCHIVE) { |
| 918 | struct inode *inode; |
| 919 | __u64 data_version = 0; |
| 920 | |
| 921 | /* Get inode for this fid */ |
| 922 | inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid); |
| 923 | if (IS_ERR(inode)) { |
| 924 | hpk.hpk_flags |= HP_FLAG_RETRY; |
| 925 | /* hpk_errval is >= 0 */ |
| 926 | hpk.hpk_errval = -PTR_ERR(inode); |
| 927 | rc = PTR_ERR(inode); |
| 928 | goto progress; |
| 929 | } |
| 930 | |
| 931 | /* Read current file data version */ |
| 932 | rc = ll_data_version(inode, &data_version, 1); |
| 933 | iput(inode); |
| 934 | if (rc != 0) { |
| 935 | CDEBUG(D_HSM, "Could not read file data version of " |
| 936 | DFID" (rc = %d). Archive request (%#llx) could not be done.\n", |
| 937 | PFID(©->hc_hai.hai_fid), rc, |
| 938 | copy->hc_hai.hai_cookie); |
| 939 | hpk.hpk_flags |= HP_FLAG_RETRY; |
| 940 | /* hpk_errval must be >= 0 */ |
| 941 | hpk.hpk_errval = -rc; |
| 942 | goto progress; |
| 943 | } |
| 944 | |
| 945 | /* Store it the hsm_copy for later copytool use. |
| 946 | * Always modified even if no lsm. */ |
| 947 | copy->hc_data_version = data_version; |
| 948 | } |
| 949 | |
| 950 | progress: |
| 951 | rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk), |
| 952 | &hpk, NULL); |
| 953 | |
| 954 | return rc; |
| 955 | } |
| 956 | |
| 957 | /** |
| 958 | * Generic handler to do any post-copy work. |
| 959 | * |
| 960 | * It will send the last hsm_progress update to coordinator to inform it |
| 961 | * that copy is finished and whether it was successful or not. |
| 962 | * |
| 963 | * Moreover, |
| 964 | * - for ARCHIVE request, it will sample the file data version and compare it |
| 965 | * with the version saved in ll_ioc_copy_start(). If they do not match, copy |
| 966 | * will be considered as failed. |
| 967 | * - for RESTORE request, it will sample the file data version and send it to |
| 968 | * coordinator which is useful if the file was imported as 'released'. |
| 969 | * |
| 970 | * \return 0 on success. |
| 971 | */ |
| 972 | static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy) |
| 973 | { |
| 974 | struct ll_sb_info *sbi = ll_s2sbi(sb); |
| 975 | struct hsm_progress_kernel hpk; |
| 976 | int rc; |
| 977 | |
| 978 | /* If you modify the logic here, also check llapi_hsm_copy_end(). */ |
| 979 | /* Take care: copy->hc_hai.hai_action, len, gid and data are not |
| 980 | * initialized if copy_end was called with copy == NULL. |
| 981 | */ |
| 982 | |
| 983 | /* Forge a hsm_progress based on data from copy. */ |
| 984 | hpk.hpk_fid = copy->hc_hai.hai_fid; |
| 985 | hpk.hpk_cookie = copy->hc_hai.hai_cookie; |
| 986 | hpk.hpk_extent = copy->hc_hai.hai_extent; |
| 987 | hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED; |
| 988 | hpk.hpk_errval = copy->hc_errval; |
| 989 | hpk.hpk_data_version = 0; |
| 990 | |
| 991 | /* For archive request, we need to check the file data was not changed. |
| 992 | * |
| 993 | * For restore request, we need to send the file data version, this is |
| 994 | * useful when the file was created using hsm_import. |
| 995 | */ |
| 996 | if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) || |
| 997 | (copy->hc_hai.hai_action == HSMA_RESTORE)) && |
| 998 | (copy->hc_errval == 0)) { |
| 999 | struct inode *inode; |
| 1000 | __u64 data_version = 0; |
| 1001 | |
| 1002 | /* Get lsm for this fid */ |
| 1003 | inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid); |
| 1004 | if (IS_ERR(inode)) { |
| 1005 | hpk.hpk_flags |= HP_FLAG_RETRY; |
| 1006 | /* hpk_errval must be >= 0 */ |
| 1007 | hpk.hpk_errval = -PTR_ERR(inode); |
| 1008 | rc = PTR_ERR(inode); |
| 1009 | goto progress; |
| 1010 | } |
| 1011 | |
| 1012 | rc = ll_data_version(inode, &data_version, |
| 1013 | copy->hc_hai.hai_action == HSMA_ARCHIVE); |
| 1014 | iput(inode); |
| 1015 | if (rc) { |
| 1016 | CDEBUG(D_HSM, "Could not read file data version. " |
| 1017 | "Request could not be confirmed.\n"); |
| 1018 | if (hpk.hpk_errval == 0) |
| 1019 | hpk.hpk_errval = -rc; |
| 1020 | goto progress; |
| 1021 | } |
| 1022 | |
| 1023 | /* Store it the hsm_copy for later copytool use. |
| 1024 | * Always modified even if no lsm. */ |
| 1025 | hpk.hpk_data_version = data_version; |
| 1026 | |
| 1027 | /* File could have been stripped during archiving, so we need |
| 1028 | * to check anyway. */ |
| 1029 | if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) && |
| 1030 | (copy->hc_data_version != data_version)) { |
| 1031 | CDEBUG(D_HSM, "File data version mismatched. " |
| 1032 | "File content was changed during archiving. " |
| 1033 | DFID", start:%#llx current:%#llx\n", |
| 1034 | PFID(©->hc_hai.hai_fid), |
| 1035 | copy->hc_data_version, data_version); |
| 1036 | /* File was changed, send error to cdt. Do not ask for |
| 1037 | * retry because if a file is modified frequently, |
| 1038 | * the cdt will loop on retried archive requests. |
| 1039 | * The policy engine will ask for a new archive later |
| 1040 | * when the file will not be modified for some tunable |
| 1041 | * time */ |
| 1042 | /* we do not notify caller */ |
| 1043 | hpk.hpk_flags &= ~HP_FLAG_RETRY; |
| 1044 | /* hpk_errval must be >= 0 */ |
| 1045 | hpk.hpk_errval = EBUSY; |
| 1046 | } |
| 1047 | |
| 1048 | } |
| 1049 | |
| 1050 | progress: |
| 1051 | rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk), |
| 1052 | &hpk, NULL); |
| 1053 | |
| 1054 | return rc; |
| 1055 | } |
| 1056 | |
| 1057 | |
| 1058 | static int copy_and_ioctl(int cmd, struct obd_export *exp, |
| 1059 | const void __user *data, size_t size) |
| 1060 | { |
| 1061 | void *copy; |
| 1062 | int rc; |
| 1063 | |
| 1064 | copy = kzalloc(size, GFP_NOFS); |
| 1065 | if (!copy) |
| 1066 | return -ENOMEM; |
| 1067 | |
| 1068 | if (copy_from_user(copy, data, size)) { |
| 1069 | rc = -EFAULT; |
| 1070 | goto out; |
| 1071 | } |
| 1072 | |
| 1073 | rc = obd_iocontrol(cmd, exp, size, copy, NULL); |
| 1074 | out: |
| 1075 | OBD_FREE(copy, size); |
| 1076 | |
| 1077 | return rc; |
| 1078 | } |
| 1079 | |
| 1080 | static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl) |
| 1081 | { |
| 1082 | int cmd = qctl->qc_cmd; |
| 1083 | int type = qctl->qc_type; |
| 1084 | int id = qctl->qc_id; |
| 1085 | int valid = qctl->qc_valid; |
| 1086 | int rc = 0; |
| 1087 | |
| 1088 | switch (cmd) { |
| 1089 | case LUSTRE_Q_INVALIDATE: |
| 1090 | case LUSTRE_Q_FINVALIDATE: |
| 1091 | case Q_QUOTAON: |
| 1092 | case Q_QUOTAOFF: |
| 1093 | case Q_SETQUOTA: |
| 1094 | case Q_SETINFO: |
| 1095 | if (!capable(CFS_CAP_SYS_ADMIN) || |
| 1096 | sbi->ll_flags & LL_SBI_RMT_CLIENT) |
| 1097 | return -EPERM; |
| 1098 | break; |
| 1099 | case Q_GETQUOTA: |
| 1100 | if (((type == USRQUOTA && |
| 1101 | !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) || |
| 1102 | (type == GRPQUOTA && |
| 1103 | !in_egroup_p(make_kgid(&init_user_ns, id)))) && |
| 1104 | (!capable(CFS_CAP_SYS_ADMIN) || |
| 1105 | sbi->ll_flags & LL_SBI_RMT_CLIENT)) |
| 1106 | return -EPERM; |
| 1107 | break; |
| 1108 | case Q_GETINFO: |
| 1109 | break; |
| 1110 | default: |
| 1111 | CERROR("unsupported quotactl op: %#x\n", cmd); |
| 1112 | return -ENOTTY; |
| 1113 | } |
| 1114 | |
| 1115 | if (valid != QC_GENERAL) { |
| 1116 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) |
| 1117 | return -EOPNOTSUPP; |
| 1118 | |
| 1119 | if (cmd == Q_GETINFO) |
| 1120 | qctl->qc_cmd = Q_GETOINFO; |
| 1121 | else if (cmd == Q_GETQUOTA) |
| 1122 | qctl->qc_cmd = Q_GETOQUOTA; |
| 1123 | else |
| 1124 | return -EINVAL; |
| 1125 | |
| 1126 | switch (valid) { |
| 1127 | case QC_MDTIDX: |
| 1128 | rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp, |
| 1129 | sizeof(*qctl), qctl, NULL); |
| 1130 | break; |
| 1131 | case QC_OSTIDX: |
| 1132 | rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp, |
| 1133 | sizeof(*qctl), qctl, NULL); |
| 1134 | break; |
| 1135 | case QC_UUID: |
| 1136 | rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp, |
| 1137 | sizeof(*qctl), qctl, NULL); |
| 1138 | if (rc == -EAGAIN) |
| 1139 | rc = obd_iocontrol(OBD_IOC_QUOTACTL, |
| 1140 | sbi->ll_dt_exp, |
| 1141 | sizeof(*qctl), qctl, NULL); |
| 1142 | break; |
| 1143 | default: |
| 1144 | rc = -EINVAL; |
| 1145 | break; |
| 1146 | } |
| 1147 | |
| 1148 | if (rc) |
| 1149 | return rc; |
| 1150 | |
| 1151 | qctl->qc_cmd = cmd; |
| 1152 | } else { |
| 1153 | struct obd_quotactl *oqctl; |
| 1154 | |
| 1155 | oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS); |
| 1156 | if (!oqctl) |
| 1157 | return -ENOMEM; |
| 1158 | |
| 1159 | QCTL_COPY(oqctl, qctl); |
| 1160 | rc = obd_quotactl(sbi->ll_md_exp, oqctl); |
| 1161 | if (rc) { |
| 1162 | if (rc != -EALREADY && cmd == Q_QUOTAON) { |
| 1163 | oqctl->qc_cmd = Q_QUOTAOFF; |
| 1164 | obd_quotactl(sbi->ll_md_exp, oqctl); |
| 1165 | } |
| 1166 | OBD_FREE_PTR(oqctl); |
| 1167 | return rc; |
| 1168 | } |
| 1169 | /* If QIF_SPACE is not set, client should collect the |
| 1170 | * space usage from OSSs by itself */ |
| 1171 | if (cmd == Q_GETQUOTA && |
| 1172 | !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) && |
| 1173 | !oqctl->qc_dqblk.dqb_curspace) { |
| 1174 | struct obd_quotactl *oqctl_tmp; |
| 1175 | |
| 1176 | oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS); |
| 1177 | if (!oqctl_tmp) { |
| 1178 | rc = -ENOMEM; |
| 1179 | goto out; |
| 1180 | } |
| 1181 | |
| 1182 | oqctl_tmp->qc_cmd = Q_GETOQUOTA; |
| 1183 | oqctl_tmp->qc_id = oqctl->qc_id; |
| 1184 | oqctl_tmp->qc_type = oqctl->qc_type; |
| 1185 | |
| 1186 | /* collect space usage from OSTs */ |
| 1187 | oqctl_tmp->qc_dqblk.dqb_curspace = 0; |
| 1188 | rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp); |
| 1189 | if (!rc || rc == -EREMOTEIO) { |
| 1190 | oqctl->qc_dqblk.dqb_curspace = |
| 1191 | oqctl_tmp->qc_dqblk.dqb_curspace; |
| 1192 | oqctl->qc_dqblk.dqb_valid |= QIF_SPACE; |
| 1193 | } |
| 1194 | |
| 1195 | /* collect space & inode usage from MDTs */ |
| 1196 | oqctl_tmp->qc_dqblk.dqb_curspace = 0; |
| 1197 | oqctl_tmp->qc_dqblk.dqb_curinodes = 0; |
| 1198 | rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp); |
| 1199 | if (!rc || rc == -EREMOTEIO) { |
| 1200 | oqctl->qc_dqblk.dqb_curspace += |
| 1201 | oqctl_tmp->qc_dqblk.dqb_curspace; |
| 1202 | oqctl->qc_dqblk.dqb_curinodes = |
| 1203 | oqctl_tmp->qc_dqblk.dqb_curinodes; |
| 1204 | oqctl->qc_dqblk.dqb_valid |= QIF_INODES; |
| 1205 | } else { |
| 1206 | oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE; |
| 1207 | } |
| 1208 | |
| 1209 | OBD_FREE_PTR(oqctl_tmp); |
| 1210 | } |
| 1211 | out: |
| 1212 | QCTL_COPY(qctl, oqctl); |
| 1213 | OBD_FREE_PTR(oqctl); |
| 1214 | } |
| 1215 | |
| 1216 | return rc; |
| 1217 | } |
| 1218 | |
| 1219 | static char * |
| 1220 | ll_getname(const char __user *filename) |
| 1221 | { |
| 1222 | int ret = 0, len; |
| 1223 | char *tmp = __getname(); |
| 1224 | |
| 1225 | if (!tmp) |
| 1226 | return ERR_PTR(-ENOMEM); |
| 1227 | |
| 1228 | len = strncpy_from_user(tmp, filename, PATH_MAX); |
| 1229 | if (len == 0) |
| 1230 | ret = -ENOENT; |
| 1231 | else if (len > PATH_MAX) |
| 1232 | ret = -ENAMETOOLONG; |
| 1233 | |
| 1234 | if (ret) { |
| 1235 | __putname(tmp); |
| 1236 | tmp = ERR_PTR(ret); |
| 1237 | } |
| 1238 | return tmp; |
| 1239 | } |
| 1240 | |
| 1241 | #define ll_putname(filename) __putname(filename) |
| 1242 | |
| 1243 | static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg) |
| 1244 | { |
| 1245 | struct inode *inode = file->f_dentry->d_inode; |
| 1246 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
| 1247 | struct obd_ioctl_data *data; |
| 1248 | int rc = 0; |
| 1249 | |
| 1250 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n", |
| 1251 | inode->i_ino, inode->i_generation, inode, cmd); |
| 1252 | |
| 1253 | /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */ |
| 1254 | if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */ |
| 1255 | return -ENOTTY; |
| 1256 | |
| 1257 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1); |
| 1258 | switch (cmd) { |
| 1259 | case FSFILT_IOC_GETFLAGS: |
| 1260 | case FSFILT_IOC_SETFLAGS: |
| 1261 | return ll_iocontrol(inode, file, cmd, arg); |
| 1262 | case FSFILT_IOC_GETVERSION_OLD: |
| 1263 | case FSFILT_IOC_GETVERSION: |
| 1264 | return put_user(inode->i_generation, (int *)arg); |
| 1265 | /* We need to special case any other ioctls we want to handle, |
| 1266 | * to send them to the MDS/OST as appropriate and to properly |
| 1267 | * network encode the arg field. |
| 1268 | case FSFILT_IOC_SETVERSION_OLD: |
| 1269 | case FSFILT_IOC_SETVERSION: |
| 1270 | */ |
| 1271 | case LL_IOC_GET_MDTIDX: { |
| 1272 | int mdtidx; |
| 1273 | |
| 1274 | mdtidx = ll_get_mdt_idx(inode); |
| 1275 | if (mdtidx < 0) |
| 1276 | return mdtidx; |
| 1277 | |
| 1278 | if (put_user((int)mdtidx, (int *)arg)) |
| 1279 | return -EFAULT; |
| 1280 | |
| 1281 | return 0; |
| 1282 | } |
| 1283 | case IOC_MDC_LOOKUP: { |
| 1284 | struct ptlrpc_request *request = NULL; |
| 1285 | int namelen, len = 0; |
| 1286 | char *buf = NULL; |
| 1287 | char *filename; |
| 1288 | struct md_op_data *op_data; |
| 1289 | |
| 1290 | rc = obd_ioctl_getdata(&buf, &len, (void *)arg); |
| 1291 | if (rc) |
| 1292 | return rc; |
| 1293 | data = (void *)buf; |
| 1294 | |
| 1295 | filename = data->ioc_inlbuf1; |
| 1296 | namelen = strlen(filename); |
| 1297 | |
| 1298 | if (namelen < 1) { |
| 1299 | CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n"); |
| 1300 | rc = -EINVAL; |
| 1301 | goto out_free; |
| 1302 | } |
| 1303 | |
| 1304 | op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen, |
| 1305 | 0, LUSTRE_OPC_ANY, NULL); |
| 1306 | if (IS_ERR(op_data)) { |
| 1307 | rc = PTR_ERR(op_data); |
| 1308 | goto out_free; |
| 1309 | } |
| 1310 | |
| 1311 | op_data->op_valid = OBD_MD_FLID; |
| 1312 | rc = md_getattr_name(sbi->ll_md_exp, op_data, &request); |
| 1313 | ll_finish_md_op_data(op_data); |
| 1314 | if (rc < 0) { |
| 1315 | CDEBUG(D_INFO, "md_getattr_name: %d\n", rc); |
| 1316 | goto out_free; |
| 1317 | } |
| 1318 | ptlrpc_req_finished(request); |
| 1319 | out_free: |
| 1320 | obd_ioctl_freedata(buf, len); |
| 1321 | return rc; |
| 1322 | } |
| 1323 | case LL_IOC_LMV_SETSTRIPE: { |
| 1324 | struct lmv_user_md *lum; |
| 1325 | char *buf = NULL; |
| 1326 | char *filename; |
| 1327 | int namelen = 0; |
| 1328 | int lumlen = 0; |
| 1329 | int len; |
| 1330 | int rc; |
| 1331 | |
| 1332 | rc = obd_ioctl_getdata(&buf, &len, (void *)arg); |
| 1333 | if (rc) |
| 1334 | return rc; |
| 1335 | |
| 1336 | data = (void *)buf; |
| 1337 | if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL || |
| 1338 | data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) { |
| 1339 | rc = -EINVAL; |
| 1340 | goto lmv_out_free; |
| 1341 | } |
| 1342 | |
| 1343 | filename = data->ioc_inlbuf1; |
| 1344 | namelen = data->ioc_inllen1; |
| 1345 | |
| 1346 | if (namelen < 1) { |
| 1347 | CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n"); |
| 1348 | rc = -EINVAL; |
| 1349 | goto lmv_out_free; |
| 1350 | } |
| 1351 | lum = (struct lmv_user_md *)data->ioc_inlbuf2; |
| 1352 | lumlen = data->ioc_inllen2; |
| 1353 | |
| 1354 | if (lum->lum_magic != LMV_USER_MAGIC || |
| 1355 | lumlen != sizeof(*lum)) { |
| 1356 | CERROR("%s: wrong lum magic %x or size %d: rc = %d\n", |
| 1357 | filename, lum->lum_magic, lumlen, -EFAULT); |
| 1358 | rc = -EINVAL; |
| 1359 | goto lmv_out_free; |
| 1360 | } |
| 1361 | |
| 1362 | /** |
| 1363 | * ll_dir_setdirstripe will be used to set dir stripe |
| 1364 | * mdc_create--->mdt_reint_create (with dirstripe) |
| 1365 | */ |
| 1366 | rc = ll_dir_setdirstripe(inode, lum, filename); |
| 1367 | lmv_out_free: |
| 1368 | obd_ioctl_freedata(buf, len); |
| 1369 | return rc; |
| 1370 | |
| 1371 | } |
| 1372 | case LL_IOC_LOV_SETSTRIPE: { |
| 1373 | struct lov_user_md_v3 lumv3; |
| 1374 | struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; |
| 1375 | struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg; |
| 1376 | struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg; |
| 1377 | |
| 1378 | int set_default = 0; |
| 1379 | |
| 1380 | LASSERT(sizeof(lumv3) == sizeof(*lumv3p)); |
| 1381 | LASSERT(sizeof(lumv3.lmm_objects[0]) == |
| 1382 | sizeof(lumv3p->lmm_objects[0])); |
| 1383 | /* first try with v1 which is smaller than v3 */ |
| 1384 | if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1))) |
| 1385 | return -EFAULT; |
| 1386 | |
| 1387 | if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { |
| 1388 | if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3))) |
| 1389 | return -EFAULT; |
| 1390 | } |
| 1391 | |
| 1392 | if (inode->i_sb->s_root == file->f_dentry) |
| 1393 | set_default = 1; |
| 1394 | |
| 1395 | /* in v1 and v3 cases lumv1 points to data */ |
| 1396 | rc = ll_dir_setstripe(inode, lumv1, set_default); |
| 1397 | |
| 1398 | return rc; |
| 1399 | } |
| 1400 | case LL_IOC_LMV_GETSTRIPE: { |
| 1401 | struct lmv_user_md *lump = (struct lmv_user_md *)arg; |
| 1402 | struct lmv_user_md lum; |
| 1403 | struct lmv_user_md *tmp; |
| 1404 | int lum_size; |
| 1405 | int rc = 0; |
| 1406 | int mdtindex; |
| 1407 | |
| 1408 | if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md))) |
| 1409 | return -EFAULT; |
| 1410 | |
| 1411 | if (lum.lum_magic != LMV_MAGIC_V1) |
| 1412 | return -EINVAL; |
| 1413 | |
| 1414 | lum_size = lmv_user_md_size(1, LMV_MAGIC_V1); |
| 1415 | tmp = kzalloc(lum_size, GFP_NOFS); |
| 1416 | if (!tmp) { |
| 1417 | rc = -ENOMEM; |
| 1418 | goto free_lmv; |
| 1419 | } |
| 1420 | |
| 1421 | *tmp = lum; |
| 1422 | tmp->lum_type = LMV_STRIPE_TYPE; |
| 1423 | tmp->lum_stripe_count = 1; |
| 1424 | mdtindex = ll_get_mdt_idx(inode); |
| 1425 | if (mdtindex < 0) { |
| 1426 | rc = -ENOMEM; |
| 1427 | goto free_lmv; |
| 1428 | } |
| 1429 | |
| 1430 | tmp->lum_stripe_offset = mdtindex; |
| 1431 | tmp->lum_objects[0].lum_mds = mdtindex; |
| 1432 | memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode), |
| 1433 | sizeof(struct lu_fid)); |
| 1434 | if (copy_to_user((void *)arg, tmp, lum_size)) { |
| 1435 | rc = -EFAULT; |
| 1436 | goto free_lmv; |
| 1437 | } |
| 1438 | free_lmv: |
| 1439 | if (tmp) |
| 1440 | OBD_FREE(tmp, lum_size); |
| 1441 | return rc; |
| 1442 | } |
| 1443 | case LL_IOC_REMOVE_ENTRY: { |
| 1444 | char *filename = NULL; |
| 1445 | int namelen = 0; |
| 1446 | int rc; |
| 1447 | |
| 1448 | /* Here is a little hack to avoid sending REINT_RMENTRY to |
| 1449 | * unsupported server, which might crash the server(LU-2730), |
| 1450 | * Because both LVB_TYPE and REINT_RMENTRY will be supported |
| 1451 | * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the |
| 1452 | * server will support REINT_RMENTRY XXX*/ |
| 1453 | if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE)) |
| 1454 | return -ENOTSUPP; |
| 1455 | |
| 1456 | filename = ll_getname((const char *)arg); |
| 1457 | if (IS_ERR(filename)) |
| 1458 | return PTR_ERR(filename); |
| 1459 | |
| 1460 | namelen = strlen(filename); |
| 1461 | if (namelen < 1) { |
| 1462 | rc = -EINVAL; |
| 1463 | goto out_rmdir; |
| 1464 | } |
| 1465 | |
| 1466 | rc = ll_rmdir_entry(inode, filename, namelen); |
| 1467 | out_rmdir: |
| 1468 | if (filename) |
| 1469 | ll_putname(filename); |
| 1470 | return rc; |
| 1471 | } |
| 1472 | case LL_IOC_LOV_SWAP_LAYOUTS: |
| 1473 | return -EPERM; |
| 1474 | case LL_IOC_OBD_STATFS: |
| 1475 | return ll_obd_statfs(inode, (void *)arg); |
| 1476 | case LL_IOC_LOV_GETSTRIPE: |
| 1477 | case LL_IOC_MDC_GETINFO: |
| 1478 | case IOC_MDC_GETFILEINFO: |
| 1479 | case IOC_MDC_GETFILESTRIPE: { |
| 1480 | struct ptlrpc_request *request = NULL; |
| 1481 | struct lov_user_md *lump; |
| 1482 | struct lov_mds_md *lmm = NULL; |
| 1483 | struct mdt_body *body; |
| 1484 | char *filename = NULL; |
| 1485 | int lmmsize; |
| 1486 | |
| 1487 | if (cmd == IOC_MDC_GETFILEINFO || |
| 1488 | cmd == IOC_MDC_GETFILESTRIPE) { |
| 1489 | filename = ll_getname((const char *)arg); |
| 1490 | if (IS_ERR(filename)) |
| 1491 | return PTR_ERR(filename); |
| 1492 | |
| 1493 | rc = ll_lov_getstripe_ea_info(inode, filename, &lmm, |
| 1494 | &lmmsize, &request); |
| 1495 | } else { |
| 1496 | rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request); |
| 1497 | } |
| 1498 | |
| 1499 | if (request) { |
| 1500 | body = req_capsule_server_get(&request->rq_pill, |
| 1501 | &RMF_MDT_BODY); |
| 1502 | LASSERT(body != NULL); |
| 1503 | } else { |
| 1504 | goto out_req; |
| 1505 | } |
| 1506 | |
| 1507 | if (rc < 0) { |
| 1508 | if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO || |
| 1509 | cmd == LL_IOC_MDC_GETINFO)) { |
| 1510 | rc = 0; |
| 1511 | goto skip_lmm; |
| 1512 | } else |
| 1513 | goto out_req; |
| 1514 | } |
| 1515 | |
| 1516 | if (cmd == IOC_MDC_GETFILESTRIPE || |
| 1517 | cmd == LL_IOC_LOV_GETSTRIPE) { |
| 1518 | lump = (struct lov_user_md *)arg; |
| 1519 | } else { |
| 1520 | struct lov_user_mds_data *lmdp; |
| 1521 | lmdp = (struct lov_user_mds_data *)arg; |
| 1522 | lump = &lmdp->lmd_lmm; |
| 1523 | } |
| 1524 | if (copy_to_user(lump, lmm, lmmsize)) { |
| 1525 | if (copy_to_user(lump, lmm, sizeof(*lump))) { |
| 1526 | rc = -EFAULT; |
| 1527 | goto out_req; |
| 1528 | } |
| 1529 | rc = -EOVERFLOW; |
| 1530 | } |
| 1531 | skip_lmm: |
| 1532 | if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) { |
| 1533 | struct lov_user_mds_data *lmdp; |
| 1534 | lstat_t st = { 0 }; |
| 1535 | |
| 1536 | st.st_dev = inode->i_sb->s_dev; |
| 1537 | st.st_mode = body->mode; |
| 1538 | st.st_nlink = body->nlink; |
| 1539 | st.st_uid = body->uid; |
| 1540 | st.st_gid = body->gid; |
| 1541 | st.st_rdev = body->rdev; |
| 1542 | st.st_size = body->size; |
| 1543 | st.st_blksize = PAGE_CACHE_SIZE; |
| 1544 | st.st_blocks = body->blocks; |
| 1545 | st.st_atime = body->atime; |
| 1546 | st.st_mtime = body->mtime; |
| 1547 | st.st_ctime = body->ctime; |
| 1548 | st.st_ino = inode->i_ino; |
| 1549 | |
| 1550 | lmdp = (struct lov_user_mds_data *)arg; |
| 1551 | if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) { |
| 1552 | rc = -EFAULT; |
| 1553 | goto out_req; |
| 1554 | } |
| 1555 | } |
| 1556 | |
| 1557 | out_req: |
| 1558 | ptlrpc_req_finished(request); |
| 1559 | if (filename) |
| 1560 | ll_putname(filename); |
| 1561 | return rc; |
| 1562 | } |
| 1563 | case IOC_LOV_GETINFO: { |
| 1564 | struct lov_user_mds_data *lumd; |
| 1565 | struct lov_stripe_md *lsm; |
| 1566 | struct lov_user_md *lum; |
| 1567 | struct lov_mds_md *lmm; |
| 1568 | int lmmsize; |
| 1569 | lstat_t st; |
| 1570 | |
| 1571 | lumd = (struct lov_user_mds_data *)arg; |
| 1572 | lum = &lumd->lmd_lmm; |
| 1573 | |
| 1574 | rc = ll_get_max_mdsize(sbi, &lmmsize); |
| 1575 | if (rc) |
| 1576 | return rc; |
| 1577 | |
| 1578 | OBD_ALLOC_LARGE(lmm, lmmsize); |
| 1579 | if (lmm == NULL) |
| 1580 | return -ENOMEM; |
| 1581 | if (copy_from_user(lmm, lum, lmmsize)) { |
| 1582 | rc = -EFAULT; |
| 1583 | goto free_lmm; |
| 1584 | } |
| 1585 | |
| 1586 | switch (lmm->lmm_magic) { |
| 1587 | case LOV_USER_MAGIC_V1: |
| 1588 | if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1)) |
| 1589 | break; |
| 1590 | /* swab objects first so that stripes num will be sane */ |
| 1591 | lustre_swab_lov_user_md_objects( |
| 1592 | ((struct lov_user_md_v1 *)lmm)->lmm_objects, |
| 1593 | ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count); |
| 1594 | lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); |
| 1595 | break; |
| 1596 | case LOV_USER_MAGIC_V3: |
| 1597 | if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3)) |
| 1598 | break; |
| 1599 | /* swab objects first so that stripes num will be sane */ |
| 1600 | lustre_swab_lov_user_md_objects( |
| 1601 | ((struct lov_user_md_v3 *)lmm)->lmm_objects, |
| 1602 | ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count); |
| 1603 | lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); |
| 1604 | break; |
| 1605 | default: |
| 1606 | rc = -EINVAL; |
| 1607 | goto free_lmm; |
| 1608 | } |
| 1609 | |
| 1610 | rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize); |
| 1611 | if (rc < 0) { |
| 1612 | rc = -ENOMEM; |
| 1613 | goto free_lmm; |
| 1614 | } |
| 1615 | |
| 1616 | /* Perform glimpse_size operation. */ |
| 1617 | memset(&st, 0, sizeof(st)); |
| 1618 | |
| 1619 | rc = ll_glimpse_ioctl(sbi, lsm, &st); |
| 1620 | if (rc) |
| 1621 | goto free_lsm; |
| 1622 | |
| 1623 | if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) { |
| 1624 | rc = -EFAULT; |
| 1625 | goto free_lsm; |
| 1626 | } |
| 1627 | |
| 1628 | free_lsm: |
| 1629 | obd_free_memmd(sbi->ll_dt_exp, &lsm); |
| 1630 | free_lmm: |
| 1631 | OBD_FREE_LARGE(lmm, lmmsize); |
| 1632 | return rc; |
| 1633 | } |
| 1634 | case OBD_IOC_LLOG_CATINFO: { |
| 1635 | return -EOPNOTSUPP; |
| 1636 | } |
| 1637 | case OBD_IOC_QUOTACHECK: { |
| 1638 | struct obd_quotactl *oqctl; |
| 1639 | int error = 0; |
| 1640 | |
| 1641 | if (!capable(CFS_CAP_SYS_ADMIN) || |
| 1642 | sbi->ll_flags & LL_SBI_RMT_CLIENT) |
| 1643 | return -EPERM; |
| 1644 | |
| 1645 | oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS); |
| 1646 | if (!oqctl) |
| 1647 | return -ENOMEM; |
| 1648 | oqctl->qc_type = arg; |
| 1649 | rc = obd_quotacheck(sbi->ll_md_exp, oqctl); |
| 1650 | if (rc < 0) { |
| 1651 | CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc); |
| 1652 | error = rc; |
| 1653 | } |
| 1654 | |
| 1655 | rc = obd_quotacheck(sbi->ll_dt_exp, oqctl); |
| 1656 | if (rc < 0) |
| 1657 | CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc); |
| 1658 | |
| 1659 | OBD_FREE_PTR(oqctl); |
| 1660 | return error ?: rc; |
| 1661 | } |
| 1662 | case OBD_IOC_POLL_QUOTACHECK: { |
| 1663 | struct if_quotacheck *check; |
| 1664 | |
| 1665 | if (!capable(CFS_CAP_SYS_ADMIN) || |
| 1666 | sbi->ll_flags & LL_SBI_RMT_CLIENT) |
| 1667 | return -EPERM; |
| 1668 | |
| 1669 | check = kzalloc(sizeof(*check), GFP_NOFS); |
| 1670 | if (!check) |
| 1671 | return -ENOMEM; |
| 1672 | |
| 1673 | rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check, |
| 1674 | NULL); |
| 1675 | if (rc) { |
| 1676 | CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc); |
| 1677 | if (copy_to_user((void *)arg, check, |
| 1678 | sizeof(*check))) |
| 1679 | CDEBUG(D_QUOTA, "copy_to_user failed\n"); |
| 1680 | goto out_poll; |
| 1681 | } |
| 1682 | |
| 1683 | rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check, |
| 1684 | NULL); |
| 1685 | if (rc) { |
| 1686 | CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc); |
| 1687 | if (copy_to_user((void *)arg, check, |
| 1688 | sizeof(*check))) |
| 1689 | CDEBUG(D_QUOTA, "copy_to_user failed\n"); |
| 1690 | goto out_poll; |
| 1691 | } |
| 1692 | out_poll: |
| 1693 | OBD_FREE_PTR(check); |
| 1694 | return rc; |
| 1695 | } |
| 1696 | case LL_IOC_QUOTACTL: { |
| 1697 | struct if_quotactl *qctl; |
| 1698 | |
| 1699 | qctl = kzalloc(sizeof(*qctl), GFP_NOFS); |
| 1700 | if (!qctl) |
| 1701 | return -ENOMEM; |
| 1702 | |
| 1703 | if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) { |
| 1704 | rc = -EFAULT; |
| 1705 | goto out_quotactl; |
| 1706 | } |
| 1707 | |
| 1708 | rc = quotactl_ioctl(sbi, qctl); |
| 1709 | |
| 1710 | if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl))) |
| 1711 | rc = -EFAULT; |
| 1712 | |
| 1713 | out_quotactl: |
| 1714 | OBD_FREE_PTR(qctl); |
| 1715 | return rc; |
| 1716 | } |
| 1717 | case OBD_IOC_GETDTNAME: |
| 1718 | case OBD_IOC_GETMDNAME: |
| 1719 | return ll_get_obd_name(inode, cmd, arg); |
| 1720 | case LL_IOC_FLUSHCTX: |
| 1721 | return ll_flush_ctx(inode); |
| 1722 | #ifdef CONFIG_FS_POSIX_ACL |
| 1723 | case LL_IOC_RMTACL: { |
| 1724 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT && |
| 1725 | inode == inode->i_sb->s_root->d_inode) { |
| 1726 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
| 1727 | |
| 1728 | LASSERT(fd != NULL); |
| 1729 | rc = rct_add(&sbi->ll_rct, current_pid(), arg); |
| 1730 | if (!rc) |
| 1731 | fd->fd_flags |= LL_FILE_RMTACL; |
| 1732 | return rc; |
| 1733 | } else |
| 1734 | return 0; |
| 1735 | } |
| 1736 | #endif |
| 1737 | case LL_IOC_GETOBDCOUNT: { |
| 1738 | int count, vallen; |
| 1739 | struct obd_export *exp; |
| 1740 | |
| 1741 | if (copy_from_user(&count, (int *)arg, sizeof(int))) |
| 1742 | return -EFAULT; |
| 1743 | |
| 1744 | /* get ost count when count is zero, get mdt count otherwise */ |
| 1745 | exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp; |
| 1746 | vallen = sizeof(count); |
| 1747 | rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT), |
| 1748 | KEY_TGT_COUNT, &vallen, &count, NULL); |
| 1749 | if (rc) { |
| 1750 | CERROR("get target count failed: %d\n", rc); |
| 1751 | return rc; |
| 1752 | } |
| 1753 | |
| 1754 | if (copy_to_user((int *)arg, &count, sizeof(int))) |
| 1755 | return -EFAULT; |
| 1756 | |
| 1757 | return 0; |
| 1758 | } |
| 1759 | case LL_IOC_PATH2FID: |
| 1760 | if (copy_to_user((void *)arg, ll_inode2fid(inode), |
| 1761 | sizeof(struct lu_fid))) |
| 1762 | return -EFAULT; |
| 1763 | return 0; |
| 1764 | case LL_IOC_GET_CONNECT_FLAGS: { |
| 1765 | return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg); |
| 1766 | } |
| 1767 | case OBD_IOC_CHANGELOG_SEND: |
| 1768 | case OBD_IOC_CHANGELOG_CLEAR: |
| 1769 | rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg, |
| 1770 | sizeof(struct ioc_changelog)); |
| 1771 | return rc; |
| 1772 | case OBD_IOC_FID2PATH: |
| 1773 | return ll_fid2path(inode, (void *)arg); |
| 1774 | case LL_IOC_HSM_REQUEST: { |
| 1775 | struct hsm_user_request *hur; |
| 1776 | ssize_t totalsize; |
| 1777 | |
| 1778 | hur = kzalloc(sizeof(*hur), GFP_NOFS); |
| 1779 | if (!hur) |
| 1780 | return -ENOMEM; |
| 1781 | |
| 1782 | /* We don't know the true size yet; copy the fixed-size part */ |
| 1783 | if (copy_from_user(hur, (void *)arg, sizeof(*hur))) { |
| 1784 | OBD_FREE_PTR(hur); |
| 1785 | return -EFAULT; |
| 1786 | } |
| 1787 | |
| 1788 | /* Compute the whole struct size */ |
| 1789 | totalsize = hur_len(hur); |
| 1790 | OBD_FREE_PTR(hur); |
| 1791 | if (totalsize < 0) |
| 1792 | return -E2BIG; |
| 1793 | |
| 1794 | /* Final size will be more than double totalsize */ |
| 1795 | if (totalsize >= MDS_MAXREQSIZE / 3) |
| 1796 | return -E2BIG; |
| 1797 | |
| 1798 | OBD_ALLOC_LARGE(hur, totalsize); |
| 1799 | if (hur == NULL) |
| 1800 | return -ENOMEM; |
| 1801 | |
| 1802 | /* Copy the whole struct */ |
| 1803 | if (copy_from_user(hur, (void *)arg, totalsize)) { |
| 1804 | OBD_FREE_LARGE(hur, totalsize); |
| 1805 | return -EFAULT; |
| 1806 | } |
| 1807 | |
| 1808 | if (hur->hur_request.hr_action == HUA_RELEASE) { |
| 1809 | const struct lu_fid *fid; |
| 1810 | struct inode *f; |
| 1811 | int i; |
| 1812 | |
| 1813 | for (i = 0; i < hur->hur_request.hr_itemcount; i++) { |
| 1814 | fid = &hur->hur_user_item[i].hui_fid; |
| 1815 | f = search_inode_for_lustre(inode->i_sb, fid); |
| 1816 | if (IS_ERR(f)) { |
| 1817 | rc = PTR_ERR(f); |
| 1818 | break; |
| 1819 | } |
| 1820 | |
| 1821 | rc = ll_hsm_release(f); |
| 1822 | iput(f); |
| 1823 | if (rc != 0) |
| 1824 | break; |
| 1825 | } |
| 1826 | } else { |
| 1827 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize, |
| 1828 | hur, NULL); |
| 1829 | } |
| 1830 | |
| 1831 | OBD_FREE_LARGE(hur, totalsize); |
| 1832 | |
| 1833 | return rc; |
| 1834 | } |
| 1835 | case LL_IOC_HSM_PROGRESS: { |
| 1836 | struct hsm_progress_kernel hpk; |
| 1837 | struct hsm_progress hp; |
| 1838 | |
| 1839 | if (copy_from_user(&hp, (void *)arg, sizeof(hp))) |
| 1840 | return -EFAULT; |
| 1841 | |
| 1842 | hpk.hpk_fid = hp.hp_fid; |
| 1843 | hpk.hpk_cookie = hp.hp_cookie; |
| 1844 | hpk.hpk_extent = hp.hp_extent; |
| 1845 | hpk.hpk_flags = hp.hp_flags; |
| 1846 | hpk.hpk_errval = hp.hp_errval; |
| 1847 | hpk.hpk_data_version = 0; |
| 1848 | |
| 1849 | /* File may not exist in Lustre; all progress |
| 1850 | * reported to Lustre root */ |
| 1851 | rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk, |
| 1852 | NULL); |
| 1853 | return rc; |
| 1854 | } |
| 1855 | case LL_IOC_HSM_CT_START: |
| 1856 | rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg, |
| 1857 | sizeof(struct lustre_kernelcomm)); |
| 1858 | return rc; |
| 1859 | |
| 1860 | case LL_IOC_HSM_COPY_START: { |
| 1861 | struct hsm_copy *copy; |
| 1862 | int rc; |
| 1863 | |
| 1864 | copy = kzalloc(sizeof(*copy), GFP_NOFS); |
| 1865 | if (!copy) |
| 1866 | return -ENOMEM; |
| 1867 | if (copy_from_user(copy, (char *)arg, sizeof(*copy))) { |
| 1868 | OBD_FREE_PTR(copy); |
| 1869 | return -EFAULT; |
| 1870 | } |
| 1871 | |
| 1872 | rc = ll_ioc_copy_start(inode->i_sb, copy); |
| 1873 | if (copy_to_user((char *)arg, copy, sizeof(*copy))) |
| 1874 | rc = -EFAULT; |
| 1875 | |
| 1876 | OBD_FREE_PTR(copy); |
| 1877 | return rc; |
| 1878 | } |
| 1879 | case LL_IOC_HSM_COPY_END: { |
| 1880 | struct hsm_copy *copy; |
| 1881 | int rc; |
| 1882 | |
| 1883 | copy = kzalloc(sizeof(*copy), GFP_NOFS); |
| 1884 | if (!copy) |
| 1885 | return -ENOMEM; |
| 1886 | if (copy_from_user(copy, (char *)arg, sizeof(*copy))) { |
| 1887 | OBD_FREE_PTR(copy); |
| 1888 | return -EFAULT; |
| 1889 | } |
| 1890 | |
| 1891 | rc = ll_ioc_copy_end(inode->i_sb, copy); |
| 1892 | if (copy_to_user((char *)arg, copy, sizeof(*copy))) |
| 1893 | rc = -EFAULT; |
| 1894 | |
| 1895 | OBD_FREE_PTR(copy); |
| 1896 | return rc; |
| 1897 | } |
| 1898 | default: |
| 1899 | return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg); |
| 1900 | } |
| 1901 | } |
| 1902 | |
| 1903 | static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin) |
| 1904 | { |
| 1905 | struct inode *inode = file->f_mapping->host; |
| 1906 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
| 1907 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
| 1908 | int api32 = ll_need_32bit_api(sbi); |
| 1909 | loff_t ret = -EINVAL; |
| 1910 | |
| 1911 | mutex_lock(&inode->i_mutex); |
| 1912 | switch (origin) { |
| 1913 | case SEEK_SET: |
| 1914 | break; |
| 1915 | case SEEK_CUR: |
| 1916 | offset += file->f_pos; |
| 1917 | break; |
| 1918 | case SEEK_END: |
| 1919 | if (offset > 0) |
| 1920 | goto out; |
| 1921 | if (api32) |
| 1922 | offset += LL_DIR_END_OFF_32BIT; |
| 1923 | else |
| 1924 | offset += LL_DIR_END_OFF; |
| 1925 | break; |
| 1926 | default: |
| 1927 | goto out; |
| 1928 | } |
| 1929 | |
| 1930 | if (offset >= 0 && |
| 1931 | ((api32 && offset <= LL_DIR_END_OFF_32BIT) || |
| 1932 | (!api32 && offset <= LL_DIR_END_OFF))) { |
| 1933 | if (offset != file->f_pos) { |
| 1934 | if ((api32 && offset == LL_DIR_END_OFF_32BIT) || |
| 1935 | (!api32 && offset == LL_DIR_END_OFF)) |
| 1936 | fd->lfd_pos = MDS_DIR_END_OFF; |
| 1937 | else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH) |
| 1938 | fd->lfd_pos = offset << 32; |
| 1939 | else |
| 1940 | fd->lfd_pos = offset; |
| 1941 | file->f_pos = offset; |
| 1942 | file->f_version = 0; |
| 1943 | } |
| 1944 | ret = offset; |
| 1945 | } |
| 1946 | goto out; |
| 1947 | |
| 1948 | out: |
| 1949 | mutex_unlock(&inode->i_mutex); |
| 1950 | return ret; |
| 1951 | } |
| 1952 | |
| 1953 | static int ll_dir_open(struct inode *inode, struct file *file) |
| 1954 | { |
| 1955 | return ll_file_open(inode, file); |
| 1956 | } |
| 1957 | |
| 1958 | static int ll_dir_release(struct inode *inode, struct file *file) |
| 1959 | { |
| 1960 | return ll_file_release(inode, file); |
| 1961 | } |
| 1962 | |
| 1963 | const struct file_operations ll_dir_operations = { |
| 1964 | .llseek = ll_dir_seek, |
| 1965 | .open = ll_dir_open, |
| 1966 | .release = ll_dir_release, |
| 1967 | .read = generic_read_dir, |
| 1968 | .iterate = ll_readdir, |
| 1969 | .unlocked_ioctl = ll_dir_ioctl, |
| 1970 | .fsync = ll_fsync, |
| 1971 | }; |