Commit | Line | Data |
---|---|---|
d7e09d03 PT |
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 | * | |
1dc563a6 | 30 | * Copyright (c) 2011, 2015, Intel Corporation. |
d7e09d03 PT |
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> | |
31982482 | 44 | #include <linux/uaccess.h> |
995c8b4a | 45 | #include <linux/buffer_head.h> /* for wait_on_buffer */ |
d7e09d03 | 46 | #include <linux/pagevec.h> |
2870cd10 | 47 | #include <linux/prefetch.h> |
d7e09d03 PT |
48 | |
49 | #define DEBUG_SUBSYSTEM S_LLITE | |
50 | ||
67a235f5 GKH |
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" | |
e2780478 | 58 | #include "../include/lustre_kernelcomm.h" |
d7e09d03 PT |
59 | #include "llite_internal.h" |
60 | ||
61 | /* | |
62 | * (new) readdir implementation overview. | |
63 | * | |
64 | * Original lustre readdir implementation cached exact copy of raw directory | |
65 | * pages on the client. These pages were indexed in client page cache by | |
66 | * logical offset in the directory file. This design, while very simple and | |
67 | * intuitive had some inherent problems: | |
68 | * | |
69 | * . it implies that byte offset to the directory entry serves as a | |
70 | * telldir(3)/seekdir(3) cookie, but that offset is not stable: in | |
71 | * ext3/htree directory entries may move due to splits, and more | |
72 | * importantly, | |
73 | * | |
74 | * . it is incompatible with the design of split directories for cmd3, | |
75 | * that assumes that names are distributed across nodes based on their | |
76 | * hash, and so readdir should be done in hash order. | |
77 | * | |
78 | * New readdir implementation does readdir in hash order, and uses hash of a | |
79 | * file name as a telldir/seekdir cookie. This led to number of complications: | |
80 | * | |
81 | * . hash is not unique, so it cannot be used to index cached directory | |
82 | * pages on the client (note, that it requires a whole pageful of hash | |
83 | * collided entries to cause two pages to have identical hashes); | |
84 | * | |
85 | * . hash is not unique, so it cannot, strictly speaking, be used as an | |
86 | * entry cookie. ext3/htree has the same problem and lustre implementation | |
87 | * mimics their solution: seekdir(hash) positions directory at the first | |
88 | * entry with the given hash. | |
89 | * | |
90 | * Client side. | |
91 | * | |
92 | * 0. caching | |
93 | * | |
94 | * Client caches directory pages using hash of the first entry as an index. As | |
95 | * noted above hash is not unique, so this solution doesn't work as is: | |
96 | * special processing is needed for "page hash chains" (i.e., sequences of | |
97 | * pages filled with entries all having the same hash value). | |
98 | * | |
99 | * First, such chains have to be detected. To this end, server returns to the | |
100 | * client the hash of the first entry on the page next to one returned. When | |
101 | * client detects that this hash is the same as hash of the first entry on the | |
102 | * returned page, page hash collision has to be handled. Pages in the | |
103 | * hash chain, except first one, are termed "overflow pages". | |
104 | * | |
105 | * Solution to index uniqueness problem is to not cache overflow | |
106 | * pages. Instead, when page hash collision is detected, all overflow pages | |
107 | * from emerging chain are immediately requested from the server and placed in | |
108 | * a special data structure (struct ll_dir_chain). This data structure is used | |
109 | * by ll_readdir() to process entries from overflow pages. When readdir | |
110 | * invocation finishes, overflow pages are discarded. If page hash collision | |
111 | * chain weren't completely processed, next call to readdir will again detect | |
112 | * page hash collision, again read overflow pages in, process next portion of | |
113 | * entries and again discard the pages. This is not as wasteful as it looks, | |
114 | * because, given reasonable hash, page hash collisions are extremely rare. | |
115 | * | |
116 | * 1. directory positioning | |
117 | * | |
118 | * When seekdir(hash) is called, original | |
119 | * | |
120 | * | |
121 | * | |
122 | * | |
123 | * | |
124 | * | |
125 | * | |
126 | * | |
127 | * Server. | |
128 | * | |
129 | * identification of and access to overflow pages | |
130 | * | |
131 | * page format | |
132 | * | |
133 | * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains | |
134 | * a header lu_dirpage which describes the start/end hash, and whether this | |
135 | * page is empty (contains no dir entry) or hash collide with next page. | |
136 | * After client receives reply, several pages will be integrated into dir page | |
137 | * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the | |
138 | * lu_dirpage for this integrated page will be adjusted. See | |
139 | * lmv_adjust_dirpages(). | |
140 | * | |
141 | */ | |
142 | ||
143 | /* returns the page unlocked, but with a reference */ | |
144 | static int ll_dir_filler(void *_hash, struct page *page0) | |
145 | { | |
146 | struct inode *inode = page0->mapping->host; | |
147 | int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH; | |
148 | struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp; | |
149 | struct ptlrpc_request *request; | |
150 | struct mdt_body *body; | |
151 | struct md_op_data *op_data; | |
152 | __u64 hash = *((__u64 *)_hash); | |
153 | struct page **page_pool; | |
154 | struct page *page; | |
155 | struct lu_dirpage *dp; | |
156 | int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT; | |
157 | int nrdpgs = 0; /* number of pages read actually */ | |
158 | int npages; | |
159 | int i; | |
160 | int rc; | |
d7e09d03 | 161 | |
b0f5aad5 | 162 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash %llu\n", |
d7e09d03 PT |
163 | inode->i_ino, inode->i_generation, inode, hash); |
164 | ||
165 | LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES); | |
166 | ||
0fa3b9d3 | 167 | page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS); |
496a51bd | 168 | if (page_pool) { |
d7e09d03 PT |
169 | page_pool[0] = page0; |
170 | } else { | |
171 | page_pool = &page0; | |
172 | max_pages = 1; | |
173 | } | |
174 | for (npages = 1; npages < max_pages; npages++) { | |
175 | page = page_cache_alloc_cold(inode->i_mapping); | |
176 | if (!page) | |
177 | break; | |
178 | page_pool[npages] = page; | |
179 | } | |
180 | ||
181 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
182 | LUSTRE_OPC_ANY, NULL); | |
183 | op_data->op_npages = npages; | |
184 | op_data->op_offset = hash; | |
185 | rc = md_readpage(exp, op_data, page_pool, &request); | |
186 | ll_finish_md_op_data(op_data); | |
c67587a7 LS |
187 | if (rc < 0) { |
188 | /* page0 is special, which was added into page cache early */ | |
189 | delete_from_page_cache(page0); | |
190 | } else if (rc == 0) { | |
d7e09d03 PT |
191 | body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY); |
192 | /* Checked by mdc_readpage() */ | |
193 | LASSERT(body != NULL); | |
194 | ||
195 | if (body->valid & OBD_MD_FLSIZE) | |
196 | cl_isize_write(inode, body->size); | |
197 | ||
198 | nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1) | |
199 | >> PAGE_CACHE_SHIFT; | |
200 | SetPageUptodate(page0); | |
201 | } | |
202 | unlock_page(page0); | |
203 | ptlrpc_req_finished(request); | |
204 | ||
205 | CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages); | |
206 | ||
d7e09d03 PT |
207 | for (i = 1; i < npages; i++) { |
208 | unsigned long offset; | |
209 | int ret; | |
210 | ||
211 | page = page_pool[i]; | |
212 | ||
213 | if (rc < 0 || i >= nrdpgs) { | |
214 | page_cache_release(page); | |
215 | continue; | |
216 | } | |
217 | ||
218 | SetPageUptodate(page); | |
219 | ||
220 | dp = kmap(page); | |
221 | hash = le64_to_cpu(dp->ldp_hash_start); | |
222 | kunmap(page); | |
223 | ||
224 | offset = hash_x_index(hash, hash64); | |
225 | ||
226 | prefetchw(&page->flags); | |
227 | ret = add_to_page_cache_lru(page, inode->i_mapping, offset, | |
063d99b4 | 228 | GFP_NOFS); |
d7e09d03 PT |
229 | if (ret == 0) { |
230 | unlock_page(page); | |
d7e09d03 | 231 | } else { |
2d00bd17 JP |
232 | CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n", |
233 | offset, ret); | |
d7e09d03 PT |
234 | } |
235 | page_cache_release(page); | |
236 | } | |
d7e09d03 PT |
237 | |
238 | if (page_pool != &page0) | |
97903a26 | 239 | kfree(page_pool); |
d7e09d03 PT |
240 | return rc; |
241 | } | |
242 | ||
d7e09d03 PT |
243 | void ll_release_page(struct page *page, int remove) |
244 | { | |
245 | kunmap(page); | |
246 | if (remove) { | |
247 | lock_page(page); | |
248 | if (likely(page->mapping != NULL)) | |
249 | truncate_complete_page(page->mapping, page); | |
250 | unlock_page(page); | |
251 | } | |
252 | page_cache_release(page); | |
253 | } | |
254 | ||
255 | /* | |
256 | * Find, kmap and return page that contains given hash. | |
257 | */ | |
258 | static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash, | |
259 | __u64 *start, __u64 *end) | |
260 | { | |
261 | int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH; | |
262 | struct address_space *mapping = dir->i_mapping; | |
263 | /* | |
264 | * Complement of hash is used as an index so that | |
265 | * radix_tree_gang_lookup() can be used to find a page with starting | |
266 | * hash _smaller_ than one we are looking for. | |
267 | */ | |
268 | unsigned long offset = hash_x_index(*hash, hash64); | |
269 | struct page *page; | |
270 | int found; | |
271 | ||
12afe493 | 272 | spin_lock_irq(&mapping->tree_lock); |
d7e09d03 PT |
273 | found = radix_tree_gang_lookup(&mapping->page_tree, |
274 | (void **)&page, offset, 1); | |
437dfb20 | 275 | if (found > 0 && !radix_tree_exceptional_entry(page)) { |
d7e09d03 PT |
276 | struct lu_dirpage *dp; |
277 | ||
278 | page_cache_get(page); | |
12afe493 | 279 | spin_unlock_irq(&mapping->tree_lock); |
d7e09d03 PT |
280 | /* |
281 | * In contrast to find_lock_page() we are sure that directory | |
282 | * page cannot be truncated (while DLM lock is held) and, | |
283 | * hence, can avoid restart. | |
284 | * | |
285 | * In fact, page cannot be locked here at all, because | |
286 | * ll_dir_filler() does synchronous io. | |
287 | */ | |
288 | wait_on_page_locked(page); | |
289 | if (PageUptodate(page)) { | |
290 | dp = kmap(page); | |
291 | if (BITS_PER_LONG == 32 && hash64) { | |
292 | *start = le64_to_cpu(dp->ldp_hash_start) >> 32; | |
293 | *end = le64_to_cpu(dp->ldp_hash_end) >> 32; | |
294 | *hash = *hash >> 32; | |
295 | } else { | |
296 | *start = le64_to_cpu(dp->ldp_hash_start); | |
297 | *end = le64_to_cpu(dp->ldp_hash_end); | |
298 | } | |
55f5a824 GKH |
299 | LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n", |
300 | *start, *end, *hash); | |
b0f5aad5 | 301 | CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n", |
d7e09d03 PT |
302 | offset, *start, *end, *hash); |
303 | if (*hash > *end) { | |
304 | ll_release_page(page, 0); | |
305 | page = NULL; | |
306 | } else if (*end != *start && *hash == *end) { | |
307 | /* | |
308 | * upon hash collision, remove this page, | |
309 | * otherwise put page reference, and | |
310 | * ll_get_dir_page() will issue RPC to fetch | |
311 | * the page we want. | |
312 | */ | |
313 | ll_release_page(page, | |
314 | le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE); | |
315 | page = NULL; | |
316 | } | |
317 | } else { | |
318 | page_cache_release(page); | |
319 | page = ERR_PTR(-EIO); | |
320 | } | |
321 | ||
322 | } else { | |
12afe493 | 323 | spin_unlock_irq(&mapping->tree_lock); |
d7e09d03 PT |
324 | page = NULL; |
325 | } | |
326 | return page; | |
327 | } | |
328 | ||
329 | struct page *ll_get_dir_page(struct inode *dir, __u64 hash, | |
330 | struct ll_dir_chain *chain) | |
331 | { | |
332 | ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} }; | |
333 | struct address_space *mapping = dir->i_mapping; | |
334 | struct lustre_handle lockh; | |
335 | struct lu_dirpage *dp; | |
336 | struct page *page; | |
337 | ldlm_mode_t mode; | |
338 | int rc; | |
339 | __u64 start = 0; | |
340 | __u64 end = 0; | |
341 | __u64 lhash = hash; | |
342 | struct ll_inode_info *lli = ll_i2info(dir); | |
343 | int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH; | |
344 | ||
345 | mode = LCK_PR; | |
346 | rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED, | |
347 | ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh); | |
348 | if (!rc) { | |
f2145eae BK |
349 | struct ldlm_enqueue_info einfo = { |
350 | .ei_type = LDLM_IBITS, | |
351 | .ei_mode = mode, | |
352 | .ei_cb_bl = ll_md_blocking_ast, | |
353 | .ei_cb_cp = ldlm_completion_ast, | |
354 | }; | |
d7e09d03 PT |
355 | struct lookup_intent it = { .it_op = IT_READDIR }; |
356 | struct ptlrpc_request *request; | |
357 | struct md_op_data *op_data; | |
358 | ||
588de43a | 359 | op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0, |
d7e09d03 PT |
360 | LUSTRE_OPC_ANY, NULL); |
361 | if (IS_ERR(op_data)) | |
362 | return (void *)op_data; | |
363 | ||
364 | rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it, | |
365 | op_data, &lockh, NULL, 0, NULL, 0); | |
366 | ||
367 | ll_finish_md_op_data(op_data); | |
368 | ||
369 | request = (struct ptlrpc_request *)it.d.lustre.it_data; | |
370 | if (request) | |
371 | ptlrpc_req_finished(request); | |
372 | if (rc < 0) { | |
b0f5aad5 | 373 | CERROR("lock enqueue: "DFID" at %llu: rc %d\n", |
d7e09d03 PT |
374 | PFID(ll_inode2fid(dir)), hash, rc); |
375 | return ERR_PTR(rc); | |
376 | } | |
377 | ||
378 | CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n", | |
379 | dir, dir->i_ino, dir->i_generation); | |
380 | md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, | |
381 | &it.d.lustre.it_lock_handle, dir, NULL); | |
382 | } else { | |
383 | /* for cross-ref object, l_ast_data of the lock may not be set, | |
384 | * we reset it here */ | |
385 | md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie, | |
386 | dir, NULL); | |
387 | } | |
388 | ldlm_lock_dump_handle(D_OTHER, &lockh); | |
389 | ||
390 | mutex_lock(&lli->lli_readdir_mutex); | |
391 | page = ll_dir_page_locate(dir, &lhash, &start, &end); | |
392 | if (IS_ERR(page)) { | |
b0f5aad5 | 393 | CERROR("dir page locate: "DFID" at %llu: rc %ld\n", |
d7e09d03 | 394 | PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page)); |
34e1f2bb | 395 | goto out_unlock; |
d7e09d03 PT |
396 | } else if (page != NULL) { |
397 | /* | |
398 | * XXX nikita: not entirely correct handling of a corner case: | |
399 | * suppose hash chain of entries with hash value HASH crosses | |
400 | * border between pages P0 and P1. First both P0 and P1 are | |
401 | * cached, seekdir() is called for some entry from the P0 part | |
402 | * of the chain. Later P0 goes out of cache. telldir(HASH) | |
403 | * happens and finds P1, as it starts with matching hash | |
404 | * value. Remaining entries from P0 part of the chain are | |
405 | * skipped. (Is that really a bug?) | |
406 | * | |
407 | * Possible solutions: 0. don't cache P1 is such case, handle | |
408 | * it as an "overflow" page. 1. invalidate all pages at | |
409 | * once. 2. use HASH|1 as an index for P1. | |
410 | */ | |
34e1f2bb | 411 | goto hash_collision; |
d7e09d03 PT |
412 | } |
413 | ||
414 | page = read_cache_page(mapping, hash_x_index(hash, hash64), | |
415 | ll_dir_filler, &lhash); | |
416 | if (IS_ERR(page)) { | |
b0f5aad5 | 417 | CERROR("read cache page: "DFID" at %llu: rc %ld\n", |
d7e09d03 | 418 | PFID(ll_inode2fid(dir)), hash, PTR_ERR(page)); |
34e1f2bb | 419 | goto out_unlock; |
d7e09d03 PT |
420 | } |
421 | ||
422 | wait_on_page_locked(page); | |
423 | (void)kmap(page); | |
424 | if (!PageUptodate(page)) { | |
b0f5aad5 | 425 | CERROR("page not updated: "DFID" at %llu: rc %d\n", |
d7e09d03 PT |
426 | PFID(ll_inode2fid(dir)), hash, -5); |
427 | goto fail; | |
428 | } | |
429 | if (!PageChecked(page)) | |
c6ef5b91 SB |
430 | /* XXX: check page format later */ |
431 | SetPageChecked(page); | |
d7e09d03 | 432 | if (PageError(page)) { |
b0f5aad5 | 433 | CERROR("page error: "DFID" at %llu: rc %d\n", |
d7e09d03 PT |
434 | PFID(ll_inode2fid(dir)), hash, -5); |
435 | goto fail; | |
436 | } | |
437 | hash_collision: | |
438 | dp = page_address(page); | |
439 | if (BITS_PER_LONG == 32 && hash64) { | |
440 | start = le64_to_cpu(dp->ldp_hash_start) >> 32; | |
441 | end = le64_to_cpu(dp->ldp_hash_end) >> 32; | |
442 | lhash = hash >> 32; | |
443 | } else { | |
444 | start = le64_to_cpu(dp->ldp_hash_start); | |
445 | end = le64_to_cpu(dp->ldp_hash_end); | |
446 | lhash = hash; | |
447 | } | |
448 | if (end == start) { | |
449 | LASSERT(start == lhash); | |
b0f5aad5 | 450 | CWARN("Page-wide hash collision: %llu\n", end); |
d7e09d03 | 451 | if (BITS_PER_LONG == 32 && hash64) |
b0f5aad5 | 452 | CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n", |
d7e09d03 PT |
453 | le64_to_cpu(dp->ldp_hash_start), |
454 | le64_to_cpu(dp->ldp_hash_end), hash); | |
455 | /* | |
456 | * Fetch whole overflow chain... | |
457 | * | |
458 | * XXX not yet. | |
459 | */ | |
460 | goto fail; | |
461 | } | |
462 | out_unlock: | |
463 | mutex_unlock(&lli->lli_readdir_mutex); | |
464 | ldlm_lock_decref(&lockh, mode); | |
465 | return page; | |
466 | ||
467 | fail: | |
468 | ll_release_page(page, 1); | |
469 | page = ERR_PTR(-EIO); | |
470 | goto out_unlock; | |
471 | } | |
472 | ||
0b09d381 | 473 | int ll_dir_read(struct inode *inode, struct dir_context *ctx) |
d7e09d03 PT |
474 | { |
475 | struct ll_inode_info *info = ll_i2info(inode); | |
476 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
0b09d381 | 477 | __u64 pos = ctx->pos; |
d7e09d03 PT |
478 | int api32 = ll_need_32bit_api(sbi); |
479 | int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH; | |
480 | struct page *page; | |
481 | struct ll_dir_chain chain; | |
482 | int done = 0; | |
483 | int rc = 0; | |
d7e09d03 PT |
484 | |
485 | ll_dir_chain_init(&chain); | |
486 | ||
487 | page = ll_get_dir_page(inode, pos, &chain); | |
488 | ||
489 | while (rc == 0 && !done) { | |
490 | struct lu_dirpage *dp; | |
491 | struct lu_dirent *ent; | |
492 | ||
493 | if (!IS_ERR(page)) { | |
494 | /* | |
495 | * If page is empty (end of directory is reached), | |
496 | * use this value. | |
497 | */ | |
498 | __u64 hash = MDS_DIR_END_OFF; | |
499 | __u64 next; | |
500 | ||
501 | dp = page_address(page); | |
502 | for (ent = lu_dirent_start(dp); ent != NULL && !done; | |
503 | ent = lu_dirent_next(ent)) { | |
504 | __u16 type; | |
505 | int namelen; | |
506 | struct lu_fid fid; | |
507 | __u64 lhash; | |
508 | __u64 ino; | |
509 | ||
510 | /* | |
511 | * XXX: implement correct swabbing here. | |
512 | */ | |
513 | ||
514 | hash = le64_to_cpu(ent->lde_hash); | |
515 | if (hash < pos) | |
516 | /* | |
517 | * Skip until we find target hash | |
518 | * value. | |
519 | */ | |
520 | continue; | |
521 | ||
522 | namelen = le16_to_cpu(ent->lde_namelen); | |
523 | if (namelen == 0) | |
524 | /* | |
525 | * Skip dummy record. | |
526 | */ | |
527 | continue; | |
528 | ||
529 | if (api32 && hash64) | |
530 | lhash = hash >> 32; | |
531 | else | |
532 | lhash = hash; | |
533 | fid_le_to_cpu(&fid, &ent->lde_fid); | |
534 | ino = cl_fid_build_ino(&fid, api32); | |
535 | type = ll_dirent_type_get(ent); | |
0b09d381 | 536 | ctx->pos = lhash; |
d7e09d03 PT |
537 | /* For 'll_nfs_get_name_filldir()', it will try |
538 | * to access the 'ent' through its 'lde_name', | |
0b09d381 PT |
539 | * so the parameter 'name' for 'ctx->actor()' |
540 | * must be part of the 'ent'. | |
541 | */ | |
542 | done = !dir_emit(ctx, ent->lde_name, | |
543 | namelen, ino, type); | |
d7e09d03 PT |
544 | } |
545 | next = le64_to_cpu(dp->ldp_hash_end); | |
546 | if (!done) { | |
547 | pos = next; | |
548 | if (pos == MDS_DIR_END_OFF) { | |
549 | /* | |
550 | * End of directory reached. | |
551 | */ | |
552 | done = 1; | |
553 | ll_release_page(page, 0); | |
554 | } else if (1 /* chain is exhausted*/) { | |
555 | /* | |
556 | * Normal case: continue to the next | |
557 | * page. | |
558 | */ | |
559 | ll_release_page(page, | |
560 | le32_to_cpu(dp->ldp_flags) & | |
561 | LDF_COLLIDE); | |
562 | next = pos; | |
563 | page = ll_get_dir_page(inode, pos, | |
564 | &chain); | |
565 | } else { | |
566 | /* | |
567 | * go into overflow page. | |
568 | */ | |
569 | LASSERT(le32_to_cpu(dp->ldp_flags) & | |
570 | LDF_COLLIDE); | |
571 | ll_release_page(page, 1); | |
572 | } | |
573 | } else { | |
574 | pos = hash; | |
575 | ll_release_page(page, 0); | |
576 | } | |
577 | } else { | |
578 | rc = PTR_ERR(page); | |
579 | CERROR("error reading dir "DFID" at %lu: rc %d\n", | |
580 | PFID(&info->lli_fid), (unsigned long)pos, rc); | |
581 | } | |
582 | } | |
583 | ||
0b09d381 | 584 | ctx->pos = pos; |
d7e09d03 | 585 | ll_dir_chain_fini(&chain); |
0a3bdb00 | 586 | return rc; |
d7e09d03 PT |
587 | } |
588 | ||
0b09d381 | 589 | static int ll_readdir(struct file *filp, struct dir_context *ctx) |
d7e09d03 | 590 | { |
2a8a3597 | 591 | struct inode *inode = file_inode(filp); |
d7e09d03 PT |
592 | struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp); |
593 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
d7e09d03 PT |
594 | int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH; |
595 | int api32 = ll_need_32bit_api(sbi); | |
596 | int rc; | |
d7e09d03 | 597 | |
2d00bd17 JP |
598 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n", |
599 | inode->i_ino, inode->i_generation, | |
0b09d381 | 600 | inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32); |
d7e09d03 | 601 | |
34e1f2bb | 602 | if (lfd->lfd_pos == MDS_DIR_END_OFF) { |
d7e09d03 PT |
603 | /* |
604 | * end-of-file. | |
605 | */ | |
34e1f2bb JL |
606 | rc = 0; |
607 | goto out; | |
608 | } | |
d7e09d03 | 609 | |
0b09d381 PT |
610 | ctx->pos = lfd->lfd_pos; |
611 | rc = ll_dir_read(inode, ctx); | |
612 | lfd->lfd_pos = ctx->pos; | |
613 | if (ctx->pos == MDS_DIR_END_OFF) { | |
d7e09d03 | 614 | if (api32) |
0b09d381 | 615 | ctx->pos = LL_DIR_END_OFF_32BIT; |
d7e09d03 | 616 | else |
0b09d381 | 617 | ctx->pos = LL_DIR_END_OFF; |
d7e09d03 PT |
618 | } else { |
619 | if (api32 && hash64) | |
0b09d381 | 620 | ctx->pos >>= 32; |
d7e09d03 PT |
621 | } |
622 | filp->f_version = inode->i_version; | |
d7e09d03 PT |
623 | |
624 | out: | |
625 | if (!rc) | |
626 | ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1); | |
627 | ||
0a3bdb00 | 628 | return rc; |
d7e09d03 PT |
629 | } |
630 | ||
2d95f10e | 631 | static int ll_send_mgc_param(struct obd_export *mgc, char *string) |
d7e09d03 PT |
632 | { |
633 | struct mgs_send_param *msp; | |
634 | int rc = 0; | |
635 | ||
496a51bd | 636 | msp = kzalloc(sizeof(*msp), GFP_NOFS); |
d7e09d03 PT |
637 | if (!msp) |
638 | return -ENOMEM; | |
639 | ||
9563fe8a | 640 | strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param)); |
d7e09d03 PT |
641 | rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO, |
642 | sizeof(struct mgs_send_param), msp, NULL); | |
643 | if (rc) | |
644 | CERROR("Failed to set parameter: %d\n", rc); | |
97903a26 | 645 | kfree(msp); |
d7e09d03 PT |
646 | |
647 | return rc; | |
648 | } | |
649 | ||
920b4f2e LC |
650 | static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump, |
651 | char *filename) | |
d7e09d03 PT |
652 | { |
653 | struct ptlrpc_request *request = NULL; | |
654 | struct md_op_data *op_data; | |
655 | struct ll_sb_info *sbi = ll_i2sbi(dir); | |
656 | int mode; | |
657 | int err; | |
658 | ||
1f6eaf83 | 659 | mode = (~current_umask() & 0755) | S_IFDIR; |
d7e09d03 PT |
660 | op_data = ll_prep_md_op_data(NULL, dir, NULL, filename, |
661 | strlen(filename), mode, LUSTRE_OPC_MKDIR, | |
662 | lump); | |
34e1f2bb JL |
663 | if (IS_ERR(op_data)) { |
664 | err = PTR_ERR(op_data); | |
665 | goto err_exit; | |
666 | } | |
d7e09d03 PT |
667 | |
668 | op_data->op_cli_flags |= CLI_SET_MEA; | |
669 | err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode, | |
4b1a25f0 PT |
670 | from_kuid(&init_user_ns, current_fsuid()), |
671 | from_kgid(&init_user_ns, current_fsgid()), | |
d7e09d03 PT |
672 | cfs_curproc_cap_pack(), 0, &request); |
673 | ll_finish_md_op_data(op_data); | |
674 | if (err) | |
34e1f2bb | 675 | goto err_exit; |
d7e09d03 PT |
676 | err_exit: |
677 | ptlrpc_req_finished(request); | |
678 | return err; | |
679 | } | |
680 | ||
681 | int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump, | |
682 | int set_default) | |
683 | { | |
684 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
685 | struct md_op_data *op_data; | |
686 | struct ptlrpc_request *req = NULL; | |
687 | int rc = 0; | |
688 | struct lustre_sb_info *lsi = s2lsi(inode->i_sb); | |
689 | struct obd_device *mgc = lsi->lsi_mgc; | |
690 | int lum_size; | |
d7e09d03 PT |
691 | |
692 | if (lump != NULL) { | |
693 | /* | |
694 | * This is coming from userspace, so should be in | |
695 | * local endian. But the MDS would like it in little | |
696 | * endian, so we swab it before we send it. | |
697 | */ | |
698 | switch (lump->lmm_magic) { | |
699 | case LOV_USER_MAGIC_V1: { | |
700 | if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1)) | |
701 | lustre_swab_lov_user_md_v1(lump); | |
702 | lum_size = sizeof(struct lov_user_md_v1); | |
703 | break; | |
704 | } | |
705 | case LOV_USER_MAGIC_V3: { | |
706 | if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3)) | |
707 | lustre_swab_lov_user_md_v3( | |
708 | (struct lov_user_md_v3 *)lump); | |
709 | lum_size = sizeof(struct lov_user_md_v3); | |
710 | break; | |
711 | } | |
712 | default: { | |
2d00bd17 JP |
713 | CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n", |
714 | lump->lmm_magic, LOV_USER_MAGIC_V1, | |
715 | LOV_USER_MAGIC_V3); | |
0a3bdb00 | 716 | return -EINVAL; |
d7e09d03 PT |
717 | } |
718 | } | |
719 | } else { | |
720 | lum_size = sizeof(struct lov_user_md_v1); | |
721 | } | |
722 | ||
723 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
724 | LUSTRE_OPC_ANY, NULL); | |
725 | if (IS_ERR(op_data)) | |
0a3bdb00 | 726 | return PTR_ERR(op_data); |
d7e09d03 PT |
727 | |
728 | if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC)) | |
729 | op_data->op_cli_flags |= CLI_SET_MEA; | |
730 | ||
731 | /* swabbing is done in lov_setstripe() on server side */ | |
732 | rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, | |
733 | NULL, 0, &req, NULL); | |
734 | ll_finish_md_op_data(op_data); | |
735 | ptlrpc_req_finished(req); | |
736 | if (rc) { | |
737 | if (rc != -EPERM && rc != -EACCES) | |
738 | CERROR("mdc_setattr fails: rc = %d\n", rc); | |
739 | } | |
740 | ||
741 | /* In the following we use the fact that LOV_USER_MAGIC_V1 and | |
742 | LOV_USER_MAGIC_V3 have the same initial fields so we do not | |
bef31c78 | 743 | need to make the distinction between the 2 versions */ |
d7e09d03 PT |
744 | if (set_default && mgc->u.cli.cl_mgc_mgsexp) { |
745 | char *param = NULL; | |
746 | char *buf; | |
747 | ||
496a51bd | 748 | param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS); |
57876fd8 JL |
749 | if (!param) |
750 | return -ENOMEM; | |
d7e09d03 PT |
751 | |
752 | buf = param; | |
753 | /* Get fsname and assume devname to be -MDT0000. */ | |
754 | ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN); | |
755 | strcat(buf, "-MDT0000.lov"); | |
756 | buf += strlen(buf); | |
757 | ||
758 | /* Set root stripesize */ | |
759 | sprintf(buf, ".stripesize=%u", | |
760 | lump ? le32_to_cpu(lump->lmm_stripe_size) : 0); | |
761 | rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param); | |
762 | if (rc) | |
34e1f2bb | 763 | goto end; |
d7e09d03 PT |
764 | |
765 | /* Set root stripecount */ | |
766 | sprintf(buf, ".stripecount=%hd", | |
767 | lump ? le16_to_cpu(lump->lmm_stripe_count) : 0); | |
768 | rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param); | |
769 | if (rc) | |
34e1f2bb | 770 | goto end; |
d7e09d03 PT |
771 | |
772 | /* Set root stripeoffset */ | |
773 | sprintf(buf, ".stripeoffset=%hd", | |
774 | lump ? le16_to_cpu(lump->lmm_stripe_offset) : | |
775 | (typeof(lump->lmm_stripe_offset))(-1)); | |
776 | rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param); | |
777 | ||
778 | end: | |
57876fd8 | 779 | kfree(param); |
d7e09d03 | 780 | } |
0a3bdb00 | 781 | return rc; |
d7e09d03 PT |
782 | } |
783 | ||
784 | int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp, | |
785 | int *lmm_size, struct ptlrpc_request **request) | |
786 | { | |
787 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
788 | struct mdt_body *body; | |
789 | struct lov_mds_md *lmm = NULL; | |
790 | struct ptlrpc_request *req = NULL; | |
791 | int rc, lmmsize; | |
792 | struct md_op_data *op_data; | |
793 | ||
44779340 | 794 | rc = ll_get_default_mdsize(sbi, &lmmsize); |
d7e09d03 | 795 | if (rc) |
0a3bdb00 | 796 | return rc; |
d7e09d03 PT |
797 | |
798 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, | |
799 | 0, lmmsize, LUSTRE_OPC_ANY, | |
800 | NULL); | |
801 | if (IS_ERR(op_data)) | |
0a3bdb00 | 802 | return PTR_ERR(op_data); |
d7e09d03 PT |
803 | |
804 | op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; | |
805 | rc = md_getattr(sbi->ll_md_exp, op_data, &req); | |
806 | ll_finish_md_op_data(op_data); | |
807 | if (rc < 0) { | |
2d00bd17 JP |
808 | CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n", |
809 | inode->i_ino, | |
d7e09d03 | 810 | inode->i_generation, rc); |
34e1f2bb | 811 | goto out; |
d7e09d03 PT |
812 | } |
813 | ||
814 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
815 | LASSERT(body != NULL); | |
816 | ||
817 | lmmsize = body->eadatasize; | |
818 | ||
819 | if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) || | |
820 | lmmsize == 0) { | |
34e1f2bb JL |
821 | rc = -ENODATA; |
822 | goto out; | |
d7e09d03 PT |
823 | } |
824 | ||
825 | lmm = req_capsule_server_sized_get(&req->rq_pill, | |
826 | &RMF_MDT_MD, lmmsize); | |
827 | LASSERT(lmm != NULL); | |
828 | ||
829 | /* | |
830 | * This is coming from the MDS, so is probably in | |
831 | * little endian. We convert it to host endian before | |
832 | * passing it to userspace. | |
833 | */ | |
834 | /* We don't swab objects for directories */ | |
835 | switch (le32_to_cpu(lmm->lmm_magic)) { | |
836 | case LOV_MAGIC_V1: | |
1f6eaf83 | 837 | if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) |
d7e09d03 PT |
838 | lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); |
839 | break; | |
840 | case LOV_MAGIC_V3: | |
1f6eaf83 | 841 | if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) |
d7e09d03 PT |
842 | lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); |
843 | break; | |
844 | default: | |
845 | CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic); | |
846 | rc = -EPROTO; | |
847 | } | |
848 | out: | |
849 | *lmmp = lmm; | |
850 | *lmm_size = lmmsize; | |
851 | *request = req; | |
852 | return rc; | |
853 | } | |
854 | ||
855 | /* | |
856 | * Get MDT index for the inode. | |
857 | */ | |
858 | int ll_get_mdt_idx(struct inode *inode) | |
859 | { | |
860 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
861 | struct md_op_data *op_data; | |
862 | int rc, mdtidx; | |
d7e09d03 PT |
863 | |
864 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, | |
865 | 0, LUSTRE_OPC_ANY, NULL); | |
866 | if (IS_ERR(op_data)) | |
0a3bdb00 | 867 | return PTR_ERR(op_data); |
d7e09d03 PT |
868 | |
869 | op_data->op_flags |= MF_GET_MDT_IDX; | |
870 | rc = md_getattr(sbi->ll_md_exp, op_data, NULL); | |
871 | mdtidx = op_data->op_mds; | |
872 | ll_finish_md_op_data(op_data); | |
873 | if (rc < 0) { | |
874 | CDEBUG(D_INFO, "md_getattr_name: %d\n", rc); | |
0a3bdb00 | 875 | return rc; |
d7e09d03 PT |
876 | } |
877 | return mdtidx; | |
878 | } | |
879 | ||
880 | /** | |
881 | * Generic handler to do any pre-copy work. | |
882 | * | |
883 | * It send a first hsm_progress (with extent length == 0) to coordinator as a | |
884 | * first information for it that real work has started. | |
885 | * | |
886 | * Moreover, for a ARCHIVE request, it will sample the file data version and | |
887 | * store it in \a copy. | |
888 | * | |
889 | * \return 0 on success. | |
890 | */ | |
891 | static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy) | |
892 | { | |
893 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
894 | struct hsm_progress_kernel hpk; | |
895 | int rc; | |
d7e09d03 PT |
896 | |
897 | /* Forge a hsm_progress based on data from copy. */ | |
898 | hpk.hpk_fid = copy->hc_hai.hai_fid; | |
899 | hpk.hpk_cookie = copy->hc_hai.hai_cookie; | |
900 | hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset; | |
901 | hpk.hpk_extent.length = 0; | |
902 | hpk.hpk_flags = 0; | |
903 | hpk.hpk_errval = 0; | |
904 | hpk.hpk_data_version = 0; | |
905 | ||
d7e09d03 PT |
906 | /* For archive request, we need to read the current file version. */ |
907 | if (copy->hc_hai.hai_action == HSMA_ARCHIVE) { | |
908 | struct inode *inode; | |
909 | __u64 data_version = 0; | |
910 | ||
911 | /* Get inode for this fid */ | |
912 | inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid); | |
913 | if (IS_ERR(inode)) { | |
914 | hpk.hpk_flags |= HP_FLAG_RETRY; | |
915 | /* hpk_errval is >= 0 */ | |
916 | hpk.hpk_errval = -PTR_ERR(inode); | |
34e1f2bb JL |
917 | rc = PTR_ERR(inode); |
918 | goto progress; | |
d7e09d03 PT |
919 | } |
920 | ||
921 | /* Read current file data version */ | |
922 | rc = ll_data_version(inode, &data_version, 1); | |
923 | iput(inode); | |
924 | if (rc != 0) { | |
925 | CDEBUG(D_HSM, "Could not read file data version of " | |
55f5a824 | 926 | DFID" (rc = %d). Archive request (%#llx) could not be done.\n", |
d7e09d03 PT |
927 | PFID(©->hc_hai.hai_fid), rc, |
928 | copy->hc_hai.hai_cookie); | |
929 | hpk.hpk_flags |= HP_FLAG_RETRY; | |
930 | /* hpk_errval must be >= 0 */ | |
931 | hpk.hpk_errval = -rc; | |
34e1f2bb | 932 | goto progress; |
d7e09d03 PT |
933 | } |
934 | ||
935 | /* Store it the hsm_copy for later copytool use. | |
936 | * Always modified even if no lsm. */ | |
937 | copy->hc_data_version = data_version; | |
938 | } | |
939 | ||
940 | progress: | |
941 | rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk), | |
942 | &hpk, NULL); | |
943 | ||
0a3bdb00 | 944 | return rc; |
d7e09d03 PT |
945 | } |
946 | ||
947 | /** | |
948 | * Generic handler to do any post-copy work. | |
949 | * | |
950 | * It will send the last hsm_progress update to coordinator to inform it | |
951 | * that copy is finished and whether it was successful or not. | |
952 | * | |
953 | * Moreover, | |
954 | * - for ARCHIVE request, it will sample the file data version and compare it | |
955 | * with the version saved in ll_ioc_copy_start(). If they do not match, copy | |
956 | * will be considered as failed. | |
957 | * - for RESTORE request, it will sample the file data version and send it to | |
958 | * coordinator which is useful if the file was imported as 'released'. | |
959 | * | |
960 | * \return 0 on success. | |
961 | */ | |
962 | static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy) | |
963 | { | |
964 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
965 | struct hsm_progress_kernel hpk; | |
966 | int rc; | |
d7e09d03 PT |
967 | |
968 | /* If you modify the logic here, also check llapi_hsm_copy_end(). */ | |
969 | /* Take care: copy->hc_hai.hai_action, len, gid and data are not | |
970 | * initialized if copy_end was called with copy == NULL. | |
971 | */ | |
972 | ||
973 | /* Forge a hsm_progress based on data from copy. */ | |
974 | hpk.hpk_fid = copy->hc_hai.hai_fid; | |
975 | hpk.hpk_cookie = copy->hc_hai.hai_cookie; | |
976 | hpk.hpk_extent = copy->hc_hai.hai_extent; | |
977 | hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED; | |
978 | hpk.hpk_errval = copy->hc_errval; | |
979 | hpk.hpk_data_version = 0; | |
980 | ||
981 | /* For archive request, we need to check the file data was not changed. | |
982 | * | |
983 | * For restore request, we need to send the file data version, this is | |
984 | * useful when the file was created using hsm_import. | |
985 | */ | |
986 | if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) || | |
987 | (copy->hc_hai.hai_action == HSMA_RESTORE)) && | |
988 | (copy->hc_errval == 0)) { | |
989 | struct inode *inode; | |
990 | __u64 data_version = 0; | |
991 | ||
992 | /* Get lsm for this fid */ | |
993 | inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid); | |
994 | if (IS_ERR(inode)) { | |
995 | hpk.hpk_flags |= HP_FLAG_RETRY; | |
996 | /* hpk_errval must be >= 0 */ | |
997 | hpk.hpk_errval = -PTR_ERR(inode); | |
34e1f2bb JL |
998 | rc = PTR_ERR(inode); |
999 | goto progress; | |
d7e09d03 PT |
1000 | } |
1001 | ||
1002 | rc = ll_data_version(inode, &data_version, | |
1003 | copy->hc_hai.hai_action == HSMA_ARCHIVE); | |
1004 | iput(inode); | |
1005 | if (rc) { | |
2d00bd17 | 1006 | CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n"); |
d7e09d03 PT |
1007 | if (hpk.hpk_errval == 0) |
1008 | hpk.hpk_errval = -rc; | |
34e1f2bb | 1009 | goto progress; |
d7e09d03 PT |
1010 | } |
1011 | ||
1012 | /* Store it the hsm_copy for later copytool use. | |
1013 | * Always modified even if no lsm. */ | |
1014 | hpk.hpk_data_version = data_version; | |
1015 | ||
1016 | /* File could have been stripped during archiving, so we need | |
1017 | * to check anyway. */ | |
1018 | if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) && | |
1019 | (copy->hc_data_version != data_version)) { | |
2d00bd17 | 1020 | CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. " |
55f5a824 | 1021 | DFID", start:%#llx current:%#llx\n", |
d7e09d03 PT |
1022 | PFID(©->hc_hai.hai_fid), |
1023 | copy->hc_data_version, data_version); | |
1024 | /* File was changed, send error to cdt. Do not ask for | |
1025 | * retry because if a file is modified frequently, | |
1026 | * the cdt will loop on retried archive requests. | |
1027 | * The policy engine will ask for a new archive later | |
1028 | * when the file will not be modified for some tunable | |
1029 | * time */ | |
1030 | /* we do not notify caller */ | |
1031 | hpk.hpk_flags &= ~HP_FLAG_RETRY; | |
1032 | /* hpk_errval must be >= 0 */ | |
1033 | hpk.hpk_errval = EBUSY; | |
1034 | } | |
1035 | ||
1036 | } | |
1037 | ||
1038 | progress: | |
1039 | rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk), | |
1040 | &hpk, NULL); | |
1041 | ||
0a3bdb00 | 1042 | return rc; |
d7e09d03 PT |
1043 | } |
1044 | ||
b0337d6c JH |
1045 | static int copy_and_ioctl(int cmd, struct obd_export *exp, |
1046 | const void __user *data, size_t size) | |
d7e09d03 | 1047 | { |
b0337d6c | 1048 | void *copy; |
d7e09d03 PT |
1049 | int rc; |
1050 | ||
496a51bd JL |
1051 | copy = kzalloc(size, GFP_NOFS); |
1052 | if (!copy) | |
d7e09d03 | 1053 | return -ENOMEM; |
b0337d6c JH |
1054 | |
1055 | if (copy_from_user(copy, data, size)) { | |
1056 | rc = -EFAULT; | |
1057 | goto out; | |
d7e09d03 | 1058 | } |
b0337d6c JH |
1059 | |
1060 | rc = obd_iocontrol(cmd, exp, size, copy, NULL); | |
1061 | out: | |
97903a26 | 1062 | kfree(copy); |
b0337d6c | 1063 | |
d7e09d03 PT |
1064 | return rc; |
1065 | } | |
1066 | ||
1067 | static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl) | |
1068 | { | |
1069 | int cmd = qctl->qc_cmd; | |
1070 | int type = qctl->qc_type; | |
1071 | int id = qctl->qc_id; | |
1072 | int valid = qctl->qc_valid; | |
1073 | int rc = 0; | |
d7e09d03 PT |
1074 | |
1075 | switch (cmd) { | |
1076 | case LUSTRE_Q_INVALIDATE: | |
1077 | case LUSTRE_Q_FINVALIDATE: | |
1078 | case Q_QUOTAON: | |
1079 | case Q_QUOTAOFF: | |
1080 | case Q_SETQUOTA: | |
1081 | case Q_SETINFO: | |
2eb90a75 | 1082 | if (!capable(CFS_CAP_SYS_ADMIN) || |
d7e09d03 | 1083 | sbi->ll_flags & LL_SBI_RMT_CLIENT) |
0a3bdb00 | 1084 | return -EPERM; |
d7e09d03 PT |
1085 | break; |
1086 | case Q_GETQUOTA: | |
4b1a25f0 | 1087 | if (((type == USRQUOTA && |
8b9e418c | 1088 | !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) || |
4b1a25f0 PT |
1089 | (type == GRPQUOTA && |
1090 | !in_egroup_p(make_kgid(&init_user_ns, id)))) && | |
2eb90a75 | 1091 | (!capable(CFS_CAP_SYS_ADMIN) || |
d7e09d03 | 1092 | sbi->ll_flags & LL_SBI_RMT_CLIENT)) |
0a3bdb00 | 1093 | return -EPERM; |
d7e09d03 PT |
1094 | break; |
1095 | case Q_GETINFO: | |
1096 | break; | |
1097 | default: | |
1098 | CERROR("unsupported quotactl op: %#x\n", cmd); | |
0a3bdb00 | 1099 | return -ENOTTY; |
d7e09d03 PT |
1100 | } |
1101 | ||
1102 | if (valid != QC_GENERAL) { | |
1103 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) | |
0a3bdb00 | 1104 | return -EOPNOTSUPP; |
d7e09d03 PT |
1105 | |
1106 | if (cmd == Q_GETINFO) | |
1107 | qctl->qc_cmd = Q_GETOINFO; | |
1108 | else if (cmd == Q_GETQUOTA) | |
1109 | qctl->qc_cmd = Q_GETOQUOTA; | |
1110 | else | |
0a3bdb00 | 1111 | return -EINVAL; |
d7e09d03 PT |
1112 | |
1113 | switch (valid) { | |
1114 | case QC_MDTIDX: | |
1115 | rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp, | |
1116 | sizeof(*qctl), qctl, NULL); | |
1117 | break; | |
1118 | case QC_OSTIDX: | |
1119 | rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp, | |
1120 | sizeof(*qctl), qctl, NULL); | |
1121 | break; | |
1122 | case QC_UUID: | |
1123 | rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp, | |
1124 | sizeof(*qctl), qctl, NULL); | |
1125 | if (rc == -EAGAIN) | |
1126 | rc = obd_iocontrol(OBD_IOC_QUOTACTL, | |
1127 | sbi->ll_dt_exp, | |
1128 | sizeof(*qctl), qctl, NULL); | |
1129 | break; | |
1130 | default: | |
1131 | rc = -EINVAL; | |
1132 | break; | |
1133 | } | |
1134 | ||
1135 | if (rc) | |
0a3bdb00 | 1136 | return rc; |
d7e09d03 PT |
1137 | |
1138 | qctl->qc_cmd = cmd; | |
1139 | } else { | |
1140 | struct obd_quotactl *oqctl; | |
1141 | ||
496a51bd JL |
1142 | oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS); |
1143 | if (!oqctl) | |
0a3bdb00 | 1144 | return -ENOMEM; |
d7e09d03 PT |
1145 | |
1146 | QCTL_COPY(oqctl, qctl); | |
1147 | rc = obd_quotactl(sbi->ll_md_exp, oqctl); | |
1148 | if (rc) { | |
1149 | if (rc != -EALREADY && cmd == Q_QUOTAON) { | |
1150 | oqctl->qc_cmd = Q_QUOTAOFF; | |
1151 | obd_quotactl(sbi->ll_md_exp, oqctl); | |
1152 | } | |
97903a26 | 1153 | kfree(oqctl); |
0a3bdb00 | 1154 | return rc; |
d7e09d03 PT |
1155 | } |
1156 | /* If QIF_SPACE is not set, client should collect the | |
1157 | * space usage from OSSs by itself */ | |
1158 | if (cmd == Q_GETQUOTA && | |
1159 | !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) && | |
1160 | !oqctl->qc_dqblk.dqb_curspace) { | |
1161 | struct obd_quotactl *oqctl_tmp; | |
1162 | ||
496a51bd JL |
1163 | oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS); |
1164 | if (!oqctl_tmp) { | |
34e1f2bb JL |
1165 | rc = -ENOMEM; |
1166 | goto out; | |
1167 | } | |
d7e09d03 PT |
1168 | |
1169 | oqctl_tmp->qc_cmd = Q_GETOQUOTA; | |
1170 | oqctl_tmp->qc_id = oqctl->qc_id; | |
1171 | oqctl_tmp->qc_type = oqctl->qc_type; | |
1172 | ||
1173 | /* collect space usage from OSTs */ | |
1174 | oqctl_tmp->qc_dqblk.dqb_curspace = 0; | |
1175 | rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp); | |
1176 | if (!rc || rc == -EREMOTEIO) { | |
1177 | oqctl->qc_dqblk.dqb_curspace = | |
1178 | oqctl_tmp->qc_dqblk.dqb_curspace; | |
1179 | oqctl->qc_dqblk.dqb_valid |= QIF_SPACE; | |
1180 | } | |
1181 | ||
1182 | /* collect space & inode usage from MDTs */ | |
1183 | oqctl_tmp->qc_dqblk.dqb_curspace = 0; | |
1184 | oqctl_tmp->qc_dqblk.dqb_curinodes = 0; | |
1185 | rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp); | |
1186 | if (!rc || rc == -EREMOTEIO) { | |
1187 | oqctl->qc_dqblk.dqb_curspace += | |
1188 | oqctl_tmp->qc_dqblk.dqb_curspace; | |
1189 | oqctl->qc_dqblk.dqb_curinodes = | |
1190 | oqctl_tmp->qc_dqblk.dqb_curinodes; | |
1191 | oqctl->qc_dqblk.dqb_valid |= QIF_INODES; | |
1192 | } else { | |
1193 | oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE; | |
1194 | } | |
1195 | ||
97903a26 | 1196 | kfree(oqctl_tmp); |
d7e09d03 PT |
1197 | } |
1198 | out: | |
1199 | QCTL_COPY(qctl, oqctl); | |
97903a26 | 1200 | kfree(oqctl); |
d7e09d03 PT |
1201 | } |
1202 | ||
0a3bdb00 | 1203 | return rc; |
d7e09d03 PT |
1204 | } |
1205 | ||
a7503434 OD |
1206 | /* This function tries to get a single name component, |
1207 | * to send to the server. No actual path traversal involved, | |
1208 | * so we limit to NAME_MAX */ | |
1209 | static char *ll_getname(const char __user *filename) | |
d7e09d03 PT |
1210 | { |
1211 | int ret = 0, len; | |
a7503434 | 1212 | char *tmp; |
d7e09d03 | 1213 | |
a7503434 | 1214 | tmp = kzalloc(NAME_MAX + 1, GFP_KERNEL); |
d7e09d03 PT |
1215 | if (!tmp) |
1216 | return ERR_PTR(-ENOMEM); | |
1217 | ||
a7503434 OD |
1218 | len = strncpy_from_user(tmp, filename, NAME_MAX + 1); |
1219 | if (len < 0) | |
1220 | ret = len; | |
1221 | else if (len == 0) | |
d7e09d03 | 1222 | ret = -ENOENT; |
a7503434 | 1223 | else if (len > NAME_MAX && tmp[NAME_MAX] != 0) |
d7e09d03 PT |
1224 | ret = -ENAMETOOLONG; |
1225 | ||
1226 | if (ret) { | |
a7503434 | 1227 | kfree(tmp); |
d7e09d03 PT |
1228 | tmp = ERR_PTR(ret); |
1229 | } | |
1230 | return tmp; | |
1231 | } | |
1232 | ||
a7503434 | 1233 | #define ll_putname(filename) kfree(filename) |
d7e09d03 PT |
1234 | |
1235 | static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
1236 | { | |
2a8a3597 | 1237 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
1238 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
1239 | struct obd_ioctl_data *data; | |
1240 | int rc = 0; | |
d7e09d03 PT |
1241 | |
1242 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n", | |
1243 | inode->i_ino, inode->i_generation, inode, cmd); | |
1244 | ||
1245 | /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */ | |
1246 | if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */ | |
1247 | return -ENOTTY; | |
1248 | ||
1249 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1); | |
a58a38ac | 1250 | switch (cmd) { |
d7e09d03 PT |
1251 | case FSFILT_IOC_GETFLAGS: |
1252 | case FSFILT_IOC_SETFLAGS: | |
0a3bdb00 | 1253 | return ll_iocontrol(inode, file, cmd, arg); |
d7e09d03 PT |
1254 | case FSFILT_IOC_GETVERSION_OLD: |
1255 | case FSFILT_IOC_GETVERSION: | |
0a3bdb00 | 1256 | return put_user(inode->i_generation, (int *)arg); |
d7e09d03 PT |
1257 | /* We need to special case any other ioctls we want to handle, |
1258 | * to send them to the MDS/OST as appropriate and to properly | |
1259 | * network encode the arg field. | |
1260 | case FSFILT_IOC_SETVERSION_OLD: | |
1261 | case FSFILT_IOC_SETVERSION: | |
1262 | */ | |
1263 | case LL_IOC_GET_MDTIDX: { | |
1264 | int mdtidx; | |
1265 | ||
1266 | mdtidx = ll_get_mdt_idx(inode); | |
1267 | if (mdtidx < 0) | |
0a3bdb00 | 1268 | return mdtidx; |
d7e09d03 | 1269 | |
b0126abf | 1270 | if (put_user((int)mdtidx, (int *)arg)) |
0a3bdb00 | 1271 | return -EFAULT; |
d7e09d03 PT |
1272 | |
1273 | return 0; | |
1274 | } | |
1275 | case IOC_MDC_LOOKUP: { | |
1276 | struct ptlrpc_request *request = NULL; | |
1277 | int namelen, len = 0; | |
1278 | char *buf = NULL; | |
1279 | char *filename; | |
1280 | struct md_op_data *op_data; | |
1281 | ||
e3e8ff41 | 1282 | rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg); |
d7e09d03 | 1283 | if (rc) |
0a3bdb00 | 1284 | return rc; |
d7e09d03 PT |
1285 | data = (void *)buf; |
1286 | ||
1287 | filename = data->ioc_inlbuf1; | |
1288 | namelen = strlen(filename); | |
1289 | ||
1290 | if (namelen < 1) { | |
1291 | CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n"); | |
34e1f2bb JL |
1292 | rc = -EINVAL; |
1293 | goto out_free; | |
d7e09d03 PT |
1294 | } |
1295 | ||
1296 | op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen, | |
1297 | 0, LUSTRE_OPC_ANY, NULL); | |
34e1f2bb JL |
1298 | if (IS_ERR(op_data)) { |
1299 | rc = PTR_ERR(op_data); | |
1300 | goto out_free; | |
1301 | } | |
d7e09d03 PT |
1302 | |
1303 | op_data->op_valid = OBD_MD_FLID; | |
1304 | rc = md_getattr_name(sbi->ll_md_exp, op_data, &request); | |
1305 | ll_finish_md_op_data(op_data); | |
1306 | if (rc < 0) { | |
1307 | CDEBUG(D_INFO, "md_getattr_name: %d\n", rc); | |
34e1f2bb | 1308 | goto out_free; |
d7e09d03 PT |
1309 | } |
1310 | ptlrpc_req_finished(request); | |
d7e09d03 PT |
1311 | out_free: |
1312 | obd_ioctl_freedata(buf, len); | |
1313 | return rc; | |
1314 | } | |
1315 | case LL_IOC_LMV_SETSTRIPE: { | |
1316 | struct lmv_user_md *lum; | |
1317 | char *buf = NULL; | |
1318 | char *filename; | |
1319 | int namelen = 0; | |
1320 | int lumlen = 0; | |
1321 | int len; | |
1322 | int rc; | |
1323 | ||
e3e8ff41 | 1324 | rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg); |
d7e09d03 | 1325 | if (rc) |
0a3bdb00 | 1326 | return rc; |
d7e09d03 PT |
1327 | |
1328 | data = (void *)buf; | |
1329 | if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL || | |
34e1f2bb JL |
1330 | data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) { |
1331 | rc = -EINVAL; | |
1332 | goto lmv_out_free; | |
1333 | } | |
d7e09d03 PT |
1334 | |
1335 | filename = data->ioc_inlbuf1; | |
1336 | namelen = data->ioc_inllen1; | |
1337 | ||
1338 | if (namelen < 1) { | |
1339 | CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n"); | |
34e1f2bb JL |
1340 | rc = -EINVAL; |
1341 | goto lmv_out_free; | |
d7e09d03 PT |
1342 | } |
1343 | lum = (struct lmv_user_md *)data->ioc_inlbuf2; | |
1344 | lumlen = data->ioc_inllen2; | |
1345 | ||
1346 | if (lum->lum_magic != LMV_USER_MAGIC || | |
1347 | lumlen != sizeof(*lum)) { | |
1348 | CERROR("%s: wrong lum magic %x or size %d: rc = %d\n", | |
1349 | filename, lum->lum_magic, lumlen, -EFAULT); | |
34e1f2bb JL |
1350 | rc = -EINVAL; |
1351 | goto lmv_out_free; | |
d7e09d03 PT |
1352 | } |
1353 | ||
1354 | /** | |
1355 | * ll_dir_setdirstripe will be used to set dir stripe | |
1356 | * mdc_create--->mdt_reint_create (with dirstripe) | |
1357 | */ | |
1358 | rc = ll_dir_setdirstripe(inode, lum, filename); | |
1359 | lmv_out_free: | |
1360 | obd_ioctl_freedata(buf, len); | |
0a3bdb00 | 1361 | return rc; |
d7e09d03 PT |
1362 | |
1363 | } | |
1364 | case LL_IOC_LOV_SETSTRIPE: { | |
1365 | struct lov_user_md_v3 lumv3; | |
1366 | struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; | |
1367 | struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg; | |
1368 | struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg; | |
1369 | ||
1370 | int set_default = 0; | |
1371 | ||
1372 | LASSERT(sizeof(lumv3) == sizeof(*lumv3p)); | |
1373 | LASSERT(sizeof(lumv3.lmm_objects[0]) == | |
1374 | sizeof(lumv3p->lmm_objects[0])); | |
1375 | /* first try with v1 which is smaller than v3 */ | |
1376 | if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1))) | |
0a3bdb00 | 1377 | return -EFAULT; |
d7e09d03 | 1378 | |
557732ad | 1379 | if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { |
d7e09d03 | 1380 | if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3))) |
0a3bdb00 | 1381 | return -EFAULT; |
d7e09d03 PT |
1382 | } |
1383 | ||
f76c23da | 1384 | if (is_root_inode(inode)) |
d7e09d03 PT |
1385 | set_default = 1; |
1386 | ||
1387 | /* in v1 and v3 cases lumv1 points to data */ | |
1388 | rc = ll_dir_setstripe(inode, lumv1, set_default); | |
1389 | ||
0a3bdb00 | 1390 | return rc; |
d7e09d03 PT |
1391 | } |
1392 | case LL_IOC_LMV_GETSTRIPE: { | |
1393 | struct lmv_user_md *lump = (struct lmv_user_md *)arg; | |
1394 | struct lmv_user_md lum; | |
1395 | struct lmv_user_md *tmp; | |
1396 | int lum_size; | |
1397 | int rc = 0; | |
1398 | int mdtindex; | |
1399 | ||
1400 | if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md))) | |
0a3bdb00 | 1401 | return -EFAULT; |
d7e09d03 PT |
1402 | |
1403 | if (lum.lum_magic != LMV_MAGIC_V1) | |
0a3bdb00 | 1404 | return -EINVAL; |
d7e09d03 PT |
1405 | |
1406 | lum_size = lmv_user_md_size(1, LMV_MAGIC_V1); | |
496a51bd JL |
1407 | tmp = kzalloc(lum_size, GFP_NOFS); |
1408 | if (!tmp) { | |
34e1f2bb JL |
1409 | rc = -ENOMEM; |
1410 | goto free_lmv; | |
1411 | } | |
d7e09d03 | 1412 | |
7f46528c | 1413 | *tmp = lum; |
d7e09d03 PT |
1414 | tmp->lum_type = LMV_STRIPE_TYPE; |
1415 | tmp->lum_stripe_count = 1; | |
1416 | mdtindex = ll_get_mdt_idx(inode); | |
34e1f2bb JL |
1417 | if (mdtindex < 0) { |
1418 | rc = -ENOMEM; | |
1419 | goto free_lmv; | |
1420 | } | |
d7e09d03 PT |
1421 | |
1422 | tmp->lum_stripe_offset = mdtindex; | |
1423 | tmp->lum_objects[0].lum_mds = mdtindex; | |
1424 | memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode), | |
1425 | sizeof(struct lu_fid)); | |
34e1f2bb JL |
1426 | if (copy_to_user((void *)arg, tmp, lum_size)) { |
1427 | rc = -EFAULT; | |
1428 | goto free_lmv; | |
1429 | } | |
d7e09d03 | 1430 | free_lmv: |
fd5e2fd0 | 1431 | kfree(tmp); |
0a3bdb00 | 1432 | return rc; |
d7e09d03 | 1433 | } |
d7e09d03 | 1434 | case LL_IOC_LOV_SWAP_LAYOUTS: |
0a3bdb00 | 1435 | return -EPERM; |
d7e09d03 | 1436 | case LL_IOC_OBD_STATFS: |
4c6243ec | 1437 | return ll_obd_statfs(inode, (void __user *)arg); |
d7e09d03 PT |
1438 | case LL_IOC_LOV_GETSTRIPE: |
1439 | case LL_IOC_MDC_GETINFO: | |
1440 | case IOC_MDC_GETFILEINFO: | |
1441 | case IOC_MDC_GETFILESTRIPE: { | |
1442 | struct ptlrpc_request *request = NULL; | |
1443 | struct lov_user_md *lump; | |
1444 | struct lov_mds_md *lmm = NULL; | |
1445 | struct mdt_body *body; | |
1446 | char *filename = NULL; | |
1447 | int lmmsize; | |
1448 | ||
1449 | if (cmd == IOC_MDC_GETFILEINFO || | |
1450 | cmd == IOC_MDC_GETFILESTRIPE) { | |
1451 | filename = ll_getname((const char *)arg); | |
1452 | if (IS_ERR(filename)) | |
0a3bdb00 | 1453 | return PTR_ERR(filename); |
d7e09d03 PT |
1454 | |
1455 | rc = ll_lov_getstripe_ea_info(inode, filename, &lmm, | |
1456 | &lmmsize, &request); | |
1457 | } else { | |
1458 | rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request); | |
1459 | } | |
1460 | ||
1461 | if (request) { | |
1462 | body = req_capsule_server_get(&request->rq_pill, | |
1463 | &RMF_MDT_BODY); | |
1464 | LASSERT(body != NULL); | |
1465 | } else { | |
34e1f2bb | 1466 | goto out_req; |
d7e09d03 PT |
1467 | } |
1468 | ||
1469 | if (rc < 0) { | |
1470 | if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO || | |
34e1f2bb JL |
1471 | cmd == LL_IOC_MDC_GETINFO)) { |
1472 | rc = 0; | |
1473 | goto skip_lmm; | |
4d1d413a | 1474 | } else |
34e1f2bb | 1475 | goto out_req; |
d7e09d03 PT |
1476 | } |
1477 | ||
1478 | if (cmd == IOC_MDC_GETFILESTRIPE || | |
1479 | cmd == LL_IOC_LOV_GETSTRIPE) { | |
1480 | lump = (struct lov_user_md *)arg; | |
1481 | } else { | |
1482 | struct lov_user_mds_data *lmdp; | |
79792190 | 1483 | |
d7e09d03 PT |
1484 | lmdp = (struct lov_user_mds_data *)arg; |
1485 | lump = &lmdp->lmd_lmm; | |
1486 | } | |
1487 | if (copy_to_user(lump, lmm, lmmsize)) { | |
34e1f2bb JL |
1488 | if (copy_to_user(lump, lmm, sizeof(*lump))) { |
1489 | rc = -EFAULT; | |
1490 | goto out_req; | |
1491 | } | |
d7e09d03 PT |
1492 | rc = -EOVERFLOW; |
1493 | } | |
eb73f514 | 1494 | skip_lmm: |
d7e09d03 PT |
1495 | if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) { |
1496 | struct lov_user_mds_data *lmdp; | |
1497 | lstat_t st = { 0 }; | |
1498 | ||
1499 | st.st_dev = inode->i_sb->s_dev; | |
1500 | st.st_mode = body->mode; | |
1501 | st.st_nlink = body->nlink; | |
1502 | st.st_uid = body->uid; | |
1503 | st.st_gid = body->gid; | |
1504 | st.st_rdev = body->rdev; | |
1505 | st.st_size = body->size; | |
1506 | st.st_blksize = PAGE_CACHE_SIZE; | |
1507 | st.st_blocks = body->blocks; | |
1508 | st.st_atime = body->atime; | |
1509 | st.st_mtime = body->mtime; | |
1510 | st.st_ctime = body->ctime; | |
1511 | st.st_ino = inode->i_ino; | |
1512 | ||
1513 | lmdp = (struct lov_user_mds_data *)arg; | |
34e1f2bb JL |
1514 | if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) { |
1515 | rc = -EFAULT; | |
1516 | goto out_req; | |
1517 | } | |
d7e09d03 PT |
1518 | } |
1519 | ||
eb73f514 | 1520 | out_req: |
d7e09d03 PT |
1521 | ptlrpc_req_finished(request); |
1522 | if (filename) | |
1523 | ll_putname(filename); | |
1524 | return rc; | |
1525 | } | |
1526 | case IOC_LOV_GETINFO: { | |
1527 | struct lov_user_mds_data *lumd; | |
1528 | struct lov_stripe_md *lsm; | |
1529 | struct lov_user_md *lum; | |
1530 | struct lov_mds_md *lmm; | |
1531 | int lmmsize; | |
1532 | lstat_t st; | |
1533 | ||
1534 | lumd = (struct lov_user_mds_data *)arg; | |
1535 | lum = &lumd->lmd_lmm; | |
1536 | ||
1537 | rc = ll_get_max_mdsize(sbi, &lmmsize); | |
1538 | if (rc) | |
0a3bdb00 | 1539 | return rc; |
d7e09d03 | 1540 | |
e958f49b | 1541 | lmm = libcfs_kvzalloc(lmmsize, GFP_NOFS); |
73863d83 | 1542 | if (lmm == NULL) |
0a3bdb00 | 1543 | return -ENOMEM; |
34e1f2bb JL |
1544 | if (copy_from_user(lmm, lum, lmmsize)) { |
1545 | rc = -EFAULT; | |
1546 | goto free_lmm; | |
1547 | } | |
d7e09d03 PT |
1548 | |
1549 | switch (lmm->lmm_magic) { | |
1550 | case LOV_USER_MAGIC_V1: | |
1f6eaf83 | 1551 | if (cpu_to_le32(LOV_USER_MAGIC_V1) == LOV_USER_MAGIC_V1) |
d7e09d03 PT |
1552 | break; |
1553 | /* swab objects first so that stripes num will be sane */ | |
1554 | lustre_swab_lov_user_md_objects( | |
1555 | ((struct lov_user_md_v1 *)lmm)->lmm_objects, | |
1556 | ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count); | |
1557 | lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); | |
1558 | break; | |
1559 | case LOV_USER_MAGIC_V3: | |
1f6eaf83 | 1560 | if (cpu_to_le32(LOV_USER_MAGIC_V3) == LOV_USER_MAGIC_V3) |
d7e09d03 PT |
1561 | break; |
1562 | /* swab objects first so that stripes num will be sane */ | |
1563 | lustre_swab_lov_user_md_objects( | |
1564 | ((struct lov_user_md_v3 *)lmm)->lmm_objects, | |
1565 | ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count); | |
1566 | lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); | |
1567 | break; | |
1568 | default: | |
34e1f2bb JL |
1569 | rc = -EINVAL; |
1570 | goto free_lmm; | |
d7e09d03 PT |
1571 | } |
1572 | ||
1573 | rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize); | |
34e1f2bb JL |
1574 | if (rc < 0) { |
1575 | rc = -ENOMEM; | |
1576 | goto free_lmm; | |
1577 | } | |
d7e09d03 PT |
1578 | |
1579 | /* Perform glimpse_size operation. */ | |
1580 | memset(&st, 0, sizeof(st)); | |
1581 | ||
1582 | rc = ll_glimpse_ioctl(sbi, lsm, &st); | |
1583 | if (rc) | |
34e1f2bb | 1584 | goto free_lsm; |
d7e09d03 | 1585 | |
34e1f2bb JL |
1586 | if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) { |
1587 | rc = -EFAULT; | |
1588 | goto free_lsm; | |
1589 | } | |
d7e09d03 | 1590 | |
eb73f514 | 1591 | free_lsm: |
d7e09d03 | 1592 | obd_free_memmd(sbi->ll_dt_exp, &lsm); |
eb73f514 | 1593 | free_lmm: |
e958f49b | 1594 | kvfree(lmm); |
d7e09d03 PT |
1595 | return rc; |
1596 | } | |
1597 | case OBD_IOC_LLOG_CATINFO: { | |
0a3bdb00 | 1598 | return -EOPNOTSUPP; |
d7e09d03 PT |
1599 | } |
1600 | case OBD_IOC_QUOTACHECK: { | |
1601 | struct obd_quotactl *oqctl; | |
1602 | int error = 0; | |
1603 | ||
2eb90a75 | 1604 | if (!capable(CFS_CAP_SYS_ADMIN) || |
d7e09d03 | 1605 | sbi->ll_flags & LL_SBI_RMT_CLIENT) |
0a3bdb00 | 1606 | return -EPERM; |
d7e09d03 | 1607 | |
496a51bd | 1608 | oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS); |
d7e09d03 | 1609 | if (!oqctl) |
0a3bdb00 | 1610 | return -ENOMEM; |
d7e09d03 PT |
1611 | oqctl->qc_type = arg; |
1612 | rc = obd_quotacheck(sbi->ll_md_exp, oqctl); | |
1613 | if (rc < 0) { | |
1614 | CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc); | |
1615 | error = rc; | |
1616 | } | |
1617 | ||
1618 | rc = obd_quotacheck(sbi->ll_dt_exp, oqctl); | |
1619 | if (rc < 0) | |
1620 | CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc); | |
1621 | ||
97903a26 | 1622 | kfree(oqctl); |
d7e09d03 PT |
1623 | return error ?: rc; |
1624 | } | |
1625 | case OBD_IOC_POLL_QUOTACHECK: { | |
1626 | struct if_quotacheck *check; | |
1627 | ||
2eb90a75 | 1628 | if (!capable(CFS_CAP_SYS_ADMIN) || |
d7e09d03 | 1629 | sbi->ll_flags & LL_SBI_RMT_CLIENT) |
0a3bdb00 | 1630 | return -EPERM; |
d7e09d03 | 1631 | |
496a51bd | 1632 | check = kzalloc(sizeof(*check), GFP_NOFS); |
d7e09d03 | 1633 | if (!check) |
0a3bdb00 | 1634 | return -ENOMEM; |
d7e09d03 PT |
1635 | |
1636 | rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check, | |
1637 | NULL); | |
1638 | if (rc) { | |
1639 | CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc); | |
1640 | if (copy_to_user((void *)arg, check, | |
1641 | sizeof(*check))) | |
1642 | CDEBUG(D_QUOTA, "copy_to_user failed\n"); | |
34e1f2bb | 1643 | goto out_poll; |
d7e09d03 PT |
1644 | } |
1645 | ||
1646 | rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check, | |
1647 | NULL); | |
1648 | if (rc) { | |
1649 | CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc); | |
1650 | if (copy_to_user((void *)arg, check, | |
1651 | sizeof(*check))) | |
1652 | CDEBUG(D_QUOTA, "copy_to_user failed\n"); | |
34e1f2bb | 1653 | goto out_poll; |
d7e09d03 | 1654 | } |
eb73f514 | 1655 | out_poll: |
97903a26 | 1656 | kfree(check); |
0a3bdb00 | 1657 | return rc; |
d7e09d03 | 1658 | } |
d7e09d03 PT |
1659 | case LL_IOC_QUOTACTL: { |
1660 | struct if_quotactl *qctl; | |
1661 | ||
496a51bd | 1662 | qctl = kzalloc(sizeof(*qctl), GFP_NOFS); |
d7e09d03 | 1663 | if (!qctl) |
0a3bdb00 | 1664 | return -ENOMEM; |
d7e09d03 | 1665 | |
34e1f2bb JL |
1666 | if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) { |
1667 | rc = -EFAULT; | |
1668 | goto out_quotactl; | |
1669 | } | |
d7e09d03 PT |
1670 | |
1671 | rc = quotactl_ioctl(sbi, qctl); | |
1672 | ||
1d8cb70c | 1673 | if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl))) |
d7e09d03 PT |
1674 | rc = -EFAULT; |
1675 | ||
eb73f514 | 1676 | out_quotactl: |
97903a26 | 1677 | kfree(qctl); |
0a3bdb00 | 1678 | return rc; |
d7e09d03 PT |
1679 | } |
1680 | case OBD_IOC_GETDTNAME: | |
1681 | case OBD_IOC_GETMDNAME: | |
0a3bdb00 | 1682 | return ll_get_obd_name(inode, cmd, arg); |
d7e09d03 | 1683 | case LL_IOC_FLUSHCTX: |
0a3bdb00 | 1684 | return ll_flush_ctx(inode); |
d7e09d03 PT |
1685 | #ifdef CONFIG_FS_POSIX_ACL |
1686 | case LL_IOC_RMTACL: { | |
f76c23da | 1687 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) { |
d7e09d03 PT |
1688 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
1689 | ||
1690 | LASSERT(fd != NULL); | |
1691 | rc = rct_add(&sbi->ll_rct, current_pid(), arg); | |
1692 | if (!rc) | |
1693 | fd->fd_flags |= LL_FILE_RMTACL; | |
0a3bdb00 | 1694 | return rc; |
d7e09d03 | 1695 | } else |
0a3bdb00 | 1696 | return 0; |
d7e09d03 PT |
1697 | } |
1698 | #endif | |
1699 | case LL_IOC_GETOBDCOUNT: { | |
1700 | int count, vallen; | |
1701 | struct obd_export *exp; | |
1702 | ||
1703 | if (copy_from_user(&count, (int *)arg, sizeof(int))) | |
0a3bdb00 | 1704 | return -EFAULT; |
d7e09d03 PT |
1705 | |
1706 | /* get ost count when count is zero, get mdt count otherwise */ | |
1707 | exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp; | |
1708 | vallen = sizeof(count); | |
1709 | rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT), | |
1710 | KEY_TGT_COUNT, &vallen, &count, NULL); | |
1711 | if (rc) { | |
1712 | CERROR("get target count failed: %d\n", rc); | |
0a3bdb00 | 1713 | return rc; |
d7e09d03 PT |
1714 | } |
1715 | ||
1716 | if (copy_to_user((int *)arg, &count, sizeof(int))) | |
0a3bdb00 | 1717 | return -EFAULT; |
d7e09d03 | 1718 | |
0a3bdb00 | 1719 | return 0; |
d7e09d03 PT |
1720 | } |
1721 | case LL_IOC_PATH2FID: | |
1722 | if (copy_to_user((void *)arg, ll_inode2fid(inode), | |
1723 | sizeof(struct lu_fid))) | |
0a3bdb00 GKH |
1724 | return -EFAULT; |
1725 | return 0; | |
d7e09d03 | 1726 | case LL_IOC_GET_CONNECT_FLAGS: { |
e09bee34 OD |
1727 | return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, |
1728 | (void __user *)arg); | |
d7e09d03 PT |
1729 | } |
1730 | case OBD_IOC_CHANGELOG_SEND: | |
1731 | case OBD_IOC_CHANGELOG_CLEAR: | |
bbaa9c10 NY |
1732 | if (!capable(CFS_CAP_SYS_ADMIN)) |
1733 | return -EPERM; | |
1734 | ||
d7e09d03 PT |
1735 | rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg, |
1736 | sizeof(struct ioc_changelog)); | |
0a3bdb00 | 1737 | return rc; |
d7e09d03 | 1738 | case OBD_IOC_FID2PATH: |
7ec89fa5 | 1739 | return ll_fid2path(inode, (void __user *)arg); |
d7e09d03 PT |
1740 | case LL_IOC_HSM_REQUEST: { |
1741 | struct hsm_user_request *hur; | |
6b2eb32e | 1742 | ssize_t totalsize; |
d7e09d03 | 1743 | |
4a07594e AH |
1744 | hur = memdup_user((void *)arg, sizeof(*hur)); |
1745 | if (IS_ERR(hur)) | |
1746 | return PTR_ERR(hur); | |
d7e09d03 PT |
1747 | |
1748 | /* Compute the whole struct size */ | |
1749 | totalsize = hur_len(hur); | |
97903a26 | 1750 | kfree(hur); |
6b2eb32e NC |
1751 | if (totalsize < 0) |
1752 | return -E2BIG; | |
e55c4476 JN |
1753 | |
1754 | /* Final size will be more than double totalsize */ | |
1755 | if (totalsize >= MDS_MAXREQSIZE / 3) | |
1756 | return -E2BIG; | |
1757 | ||
e958f49b | 1758 | hur = libcfs_kvzalloc(totalsize, GFP_NOFS); |
d7e09d03 | 1759 | if (hur == NULL) |
0a3bdb00 | 1760 | return -ENOMEM; |
d7e09d03 PT |
1761 | |
1762 | /* Copy the whole struct */ | |
1763 | if (copy_from_user(hur, (void *)arg, totalsize)) { | |
e958f49b | 1764 | kvfree(hur); |
0a3bdb00 | 1765 | return -EFAULT; |
d7e09d03 PT |
1766 | } |
1767 | ||
48d23e61 JX |
1768 | if (hur->hur_request.hr_action == HUA_RELEASE) { |
1769 | const struct lu_fid *fid; | |
1770 | struct inode *f; | |
1771 | int i; | |
1772 | ||
1773 | for (i = 0; i < hur->hur_request.hr_itemcount; i++) { | |
1774 | fid = &hur->hur_user_item[i].hui_fid; | |
1775 | f = search_inode_for_lustre(inode->i_sb, fid); | |
1776 | if (IS_ERR(f)) { | |
1777 | rc = PTR_ERR(f); | |
1778 | break; | |
1779 | } | |
1780 | ||
1781 | rc = ll_hsm_release(f); | |
1782 | iput(f); | |
1783 | if (rc != 0) | |
1784 | break; | |
1785 | } | |
1786 | } else { | |
1787 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize, | |
1788 | hur, NULL); | |
1789 | } | |
d7e09d03 | 1790 | |
e958f49b | 1791 | kvfree(hur); |
d7e09d03 | 1792 | |
0a3bdb00 | 1793 | return rc; |
d7e09d03 PT |
1794 | } |
1795 | case LL_IOC_HSM_PROGRESS: { | |
1796 | struct hsm_progress_kernel hpk; | |
1797 | struct hsm_progress hp; | |
1798 | ||
1799 | if (copy_from_user(&hp, (void *)arg, sizeof(hp))) | |
0a3bdb00 | 1800 | return -EFAULT; |
d7e09d03 PT |
1801 | |
1802 | hpk.hpk_fid = hp.hp_fid; | |
1803 | hpk.hpk_cookie = hp.hp_cookie; | |
1804 | hpk.hpk_extent = hp.hp_extent; | |
1805 | hpk.hpk_flags = hp.hp_flags; | |
1806 | hpk.hpk_errval = hp.hp_errval; | |
1807 | hpk.hpk_data_version = 0; | |
1808 | ||
1809 | /* File may not exist in Lustre; all progress | |
1810 | * reported to Lustre root */ | |
1811 | rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk, | |
1812 | NULL); | |
0a3bdb00 | 1813 | return rc; |
d7e09d03 PT |
1814 | } |
1815 | case LL_IOC_HSM_CT_START: | |
1816 | rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg, | |
1817 | sizeof(struct lustre_kernelcomm)); | |
0a3bdb00 | 1818 | return rc; |
d7e09d03 PT |
1819 | |
1820 | case LL_IOC_HSM_COPY_START: { | |
1821 | struct hsm_copy *copy; | |
1822 | int rc; | |
1823 | ||
4a07594e AH |
1824 | copy = memdup_user((char *)arg, sizeof(*copy)); |
1825 | if (IS_ERR(copy)) | |
1826 | return PTR_ERR(copy); | |
d7e09d03 PT |
1827 | |
1828 | rc = ll_ioc_copy_start(inode->i_sb, copy); | |
1829 | if (copy_to_user((char *)arg, copy, sizeof(*copy))) | |
1830 | rc = -EFAULT; | |
1831 | ||
97903a26 | 1832 | kfree(copy); |
0a3bdb00 | 1833 | return rc; |
d7e09d03 PT |
1834 | } |
1835 | case LL_IOC_HSM_COPY_END: { | |
1836 | struct hsm_copy *copy; | |
1837 | int rc; | |
1838 | ||
4a07594e AH |
1839 | copy = memdup_user((char *)arg, sizeof(*copy)); |
1840 | if (IS_ERR(copy)) | |
1841 | return PTR_ERR(copy); | |
d7e09d03 PT |
1842 | |
1843 | rc = ll_ioc_copy_end(inode->i_sb, copy); | |
1844 | if (copy_to_user((char *)arg, copy, sizeof(*copy))) | |
1845 | rc = -EFAULT; | |
1846 | ||
97903a26 | 1847 | kfree(copy); |
0a3bdb00 | 1848 | return rc; |
d7e09d03 PT |
1849 | } |
1850 | default: | |
e09bee34 OD |
1851 | return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, |
1852 | (void __user *)arg); | |
d7e09d03 PT |
1853 | } |
1854 | } | |
1855 | ||
1856 | static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin) | |
1857 | { | |
1858 | struct inode *inode = file->f_mapping->host; | |
1859 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
1860 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1861 | int api32 = ll_need_32bit_api(sbi); | |
1862 | loff_t ret = -EINVAL; | |
d7e09d03 | 1863 | |
5955102c | 1864 | inode_lock(inode); |
d7e09d03 | 1865 | switch (origin) { |
e17f5594 | 1866 | case SEEK_SET: |
1867 | break; | |
1868 | case SEEK_CUR: | |
1869 | offset += file->f_pos; | |
1870 | break; | |
1871 | case SEEK_END: | |
1872 | if (offset > 0) | |
34e1f2bb | 1873 | goto out; |
e17f5594 | 1874 | if (api32) |
1875 | offset += LL_DIR_END_OFF_32BIT; | |
1876 | else | |
1877 | offset += LL_DIR_END_OFF; | |
1878 | break; | |
1879 | default: | |
1880 | goto out; | |
d7e09d03 PT |
1881 | } |
1882 | ||
1883 | if (offset >= 0 && | |
1884 | ((api32 && offset <= LL_DIR_END_OFF_32BIT) || | |
1885 | (!api32 && offset <= LL_DIR_END_OFF))) { | |
1886 | if (offset != file->f_pos) { | |
1887 | if ((api32 && offset == LL_DIR_END_OFF_32BIT) || | |
1888 | (!api32 && offset == LL_DIR_END_OFF)) | |
1889 | fd->lfd_pos = MDS_DIR_END_OFF; | |
1890 | else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH) | |
1891 | fd->lfd_pos = offset << 32; | |
1892 | else | |
1893 | fd->lfd_pos = offset; | |
1894 | file->f_pos = offset; | |
1895 | file->f_version = 0; | |
1896 | } | |
1897 | ret = offset; | |
1898 | } | |
34e1f2bb | 1899 | goto out; |
d7e09d03 PT |
1900 | |
1901 | out: | |
5955102c | 1902 | inode_unlock(inode); |
d7e09d03 PT |
1903 | return ret; |
1904 | } | |
1905 | ||
2d95f10e | 1906 | static int ll_dir_open(struct inode *inode, struct file *file) |
d7e09d03 | 1907 | { |
0a3bdb00 | 1908 | return ll_file_open(inode, file); |
d7e09d03 PT |
1909 | } |
1910 | ||
2d95f10e | 1911 | static int ll_dir_release(struct inode *inode, struct file *file) |
d7e09d03 | 1912 | { |
0a3bdb00 | 1913 | return ll_file_release(inode, file); |
d7e09d03 PT |
1914 | } |
1915 | ||
2d95f10e | 1916 | const struct file_operations ll_dir_operations = { |
d7e09d03 PT |
1917 | .llseek = ll_dir_seek, |
1918 | .open = ll_dir_open, | |
1919 | .release = ll_dir_release, | |
1920 | .read = generic_read_dir, | |
0b09d381 | 1921 | .iterate = ll_readdir, |
d7e09d03 PT |
1922 | .unlocked_ioctl = ll_dir_ioctl, |
1923 | .fsync = ll_fsync, | |
1924 | }; |