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) 2003, 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/llite_lib.c | |
37 | * | |
38 | * Lustre Light Super operations | |
39 | */ | |
40 | ||
41 | #define DEBUG_SUBSYSTEM S_LLITE | |
42 | ||
43 | #include <linux/module.h> | |
44 | #include <linux/types.h> | |
45 | #include <linux/version.h> | |
46 | #include <linux/mm.h> | |
47 | ||
48 | #include <lustre_lite.h> | |
49 | #include <lustre_ha.h> | |
50 | #include <lustre_dlm.h> | |
51 | #include <lprocfs_status.h> | |
52 | #include <lustre_disk.h> | |
53 | #include <lustre_param.h> | |
54 | #include <lustre_log.h> | |
55 | #include <cl_object.h> | |
56 | #include <obd_cksum.h> | |
57 | #include "llite_internal.h" | |
58 | ||
59 | struct kmem_cache *ll_file_data_slab; | |
60 | ||
61 | LIST_HEAD(ll_super_blocks); | |
62 | DEFINE_SPINLOCK(ll_sb_lock); | |
63 | ||
64 | #ifndef MS_HAS_NEW_AOPS | |
65 | extern struct address_space_operations ll_aops; | |
66 | #else | |
67 | extern struct address_space_operations_ext ll_aops; | |
68 | #endif | |
69 | ||
70 | #ifndef log2 | |
71 | #define log2(n) ffz(~(n)) | |
72 | #endif | |
73 | ||
74 | static struct ll_sb_info *ll_init_sbi(void) | |
75 | { | |
76 | struct ll_sb_info *sbi = NULL; | |
77 | unsigned long pages; | |
78 | unsigned long lru_page_max; | |
79 | struct sysinfo si; | |
80 | class_uuid_t uuid; | |
81 | int i; | |
82 | ENTRY; | |
83 | ||
84 | OBD_ALLOC(sbi, sizeof(*sbi)); | |
85 | if (!sbi) | |
86 | RETURN(NULL); | |
87 | ||
88 | spin_lock_init(&sbi->ll_lock); | |
89 | mutex_init(&sbi->ll_lco.lco_lock); | |
90 | spin_lock_init(&sbi->ll_pp_extent_lock); | |
91 | spin_lock_init(&sbi->ll_process_lock); | |
92 | sbi->ll_rw_stats_on = 0; | |
93 | ||
94 | si_meminfo(&si); | |
95 | pages = si.totalram - si.totalhigh; | |
96 | if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) { | |
97 | lru_page_max = pages / 2; | |
98 | } else { | |
99 | lru_page_max = (pages / 4) * 3; | |
100 | } | |
101 | ||
102 | /* initialize ll_cache data */ | |
103 | atomic_set(&sbi->ll_cache.ccc_users, 0); | |
104 | sbi->ll_cache.ccc_lru_max = lru_page_max; | |
105 | atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max); | |
106 | spin_lock_init(&sbi->ll_cache.ccc_lru_lock); | |
107 | INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru); | |
108 | ||
109 | atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0); | |
110 | init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq); | |
111 | ||
112 | sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32, | |
113 | SBI_DEFAULT_READAHEAD_MAX); | |
114 | sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file; | |
115 | sbi->ll_ra_info.ra_max_read_ahead_whole_pages = | |
116 | SBI_DEFAULT_READAHEAD_WHOLE_MAX; | |
117 | INIT_LIST_HEAD(&sbi->ll_conn_chain); | |
118 | INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list); | |
119 | ||
120 | ll_generate_random_uuid(uuid); | |
121 | class_uuid_unparse(uuid, &sbi->ll_sb_uuid); | |
122 | CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid); | |
123 | ||
124 | spin_lock(&ll_sb_lock); | |
125 | list_add_tail(&sbi->ll_list, &ll_super_blocks); | |
126 | spin_unlock(&ll_sb_lock); | |
127 | ||
128 | sbi->ll_flags |= LL_SBI_VERBOSE; | |
129 | sbi->ll_flags |= LL_SBI_CHECKSUM; | |
130 | ||
131 | sbi->ll_flags |= LL_SBI_LRU_RESIZE; | |
132 | ||
133 | for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) { | |
134 | spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. | |
135 | pp_r_hist.oh_lock); | |
136 | spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. | |
137 | pp_w_hist.oh_lock); | |
138 | } | |
139 | ||
140 | /* metadata statahead is enabled by default */ | |
141 | sbi->ll_sa_max = LL_SA_RPC_DEF; | |
142 | atomic_set(&sbi->ll_sa_total, 0); | |
143 | atomic_set(&sbi->ll_sa_wrong, 0); | |
144 | atomic_set(&sbi->ll_agl_total, 0); | |
145 | sbi->ll_flags |= LL_SBI_AGL_ENABLED; | |
146 | ||
147 | RETURN(sbi); | |
148 | } | |
149 | ||
150 | void ll_free_sbi(struct super_block *sb) | |
151 | { | |
152 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
153 | ENTRY; | |
154 | ||
155 | if (sbi != NULL) { | |
156 | spin_lock(&ll_sb_lock); | |
157 | list_del(&sbi->ll_list); | |
158 | spin_unlock(&ll_sb_lock); | |
159 | OBD_FREE(sbi, sizeof(*sbi)); | |
160 | } | |
161 | EXIT; | |
162 | } | |
163 | ||
164 | static struct dentry_operations ll_d_root_ops = { | |
165 | .d_compare = ll_dcompare, | |
166 | .d_revalidate = ll_revalidate_nd, | |
167 | }; | |
168 | ||
169 | static int client_common_fill_super(struct super_block *sb, char *md, char *dt, | |
170 | struct vfsmount *mnt) | |
171 | { | |
172 | struct inode *root = 0; | |
173 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
174 | struct obd_device *obd; | |
175 | struct obd_capa *oc = NULL; | |
176 | struct obd_statfs *osfs = NULL; | |
177 | struct ptlrpc_request *request = NULL; | |
178 | struct obd_connect_data *data = NULL; | |
179 | struct obd_uuid *uuid; | |
180 | struct md_op_data *op_data; | |
181 | struct lustre_md lmd; | |
182 | obd_valid valid; | |
183 | int size, err, checksum; | |
184 | ENTRY; | |
185 | ||
186 | obd = class_name2obd(md); | |
187 | if (!obd) { | |
188 | CERROR("MD %s: not setup or attached\n", md); | |
189 | RETURN(-EINVAL); | |
190 | } | |
191 | ||
192 | OBD_ALLOC_PTR(data); | |
193 | if (data == NULL) | |
194 | RETURN(-ENOMEM); | |
195 | ||
196 | OBD_ALLOC_PTR(osfs); | |
197 | if (osfs == NULL) { | |
198 | OBD_FREE_PTR(data); | |
199 | RETURN(-ENOMEM); | |
200 | } | |
201 | ||
202 | if (proc_lustre_fs_root) { | |
203 | err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb, | |
204 | dt, md); | |
205 | if (err < 0) | |
206 | CERROR("could not register mount in /proc/fs/lustre\n"); | |
207 | } | |
208 | ||
209 | /* indicate the features supported by this client */ | |
210 | data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH | | |
211 | OBD_CONNECT_ATTRFID | | |
212 | OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE | | |
213 | OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA | | |
214 | OBD_CONNECT_CANCELSET | OBD_CONNECT_FID | | |
215 | OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 | | |
216 | OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR | | |
217 | OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH| | |
218 | OBD_CONNECT_EINPROGRESS | | |
219 | OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE | | |
220 | OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS; | |
221 | ||
222 | if (sbi->ll_flags & LL_SBI_SOM_PREVIEW) | |
223 | data->ocd_connect_flags |= OBD_CONNECT_SOM; | |
224 | ||
225 | if (sbi->ll_flags & LL_SBI_LRU_RESIZE) | |
226 | data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE; | |
227 | #ifdef CONFIG_FS_POSIX_ACL | |
228 | data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK; | |
229 | #endif | |
230 | ||
231 | if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT)) | |
232 | /* flag mdc connection as lightweight, only used for test | |
233 | * purpose, use with care */ | |
234 | data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT; | |
235 | ||
236 | data->ocd_ibits_known = MDS_INODELOCK_FULL; | |
237 | data->ocd_version = LUSTRE_VERSION_CODE; | |
238 | ||
239 | if (sb->s_flags & MS_RDONLY) | |
240 | data->ocd_connect_flags |= OBD_CONNECT_RDONLY; | |
241 | if (sbi->ll_flags & LL_SBI_USER_XATTR) | |
242 | data->ocd_connect_flags |= OBD_CONNECT_XATTR; | |
243 | ||
244 | #ifdef HAVE_MS_FLOCK_LOCK | |
245 | /* force vfs to use lustre handler for flock() calls - bug 10743 */ | |
246 | sb->s_flags |= MS_FLOCK_LOCK; | |
247 | #endif | |
248 | #ifdef MS_HAS_NEW_AOPS | |
249 | sb->s_flags |= MS_HAS_NEW_AOPS; | |
250 | #endif | |
251 | ||
252 | if (sbi->ll_flags & LL_SBI_FLOCK) | |
253 | sbi->ll_fop = &ll_file_operations_flock; | |
254 | else if (sbi->ll_flags & LL_SBI_LOCALFLOCK) | |
255 | sbi->ll_fop = &ll_file_operations; | |
256 | else | |
257 | sbi->ll_fop = &ll_file_operations_noflock; | |
258 | ||
259 | /* real client */ | |
260 | data->ocd_connect_flags |= OBD_CONNECT_REAL; | |
261 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) | |
262 | data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE; | |
263 | ||
264 | data->ocd_brw_size = MD_MAX_BRW_SIZE; | |
265 | ||
266 | err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL); | |
267 | if (err == -EBUSY) { | |
268 | LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing " | |
269 | "recovery, of which this client is not a " | |
270 | "part. Please wait for recovery to complete," | |
271 | " abort, or time out.\n", md); | |
272 | GOTO(out, err); | |
273 | } else if (err) { | |
274 | CERROR("cannot connect to %s: rc = %d\n", md, err); | |
275 | GOTO(out, err); | |
276 | } | |
277 | ||
278 | sbi->ll_md_exp->exp_connect_data = *data; | |
279 | ||
280 | err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp, | |
281 | LUSTRE_SEQ_METADATA); | |
282 | if (err) { | |
283 | CERROR("%s: Can't init metadata layer FID infrastructure, " | |
284 | "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err); | |
285 | GOTO(out_md, err); | |
286 | } | |
287 | ||
288 | /* For mount, we only need fs info from MDT0, and also in DNE, it | |
289 | * can make sure the client can be mounted as long as MDT0 is | |
290 | * avaible */ | |
291 | err = obd_statfs(NULL, sbi->ll_md_exp, osfs, | |
292 | cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), | |
293 | OBD_STATFS_FOR_MDT0); | |
294 | if (err) | |
295 | GOTO(out_md_fid, err); | |
296 | ||
297 | /* This needs to be after statfs to ensure connect has finished. | |
298 | * Note that "data" does NOT contain the valid connect reply. | |
299 | * If connecting to a 1.8 server there will be no LMV device, so | |
300 | * we can access the MDC export directly and exp_connect_flags will | |
301 | * be non-zero, but if accessing an upgraded 2.1 server it will | |
302 | * have the correct flags filled in. | |
303 | * XXX: fill in the LMV exp_connect_flags from MDC(s). */ | |
304 | valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD; | |
305 | if (exp_connect_flags(sbi->ll_md_exp) != 0 && | |
306 | valid != CLIENT_CONNECT_MDT_REQD) { | |
307 | char *buf; | |
308 | ||
309 | OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE); | |
310 | obd_connect_flags2str(buf, PAGE_CACHE_SIZE, | |
311 | valid ^ CLIENT_CONNECT_MDT_REQD, ","); | |
312 | LCONSOLE_ERROR_MSG(0x170, "Server %s does not support " | |
313 | "feature(s) needed for correct operation " | |
314 | "of this client (%s). Please upgrade " | |
315 | "server or downgrade client.\n", | |
316 | sbi->ll_md_exp->exp_obd->obd_name, buf); | |
317 | OBD_FREE(buf, PAGE_CACHE_SIZE); | |
318 | GOTO(out_md_fid, err = -EPROTO); | |
319 | } | |
320 | ||
321 | size = sizeof(*data); | |
322 | err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA), | |
323 | KEY_CONN_DATA, &size, data, NULL); | |
324 | if (err) { | |
325 | CERROR("%s: Get connect data failed: rc = %d\n", | |
326 | sbi->ll_md_exp->exp_obd->obd_name, err); | |
327 | GOTO(out_md_fid, err); | |
328 | } | |
329 | ||
330 | LASSERT(osfs->os_bsize); | |
331 | sb->s_blocksize = osfs->os_bsize; | |
332 | sb->s_blocksize_bits = log2(osfs->os_bsize); | |
333 | sb->s_magic = LL_SUPER_MAGIC; | |
334 | sb->s_maxbytes = MAX_LFS_FILESIZE; | |
335 | sbi->ll_namelen = osfs->os_namelen; | |
336 | sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK; | |
337 | ||
338 | if ((sbi->ll_flags & LL_SBI_USER_XATTR) && | |
339 | !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) { | |
340 | LCONSOLE_INFO("Disabling user_xattr feature because " | |
341 | "it is not supported on the server\n"); | |
342 | sbi->ll_flags &= ~LL_SBI_USER_XATTR; | |
343 | } | |
344 | ||
345 | if (data->ocd_connect_flags & OBD_CONNECT_ACL) { | |
346 | #ifdef MS_POSIXACL | |
347 | sb->s_flags |= MS_POSIXACL; | |
348 | #endif | |
349 | sbi->ll_flags |= LL_SBI_ACL; | |
350 | } else { | |
351 | LCONSOLE_INFO("client wants to enable acl, but mdt not!\n"); | |
352 | #ifdef MS_POSIXACL | |
353 | sb->s_flags &= ~MS_POSIXACL; | |
354 | #endif | |
355 | sbi->ll_flags &= ~LL_SBI_ACL; | |
356 | } | |
357 | ||
358 | if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) { | |
359 | if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) { | |
360 | sbi->ll_flags |= LL_SBI_RMT_CLIENT; | |
361 | LCONSOLE_INFO("client is set as remote by default.\n"); | |
362 | } | |
363 | } else { | |
364 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
365 | sbi->ll_flags &= ~LL_SBI_RMT_CLIENT; | |
366 | LCONSOLE_INFO("client claims to be remote, but server " | |
367 | "rejected, forced to be local.\n"); | |
368 | } | |
369 | } | |
370 | ||
371 | if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) { | |
372 | LCONSOLE_INFO("client enabled MDS capability!\n"); | |
373 | sbi->ll_flags |= LL_SBI_MDS_CAPA; | |
374 | } | |
375 | ||
376 | if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) { | |
377 | LCONSOLE_INFO("client enabled OSS capability!\n"); | |
378 | sbi->ll_flags |= LL_SBI_OSS_CAPA; | |
379 | } | |
380 | ||
381 | if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH) | |
382 | sbi->ll_flags |= LL_SBI_64BIT_HASH; | |
383 | ||
384 | if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) | |
385 | sbi->ll_md_brw_size = data->ocd_brw_size; | |
386 | else | |
387 | sbi->ll_md_brw_size = PAGE_CACHE_SIZE; | |
388 | ||
389 | if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) { | |
390 | LCONSOLE_INFO("Layout lock feature supported.\n"); | |
391 | sbi->ll_flags |= LL_SBI_LAYOUT_LOCK; | |
392 | } | |
393 | ||
394 | obd = class_name2obd(dt); | |
395 | if (!obd) { | |
396 | CERROR("DT %s: not setup or attached\n", dt); | |
397 | GOTO(out_md_fid, err = -ENODEV); | |
398 | } | |
399 | ||
400 | data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION | | |
401 | OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE | | |
402 | OBD_CONNECT_CANCELSET | OBD_CONNECT_FID | | |
403 | OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK| | |
404 | OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT | | |
405 | OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR| | |
406 | OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH | | |
407 | OBD_CONNECT_MAXBYTES | | |
408 | OBD_CONNECT_EINPROGRESS | | |
409 | OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE | | |
410 | OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS; | |
411 | ||
412 | if (sbi->ll_flags & LL_SBI_SOM_PREVIEW) | |
413 | data->ocd_connect_flags |= OBD_CONNECT_SOM; | |
414 | ||
415 | if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) { | |
416 | /* OBD_CONNECT_CKSUM should always be set, even if checksums are | |
417 | * disabled by default, because it can still be enabled on the | |
418 | * fly via /proc. As a consequence, we still need to come to an | |
419 | * agreement on the supported algorithms at connect time */ | |
420 | data->ocd_connect_flags |= OBD_CONNECT_CKSUM; | |
421 | ||
422 | if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY)) | |
423 | data->ocd_cksum_types = OBD_CKSUM_ADLER; | |
424 | else | |
425 | data->ocd_cksum_types = cksum_types_supported_client(); | |
426 | } | |
427 | ||
428 | data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE; | |
429 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) | |
430 | data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE; | |
431 | ||
432 | CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d " | |
433 | "ocd_grant: %d\n", data->ocd_connect_flags, | |
434 | data->ocd_version, data->ocd_grant); | |
435 | ||
436 | obd->obd_upcall.onu_owner = &sbi->ll_lco; | |
437 | obd->obd_upcall.onu_upcall = cl_ocd_update; | |
438 | ||
439 | data->ocd_brw_size = DT_MAX_BRW_SIZE; | |
440 | ||
441 | err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, | |
442 | NULL); | |
443 | if (err == -EBUSY) { | |
444 | LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing " | |
445 | "recovery, of which this client is not a " | |
446 | "part. Please wait for recovery to " | |
447 | "complete, abort, or time out.\n", dt); | |
448 | GOTO(out_md, err); | |
449 | } else if (err) { | |
450 | CERROR("%s: Cannot connect to %s: rc = %d\n", | |
451 | sbi->ll_dt_exp->exp_obd->obd_name, dt, err); | |
452 | GOTO(out_md, err); | |
453 | } | |
454 | ||
455 | sbi->ll_dt_exp->exp_connect_data = *data; | |
456 | ||
457 | err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp, | |
458 | LUSTRE_SEQ_METADATA); | |
459 | if (err) { | |
460 | CERROR("%s: Can't init data layer FID infrastructure, " | |
461 | "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err); | |
462 | GOTO(out_dt, err); | |
463 | } | |
464 | ||
465 | mutex_lock(&sbi->ll_lco.lco_lock); | |
466 | sbi->ll_lco.lco_flags = data->ocd_connect_flags; | |
467 | sbi->ll_lco.lco_md_exp = sbi->ll_md_exp; | |
468 | sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp; | |
469 | mutex_unlock(&sbi->ll_lco.lco_lock); | |
470 | ||
471 | fid_zero(&sbi->ll_root_fid); | |
472 | err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc); | |
473 | if (err) { | |
474 | CERROR("cannot mds_connect: rc = %d\n", err); | |
475 | GOTO(out_lock_cn_cb, err); | |
476 | } | |
477 | if (!fid_is_sane(&sbi->ll_root_fid)) { | |
478 | CERROR("%s: Invalid root fid "DFID" during mount\n", | |
479 | sbi->ll_md_exp->exp_obd->obd_name, | |
480 | PFID(&sbi->ll_root_fid)); | |
481 | GOTO(out_lock_cn_cb, err = -EINVAL); | |
482 | } | |
483 | CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid)); | |
484 | ||
485 | sb->s_op = &lustre_super_operations; | |
486 | #if THREAD_SIZE >= 8192 /*b=17630*/ | |
487 | sb->s_export_op = &lustre_export_operations; | |
488 | #endif | |
489 | ||
490 | /* make root inode | |
491 | * XXX: move this to after cbd setup? */ | |
492 | valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA; | |
493 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) | |
494 | valid |= OBD_MD_FLRMTPERM; | |
495 | else if (sbi->ll_flags & LL_SBI_ACL) | |
496 | valid |= OBD_MD_FLACL; | |
497 | ||
498 | OBD_ALLOC_PTR(op_data); | |
499 | if (op_data == NULL) | |
500 | GOTO(out_lock_cn_cb, err = -ENOMEM); | |
501 | ||
502 | op_data->op_fid1 = sbi->ll_root_fid; | |
503 | op_data->op_mode = 0; | |
504 | op_data->op_capa1 = oc; | |
505 | op_data->op_valid = valid; | |
506 | ||
507 | err = md_getattr(sbi->ll_md_exp, op_data, &request); | |
508 | if (oc) | |
509 | capa_put(oc); | |
510 | OBD_FREE_PTR(op_data); | |
511 | if (err) { | |
512 | CERROR("%s: md_getattr failed for root: rc = %d\n", | |
513 | sbi->ll_md_exp->exp_obd->obd_name, err); | |
514 | GOTO(out_lock_cn_cb, err); | |
515 | } | |
516 | ||
517 | err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp, | |
518 | sbi->ll_md_exp, &lmd); | |
519 | if (err) { | |
520 | CERROR("failed to understand root inode md: rc = %d\n", err); | |
521 | ptlrpc_req_finished(request); | |
522 | GOTO(out_lock_cn_cb, err); | |
523 | } | |
524 | ||
525 | LASSERT(fid_is_sane(&sbi->ll_root_fid)); | |
526 | root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, | |
527 | ll_need_32bit_api(sbi)), | |
528 | &lmd); | |
529 | md_free_lustre_md(sbi->ll_md_exp, &lmd); | |
530 | ptlrpc_req_finished(request); | |
531 | ||
532 | if (root == NULL || IS_ERR(root)) { | |
533 | if (lmd.lsm) | |
534 | obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm); | |
535 | #ifdef CONFIG_FS_POSIX_ACL | |
536 | if (lmd.posix_acl) { | |
537 | posix_acl_release(lmd.posix_acl); | |
538 | lmd.posix_acl = NULL; | |
539 | } | |
540 | #endif | |
541 | err = IS_ERR(root) ? PTR_ERR(root) : -EBADF; | |
542 | root = NULL; | |
543 | CERROR("lustre_lite: bad iget4 for root\n"); | |
544 | GOTO(out_root, err); | |
545 | } | |
546 | ||
547 | err = ll_close_thread_start(&sbi->ll_lcq); | |
548 | if (err) { | |
549 | CERROR("cannot start close thread: rc %d\n", err); | |
550 | GOTO(out_root, err); | |
551 | } | |
552 | ||
553 | #ifdef CONFIG_FS_POSIX_ACL | |
554 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
555 | rct_init(&sbi->ll_rct); | |
556 | et_init(&sbi->ll_et); | |
557 | } | |
558 | #endif | |
559 | ||
560 | checksum = sbi->ll_flags & LL_SBI_CHECKSUM; | |
561 | err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM), | |
562 | KEY_CHECKSUM, sizeof(checksum), &checksum, | |
563 | NULL); | |
564 | cl_sb_init(sb); | |
565 | ||
566 | err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET), | |
567 | KEY_CACHE_SET, sizeof(sbi->ll_cache), | |
568 | &sbi->ll_cache, NULL); | |
569 | ||
570 | sb->s_root = d_make_root(root); | |
571 | if (sb->s_root == NULL) { | |
572 | CERROR("%s: can't make root dentry\n", | |
573 | ll_get_fsname(sb, NULL, 0)); | |
574 | GOTO(out_root, err = -ENOMEM); | |
575 | } | |
576 | ||
577 | /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */ | |
578 | d_set_d_op(sb->s_root, &ll_d_root_ops); | |
579 | sb->s_d_op = &ll_d_ops; | |
580 | ||
581 | sbi->ll_sdev_orig = sb->s_dev; | |
582 | ||
583 | /* We set sb->s_dev equal on all lustre clients in order to support | |
584 | * NFS export clustering. NFSD requires that the FSID be the same | |
585 | * on all clients. */ | |
586 | /* s_dev is also used in lt_compare() to compare two fs, but that is | |
587 | * only a node-local comparison. */ | |
588 | uuid = obd_get_uuid(sbi->ll_md_exp); | |
589 | if (uuid != NULL) | |
590 | sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid)); | |
591 | ||
592 | if (data != NULL) | |
593 | OBD_FREE_PTR(data); | |
594 | if (osfs != NULL) | |
595 | OBD_FREE_PTR(osfs); | |
596 | ||
597 | RETURN(err); | |
598 | out_root: | |
599 | if (root) | |
600 | iput(root); | |
601 | out_lock_cn_cb: | |
602 | obd_fid_fini(sbi->ll_dt_exp->exp_obd); | |
603 | out_dt: | |
604 | obd_disconnect(sbi->ll_dt_exp); | |
605 | sbi->ll_dt_exp = NULL; | |
606 | /* Make sure all OScs are gone, since cl_cache is accessing sbi. */ | |
607 | obd_zombie_barrier(); | |
608 | out_md_fid: | |
609 | obd_fid_fini(sbi->ll_md_exp->exp_obd); | |
610 | out_md: | |
611 | obd_disconnect(sbi->ll_md_exp); | |
612 | sbi->ll_md_exp = NULL; | |
613 | out: | |
614 | if (data != NULL) | |
615 | OBD_FREE_PTR(data); | |
616 | if (osfs != NULL) | |
617 | OBD_FREE_PTR(osfs); | |
618 | lprocfs_unregister_mountpoint(sbi); | |
619 | return err; | |
620 | } | |
621 | ||
622 | int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize) | |
623 | { | |
624 | int size, rc; | |
625 | ||
626 | *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL); | |
627 | size = sizeof(int); | |
628 | rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE), | |
629 | KEY_MAX_EASIZE, &size, lmmsize, NULL); | |
630 | if (rc) | |
631 | CERROR("Get max mdsize error rc %d \n", rc); | |
632 | ||
633 | RETURN(rc); | |
634 | } | |
635 | ||
636 | void ll_dump_inode(struct inode *inode) | |
637 | { | |
638 | struct ll_d_hlist_node *tmp; | |
639 | int dentry_count = 0; | |
640 | ||
641 | LASSERT(inode != NULL); | |
642 | ||
643 | ll_d_hlist_for_each(tmp, &inode->i_dentry) | |
644 | dentry_count++; | |
645 | ||
646 | CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n", | |
647 | inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino, | |
648 | inode->i_mode, atomic_read(&inode->i_count), dentry_count); | |
649 | } | |
650 | ||
651 | void lustre_dump_dentry(struct dentry *dentry, int recur) | |
652 | { | |
653 | struct list_head *tmp; | |
654 | int subdirs = 0; | |
655 | ||
656 | LASSERT(dentry != NULL); | |
657 | ||
658 | list_for_each(tmp, &dentry->d_subdirs) | |
659 | subdirs++; | |
660 | ||
661 | CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u," | |
662 | " flags=0x%x, fsdata=%p, %d subdirs\n", dentry, | |
663 | dentry->d_name.len, dentry->d_name.name, | |
664 | dentry->d_parent->d_name.len, dentry->d_parent->d_name.name, | |
665 | dentry->d_parent, dentry->d_inode, d_refcount(dentry), | |
666 | dentry->d_flags, dentry->d_fsdata, subdirs); | |
667 | if (dentry->d_inode != NULL) | |
668 | ll_dump_inode(dentry->d_inode); | |
669 | ||
670 | if (recur == 0) | |
671 | return; | |
672 | ||
673 | list_for_each(tmp, &dentry->d_subdirs) { | |
674 | struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child); | |
675 | lustre_dump_dentry(d, recur - 1); | |
676 | } | |
677 | } | |
678 | ||
679 | void client_common_put_super(struct super_block *sb) | |
680 | { | |
681 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
682 | ENTRY; | |
683 | ||
684 | #ifdef CONFIG_FS_POSIX_ACL | |
685 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
686 | et_fini(&sbi->ll_et); | |
687 | rct_fini(&sbi->ll_rct); | |
688 | } | |
689 | #endif | |
690 | ||
691 | ll_close_thread_shutdown(sbi->ll_lcq); | |
692 | ||
693 | cl_sb_fini(sb); | |
694 | ||
695 | list_del(&sbi->ll_conn_chain); | |
696 | ||
697 | obd_fid_fini(sbi->ll_dt_exp->exp_obd); | |
698 | obd_disconnect(sbi->ll_dt_exp); | |
699 | sbi->ll_dt_exp = NULL; | |
700 | /* wait till all OSCs are gone, since cl_cache is accessing sbi. | |
701 | * see LU-2543. */ | |
702 | obd_zombie_barrier(); | |
703 | ||
704 | lprocfs_unregister_mountpoint(sbi); | |
705 | ||
706 | obd_fid_fini(sbi->ll_md_exp->exp_obd); | |
707 | obd_disconnect(sbi->ll_md_exp); | |
708 | sbi->ll_md_exp = NULL; | |
709 | ||
710 | EXIT; | |
711 | } | |
712 | ||
713 | void ll_kill_super(struct super_block *sb) | |
714 | { | |
715 | struct ll_sb_info *sbi; | |
716 | ||
717 | ENTRY; | |
718 | ||
719 | /* not init sb ?*/ | |
720 | if (!(sb->s_flags & MS_ACTIVE)) | |
721 | return; | |
722 | ||
723 | sbi = ll_s2sbi(sb); | |
724 | /* we need restore s_dev from changed for clustred NFS before put_super | |
725 | * because new kernels have cached s_dev and change sb->s_dev in | |
726 | * put_super not affected real removing devices */ | |
65fb55d1 | 727 | if (sbi) { |
d7e09d03 | 728 | sb->s_dev = sbi->ll_sdev_orig; |
65fb55d1 NY |
729 | sbi->ll_umounting = 1; |
730 | } | |
d7e09d03 PT |
731 | EXIT; |
732 | } | |
733 | ||
734 | char *ll_read_opt(const char *opt, char *data) | |
735 | { | |
736 | char *value; | |
737 | char *retval; | |
738 | ENTRY; | |
739 | ||
740 | CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data); | |
741 | if (strncmp(opt, data, strlen(opt))) | |
742 | RETURN(NULL); | |
743 | if ((value = strchr(data, '=')) == NULL) | |
744 | RETURN(NULL); | |
745 | ||
746 | value++; | |
747 | OBD_ALLOC(retval, strlen(value) + 1); | |
748 | if (!retval) { | |
749 | CERROR("out of memory!\n"); | |
750 | RETURN(NULL); | |
751 | } | |
752 | ||
753 | memcpy(retval, value, strlen(value)+1); | |
754 | CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval); | |
755 | RETURN(retval); | |
756 | } | |
757 | ||
758 | static inline int ll_set_opt(const char *opt, char *data, int fl) | |
759 | { | |
760 | if (strncmp(opt, data, strlen(opt)) != 0) | |
761 | return(0); | |
762 | else | |
763 | return(fl); | |
764 | } | |
765 | ||
766 | /* non-client-specific mount options are parsed in lmd_parse */ | |
767 | static int ll_options(char *options, int *flags) | |
768 | { | |
769 | int tmp; | |
770 | char *s1 = options, *s2; | |
771 | ENTRY; | |
772 | ||
773 | if (!options) | |
774 | RETURN(0); | |
775 | ||
776 | CDEBUG(D_CONFIG, "Parsing opts %s\n", options); | |
777 | ||
778 | while (*s1) { | |
779 | CDEBUG(D_SUPER, "next opt=%s\n", s1); | |
780 | tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK); | |
781 | if (tmp) { | |
782 | *flags |= tmp; | |
783 | goto next; | |
784 | } | |
785 | tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK); | |
786 | if (tmp) { | |
787 | *flags |= tmp; | |
788 | goto next; | |
789 | } | |
790 | tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK); | |
791 | if (tmp) { | |
792 | *flags |= tmp; | |
793 | goto next; | |
794 | } | |
795 | tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK); | |
796 | if (tmp) { | |
797 | *flags &= ~tmp; | |
798 | goto next; | |
799 | } | |
800 | tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR); | |
801 | if (tmp) { | |
802 | *flags |= tmp; | |
803 | goto next; | |
804 | } | |
805 | tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR); | |
806 | if (tmp) { | |
807 | *flags &= ~tmp; | |
808 | goto next; | |
809 | } | |
810 | #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0) | |
811 | tmp = ll_set_opt("acl", s1, LL_SBI_ACL); | |
812 | if (tmp) { | |
813 | /* Ignore deprecated mount option. The client will | |
814 | * always try to mount with ACL support, whether this | |
815 | * is used depends on whether server supports it. */ | |
816 | LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated " | |
817 | "mount option 'acl'.\n"); | |
818 | goto next; | |
819 | } | |
820 | tmp = ll_set_opt("noacl", s1, LL_SBI_ACL); | |
821 | if (tmp) { | |
822 | LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated " | |
823 | "mount option 'noacl'.\n"); | |
824 | goto next; | |
825 | } | |
826 | #else | |
827 | #warning "{no}acl options have been deprecated since 1.8, please remove them" | |
828 | #endif | |
829 | tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT); | |
830 | if (tmp) { | |
831 | *flags |= tmp; | |
832 | goto next; | |
833 | } | |
834 | tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH); | |
835 | if (tmp) { | |
836 | *flags |= tmp; | |
837 | goto next; | |
838 | } | |
839 | tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH); | |
840 | if (tmp) { | |
841 | *flags &= ~tmp; | |
842 | goto next; | |
843 | } | |
844 | ||
845 | tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM); | |
846 | if (tmp) { | |
847 | *flags |= tmp; | |
848 | goto next; | |
849 | } | |
850 | tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM); | |
851 | if (tmp) { | |
852 | *flags &= ~tmp; | |
853 | goto next; | |
854 | } | |
855 | tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE); | |
856 | if (tmp) { | |
857 | *flags |= tmp; | |
858 | goto next; | |
859 | } | |
860 | tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE); | |
861 | if (tmp) { | |
862 | *flags &= ~tmp; | |
863 | goto next; | |
864 | } | |
865 | tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS); | |
866 | if (tmp) { | |
867 | *flags |= tmp; | |
868 | goto next; | |
869 | } | |
870 | tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS); | |
871 | if (tmp) { | |
872 | *flags &= ~tmp; | |
873 | goto next; | |
874 | } | |
875 | tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW); | |
876 | if (tmp) { | |
877 | *flags |= tmp; | |
878 | goto next; | |
879 | } | |
880 | tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API); | |
881 | if (tmp) { | |
882 | *flags |= tmp; | |
883 | goto next; | |
884 | } | |
885 | tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE); | |
886 | if (tmp) { | |
887 | *flags |= tmp; | |
888 | goto next; | |
889 | } | |
890 | tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE); | |
891 | if (tmp) { | |
892 | *flags &= ~tmp; | |
893 | goto next; | |
894 | } | |
895 | LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n", | |
896 | s1); | |
897 | RETURN(-EINVAL); | |
898 | ||
899 | next: | |
900 | /* Find next opt */ | |
901 | s2 = strchr(s1, ','); | |
902 | if (s2 == NULL) | |
903 | break; | |
904 | s1 = s2 + 1; | |
905 | } | |
906 | RETURN(0); | |
907 | } | |
908 | ||
909 | void ll_lli_init(struct ll_inode_info *lli) | |
910 | { | |
911 | lli->lli_inode_magic = LLI_INODE_MAGIC; | |
912 | lli->lli_flags = 0; | |
913 | lli->lli_ioepoch = 0; | |
914 | lli->lli_maxbytes = MAX_LFS_FILESIZE; | |
915 | spin_lock_init(&lli->lli_lock); | |
916 | lli->lli_posix_acl = NULL; | |
917 | lli->lli_remote_perms = NULL; | |
918 | mutex_init(&lli->lli_rmtperm_mutex); | |
919 | /* Do not set lli_fid, it has been initialized already. */ | |
920 | fid_zero(&lli->lli_pfid); | |
921 | INIT_LIST_HEAD(&lli->lli_close_list); | |
922 | INIT_LIST_HEAD(&lli->lli_oss_capas); | |
923 | atomic_set(&lli->lli_open_count, 0); | |
924 | lli->lli_mds_capa = NULL; | |
925 | lli->lli_rmtperm_time = 0; | |
926 | lli->lli_pending_och = NULL; | |
927 | lli->lli_mds_read_och = NULL; | |
928 | lli->lli_mds_write_och = NULL; | |
929 | lli->lli_mds_exec_och = NULL; | |
930 | lli->lli_open_fd_read_count = 0; | |
931 | lli->lli_open_fd_write_count = 0; | |
932 | lli->lli_open_fd_exec_count = 0; | |
933 | mutex_init(&lli->lli_och_mutex); | |
934 | spin_lock_init(&lli->lli_agl_lock); | |
935 | lli->lli_has_smd = false; | |
936 | lli->lli_layout_gen = LL_LAYOUT_GEN_NONE; | |
937 | lli->lli_clob = NULL; | |
938 | ||
939 | LASSERT(lli->lli_vfs_inode.i_mode != 0); | |
940 | if (S_ISDIR(lli->lli_vfs_inode.i_mode)) { | |
941 | mutex_init(&lli->lli_readdir_mutex); | |
942 | lli->lli_opendir_key = NULL; | |
943 | lli->lli_sai = NULL; | |
944 | lli->lli_def_acl = NULL; | |
945 | spin_lock_init(&lli->lli_sa_lock); | |
946 | lli->lli_opendir_pid = 0; | |
947 | } else { | |
948 | sema_init(&lli->lli_size_sem, 1); | |
949 | lli->lli_size_sem_owner = NULL; | |
950 | lli->lli_symlink_name = NULL; | |
951 | init_rwsem(&lli->lli_trunc_sem); | |
952 | mutex_init(&lli->lli_write_mutex); | |
953 | init_rwsem(&lli->lli_glimpse_sem); | |
954 | lli->lli_glimpse_time = 0; | |
955 | INIT_LIST_HEAD(&lli->lli_agl_list); | |
956 | lli->lli_agl_index = 0; | |
957 | lli->lli_async_rc = 0; | |
958 | lli->lli_volatile = false; | |
959 | } | |
960 | mutex_init(&lli->lli_layout_mutex); | |
961 | } | |
962 | ||
963 | static inline int ll_bdi_register(struct backing_dev_info *bdi) | |
964 | { | |
965 | static atomic_t ll_bdi_num = ATOMIC_INIT(0); | |
966 | ||
967 | bdi->name = "lustre"; | |
968 | return bdi_register(bdi, NULL, "lustre-%d", | |
969 | atomic_inc_return(&ll_bdi_num)); | |
970 | } | |
971 | ||
972 | int ll_fill_super(struct super_block *sb, struct vfsmount *mnt) | |
973 | { | |
974 | struct lustre_profile *lprof = NULL; | |
975 | struct lustre_sb_info *lsi = s2lsi(sb); | |
976 | struct ll_sb_info *sbi; | |
977 | char *dt = NULL, *md = NULL; | |
978 | char *profilenm = get_profile_name(sb); | |
979 | struct config_llog_instance *cfg; | |
980 | /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */ | |
981 | const int instlen = sizeof(cfg->cfg_instance) * 2 + 2; | |
982 | int err; | |
983 | ENTRY; | |
984 | ||
985 | CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb); | |
986 | ||
987 | OBD_ALLOC_PTR(cfg); | |
988 | if (cfg == NULL) | |
989 | RETURN(-ENOMEM); | |
990 | ||
991 | try_module_get(THIS_MODULE); | |
992 | ||
993 | /* client additional sb info */ | |
994 | lsi->lsi_llsbi = sbi = ll_init_sbi(); | |
995 | if (!sbi) { | |
996 | module_put(THIS_MODULE); | |
997 | OBD_FREE_PTR(cfg); | |
998 | RETURN(-ENOMEM); | |
999 | } | |
1000 | ||
1001 | err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags); | |
1002 | if (err) | |
1003 | GOTO(out_free, err); | |
1004 | ||
1005 | err = bdi_init(&lsi->lsi_bdi); | |
1006 | if (err) | |
1007 | GOTO(out_free, err); | |
1008 | lsi->lsi_flags |= LSI_BDI_INITIALIZED; | |
1009 | lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY; | |
1010 | err = ll_bdi_register(&lsi->lsi_bdi); | |
1011 | if (err) | |
1012 | GOTO(out_free, err); | |
1013 | ||
1014 | sb->s_bdi = &lsi->lsi_bdi; | |
1015 | ||
1016 | /* Generate a string unique to this super, in case some joker tries | |
1017 | to mount the same fs at two mount points. | |
1018 | Use the address of the super itself.*/ | |
1019 | cfg->cfg_instance = sb; | |
1020 | cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid; | |
1021 | cfg->cfg_callback = class_config_llog_handler; | |
1022 | /* set up client obds */ | |
1023 | err = lustre_process_log(sb, profilenm, cfg); | |
1024 | if (err < 0) { | |
1025 | CERROR("Unable to process log: %d\n", err); | |
1026 | GOTO(out_free, err); | |
1027 | } | |
1028 | ||
1029 | /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */ | |
1030 | lprof = class_get_profile(profilenm); | |
1031 | if (lprof == NULL) { | |
1032 | LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be" | |
1033 | " read from the MGS. Does that filesystem " | |
1034 | "exist?\n", profilenm); | |
1035 | GOTO(out_free, err = -EINVAL); | |
1036 | } | |
1037 | CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm, | |
1038 | lprof->lp_md, lprof->lp_dt); | |
1039 | ||
1040 | OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2); | |
1041 | if (!dt) | |
1042 | GOTO(out_free, err = -ENOMEM); | |
1043 | sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance); | |
1044 | ||
1045 | OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2); | |
1046 | if (!md) | |
1047 | GOTO(out_free, err = -ENOMEM); | |
1048 | sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance); | |
1049 | ||
1050 | /* connections, registrations, sb setup */ | |
1051 | err = client_common_fill_super(sb, md, dt, mnt); | |
1052 | ||
1053 | out_free: | |
1054 | if (md) | |
1055 | OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2); | |
1056 | if (dt) | |
1057 | OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2); | |
1058 | if (err) | |
1059 | ll_put_super(sb); | |
1060 | else if (sbi->ll_flags & LL_SBI_VERBOSE) | |
1061 | LCONSOLE_WARN("Mounted %s\n", profilenm); | |
1062 | ||
1063 | OBD_FREE_PTR(cfg); | |
1064 | RETURN(err); | |
1065 | } /* ll_fill_super */ | |
1066 | ||
1067 | ||
1068 | void lu_context_keys_dump(void); | |
1069 | ||
1070 | void ll_put_super(struct super_block *sb) | |
1071 | { | |
1072 | struct config_llog_instance cfg; | |
1073 | struct obd_device *obd; | |
1074 | struct lustre_sb_info *lsi = s2lsi(sb); | |
1075 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
1076 | char *profilenm = get_profile_name(sb); | |
1077 | int ccc_count, next, force = 1, rc = 0; | |
1078 | ENTRY; | |
1079 | ||
1080 | CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm); | |
1081 | ||
1082 | ll_print_capa_stat(sbi); | |
1083 | ||
1084 | cfg.cfg_instance = sb; | |
1085 | lustre_end_log(sb, profilenm, &cfg); | |
1086 | ||
1087 | if (sbi->ll_md_exp) { | |
1088 | obd = class_exp2obd(sbi->ll_md_exp); | |
1089 | if (obd) | |
1090 | force = obd->obd_force; | |
1091 | } | |
1092 | ||
1093 | /* Wait for unstable pages to be committed to stable storage */ | |
1094 | if (force == 0) { | |
1095 | struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL); | |
1096 | rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq, | |
1097 | atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0, | |
1098 | &lwi); | |
1099 | } | |
1100 | ||
1101 | ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr); | |
1102 | if (force == 0 && rc != -EINTR) | |
1103 | LASSERTF(ccc_count == 0, "count: %i\n", ccc_count); | |
1104 | ||
1105 | ||
1106 | /* We need to set force before the lov_disconnect in | |
1107 | lustre_common_put_super, since l_d cleans up osc's as well. */ | |
1108 | if (force) { | |
1109 | next = 0; | |
1110 | while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, | |
1111 | &next)) != NULL) { | |
1112 | obd->obd_force = force; | |
1113 | } | |
1114 | } | |
1115 | ||
1116 | if (sbi->ll_lcq) { | |
1117 | /* Only if client_common_fill_super succeeded */ | |
1118 | client_common_put_super(sb); | |
1119 | } | |
1120 | ||
1121 | next = 0; | |
1122 | while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) { | |
1123 | class_manual_cleanup(obd); | |
1124 | } | |
1125 | ||
1126 | if (sbi->ll_flags & LL_SBI_VERBOSE) | |
1127 | LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : ""); | |
1128 | ||
1129 | if (profilenm) | |
1130 | class_del_profile(profilenm); | |
1131 | ||
1132 | if (lsi->lsi_flags & LSI_BDI_INITIALIZED) { | |
1133 | bdi_destroy(&lsi->lsi_bdi); | |
1134 | lsi->lsi_flags &= ~LSI_BDI_INITIALIZED; | |
1135 | } | |
1136 | ||
1137 | ll_free_sbi(sb); | |
1138 | lsi->lsi_llsbi = NULL; | |
1139 | ||
1140 | lustre_common_put_super(sb); | |
1141 | ||
1142 | module_put(THIS_MODULE); | |
1143 | ||
1144 | EXIT; | |
1145 | } /* client_put_super */ | |
1146 | ||
1147 | struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock) | |
1148 | { | |
1149 | struct inode *inode = NULL; | |
1150 | ||
1151 | /* NOTE: we depend on atomic igrab() -bzzz */ | |
1152 | lock_res_and_lock(lock); | |
1153 | if (lock->l_resource->lr_lvb_inode) { | |
1154 | struct ll_inode_info * lli; | |
1155 | lli = ll_i2info(lock->l_resource->lr_lvb_inode); | |
1156 | if (lli->lli_inode_magic == LLI_INODE_MAGIC) { | |
1157 | inode = igrab(lock->l_resource->lr_lvb_inode); | |
1158 | } else { | |
1159 | inode = lock->l_resource->lr_lvb_inode; | |
1160 | LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO : | |
1161 | D_WARNING, lock, "lr_lvb_inode %p is " | |
1162 | "bogus: magic %08x", | |
1163 | lock->l_resource->lr_lvb_inode, | |
1164 | lli->lli_inode_magic); | |
1165 | inode = NULL; | |
1166 | } | |
1167 | } | |
1168 | unlock_res_and_lock(lock); | |
1169 | return inode; | |
1170 | } | |
1171 | ||
1172 | struct inode *ll_inode_from_lock(struct ldlm_lock *lock) | |
1173 | { | |
1174 | struct inode *inode = NULL; | |
1175 | /* NOTE: we depend on atomic igrab() -bzzz */ | |
1176 | lock_res_and_lock(lock); | |
1177 | if (lock->l_ast_data) { | |
1178 | struct ll_inode_info *lli = ll_i2info(lock->l_ast_data); | |
1179 | if (lli->lli_inode_magic == LLI_INODE_MAGIC) { | |
1180 | inode = igrab(lock->l_ast_data); | |
1181 | } else { | |
1182 | inode = lock->l_ast_data; | |
1183 | LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO : | |
1184 | D_WARNING, lock, "l_ast_data %p is " | |
1185 | "bogus: magic %08x", lock->l_ast_data, | |
1186 | lli->lli_inode_magic); | |
1187 | inode = NULL; | |
1188 | } | |
1189 | } | |
1190 | unlock_res_and_lock(lock); | |
1191 | return inode; | |
1192 | } | |
1193 | ||
1194 | void ll_clear_inode(struct inode *inode) | |
1195 | { | |
1196 | struct ll_inode_info *lli = ll_i2info(inode); | |
1197 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1198 | ENTRY; | |
1199 | ||
1200 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, | |
1201 | inode->i_generation, inode); | |
1202 | ||
1203 | if (S_ISDIR(inode->i_mode)) { | |
1204 | /* these should have been cleared in ll_file_release */ | |
1205 | LASSERT(lli->lli_opendir_key == NULL); | |
1206 | LASSERT(lli->lli_sai == NULL); | |
1207 | LASSERT(lli->lli_opendir_pid == 0); | |
1208 | } | |
1209 | ||
1210 | ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK; | |
1211 | md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode)); | |
1212 | ||
1213 | LASSERT(!lli->lli_open_fd_write_count); | |
1214 | LASSERT(!lli->lli_open_fd_read_count); | |
1215 | LASSERT(!lli->lli_open_fd_exec_count); | |
1216 | ||
1217 | if (lli->lli_mds_write_och) | |
1218 | ll_md_real_close(inode, FMODE_WRITE); | |
1219 | if (lli->lli_mds_exec_och) | |
1220 | ll_md_real_close(inode, FMODE_EXEC); | |
1221 | if (lli->lli_mds_read_och) | |
1222 | ll_md_real_close(inode, FMODE_READ); | |
1223 | ||
1224 | if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) { | |
1225 | OBD_FREE(lli->lli_symlink_name, | |
1226 | strlen(lli->lli_symlink_name) + 1); | |
1227 | lli->lli_symlink_name = NULL; | |
1228 | } | |
1229 | ||
1230 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
1231 | LASSERT(lli->lli_posix_acl == NULL); | |
1232 | if (lli->lli_remote_perms) { | |
1233 | free_rmtperm_hash(lli->lli_remote_perms); | |
1234 | lli->lli_remote_perms = NULL; | |
1235 | } | |
1236 | } | |
1237 | #ifdef CONFIG_FS_POSIX_ACL | |
1238 | else if (lli->lli_posix_acl) { | |
1239 | LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1); | |
1240 | LASSERT(lli->lli_remote_perms == NULL); | |
1241 | posix_acl_release(lli->lli_posix_acl); | |
1242 | lli->lli_posix_acl = NULL; | |
1243 | } | |
1244 | #endif | |
1245 | lli->lli_inode_magic = LLI_INODE_DEAD; | |
1246 | ||
1247 | ll_clear_inode_capas(inode); | |
1248 | if (!S_ISDIR(inode->i_mode)) | |
1249 | LASSERT(list_empty(&lli->lli_agl_list)); | |
1250 | ||
1251 | /* | |
1252 | * XXX This has to be done before lsm is freed below, because | |
1253 | * cl_object still uses inode lsm. | |
1254 | */ | |
1255 | cl_inode_fini(inode); | |
1256 | lli->lli_has_smd = false; | |
1257 | ||
1258 | EXIT; | |
1259 | } | |
1260 | ||
1261 | int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data, | |
1262 | struct md_open_data **mod) | |
1263 | { | |
1264 | struct lustre_md md; | |
1265 | struct inode *inode = dentry->d_inode; | |
1266 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1267 | struct ptlrpc_request *request = NULL; | |
1268 | int rc, ia_valid; | |
1269 | ENTRY; | |
1270 | ||
1271 | op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, | |
1272 | LUSTRE_OPC_ANY, NULL); | |
1273 | if (IS_ERR(op_data)) | |
1274 | RETURN(PTR_ERR(op_data)); | |
1275 | ||
1276 | rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0, | |
1277 | &request, mod); | |
1278 | if (rc) { | |
1279 | ptlrpc_req_finished(request); | |
1280 | if (rc == -ENOENT) { | |
1281 | clear_nlink(inode); | |
1282 | /* Unlinked special device node? Or just a race? | |
1283 | * Pretend we done everything. */ | |
1284 | if (!S_ISREG(inode->i_mode) && | |
1285 | !S_ISDIR(inode->i_mode)) { | |
1286 | ia_valid = op_data->op_attr.ia_valid; | |
1287 | op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS; | |
1288 | rc = simple_setattr(dentry, &op_data->op_attr); | |
1289 | op_data->op_attr.ia_valid = ia_valid; | |
1290 | } | |
1291 | } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) { | |
1292 | CERROR("md_setattr fails: rc = %d\n", rc); | |
1293 | } | |
1294 | RETURN(rc); | |
1295 | } | |
1296 | ||
1297 | rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp, | |
1298 | sbi->ll_md_exp, &md); | |
1299 | if (rc) { | |
1300 | ptlrpc_req_finished(request); | |
1301 | RETURN(rc); | |
1302 | } | |
1303 | ||
1304 | ia_valid = op_data->op_attr.ia_valid; | |
1305 | /* inode size will be in ll_setattr_ost, can't do it now since dirty | |
1306 | * cache is not cleared yet. */ | |
1307 | op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE); | |
1308 | rc = simple_setattr(dentry, &op_data->op_attr); | |
1309 | op_data->op_attr.ia_valid = ia_valid; | |
1310 | ||
1311 | /* Extract epoch data if obtained. */ | |
1312 | op_data->op_handle = md.body->handle; | |
1313 | op_data->op_ioepoch = md.body->ioepoch; | |
1314 | ||
1315 | ll_update_inode(inode, &md); | |
1316 | ptlrpc_req_finished(request); | |
1317 | ||
1318 | RETURN(rc); | |
1319 | } | |
1320 | ||
1321 | /* Close IO epoch and send Size-on-MDS attribute update. */ | |
1322 | static int ll_setattr_done_writing(struct inode *inode, | |
1323 | struct md_op_data *op_data, | |
1324 | struct md_open_data *mod) | |
1325 | { | |
1326 | struct ll_inode_info *lli = ll_i2info(inode); | |
1327 | int rc = 0; | |
1328 | ENTRY; | |
1329 | ||
1330 | LASSERT(op_data != NULL); | |
1331 | if (!S_ISREG(inode->i_mode)) | |
1332 | RETURN(0); | |
1333 | ||
1334 | CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n", | |
1335 | op_data->op_ioepoch, PFID(&lli->lli_fid)); | |
1336 | ||
1337 | op_data->op_flags = MF_EPOCH_CLOSE; | |
1338 | ll_done_writing_attr(inode, op_data); | |
1339 | ll_pack_inode2opdata(inode, op_data, NULL); | |
1340 | ||
1341 | rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod); | |
1342 | if (rc == -EAGAIN) { | |
1343 | /* MDS has instructed us to obtain Size-on-MDS attribute | |
1344 | * from OSTs and send setattr to back to MDS. */ | |
1345 | rc = ll_som_update(inode, op_data); | |
1346 | } else if (rc) { | |
1347 | CERROR("inode %lu mdc truncate failed: rc = %d\n", | |
1348 | inode->i_ino, rc); | |
1349 | } | |
1350 | RETURN(rc); | |
1351 | } | |
1352 | ||
1353 | static int ll_setattr_ost(struct inode *inode, struct iattr *attr) | |
1354 | { | |
1355 | struct obd_capa *capa; | |
1356 | int rc; | |
1357 | ||
1358 | if (attr->ia_valid & ATTR_SIZE) | |
1359 | capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC); | |
1360 | else | |
1361 | capa = ll_mdscapa_get(inode); | |
1362 | ||
1363 | rc = cl_setattr_ost(inode, attr, capa); | |
1364 | ||
1365 | if (attr->ia_valid & ATTR_SIZE) | |
1366 | ll_truncate_free_capa(capa); | |
1367 | else | |
1368 | capa_put(capa); | |
1369 | ||
1370 | return rc; | |
1371 | } | |
1372 | ||
1373 | ||
1374 | /* If this inode has objects allocated to it (lsm != NULL), then the OST | |
1375 | * object(s) determine the file size and mtime. Otherwise, the MDS will | |
1376 | * keep these values until such a time that objects are allocated for it. | |
1377 | * We do the MDS operations first, as it is checking permissions for us. | |
1378 | * We don't to the MDS RPC if there is nothing that we want to store there, | |
1379 | * otherwise there is no harm in updating mtime/atime on the MDS if we are | |
1380 | * going to do an RPC anyways. | |
1381 | * | |
1382 | * If we are doing a truncate, we will send the mtime and ctime updates | |
1383 | * to the OST with the punch RPC, otherwise we do an explicit setattr RPC. | |
1384 | * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE | |
1385 | * at the same time. | |
1386 | */ | |
1387 | int ll_setattr_raw(struct dentry *dentry, struct iattr *attr) | |
1388 | { | |
1389 | struct inode *inode = dentry->d_inode; | |
1390 | struct ll_inode_info *lli = ll_i2info(inode); | |
1391 | struct md_op_data *op_data = NULL; | |
1392 | struct md_open_data *mod = NULL; | |
1393 | int rc = 0, rc1 = 0; | |
1394 | ENTRY; | |
1395 | ||
1396 | CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, " | |
1397 | "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode, | |
1398 | PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size, | |
1399 | attr->ia_valid); | |
1400 | ||
1401 | if (attr->ia_valid & ATTR_SIZE) { | |
1402 | /* Check new size against VFS/VM file size limit and rlimit */ | |
1403 | rc = inode_newsize_ok(inode, attr->ia_size); | |
1404 | if (rc) | |
1405 | RETURN(rc); | |
1406 | ||
1407 | /* The maximum Lustre file size is variable, based on the | |
1408 | * OST maximum object size and number of stripes. This | |
1409 | * needs another check in addition to the VFS check above. */ | |
1410 | if (attr->ia_size > ll_file_maxbytes(inode)) { | |
1411 | CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n", | |
1412 | PFID(&lli->lli_fid), attr->ia_size, | |
1413 | ll_file_maxbytes(inode)); | |
1414 | RETURN(-EFBIG); | |
1415 | } | |
1416 | ||
1417 | attr->ia_valid |= ATTR_MTIME | ATTR_CTIME; | |
1418 | } | |
1419 | ||
1420 | /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */ | |
1421 | if (attr->ia_valid & TIMES_SET_FLAGS) { | |
1422 | if (current_fsuid() != inode->i_uid && | |
1423 | !cfs_capable(CFS_CAP_FOWNER)) | |
1424 | RETURN(-EPERM); | |
1425 | } | |
1426 | ||
1427 | /* We mark all of the fields "set" so MDS/OST does not re-set them */ | |
1428 | if (attr->ia_valid & ATTR_CTIME) { | |
1429 | attr->ia_ctime = CFS_CURRENT_TIME; | |
1430 | attr->ia_valid |= ATTR_CTIME_SET; | |
1431 | } | |
1432 | if (!(attr->ia_valid & ATTR_ATIME_SET) && | |
1433 | (attr->ia_valid & ATTR_ATIME)) { | |
1434 | attr->ia_atime = CFS_CURRENT_TIME; | |
1435 | attr->ia_valid |= ATTR_ATIME_SET; | |
1436 | } | |
1437 | if (!(attr->ia_valid & ATTR_MTIME_SET) && | |
1438 | (attr->ia_valid & ATTR_MTIME)) { | |
1439 | attr->ia_mtime = CFS_CURRENT_TIME; | |
1440 | attr->ia_valid |= ATTR_MTIME_SET; | |
1441 | } | |
1442 | ||
1443 | if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME)) | |
1444 | CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n", | |
1445 | LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime), | |
1446 | cfs_time_current_sec()); | |
1447 | ||
1448 | /* If we are changing file size, file content is modified, flag it. */ | |
1449 | if (attr->ia_valid & ATTR_SIZE) { | |
1450 | attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE; | |
1451 | spin_lock(&lli->lli_lock); | |
1452 | lli->lli_flags |= LLIF_DATA_MODIFIED; | |
1453 | spin_unlock(&lli->lli_lock); | |
1454 | } | |
1455 | ||
1456 | /* We always do an MDS RPC, even if we're only changing the size; | |
1457 | * only the MDS knows whether truncate() should fail with -ETXTBUSY */ | |
1458 | ||
1459 | OBD_ALLOC_PTR(op_data); | |
1460 | if (op_data == NULL) | |
1461 | RETURN(-ENOMEM); | |
1462 | ||
1463 | if (!S_ISDIR(inode->i_mode)) { | |
1464 | if (attr->ia_valid & ATTR_SIZE) | |
1465 | inode_dio_write_done(inode); | |
1466 | mutex_unlock(&inode->i_mutex); | |
1467 | down_write(&lli->lli_trunc_sem); | |
1468 | } | |
1469 | ||
1470 | memcpy(&op_data->op_attr, attr, sizeof(*attr)); | |
1471 | ||
1472 | /* Open epoch for truncate. */ | |
1473 | if (exp_connect_som(ll_i2mdexp(inode)) && | |
1474 | (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET))) | |
1475 | op_data->op_flags = MF_EPOCH_OPEN; | |
1476 | ||
1477 | rc = ll_md_setattr(dentry, op_data, &mod); | |
1478 | if (rc) | |
1479 | GOTO(out, rc); | |
1480 | ||
1481 | /* RPC to MDT is sent, cancel data modification flag */ | |
1482 | if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) { | |
1483 | spin_lock(&lli->lli_lock); | |
1484 | lli->lli_flags &= ~LLIF_DATA_MODIFIED; | |
1485 | spin_unlock(&lli->lli_lock); | |
1486 | } | |
1487 | ||
1488 | ll_ioepoch_open(lli, op_data->op_ioepoch); | |
1489 | if (!S_ISREG(inode->i_mode)) | |
1490 | GOTO(out, rc = 0); | |
1491 | ||
1492 | if (attr->ia_valid & (ATTR_SIZE | | |
1493 | ATTR_ATIME | ATTR_ATIME_SET | | |
1494 | ATTR_MTIME | ATTR_MTIME_SET)) | |
1495 | /* For truncate and utimes sending attributes to OSTs, setting | |
1496 | * mtime/atime to the past will be performed under PW [0:EOF] | |
1497 | * extent lock (new_size:EOF for truncate). It may seem | |
1498 | * excessive to send mtime/atime updates to OSTs when not | |
1499 | * setting times to past, but it is necessary due to possible | |
1500 | * time de-synchronization between MDT inode and OST objects */ | |
1501 | rc = ll_setattr_ost(inode, attr); | |
1502 | EXIT; | |
1503 | out: | |
1504 | if (op_data) { | |
1505 | if (op_data->op_ioepoch) { | |
1506 | rc1 = ll_setattr_done_writing(inode, op_data, mod); | |
1507 | if (!rc) | |
1508 | rc = rc1; | |
1509 | } | |
1510 | ll_finish_md_op_data(op_data); | |
1511 | } | |
1512 | if (!S_ISDIR(inode->i_mode)) { | |
1513 | up_write(&lli->lli_trunc_sem); | |
1514 | mutex_lock(&inode->i_mutex); | |
1515 | if (attr->ia_valid & ATTR_SIZE) | |
1516 | inode_dio_wait(inode); | |
1517 | } | |
1518 | ||
1519 | ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ? | |
1520 | LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1); | |
1521 | ||
1522 | return rc; | |
1523 | } | |
1524 | ||
1525 | int ll_setattr(struct dentry *de, struct iattr *attr) | |
1526 | { | |
1527 | int mode = de->d_inode->i_mode; | |
1528 | ||
1529 | if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) == | |
1530 | (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) | |
1531 | attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE; | |
1532 | ||
1533 | if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) == | |
1534 | (ATTR_SIZE|ATTR_MODE)) && | |
1535 | (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) || | |
1536 | (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) && | |
1537 | !(attr->ia_mode & S_ISGID)))) | |
1538 | attr->ia_valid |= ATTR_FORCE; | |
1539 | ||
1540 | if ((mode & S_ISUID) && | |
1541 | !(attr->ia_mode & S_ISUID) && | |
1542 | !(attr->ia_valid & ATTR_KILL_SUID)) | |
1543 | attr->ia_valid |= ATTR_KILL_SUID; | |
1544 | ||
1545 | if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) && | |
1546 | !(attr->ia_mode & S_ISGID) && | |
1547 | !(attr->ia_valid & ATTR_KILL_SGID)) | |
1548 | attr->ia_valid |= ATTR_KILL_SGID; | |
1549 | ||
1550 | return ll_setattr_raw(de, attr); | |
1551 | } | |
1552 | ||
1553 | int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs, | |
1554 | __u64 max_age, __u32 flags) | |
1555 | { | |
1556 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
1557 | struct obd_statfs obd_osfs; | |
1558 | int rc; | |
1559 | ENTRY; | |
1560 | ||
1561 | rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags); | |
1562 | if (rc) { | |
1563 | CERROR("md_statfs fails: rc = %d\n", rc); | |
1564 | RETURN(rc); | |
1565 | } | |
1566 | ||
1567 | osfs->os_type = sb->s_magic; | |
1568 | ||
1569 | CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n", | |
1570 | osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files); | |
1571 | ||
1572 | if (sbi->ll_flags & LL_SBI_LAZYSTATFS) | |
1573 | flags |= OBD_STATFS_NODELAY; | |
1574 | ||
1575 | rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags); | |
1576 | if (rc) { | |
1577 | CERROR("obd_statfs fails: rc = %d\n", rc); | |
1578 | RETURN(rc); | |
1579 | } | |
1580 | ||
1581 | CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n", | |
1582 | obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree, | |
1583 | obd_osfs.os_files); | |
1584 | ||
1585 | osfs->os_bsize = obd_osfs.os_bsize; | |
1586 | osfs->os_blocks = obd_osfs.os_blocks; | |
1587 | osfs->os_bfree = obd_osfs.os_bfree; | |
1588 | osfs->os_bavail = obd_osfs.os_bavail; | |
1589 | ||
1590 | /* If we don't have as many objects free on the OST as inodes | |
1591 | * on the MDS, we reduce the total number of inodes to | |
1592 | * compensate, so that the "inodes in use" number is correct. | |
1593 | */ | |
1594 | if (obd_osfs.os_ffree < osfs->os_ffree) { | |
1595 | osfs->os_files = (osfs->os_files - osfs->os_ffree) + | |
1596 | obd_osfs.os_ffree; | |
1597 | osfs->os_ffree = obd_osfs.os_ffree; | |
1598 | } | |
1599 | ||
1600 | RETURN(rc); | |
1601 | } | |
1602 | int ll_statfs(struct dentry *de, struct kstatfs *sfs) | |
1603 | { | |
1604 | struct super_block *sb = de->d_sb; | |
1605 | struct obd_statfs osfs; | |
1606 | int rc; | |
1607 | ||
1608 | CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64()); | |
1609 | ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1); | |
1610 | ||
1611 | /* Some amount of caching on the client is allowed */ | |
1612 | rc = ll_statfs_internal(sb, &osfs, | |
1613 | cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), | |
1614 | 0); | |
1615 | if (rc) | |
1616 | return rc; | |
1617 | ||
1618 | statfs_unpack(sfs, &osfs); | |
1619 | ||
1620 | /* We need to downshift for all 32-bit kernels, because we can't | |
1621 | * tell if the kernel is being called via sys_statfs64() or not. | |
1622 | * Stop before overflowing f_bsize - in which case it is better | |
1623 | * to just risk EOVERFLOW if caller is using old sys_statfs(). */ | |
1624 | if (sizeof(long) < 8) { | |
1625 | while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) { | |
1626 | sfs->f_bsize <<= 1; | |
1627 | ||
1628 | osfs.os_blocks >>= 1; | |
1629 | osfs.os_bfree >>= 1; | |
1630 | osfs.os_bavail >>= 1; | |
1631 | } | |
1632 | } | |
1633 | ||
1634 | sfs->f_blocks = osfs.os_blocks; | |
1635 | sfs->f_bfree = osfs.os_bfree; | |
1636 | sfs->f_bavail = osfs.os_bavail; | |
1637 | ||
1638 | return 0; | |
1639 | } | |
1640 | ||
1641 | void ll_inode_size_lock(struct inode *inode) | |
1642 | { | |
1643 | struct ll_inode_info *lli; | |
1644 | ||
1645 | LASSERT(!S_ISDIR(inode->i_mode)); | |
1646 | ||
1647 | lli = ll_i2info(inode); | |
1648 | LASSERT(lli->lli_size_sem_owner != current); | |
1649 | down(&lli->lli_size_sem); | |
1650 | LASSERT(lli->lli_size_sem_owner == NULL); | |
1651 | lli->lli_size_sem_owner = current; | |
1652 | } | |
1653 | ||
1654 | void ll_inode_size_unlock(struct inode *inode) | |
1655 | { | |
1656 | struct ll_inode_info *lli; | |
1657 | ||
1658 | lli = ll_i2info(inode); | |
1659 | LASSERT(lli->lli_size_sem_owner == current); | |
1660 | lli->lli_size_sem_owner = NULL; | |
1661 | up(&lli->lli_size_sem); | |
1662 | } | |
1663 | ||
1664 | void ll_update_inode(struct inode *inode, struct lustre_md *md) | |
1665 | { | |
1666 | struct ll_inode_info *lli = ll_i2info(inode); | |
1667 | struct mdt_body *body = md->body; | |
1668 | struct lov_stripe_md *lsm = md->lsm; | |
1669 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1670 | ||
1671 | LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0)); | |
1672 | if (lsm != NULL) { | |
1673 | if (!lli->lli_has_smd && | |
1674 | !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK)) | |
1675 | cl_file_inode_init(inode, md); | |
1676 | ||
1677 | lli->lli_maxbytes = lsm->lsm_maxbytes; | |
1678 | if (lli->lli_maxbytes > MAX_LFS_FILESIZE) | |
1679 | lli->lli_maxbytes = MAX_LFS_FILESIZE; | |
1680 | } | |
1681 | ||
1682 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
1683 | if (body->valid & OBD_MD_FLRMTPERM) | |
1684 | ll_update_remote_perm(inode, md->remote_perm); | |
1685 | } | |
1686 | #ifdef CONFIG_FS_POSIX_ACL | |
1687 | else if (body->valid & OBD_MD_FLACL) { | |
1688 | spin_lock(&lli->lli_lock); | |
1689 | if (lli->lli_posix_acl) | |
1690 | posix_acl_release(lli->lli_posix_acl); | |
1691 | lli->lli_posix_acl = md->posix_acl; | |
1692 | spin_unlock(&lli->lli_lock); | |
1693 | } | |
1694 | #endif | |
1695 | inode->i_ino = cl_fid_build_ino(&body->fid1, ll_need_32bit_api(sbi)); | |
1696 | inode->i_generation = cl_fid_build_gen(&body->fid1); | |
1697 | ||
1698 | if (body->valid & OBD_MD_FLATIME) { | |
1699 | if (body->atime > LTIME_S(inode->i_atime)) | |
1700 | LTIME_S(inode->i_atime) = body->atime; | |
1701 | lli->lli_lvb.lvb_atime = body->atime; | |
1702 | } | |
1703 | if (body->valid & OBD_MD_FLMTIME) { | |
1704 | if (body->mtime > LTIME_S(inode->i_mtime)) { | |
1705 | CDEBUG(D_INODE, "setting ino %lu mtime from %lu " | |
1706 | "to "LPU64"\n", inode->i_ino, | |
1707 | LTIME_S(inode->i_mtime), body->mtime); | |
1708 | LTIME_S(inode->i_mtime) = body->mtime; | |
1709 | } | |
1710 | lli->lli_lvb.lvb_mtime = body->mtime; | |
1711 | } | |
1712 | if (body->valid & OBD_MD_FLCTIME) { | |
1713 | if (body->ctime > LTIME_S(inode->i_ctime)) | |
1714 | LTIME_S(inode->i_ctime) = body->ctime; | |
1715 | lli->lli_lvb.lvb_ctime = body->ctime; | |
1716 | } | |
1717 | if (body->valid & OBD_MD_FLMODE) | |
1718 | inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT); | |
1719 | if (body->valid & OBD_MD_FLTYPE) | |
1720 | inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT); | |
1721 | LASSERT(inode->i_mode != 0); | |
1722 | if (S_ISREG(inode->i_mode)) { | |
1723 | inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS); | |
1724 | } else { | |
1725 | inode->i_blkbits = inode->i_sb->s_blocksize_bits; | |
1726 | } | |
1727 | if (body->valid & OBD_MD_FLUID) | |
1728 | inode->i_uid = body->uid; | |
1729 | if (body->valid & OBD_MD_FLGID) | |
1730 | inode->i_gid = body->gid; | |
1731 | if (body->valid & OBD_MD_FLFLAGS) | |
1732 | inode->i_flags = ll_ext_to_inode_flags(body->flags); | |
1733 | if (body->valid & OBD_MD_FLNLINK) | |
1734 | set_nlink(inode, body->nlink); | |
1735 | if (body->valid & OBD_MD_FLRDEV) | |
1736 | inode->i_rdev = old_decode_dev(body->rdev); | |
1737 | ||
1738 | if (body->valid & OBD_MD_FLID) { | |
1739 | /* FID shouldn't be changed! */ | |
1740 | if (fid_is_sane(&lli->lli_fid)) { | |
1741 | LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1), | |
1742 | "Trying to change FID "DFID | |
1743 | " to the "DFID", inode %lu/%u(%p)\n", | |
1744 | PFID(&lli->lli_fid), PFID(&body->fid1), | |
1745 | inode->i_ino, inode->i_generation, inode); | |
1746 | } else | |
1747 | lli->lli_fid = body->fid1; | |
1748 | } | |
1749 | ||
1750 | LASSERT(fid_seq(&lli->lli_fid) != 0); | |
1751 | ||
1752 | if (body->valid & OBD_MD_FLSIZE) { | |
1753 | if (exp_connect_som(ll_i2mdexp(inode)) && | |
1754 | S_ISREG(inode->i_mode)) { | |
1755 | struct lustre_handle lockh; | |
1756 | ldlm_mode_t mode; | |
1757 | ||
1758 | /* As it is possible a blocking ast has been processed | |
1759 | * by this time, we need to check there is an UPDATE | |
1760 | * lock on the client and set LLIF_MDS_SIZE_LOCK holding | |
1761 | * it. */ | |
1762 | mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE, | |
1763 | &lockh, LDLM_FL_CBPENDING); | |
1764 | if (mode) { | |
1765 | if (lli->lli_flags & (LLIF_DONE_WRITING | | |
1766 | LLIF_EPOCH_PENDING | | |
1767 | LLIF_SOM_DIRTY)) { | |
1768 | CERROR("ino %lu flags %u still has " | |
1769 | "size authority! do not trust " | |
1770 | "the size got from MDS\n", | |
1771 | inode->i_ino, lli->lli_flags); | |
1772 | } else { | |
1773 | /* Use old size assignment to avoid | |
1774 | * deadlock bz14138 & bz14326 */ | |
1775 | i_size_write(inode, body->size); | |
1776 | lli->lli_flags |= LLIF_MDS_SIZE_LOCK; | |
1777 | } | |
1778 | ldlm_lock_decref(&lockh, mode); | |
1779 | } | |
1780 | } else { | |
1781 | /* Use old size assignment to avoid | |
1782 | * deadlock bz14138 & bz14326 */ | |
1783 | i_size_write(inode, body->size); | |
1784 | ||
1785 | CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n", | |
1786 | inode->i_ino, (unsigned long long)body->size); | |
1787 | } | |
1788 | ||
1789 | if (body->valid & OBD_MD_FLBLOCKS) | |
1790 | inode->i_blocks = body->blocks; | |
1791 | } | |
1792 | ||
1793 | if (body->valid & OBD_MD_FLMDSCAPA) { | |
1794 | LASSERT(md->mds_capa); | |
1795 | ll_add_capa(inode, md->mds_capa); | |
1796 | } | |
1797 | if (body->valid & OBD_MD_FLOSSCAPA) { | |
1798 | LASSERT(md->oss_capa); | |
1799 | ll_add_capa(inode, md->oss_capa); | |
1800 | } | |
1801 | } | |
1802 | ||
1803 | void ll_read_inode2(struct inode *inode, void *opaque) | |
1804 | { | |
1805 | struct lustre_md *md = opaque; | |
1806 | struct ll_inode_info *lli = ll_i2info(inode); | |
1807 | ENTRY; | |
1808 | ||
1809 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n", | |
1810 | PFID(&lli->lli_fid), inode); | |
1811 | ||
1812 | LASSERT(!lli->lli_has_smd); | |
1813 | ||
1814 | /* Core attributes from the MDS first. This is a new inode, and | |
1815 | * the VFS doesn't zero times in the core inode so we have to do | |
1816 | * it ourselves. They will be overwritten by either MDS or OST | |
1817 | * attributes - we just need to make sure they aren't newer. */ | |
1818 | LTIME_S(inode->i_mtime) = 0; | |
1819 | LTIME_S(inode->i_atime) = 0; | |
1820 | LTIME_S(inode->i_ctime) = 0; | |
1821 | inode->i_rdev = 0; | |
1822 | ll_update_inode(inode, md); | |
1823 | ||
1824 | /* OIDEBUG(inode); */ | |
1825 | ||
1826 | /* initializing backing dev info. */ | |
1827 | inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi; | |
1828 | ||
1829 | ||
1830 | if (S_ISREG(inode->i_mode)) { | |
1831 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1832 | inode->i_op = &ll_file_inode_operations; | |
1833 | inode->i_fop = sbi->ll_fop; | |
1834 | inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops; | |
1835 | EXIT; | |
1836 | } else if (S_ISDIR(inode->i_mode)) { | |
1837 | inode->i_op = &ll_dir_inode_operations; | |
1838 | inode->i_fop = &ll_dir_operations; | |
1839 | EXIT; | |
1840 | } else if (S_ISLNK(inode->i_mode)) { | |
1841 | inode->i_op = &ll_fast_symlink_inode_operations; | |
1842 | EXIT; | |
1843 | } else { | |
1844 | inode->i_op = &ll_special_inode_operations; | |
1845 | ||
1846 | init_special_inode(inode, inode->i_mode, | |
1847 | inode->i_rdev); | |
1848 | ||
1849 | EXIT; | |
1850 | } | |
1851 | } | |
1852 | ||
1853 | void ll_delete_inode(struct inode *inode) | |
1854 | { | |
1855 | struct cl_inode_info *lli = cl_i2info(inode); | |
1856 | ENTRY; | |
1857 | ||
1858 | if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) | |
1859 | /* discard all dirty pages before truncating them, required by | |
1860 | * osc_extent implementation at LU-1030. */ | |
65fb55d1 NY |
1861 | cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, |
1862 | CL_FSYNC_DISCARD, 1); | |
d7e09d03 PT |
1863 | |
1864 | truncate_inode_pages(&inode->i_data, 0); | |
1865 | ||
1866 | /* Workaround for LU-118 */ | |
1867 | if (inode->i_data.nrpages) { | |
1868 | TREE_READ_LOCK_IRQ(&inode->i_data); | |
1869 | TREE_READ_UNLOCK_IRQ(&inode->i_data); | |
1870 | LASSERTF(inode->i_data.nrpages == 0, | |
1871 | "inode=%lu/%u(%p) nrpages=%lu, see " | |
1872 | "http://jira.whamcloud.com/browse/LU-118\n", | |
1873 | inode->i_ino, inode->i_generation, inode, | |
1874 | inode->i_data.nrpages); | |
1875 | } | |
1876 | /* Workaround end */ | |
1877 | ||
1878 | ll_clear_inode(inode); | |
1879 | clear_inode(inode); | |
1880 | ||
1881 | EXIT; | |
1882 | } | |
1883 | ||
1884 | int ll_iocontrol(struct inode *inode, struct file *file, | |
1885 | unsigned int cmd, unsigned long arg) | |
1886 | { | |
1887 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1888 | struct ptlrpc_request *req = NULL; | |
1889 | int rc, flags = 0; | |
1890 | ENTRY; | |
1891 | ||
1892 | switch(cmd) { | |
1893 | case FSFILT_IOC_GETFLAGS: { | |
1894 | struct mdt_body *body; | |
1895 | struct md_op_data *op_data; | |
1896 | ||
1897 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, | |
1898 | 0, 0, LUSTRE_OPC_ANY, | |
1899 | NULL); | |
1900 | if (IS_ERR(op_data)) | |
1901 | RETURN(PTR_ERR(op_data)); | |
1902 | ||
1903 | op_data->op_valid = OBD_MD_FLFLAGS; | |
1904 | rc = md_getattr(sbi->ll_md_exp, op_data, &req); | |
1905 | ll_finish_md_op_data(op_data); | |
1906 | if (rc) { | |
1907 | CERROR("failure %d inode %lu\n", rc, inode->i_ino); | |
1908 | RETURN(-abs(rc)); | |
1909 | } | |
1910 | ||
1911 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
1912 | ||
1913 | flags = body->flags; | |
1914 | ||
1915 | ptlrpc_req_finished(req); | |
1916 | ||
1917 | RETURN(put_user(flags, (int *)arg)); | |
1918 | } | |
1919 | case FSFILT_IOC_SETFLAGS: { | |
1920 | struct lov_stripe_md *lsm; | |
1921 | struct obd_info oinfo = { { { 0 } } }; | |
1922 | struct md_op_data *op_data; | |
1923 | ||
1924 | if (get_user(flags, (int *)arg)) | |
1925 | RETURN(-EFAULT); | |
1926 | ||
1927 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
1928 | LUSTRE_OPC_ANY, NULL); | |
1929 | if (IS_ERR(op_data)) | |
1930 | RETURN(PTR_ERR(op_data)); | |
1931 | ||
1932 | ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags; | |
1933 | op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG; | |
1934 | rc = md_setattr(sbi->ll_md_exp, op_data, | |
1935 | NULL, 0, NULL, 0, &req, NULL); | |
1936 | ll_finish_md_op_data(op_data); | |
1937 | ptlrpc_req_finished(req); | |
1938 | if (rc) | |
1939 | RETURN(rc); | |
1940 | ||
1941 | inode->i_flags = ll_ext_to_inode_flags(flags); | |
1942 | ||
1943 | lsm = ccc_inode_lsm_get(inode); | |
1944 | if (lsm == NULL) | |
1945 | RETURN(0); | |
1946 | ||
1947 | OBDO_ALLOC(oinfo.oi_oa); | |
1948 | if (!oinfo.oi_oa) { | |
1949 | ccc_inode_lsm_put(inode, lsm); | |
1950 | RETURN(-ENOMEM); | |
1951 | } | |
1952 | oinfo.oi_md = lsm; | |
1953 | oinfo.oi_oa->o_oi = lsm->lsm_oi; | |
1954 | oinfo.oi_oa->o_flags = flags; | |
1955 | oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | | |
1956 | OBD_MD_FLGROUP; | |
1957 | oinfo.oi_capa = ll_mdscapa_get(inode); | |
1958 | obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid); | |
1959 | rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL); | |
1960 | capa_put(oinfo.oi_capa); | |
1961 | OBDO_FREE(oinfo.oi_oa); | |
1962 | ccc_inode_lsm_put(inode, lsm); | |
1963 | ||
1964 | if (rc && rc != -EPERM && rc != -EACCES) | |
1965 | CERROR("osc_setattr_async fails: rc = %d\n", rc); | |
1966 | ||
1967 | RETURN(rc); | |
1968 | } | |
1969 | default: | |
1970 | RETURN(-ENOSYS); | |
1971 | } | |
1972 | ||
1973 | RETURN(0); | |
1974 | } | |
1975 | ||
1976 | int ll_flush_ctx(struct inode *inode) | |
1977 | { | |
1978 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1979 | ||
1980 | CDEBUG(D_SEC, "flush context for user %d\n", current_uid()); | |
1981 | ||
1982 | obd_set_info_async(NULL, sbi->ll_md_exp, | |
1983 | sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX, | |
1984 | 0, NULL, NULL); | |
1985 | obd_set_info_async(NULL, sbi->ll_dt_exp, | |
1986 | sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX, | |
1987 | 0, NULL, NULL); | |
1988 | return 0; | |
1989 | } | |
1990 | ||
1991 | /* umount -f client means force down, don't save state */ | |
1992 | void ll_umount_begin(struct super_block *sb) | |
1993 | { | |
1994 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
1995 | struct obd_device *obd; | |
1996 | struct obd_ioctl_data *ioc_data; | |
1997 | ENTRY; | |
1998 | ||
1999 | ||
2000 | CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb, | |
2001 | sb->s_count, atomic_read(&sb->s_active)); | |
2002 | ||
2003 | obd = class_exp2obd(sbi->ll_md_exp); | |
2004 | if (obd == NULL) { | |
2005 | CERROR("Invalid MDC connection handle "LPX64"\n", | |
2006 | sbi->ll_md_exp->exp_handle.h_cookie); | |
2007 | EXIT; | |
2008 | return; | |
2009 | } | |
2010 | obd->obd_force = 1; | |
2011 | ||
2012 | obd = class_exp2obd(sbi->ll_dt_exp); | |
2013 | if (obd == NULL) { | |
2014 | CERROR("Invalid LOV connection handle "LPX64"\n", | |
2015 | sbi->ll_dt_exp->exp_handle.h_cookie); | |
2016 | EXIT; | |
2017 | return; | |
2018 | } | |
2019 | obd->obd_force = 1; | |
2020 | ||
2021 | OBD_ALLOC_PTR(ioc_data); | |
2022 | if (ioc_data) { | |
2023 | obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, | |
2024 | sizeof *ioc_data, ioc_data, NULL); | |
2025 | ||
2026 | obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, | |
2027 | sizeof *ioc_data, ioc_data, NULL); | |
2028 | ||
2029 | OBD_FREE_PTR(ioc_data); | |
2030 | } | |
2031 | ||
d7e09d03 PT |
2032 | /* Really, we'd like to wait until there are no requests outstanding, |
2033 | * and then continue. For now, we just invalidate the requests, | |
2034 | * schedule() and sleep one second if needed, and hope. | |
2035 | */ | |
2036 | schedule(); | |
2037 | ||
2038 | EXIT; | |
2039 | } | |
2040 | ||
2041 | int ll_remount_fs(struct super_block *sb, int *flags, char *data) | |
2042 | { | |
2043 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
2044 | char *profilenm = get_profile_name(sb); | |
2045 | int err; | |
2046 | __u32 read_only; | |
2047 | ||
2048 | if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) { | |
2049 | read_only = *flags & MS_RDONLY; | |
2050 | err = obd_set_info_async(NULL, sbi->ll_md_exp, | |
2051 | sizeof(KEY_READ_ONLY), | |
2052 | KEY_READ_ONLY, sizeof(read_only), | |
2053 | &read_only, NULL); | |
2054 | if (err) { | |
2055 | LCONSOLE_WARN("Failed to remount %s %s (%d)\n", | |
2056 | profilenm, read_only ? | |
2057 | "read-only" : "read-write", err); | |
2058 | return err; | |
2059 | } | |
2060 | ||
2061 | if (read_only) | |
2062 | sb->s_flags |= MS_RDONLY; | |
2063 | else | |
2064 | sb->s_flags &= ~MS_RDONLY; | |
2065 | ||
2066 | if (sbi->ll_flags & LL_SBI_VERBOSE) | |
2067 | LCONSOLE_WARN("Remounted %s %s\n", profilenm, | |
2068 | read_only ? "read-only" : "read-write"); | |
2069 | } | |
2070 | return 0; | |
2071 | } | |
2072 | ||
2073 | int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req, | |
2074 | struct super_block *sb, struct lookup_intent *it) | |
2075 | { | |
2076 | struct ll_sb_info *sbi = NULL; | |
2077 | struct lustre_md md; | |
2078 | int rc; | |
2079 | ENTRY; | |
2080 | ||
2081 | LASSERT(*inode || sb); | |
2082 | sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode); | |
2083 | rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp, | |
2084 | sbi->ll_md_exp, &md); | |
2085 | if (rc) | |
2086 | RETURN(rc); | |
2087 | ||
2088 | if (*inode) { | |
2089 | ll_update_inode(*inode, &md); | |
2090 | } else { | |
2091 | LASSERT(sb != NULL); | |
2092 | ||
2093 | /* | |
2094 | * At this point server returns to client's same fid as client | |
2095 | * generated for creating. So using ->fid1 is okay here. | |
2096 | */ | |
2097 | LASSERT(fid_is_sane(&md.body->fid1)); | |
2098 | ||
2099 | *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, | |
2100 | ll_need_32bit_api(sbi)), | |
2101 | &md); | |
2102 | if (*inode == NULL || IS_ERR(*inode)) { | |
2103 | #ifdef CONFIG_FS_POSIX_ACL | |
2104 | if (md.posix_acl) { | |
2105 | posix_acl_release(md.posix_acl); | |
2106 | md.posix_acl = NULL; | |
2107 | } | |
2108 | #endif | |
2109 | rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM; | |
2110 | *inode = NULL; | |
2111 | CERROR("new_inode -fatal: rc %d\n", rc); | |
2112 | GOTO(out, rc); | |
2113 | } | |
2114 | } | |
2115 | ||
2116 | /* Handling piggyback layout lock. | |
2117 | * Layout lock can be piggybacked by getattr and open request. | |
2118 | * The lsm can be applied to inode only if it comes with a layout lock | |
2119 | * otherwise correct layout may be overwritten, for example: | |
2120 | * 1. proc1: mdt returns a lsm but not granting layout | |
2121 | * 2. layout was changed by another client | |
2122 | * 3. proc2: refresh layout and layout lock granted | |
2123 | * 4. proc1: to apply a stale layout */ | |
2124 | if (it != NULL && it->d.lustre.it_lock_mode != 0) { | |
2125 | struct lustre_handle lockh; | |
2126 | struct ldlm_lock *lock; | |
2127 | ||
2128 | lockh.cookie = it->d.lustre.it_lock_handle; | |
2129 | lock = ldlm_handle2lock(&lockh); | |
2130 | LASSERT(lock != NULL); | |
2131 | if (ldlm_has_layout(lock)) { | |
2132 | struct cl_object_conf conf; | |
2133 | ||
2134 | memset(&conf, 0, sizeof(conf)); | |
2135 | conf.coc_opc = OBJECT_CONF_SET; | |
2136 | conf.coc_inode = *inode; | |
2137 | conf.coc_lock = lock; | |
2138 | conf.u.coc_md = &md; | |
2139 | (void)ll_layout_conf(*inode, &conf); | |
2140 | } | |
2141 | LDLM_LOCK_PUT(lock); | |
2142 | } | |
2143 | ||
2144 | out: | |
2145 | if (md.lsm != NULL) | |
2146 | obd_free_memmd(sbi->ll_dt_exp, &md.lsm); | |
2147 | md_free_lustre_md(sbi->ll_md_exp, &md); | |
2148 | RETURN(rc); | |
2149 | } | |
2150 | ||
2151 | int ll_obd_statfs(struct inode *inode, void *arg) | |
2152 | { | |
2153 | struct ll_sb_info *sbi = NULL; | |
2154 | struct obd_export *exp; | |
2155 | char *buf = NULL; | |
2156 | struct obd_ioctl_data *data = NULL; | |
2157 | __u32 type; | |
2158 | __u32 flags; | |
2159 | int len = 0, rc; | |
2160 | ||
2161 | if (!inode || !(sbi = ll_i2sbi(inode))) | |
2162 | GOTO(out_statfs, rc = -EINVAL); | |
2163 | ||
2164 | rc = obd_ioctl_getdata(&buf, &len, arg); | |
2165 | if (rc) | |
2166 | GOTO(out_statfs, rc); | |
2167 | ||
2168 | data = (void*)buf; | |
2169 | if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 || | |
2170 | !data->ioc_pbuf1 || !data->ioc_pbuf2) | |
2171 | GOTO(out_statfs, rc = -EINVAL); | |
2172 | ||
2173 | if (data->ioc_inllen1 != sizeof(__u32) || | |
2174 | data->ioc_inllen2 != sizeof(__u32) || | |
2175 | data->ioc_plen1 != sizeof(struct obd_statfs) || | |
2176 | data->ioc_plen2 != sizeof(struct obd_uuid)) | |
2177 | GOTO(out_statfs, rc = -EINVAL); | |
2178 | ||
2179 | memcpy(&type, data->ioc_inlbuf1, sizeof(__u32)); | |
2180 | if (type & LL_STATFS_LMV) | |
2181 | exp = sbi->ll_md_exp; | |
2182 | else if (type & LL_STATFS_LOV) | |
2183 | exp = sbi->ll_dt_exp; | |
2184 | else | |
2185 | GOTO(out_statfs, rc = -ENODEV); | |
2186 | ||
2187 | flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0; | |
2188 | rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags); | |
2189 | if (rc) | |
2190 | GOTO(out_statfs, rc); | |
2191 | out_statfs: | |
2192 | if (buf) | |
2193 | obd_ioctl_freedata(buf, len); | |
2194 | return rc; | |
2195 | } | |
2196 | ||
2197 | int ll_process_config(struct lustre_cfg *lcfg) | |
2198 | { | |
2199 | char *ptr; | |
2200 | void *sb; | |
2201 | struct lprocfs_static_vars lvars; | |
2202 | unsigned long x; | |
2203 | int rc = 0; | |
2204 | ||
2205 | lprocfs_llite_init_vars(&lvars); | |
2206 | ||
2207 | /* The instance name contains the sb: lustre-client-aacfe000 */ | |
2208 | ptr = strrchr(lustre_cfg_string(lcfg, 0), '-'); | |
2209 | if (!ptr || !*(++ptr)) | |
2210 | return -EINVAL; | |
2211 | if (sscanf(ptr, "%lx", &x) != 1) | |
2212 | return -EINVAL; | |
2213 | sb = (void *)x; | |
2214 | /* This better be a real Lustre superblock! */ | |
2215 | LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC); | |
2216 | ||
2217 | /* Note we have not called client_common_fill_super yet, so | |
2218 | proc fns must be able to handle that! */ | |
2219 | rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars, | |
2220 | lcfg, sb); | |
2221 | if (rc > 0) | |
2222 | rc = 0; | |
2223 | return(rc); | |
2224 | } | |
2225 | ||
2226 | /* this function prepares md_op_data hint for passing ot down to MD stack. */ | |
2227 | struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data, | |
2228 | struct inode *i1, struct inode *i2, | |
2229 | const char *name, int namelen, | |
2230 | int mode, __u32 opc, void *data) | |
2231 | { | |
2232 | LASSERT(i1 != NULL); | |
2233 | ||
2234 | if (namelen > ll_i2sbi(i1)->ll_namelen) | |
2235 | return ERR_PTR(-ENAMETOOLONG); | |
2236 | ||
2237 | if (op_data == NULL) | |
2238 | OBD_ALLOC_PTR(op_data); | |
2239 | ||
2240 | if (op_data == NULL) | |
2241 | return ERR_PTR(-ENOMEM); | |
2242 | ||
2243 | ll_i2gids(op_data->op_suppgids, i1, i2); | |
2244 | op_data->op_fid1 = *ll_inode2fid(i1); | |
2245 | op_data->op_capa1 = ll_mdscapa_get(i1); | |
2246 | ||
2247 | if (i2) { | |
2248 | op_data->op_fid2 = *ll_inode2fid(i2); | |
2249 | op_data->op_capa2 = ll_mdscapa_get(i2); | |
2250 | } else { | |
2251 | fid_zero(&op_data->op_fid2); | |
2252 | op_data->op_capa2 = NULL; | |
2253 | } | |
2254 | ||
2255 | op_data->op_name = name; | |
2256 | op_data->op_namelen = namelen; | |
2257 | op_data->op_mode = mode; | |
2258 | op_data->op_mod_time = cfs_time_current_sec(); | |
2259 | op_data->op_fsuid = current_fsuid(); | |
2260 | op_data->op_fsgid = current_fsgid(); | |
2261 | op_data->op_cap = cfs_curproc_cap_pack(); | |
2262 | op_data->op_bias = 0; | |
2263 | op_data->op_cli_flags = 0; | |
2264 | if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) && | |
2265 | filename_is_volatile(name, namelen, NULL)) | |
2266 | op_data->op_bias |= MDS_CREATE_VOLATILE; | |
2267 | op_data->op_opc = opc; | |
2268 | op_data->op_mds = 0; | |
2269 | op_data->op_data = data; | |
2270 | ||
2271 | /* If the file is being opened after mknod() (normally due to NFS) | |
2272 | * try to use the default stripe data from parent directory for | |
2273 | * allocating OST objects. Try to pass the parent FID to MDS. */ | |
2274 | if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) && | |
2275 | !ll_i2info(i2)->lli_has_smd) { | |
2276 | struct ll_inode_info *lli = ll_i2info(i2); | |
2277 | ||
2278 | spin_lock(&lli->lli_lock); | |
2279 | if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid))) | |
2280 | op_data->op_fid1 = lli->lli_pfid; | |
2281 | spin_unlock(&lli->lli_lock); | |
2282 | /** We ignore parent's capability temporary. */ | |
2283 | } | |
2284 | ||
2285 | /* When called by ll_setattr_raw, file is i1. */ | |
2286 | if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags) | |
2287 | op_data->op_bias |= MDS_DATA_MODIFIED; | |
2288 | ||
2289 | return op_data; | |
2290 | } | |
2291 | ||
2292 | void ll_finish_md_op_data(struct md_op_data *op_data) | |
2293 | { | |
2294 | capa_put(op_data->op_capa1); | |
2295 | capa_put(op_data->op_capa2); | |
2296 | OBD_FREE_PTR(op_data); | |
2297 | } | |
2298 | ||
2299 | int ll_show_options(struct seq_file *seq, struct dentry *dentry) | |
2300 | { | |
2301 | struct ll_sb_info *sbi; | |
2302 | ||
2303 | LASSERT((seq != NULL) && (dentry != NULL)); | |
2304 | sbi = ll_s2sbi(dentry->d_sb); | |
2305 | ||
2306 | if (sbi->ll_flags & LL_SBI_NOLCK) | |
2307 | seq_puts(seq, ",nolock"); | |
2308 | ||
2309 | if (sbi->ll_flags & LL_SBI_FLOCK) | |
2310 | seq_puts(seq, ",flock"); | |
2311 | ||
2312 | if (sbi->ll_flags & LL_SBI_LOCALFLOCK) | |
2313 | seq_puts(seq, ",localflock"); | |
2314 | ||
2315 | if (sbi->ll_flags & LL_SBI_USER_XATTR) | |
2316 | seq_puts(seq, ",user_xattr"); | |
2317 | ||
2318 | if (sbi->ll_flags & LL_SBI_LAZYSTATFS) | |
2319 | seq_puts(seq, ",lazystatfs"); | |
2320 | ||
2321 | if (sbi->ll_flags & LL_SBI_USER_FID2PATH) | |
2322 | seq_puts(seq, ",user_fid2path"); | |
2323 | ||
2324 | RETURN(0); | |
2325 | } | |
2326 | ||
2327 | /** | |
2328 | * Get obd name by cmd, and copy out to user space | |
2329 | */ | |
2330 | int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg) | |
2331 | { | |
2332 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
2333 | struct obd_device *obd; | |
2334 | ENTRY; | |
2335 | ||
2336 | if (cmd == OBD_IOC_GETDTNAME) | |
2337 | obd = class_exp2obd(sbi->ll_dt_exp); | |
2338 | else if (cmd == OBD_IOC_GETMDNAME) | |
2339 | obd = class_exp2obd(sbi->ll_md_exp); | |
2340 | else | |
2341 | RETURN(-EINVAL); | |
2342 | ||
2343 | if (!obd) | |
2344 | RETURN(-ENOENT); | |
2345 | ||
2346 | if (copy_to_user((void *)arg, obd->obd_name, | |
2347 | strlen(obd->obd_name) + 1)) | |
2348 | RETURN(-EFAULT); | |
2349 | ||
2350 | RETURN(0); | |
2351 | } | |
2352 | ||
2353 | /** | |
2354 | * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the | |
2355 | * fsname will be returned in this buffer; otherwise, a static buffer will be | |
2356 | * used to store the fsname and returned to caller. | |
2357 | */ | |
2358 | char *ll_get_fsname(struct super_block *sb, char *buf, int buflen) | |
2359 | { | |
2360 | static char fsname_static[MTI_NAME_MAXLEN]; | |
2361 | struct lustre_sb_info *lsi = s2lsi(sb); | |
2362 | char *ptr; | |
2363 | int len; | |
2364 | ||
2365 | if (buf == NULL) { | |
2366 | /* this means the caller wants to use static buffer | |
2367 | * and it doesn't care about race. Usually this is | |
2368 | * in error reporting path */ | |
2369 | buf = fsname_static; | |
2370 | buflen = sizeof(fsname_static); | |
2371 | } | |
2372 | ||
2373 | len = strlen(lsi->lsi_lmd->lmd_profile); | |
2374 | ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-'); | |
2375 | if (ptr && (strcmp(ptr, "-client") == 0)) | |
2376 | len -= 7; | |
2377 | ||
2378 | if (unlikely(len >= buflen)) | |
2379 | len = buflen - 1; | |
2380 | strncpy(buf, lsi->lsi_lmd->lmd_profile, len); | |
2381 | buf[len] = '\0'; | |
2382 | ||
2383 | return buf; | |
2384 | } | |
2385 | ||
2386 | static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize) | |
2387 | { | |
2388 | char *path = NULL; | |
2389 | ||
2390 | struct path p; | |
2391 | ||
2392 | p.dentry = dentry; | |
2393 | p.mnt = current->fs->root.mnt; | |
2394 | path_get(&p); | |
2395 | path = d_path(&p, buf, bufsize); | |
2396 | path_put(&p); | |
2397 | ||
2398 | return path; | |
2399 | } | |
2400 | ||
2401 | void ll_dirty_page_discard_warn(struct page *page, int ioret) | |
2402 | { | |
2403 | char *buf, *path = NULL; | |
2404 | struct dentry *dentry = NULL; | |
2405 | struct ccc_object *obj = cl_inode2ccc(page->mapping->host); | |
2406 | ||
2407 | /* this can be called inside spin lock so use GFP_ATOMIC. */ | |
2408 | buf = (char *)__get_free_page(GFP_ATOMIC); | |
2409 | if (buf != NULL) { | |
2410 | dentry = d_find_alias(page->mapping->host); | |
2411 | if (dentry != NULL) | |
2412 | path = ll_d_path(dentry, buf, PAGE_SIZE); | |
2413 | } | |
2414 | ||
2415 | CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted " | |
2416 | "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0), | |
2417 | s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev, | |
2418 | PFID(&obj->cob_header.coh_lu.loh_fid), | |
2419 | (path && !IS_ERR(path)) ? path : "", ioret); | |
2420 | ||
2421 | if (dentry != NULL) | |
2422 | dput(dentry); | |
2423 | ||
2424 | if (buf != NULL) | |
2425 | free_page((unsigned long)buf); | |
2426 | } |