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 | ||
c52f69c5 | 102 | /* initialize lru data */ |
d7e09d03 PT |
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 | ||
d7e09d03 PT |
109 | sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32, |
110 | SBI_DEFAULT_READAHEAD_MAX); | |
111 | sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file; | |
112 | sbi->ll_ra_info.ra_max_read_ahead_whole_pages = | |
113 | SBI_DEFAULT_READAHEAD_WHOLE_MAX; | |
114 | INIT_LIST_HEAD(&sbi->ll_conn_chain); | |
115 | INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list); | |
116 | ||
117 | ll_generate_random_uuid(uuid); | |
118 | class_uuid_unparse(uuid, &sbi->ll_sb_uuid); | |
119 | CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid); | |
120 | ||
121 | spin_lock(&ll_sb_lock); | |
122 | list_add_tail(&sbi->ll_list, &ll_super_blocks); | |
123 | spin_unlock(&ll_sb_lock); | |
124 | ||
125 | sbi->ll_flags |= LL_SBI_VERBOSE; | |
126 | sbi->ll_flags |= LL_SBI_CHECKSUM; | |
127 | ||
128 | sbi->ll_flags |= LL_SBI_LRU_RESIZE; | |
129 | ||
130 | for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) { | |
131 | spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. | |
132 | pp_r_hist.oh_lock); | |
133 | spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. | |
134 | pp_w_hist.oh_lock); | |
135 | } | |
136 | ||
137 | /* metadata statahead is enabled by default */ | |
138 | sbi->ll_sa_max = LL_SA_RPC_DEF; | |
139 | atomic_set(&sbi->ll_sa_total, 0); | |
140 | atomic_set(&sbi->ll_sa_wrong, 0); | |
141 | atomic_set(&sbi->ll_agl_total, 0); | |
142 | sbi->ll_flags |= LL_SBI_AGL_ENABLED; | |
143 | ||
144 | RETURN(sbi); | |
145 | } | |
146 | ||
147 | void ll_free_sbi(struct super_block *sb) | |
148 | { | |
149 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
150 | ENTRY; | |
151 | ||
152 | if (sbi != NULL) { | |
153 | spin_lock(&ll_sb_lock); | |
154 | list_del(&sbi->ll_list); | |
155 | spin_unlock(&ll_sb_lock); | |
156 | OBD_FREE(sbi, sizeof(*sbi)); | |
157 | } | |
158 | EXIT; | |
159 | } | |
160 | ||
161 | static struct dentry_operations ll_d_root_ops = { | |
162 | .d_compare = ll_dcompare, | |
163 | .d_revalidate = ll_revalidate_nd, | |
164 | }; | |
165 | ||
166 | static int client_common_fill_super(struct super_block *sb, char *md, char *dt, | |
167 | struct vfsmount *mnt) | |
168 | { | |
169 | struct inode *root = 0; | |
170 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
171 | struct obd_device *obd; | |
172 | struct obd_capa *oc = NULL; | |
173 | struct obd_statfs *osfs = NULL; | |
174 | struct ptlrpc_request *request = NULL; | |
175 | struct obd_connect_data *data = NULL; | |
176 | struct obd_uuid *uuid; | |
177 | struct md_op_data *op_data; | |
178 | struct lustre_md lmd; | |
179 | obd_valid valid; | |
180 | int size, err, checksum; | |
181 | ENTRY; | |
182 | ||
183 | obd = class_name2obd(md); | |
184 | if (!obd) { | |
185 | CERROR("MD %s: not setup or attached\n", md); | |
186 | RETURN(-EINVAL); | |
187 | } | |
188 | ||
189 | OBD_ALLOC_PTR(data); | |
190 | if (data == NULL) | |
191 | RETURN(-ENOMEM); | |
192 | ||
193 | OBD_ALLOC_PTR(osfs); | |
194 | if (osfs == NULL) { | |
195 | OBD_FREE_PTR(data); | |
196 | RETURN(-ENOMEM); | |
197 | } | |
198 | ||
199 | if (proc_lustre_fs_root) { | |
200 | err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb, | |
201 | dt, md); | |
202 | if (err < 0) | |
203 | CERROR("could not register mount in /proc/fs/lustre\n"); | |
204 | } | |
205 | ||
206 | /* indicate the features supported by this client */ | |
207 | data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH | | |
208 | OBD_CONNECT_ATTRFID | | |
209 | OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE | | |
210 | OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA | | |
211 | OBD_CONNECT_CANCELSET | OBD_CONNECT_FID | | |
212 | OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 | | |
213 | OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR | | |
214 | OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH| | |
215 | OBD_CONNECT_EINPROGRESS | | |
216 | OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE | | |
217 | OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS; | |
218 | ||
219 | if (sbi->ll_flags & LL_SBI_SOM_PREVIEW) | |
220 | data->ocd_connect_flags |= OBD_CONNECT_SOM; | |
221 | ||
222 | if (sbi->ll_flags & LL_SBI_LRU_RESIZE) | |
223 | data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE; | |
224 | #ifdef CONFIG_FS_POSIX_ACL | |
225 | data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK; | |
226 | #endif | |
227 | ||
228 | if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT)) | |
229 | /* flag mdc connection as lightweight, only used for test | |
230 | * purpose, use with care */ | |
231 | data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT; | |
232 | ||
233 | data->ocd_ibits_known = MDS_INODELOCK_FULL; | |
234 | data->ocd_version = LUSTRE_VERSION_CODE; | |
235 | ||
236 | if (sb->s_flags & MS_RDONLY) | |
237 | data->ocd_connect_flags |= OBD_CONNECT_RDONLY; | |
238 | if (sbi->ll_flags & LL_SBI_USER_XATTR) | |
239 | data->ocd_connect_flags |= OBD_CONNECT_XATTR; | |
240 | ||
241 | #ifdef HAVE_MS_FLOCK_LOCK | |
242 | /* force vfs to use lustre handler for flock() calls - bug 10743 */ | |
243 | sb->s_flags |= MS_FLOCK_LOCK; | |
244 | #endif | |
245 | #ifdef MS_HAS_NEW_AOPS | |
246 | sb->s_flags |= MS_HAS_NEW_AOPS; | |
247 | #endif | |
248 | ||
249 | if (sbi->ll_flags & LL_SBI_FLOCK) | |
250 | sbi->ll_fop = &ll_file_operations_flock; | |
251 | else if (sbi->ll_flags & LL_SBI_LOCALFLOCK) | |
252 | sbi->ll_fop = &ll_file_operations; | |
253 | else | |
254 | sbi->ll_fop = &ll_file_operations_noflock; | |
255 | ||
256 | /* real client */ | |
257 | data->ocd_connect_flags |= OBD_CONNECT_REAL; | |
258 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) | |
259 | data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE; | |
260 | ||
261 | data->ocd_brw_size = MD_MAX_BRW_SIZE; | |
262 | ||
263 | err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL); | |
264 | if (err == -EBUSY) { | |
265 | LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing " | |
266 | "recovery, of which this client is not a " | |
267 | "part. Please wait for recovery to complete," | |
268 | " abort, or time out.\n", md); | |
269 | GOTO(out, err); | |
270 | } else if (err) { | |
271 | CERROR("cannot connect to %s: rc = %d\n", md, err); | |
272 | GOTO(out, err); | |
273 | } | |
274 | ||
275 | sbi->ll_md_exp->exp_connect_data = *data; | |
276 | ||
277 | err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp, | |
278 | LUSTRE_SEQ_METADATA); | |
279 | if (err) { | |
280 | CERROR("%s: Can't init metadata layer FID infrastructure, " | |
281 | "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err); | |
282 | GOTO(out_md, err); | |
283 | } | |
284 | ||
285 | /* For mount, we only need fs info from MDT0, and also in DNE, it | |
286 | * can make sure the client can be mounted as long as MDT0 is | |
287 | * avaible */ | |
288 | err = obd_statfs(NULL, sbi->ll_md_exp, osfs, | |
289 | cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), | |
290 | OBD_STATFS_FOR_MDT0); | |
291 | if (err) | |
292 | GOTO(out_md_fid, err); | |
293 | ||
294 | /* This needs to be after statfs to ensure connect has finished. | |
295 | * Note that "data" does NOT contain the valid connect reply. | |
296 | * If connecting to a 1.8 server there will be no LMV device, so | |
297 | * we can access the MDC export directly and exp_connect_flags will | |
298 | * be non-zero, but if accessing an upgraded 2.1 server it will | |
299 | * have the correct flags filled in. | |
300 | * XXX: fill in the LMV exp_connect_flags from MDC(s). */ | |
301 | valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD; | |
302 | if (exp_connect_flags(sbi->ll_md_exp) != 0 && | |
303 | valid != CLIENT_CONNECT_MDT_REQD) { | |
304 | char *buf; | |
305 | ||
306 | OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE); | |
307 | obd_connect_flags2str(buf, PAGE_CACHE_SIZE, | |
308 | valid ^ CLIENT_CONNECT_MDT_REQD, ","); | |
309 | LCONSOLE_ERROR_MSG(0x170, "Server %s does not support " | |
310 | "feature(s) needed for correct operation " | |
311 | "of this client (%s). Please upgrade " | |
312 | "server or downgrade client.\n", | |
313 | sbi->ll_md_exp->exp_obd->obd_name, buf); | |
314 | OBD_FREE(buf, PAGE_CACHE_SIZE); | |
315 | GOTO(out_md_fid, err = -EPROTO); | |
316 | } | |
317 | ||
318 | size = sizeof(*data); | |
319 | err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA), | |
320 | KEY_CONN_DATA, &size, data, NULL); | |
321 | if (err) { | |
322 | CERROR("%s: Get connect data failed: rc = %d\n", | |
323 | sbi->ll_md_exp->exp_obd->obd_name, err); | |
324 | GOTO(out_md_fid, err); | |
325 | } | |
326 | ||
327 | LASSERT(osfs->os_bsize); | |
328 | sb->s_blocksize = osfs->os_bsize; | |
329 | sb->s_blocksize_bits = log2(osfs->os_bsize); | |
330 | sb->s_magic = LL_SUPER_MAGIC; | |
331 | sb->s_maxbytes = MAX_LFS_FILESIZE; | |
332 | sbi->ll_namelen = osfs->os_namelen; | |
333 | sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK; | |
334 | ||
335 | if ((sbi->ll_flags & LL_SBI_USER_XATTR) && | |
336 | !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) { | |
337 | LCONSOLE_INFO("Disabling user_xattr feature because " | |
338 | "it is not supported on the server\n"); | |
339 | sbi->ll_flags &= ~LL_SBI_USER_XATTR; | |
340 | } | |
341 | ||
342 | if (data->ocd_connect_flags & OBD_CONNECT_ACL) { | |
343 | #ifdef MS_POSIXACL | |
344 | sb->s_flags |= MS_POSIXACL; | |
345 | #endif | |
346 | sbi->ll_flags |= LL_SBI_ACL; | |
347 | } else { | |
348 | LCONSOLE_INFO("client wants to enable acl, but mdt not!\n"); | |
349 | #ifdef MS_POSIXACL | |
350 | sb->s_flags &= ~MS_POSIXACL; | |
351 | #endif | |
352 | sbi->ll_flags &= ~LL_SBI_ACL; | |
353 | } | |
354 | ||
355 | if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) { | |
356 | if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) { | |
357 | sbi->ll_flags |= LL_SBI_RMT_CLIENT; | |
358 | LCONSOLE_INFO("client is set as remote by default.\n"); | |
359 | } | |
360 | } else { | |
361 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
362 | sbi->ll_flags &= ~LL_SBI_RMT_CLIENT; | |
363 | LCONSOLE_INFO("client claims to be remote, but server " | |
364 | "rejected, forced to be local.\n"); | |
365 | } | |
366 | } | |
367 | ||
368 | if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) { | |
369 | LCONSOLE_INFO("client enabled MDS capability!\n"); | |
370 | sbi->ll_flags |= LL_SBI_MDS_CAPA; | |
371 | } | |
372 | ||
373 | if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) { | |
374 | LCONSOLE_INFO("client enabled OSS capability!\n"); | |
375 | sbi->ll_flags |= LL_SBI_OSS_CAPA; | |
376 | } | |
377 | ||
378 | if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH) | |
379 | sbi->ll_flags |= LL_SBI_64BIT_HASH; | |
380 | ||
381 | if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) | |
382 | sbi->ll_md_brw_size = data->ocd_brw_size; | |
383 | else | |
384 | sbi->ll_md_brw_size = PAGE_CACHE_SIZE; | |
385 | ||
386 | if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) { | |
387 | LCONSOLE_INFO("Layout lock feature supported.\n"); | |
388 | sbi->ll_flags |= LL_SBI_LAYOUT_LOCK; | |
389 | } | |
390 | ||
391 | obd = class_name2obd(dt); | |
392 | if (!obd) { | |
393 | CERROR("DT %s: not setup or attached\n", dt); | |
394 | GOTO(out_md_fid, err = -ENODEV); | |
395 | } | |
396 | ||
397 | data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION | | |
398 | OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE | | |
399 | OBD_CONNECT_CANCELSET | OBD_CONNECT_FID | | |
400 | OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK| | |
401 | OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT | | |
402 | OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR| | |
403 | OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH | | |
404 | OBD_CONNECT_MAXBYTES | | |
405 | OBD_CONNECT_EINPROGRESS | | |
406 | OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE | | |
407 | OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS; | |
408 | ||
409 | if (sbi->ll_flags & LL_SBI_SOM_PREVIEW) | |
410 | data->ocd_connect_flags |= OBD_CONNECT_SOM; | |
411 | ||
412 | if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) { | |
413 | /* OBD_CONNECT_CKSUM should always be set, even if checksums are | |
414 | * disabled by default, because it can still be enabled on the | |
415 | * fly via /proc. As a consequence, we still need to come to an | |
416 | * agreement on the supported algorithms at connect time */ | |
417 | data->ocd_connect_flags |= OBD_CONNECT_CKSUM; | |
418 | ||
419 | if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY)) | |
420 | data->ocd_cksum_types = OBD_CKSUM_ADLER; | |
421 | else | |
422 | data->ocd_cksum_types = cksum_types_supported_client(); | |
423 | } | |
424 | ||
425 | data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE; | |
426 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) | |
427 | data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE; | |
428 | ||
429 | CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d " | |
430 | "ocd_grant: %d\n", data->ocd_connect_flags, | |
431 | data->ocd_version, data->ocd_grant); | |
432 | ||
433 | obd->obd_upcall.onu_owner = &sbi->ll_lco; | |
434 | obd->obd_upcall.onu_upcall = cl_ocd_update; | |
435 | ||
436 | data->ocd_brw_size = DT_MAX_BRW_SIZE; | |
437 | ||
438 | err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, | |
439 | NULL); | |
440 | if (err == -EBUSY) { | |
441 | LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing " | |
442 | "recovery, of which this client is not a " | |
443 | "part. Please wait for recovery to " | |
444 | "complete, abort, or time out.\n", dt); | |
445 | GOTO(out_md, err); | |
446 | } else if (err) { | |
447 | CERROR("%s: Cannot connect to %s: rc = %d\n", | |
448 | sbi->ll_dt_exp->exp_obd->obd_name, dt, err); | |
449 | GOTO(out_md, err); | |
450 | } | |
451 | ||
452 | sbi->ll_dt_exp->exp_connect_data = *data; | |
453 | ||
454 | err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp, | |
455 | LUSTRE_SEQ_METADATA); | |
456 | if (err) { | |
457 | CERROR("%s: Can't init data layer FID infrastructure, " | |
458 | "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err); | |
459 | GOTO(out_dt, err); | |
460 | } | |
461 | ||
462 | mutex_lock(&sbi->ll_lco.lco_lock); | |
463 | sbi->ll_lco.lco_flags = data->ocd_connect_flags; | |
464 | sbi->ll_lco.lco_md_exp = sbi->ll_md_exp; | |
465 | sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp; | |
466 | mutex_unlock(&sbi->ll_lco.lco_lock); | |
467 | ||
468 | fid_zero(&sbi->ll_root_fid); | |
469 | err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc); | |
470 | if (err) { | |
471 | CERROR("cannot mds_connect: rc = %d\n", err); | |
472 | GOTO(out_lock_cn_cb, err); | |
473 | } | |
474 | if (!fid_is_sane(&sbi->ll_root_fid)) { | |
475 | CERROR("%s: Invalid root fid "DFID" during mount\n", | |
476 | sbi->ll_md_exp->exp_obd->obd_name, | |
477 | PFID(&sbi->ll_root_fid)); | |
478 | GOTO(out_lock_cn_cb, err = -EINVAL); | |
479 | } | |
480 | CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid)); | |
481 | ||
482 | sb->s_op = &lustre_super_operations; | |
483 | #if THREAD_SIZE >= 8192 /*b=17630*/ | |
484 | sb->s_export_op = &lustre_export_operations; | |
485 | #endif | |
486 | ||
487 | /* make root inode | |
488 | * XXX: move this to after cbd setup? */ | |
489 | valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA; | |
490 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) | |
491 | valid |= OBD_MD_FLRMTPERM; | |
492 | else if (sbi->ll_flags & LL_SBI_ACL) | |
493 | valid |= OBD_MD_FLACL; | |
494 | ||
495 | OBD_ALLOC_PTR(op_data); | |
496 | if (op_data == NULL) | |
497 | GOTO(out_lock_cn_cb, err = -ENOMEM); | |
498 | ||
499 | op_data->op_fid1 = sbi->ll_root_fid; | |
500 | op_data->op_mode = 0; | |
501 | op_data->op_capa1 = oc; | |
502 | op_data->op_valid = valid; | |
503 | ||
504 | err = md_getattr(sbi->ll_md_exp, op_data, &request); | |
505 | if (oc) | |
506 | capa_put(oc); | |
507 | OBD_FREE_PTR(op_data); | |
508 | if (err) { | |
509 | CERROR("%s: md_getattr failed for root: rc = %d\n", | |
510 | sbi->ll_md_exp->exp_obd->obd_name, err); | |
511 | GOTO(out_lock_cn_cb, err); | |
512 | } | |
513 | ||
514 | err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp, | |
515 | sbi->ll_md_exp, &lmd); | |
516 | if (err) { | |
517 | CERROR("failed to understand root inode md: rc = %d\n", err); | |
518 | ptlrpc_req_finished(request); | |
519 | GOTO(out_lock_cn_cb, err); | |
520 | } | |
521 | ||
522 | LASSERT(fid_is_sane(&sbi->ll_root_fid)); | |
523 | root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, | |
c1e2699d | 524 | sbi->ll_flags & LL_SBI_32BIT_API), |
d7e09d03 PT |
525 | &lmd); |
526 | md_free_lustre_md(sbi->ll_md_exp, &lmd); | |
527 | ptlrpc_req_finished(request); | |
528 | ||
529 | if (root == NULL || IS_ERR(root)) { | |
530 | if (lmd.lsm) | |
531 | obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm); | |
532 | #ifdef CONFIG_FS_POSIX_ACL | |
533 | if (lmd.posix_acl) { | |
534 | posix_acl_release(lmd.posix_acl); | |
535 | lmd.posix_acl = NULL; | |
536 | } | |
537 | #endif | |
538 | err = IS_ERR(root) ? PTR_ERR(root) : -EBADF; | |
539 | root = NULL; | |
540 | CERROR("lustre_lite: bad iget4 for root\n"); | |
541 | GOTO(out_root, err); | |
542 | } | |
543 | ||
544 | err = ll_close_thread_start(&sbi->ll_lcq); | |
545 | if (err) { | |
546 | CERROR("cannot start close thread: rc %d\n", err); | |
547 | GOTO(out_root, err); | |
548 | } | |
549 | ||
550 | #ifdef CONFIG_FS_POSIX_ACL | |
551 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
552 | rct_init(&sbi->ll_rct); | |
553 | et_init(&sbi->ll_et); | |
554 | } | |
555 | #endif | |
556 | ||
557 | checksum = sbi->ll_flags & LL_SBI_CHECKSUM; | |
558 | err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM), | |
559 | KEY_CHECKSUM, sizeof(checksum), &checksum, | |
560 | NULL); | |
561 | cl_sb_init(sb); | |
562 | ||
563 | err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET), | |
564 | KEY_CACHE_SET, sizeof(sbi->ll_cache), | |
565 | &sbi->ll_cache, NULL); | |
566 | ||
567 | sb->s_root = d_make_root(root); | |
568 | if (sb->s_root == NULL) { | |
569 | CERROR("%s: can't make root dentry\n", | |
570 | ll_get_fsname(sb, NULL, 0)); | |
571 | GOTO(out_root, err = -ENOMEM); | |
572 | } | |
573 | ||
574 | /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */ | |
575 | d_set_d_op(sb->s_root, &ll_d_root_ops); | |
576 | sb->s_d_op = &ll_d_ops; | |
577 | ||
578 | sbi->ll_sdev_orig = sb->s_dev; | |
579 | ||
580 | /* We set sb->s_dev equal on all lustre clients in order to support | |
581 | * NFS export clustering. NFSD requires that the FSID be the same | |
582 | * on all clients. */ | |
583 | /* s_dev is also used in lt_compare() to compare two fs, but that is | |
584 | * only a node-local comparison. */ | |
585 | uuid = obd_get_uuid(sbi->ll_md_exp); | |
586 | if (uuid != NULL) | |
587 | sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid)); | |
588 | ||
589 | if (data != NULL) | |
590 | OBD_FREE_PTR(data); | |
591 | if (osfs != NULL) | |
592 | OBD_FREE_PTR(osfs); | |
593 | ||
594 | RETURN(err); | |
595 | out_root: | |
596 | if (root) | |
597 | iput(root); | |
598 | out_lock_cn_cb: | |
599 | obd_fid_fini(sbi->ll_dt_exp->exp_obd); | |
600 | out_dt: | |
601 | obd_disconnect(sbi->ll_dt_exp); | |
602 | sbi->ll_dt_exp = NULL; | |
603 | /* Make sure all OScs are gone, since cl_cache is accessing sbi. */ | |
604 | obd_zombie_barrier(); | |
605 | out_md_fid: | |
606 | obd_fid_fini(sbi->ll_md_exp->exp_obd); | |
607 | out_md: | |
608 | obd_disconnect(sbi->ll_md_exp); | |
609 | sbi->ll_md_exp = NULL; | |
610 | out: | |
611 | if (data != NULL) | |
612 | OBD_FREE_PTR(data); | |
613 | if (osfs != NULL) | |
614 | OBD_FREE_PTR(osfs); | |
615 | lprocfs_unregister_mountpoint(sbi); | |
616 | return err; | |
617 | } | |
618 | ||
619 | int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize) | |
620 | { | |
621 | int size, rc; | |
622 | ||
623 | *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL); | |
624 | size = sizeof(int); | |
625 | rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE), | |
626 | KEY_MAX_EASIZE, &size, lmmsize, NULL); | |
627 | if (rc) | |
628 | CERROR("Get max mdsize error rc %d \n", rc); | |
629 | ||
630 | RETURN(rc); | |
631 | } | |
632 | ||
633 | void ll_dump_inode(struct inode *inode) | |
634 | { | |
635 | struct ll_d_hlist_node *tmp; | |
636 | int dentry_count = 0; | |
637 | ||
638 | LASSERT(inode != NULL); | |
639 | ||
640 | ll_d_hlist_for_each(tmp, &inode->i_dentry) | |
641 | dentry_count++; | |
642 | ||
643 | CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n", | |
644 | inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino, | |
645 | inode->i_mode, atomic_read(&inode->i_count), dentry_count); | |
646 | } | |
647 | ||
648 | void lustre_dump_dentry(struct dentry *dentry, int recur) | |
649 | { | |
650 | struct list_head *tmp; | |
651 | int subdirs = 0; | |
652 | ||
653 | LASSERT(dentry != NULL); | |
654 | ||
655 | list_for_each(tmp, &dentry->d_subdirs) | |
656 | subdirs++; | |
657 | ||
658 | CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u," | |
659 | " flags=0x%x, fsdata=%p, %d subdirs\n", dentry, | |
660 | dentry->d_name.len, dentry->d_name.name, | |
661 | dentry->d_parent->d_name.len, dentry->d_parent->d_name.name, | |
193deee1 | 662 | dentry->d_parent, dentry->d_inode, d_count(dentry), |
d7e09d03 PT |
663 | dentry->d_flags, dentry->d_fsdata, subdirs); |
664 | if (dentry->d_inode != NULL) | |
665 | ll_dump_inode(dentry->d_inode); | |
666 | ||
667 | if (recur == 0) | |
668 | return; | |
669 | ||
670 | list_for_each(tmp, &dentry->d_subdirs) { | |
671 | struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child); | |
672 | lustre_dump_dentry(d, recur - 1); | |
673 | } | |
674 | } | |
675 | ||
676 | void client_common_put_super(struct super_block *sb) | |
677 | { | |
678 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
679 | ENTRY; | |
680 | ||
681 | #ifdef CONFIG_FS_POSIX_ACL | |
682 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
683 | et_fini(&sbi->ll_et); | |
684 | rct_fini(&sbi->ll_rct); | |
685 | } | |
686 | #endif | |
687 | ||
688 | ll_close_thread_shutdown(sbi->ll_lcq); | |
689 | ||
690 | cl_sb_fini(sb); | |
691 | ||
692 | list_del(&sbi->ll_conn_chain); | |
693 | ||
694 | obd_fid_fini(sbi->ll_dt_exp->exp_obd); | |
695 | obd_disconnect(sbi->ll_dt_exp); | |
696 | sbi->ll_dt_exp = NULL; | |
697 | /* wait till all OSCs are gone, since cl_cache is accessing sbi. | |
698 | * see LU-2543. */ | |
699 | obd_zombie_barrier(); | |
700 | ||
701 | lprocfs_unregister_mountpoint(sbi); | |
702 | ||
703 | obd_fid_fini(sbi->ll_md_exp->exp_obd); | |
704 | obd_disconnect(sbi->ll_md_exp); | |
705 | sbi->ll_md_exp = NULL; | |
706 | ||
707 | EXIT; | |
708 | } | |
709 | ||
710 | void ll_kill_super(struct super_block *sb) | |
711 | { | |
712 | struct ll_sb_info *sbi; | |
713 | ||
714 | ENTRY; | |
715 | ||
716 | /* not init sb ?*/ | |
717 | if (!(sb->s_flags & MS_ACTIVE)) | |
718 | return; | |
719 | ||
720 | sbi = ll_s2sbi(sb); | |
721 | /* we need restore s_dev from changed for clustred NFS before put_super | |
722 | * because new kernels have cached s_dev and change sb->s_dev in | |
723 | * put_super not affected real removing devices */ | |
65fb55d1 | 724 | if (sbi) { |
d7e09d03 | 725 | sb->s_dev = sbi->ll_sdev_orig; |
65fb55d1 NY |
726 | sbi->ll_umounting = 1; |
727 | } | |
d7e09d03 PT |
728 | EXIT; |
729 | } | |
730 | ||
731 | char *ll_read_opt(const char *opt, char *data) | |
732 | { | |
733 | char *value; | |
734 | char *retval; | |
735 | ENTRY; | |
736 | ||
737 | CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data); | |
738 | if (strncmp(opt, data, strlen(opt))) | |
739 | RETURN(NULL); | |
740 | if ((value = strchr(data, '=')) == NULL) | |
741 | RETURN(NULL); | |
742 | ||
743 | value++; | |
744 | OBD_ALLOC(retval, strlen(value) + 1); | |
745 | if (!retval) { | |
746 | CERROR("out of memory!\n"); | |
747 | RETURN(NULL); | |
748 | } | |
749 | ||
750 | memcpy(retval, value, strlen(value)+1); | |
751 | CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval); | |
752 | RETURN(retval); | |
753 | } | |
754 | ||
755 | static inline int ll_set_opt(const char *opt, char *data, int fl) | |
756 | { | |
757 | if (strncmp(opt, data, strlen(opt)) != 0) | |
758 | return(0); | |
759 | else | |
760 | return(fl); | |
761 | } | |
762 | ||
763 | /* non-client-specific mount options are parsed in lmd_parse */ | |
764 | static int ll_options(char *options, int *flags) | |
765 | { | |
766 | int tmp; | |
767 | char *s1 = options, *s2; | |
768 | ENTRY; | |
769 | ||
770 | if (!options) | |
771 | RETURN(0); | |
772 | ||
773 | CDEBUG(D_CONFIG, "Parsing opts %s\n", options); | |
774 | ||
775 | while (*s1) { | |
776 | CDEBUG(D_SUPER, "next opt=%s\n", s1); | |
777 | tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK); | |
778 | if (tmp) { | |
779 | *flags |= tmp; | |
780 | goto next; | |
781 | } | |
782 | tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK); | |
783 | if (tmp) { | |
784 | *flags |= tmp; | |
785 | goto next; | |
786 | } | |
787 | tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK); | |
788 | if (tmp) { | |
789 | *flags |= tmp; | |
790 | goto next; | |
791 | } | |
792 | tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK); | |
793 | if (tmp) { | |
794 | *flags &= ~tmp; | |
795 | goto next; | |
796 | } | |
797 | tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR); | |
798 | if (tmp) { | |
799 | *flags |= tmp; | |
800 | goto next; | |
801 | } | |
802 | tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR); | |
803 | if (tmp) { | |
804 | *flags &= ~tmp; | |
805 | goto next; | |
806 | } | |
807 | #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0) | |
808 | tmp = ll_set_opt("acl", s1, LL_SBI_ACL); | |
809 | if (tmp) { | |
810 | /* Ignore deprecated mount option. The client will | |
811 | * always try to mount with ACL support, whether this | |
812 | * is used depends on whether server supports it. */ | |
813 | LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated " | |
814 | "mount option 'acl'.\n"); | |
815 | goto next; | |
816 | } | |
817 | tmp = ll_set_opt("noacl", s1, LL_SBI_ACL); | |
818 | if (tmp) { | |
819 | LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated " | |
820 | "mount option 'noacl'.\n"); | |
821 | goto next; | |
822 | } | |
823 | #else | |
824 | #warning "{no}acl options have been deprecated since 1.8, please remove them" | |
825 | #endif | |
826 | tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT); | |
827 | if (tmp) { | |
828 | *flags |= tmp; | |
829 | goto next; | |
830 | } | |
831 | tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH); | |
832 | if (tmp) { | |
833 | *flags |= tmp; | |
834 | goto next; | |
835 | } | |
836 | tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH); | |
837 | if (tmp) { | |
838 | *flags &= ~tmp; | |
839 | goto next; | |
840 | } | |
841 | ||
842 | tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM); | |
843 | if (tmp) { | |
844 | *flags |= tmp; | |
845 | goto next; | |
846 | } | |
847 | tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM); | |
848 | if (tmp) { | |
849 | *flags &= ~tmp; | |
850 | goto next; | |
851 | } | |
852 | tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE); | |
853 | if (tmp) { | |
854 | *flags |= tmp; | |
855 | goto next; | |
856 | } | |
857 | tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE); | |
858 | if (tmp) { | |
859 | *flags &= ~tmp; | |
860 | goto next; | |
861 | } | |
862 | tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS); | |
863 | if (tmp) { | |
864 | *flags |= tmp; | |
865 | goto next; | |
866 | } | |
867 | tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS); | |
868 | if (tmp) { | |
869 | *flags &= ~tmp; | |
870 | goto next; | |
871 | } | |
872 | tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW); | |
873 | if (tmp) { | |
874 | *flags |= tmp; | |
875 | goto next; | |
876 | } | |
877 | tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API); | |
878 | if (tmp) { | |
879 | *flags |= tmp; | |
880 | goto next; | |
881 | } | |
882 | tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE); | |
883 | if (tmp) { | |
884 | *flags |= tmp; | |
885 | goto next; | |
886 | } | |
887 | tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE); | |
888 | if (tmp) { | |
889 | *flags &= ~tmp; | |
890 | goto next; | |
891 | } | |
892 | LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n", | |
893 | s1); | |
894 | RETURN(-EINVAL); | |
895 | ||
896 | next: | |
897 | /* Find next opt */ | |
898 | s2 = strchr(s1, ','); | |
899 | if (s2 == NULL) | |
900 | break; | |
901 | s1 = s2 + 1; | |
902 | } | |
903 | RETURN(0); | |
904 | } | |
905 | ||
906 | void ll_lli_init(struct ll_inode_info *lli) | |
907 | { | |
908 | lli->lli_inode_magic = LLI_INODE_MAGIC; | |
909 | lli->lli_flags = 0; | |
910 | lli->lli_ioepoch = 0; | |
911 | lli->lli_maxbytes = MAX_LFS_FILESIZE; | |
912 | spin_lock_init(&lli->lli_lock); | |
913 | lli->lli_posix_acl = NULL; | |
914 | lli->lli_remote_perms = NULL; | |
915 | mutex_init(&lli->lli_rmtperm_mutex); | |
916 | /* Do not set lli_fid, it has been initialized already. */ | |
917 | fid_zero(&lli->lli_pfid); | |
918 | INIT_LIST_HEAD(&lli->lli_close_list); | |
919 | INIT_LIST_HEAD(&lli->lli_oss_capas); | |
920 | atomic_set(&lli->lli_open_count, 0); | |
921 | lli->lli_mds_capa = NULL; | |
922 | lli->lli_rmtperm_time = 0; | |
923 | lli->lli_pending_och = NULL; | |
924 | lli->lli_mds_read_och = NULL; | |
925 | lli->lli_mds_write_och = NULL; | |
926 | lli->lli_mds_exec_och = NULL; | |
927 | lli->lli_open_fd_read_count = 0; | |
928 | lli->lli_open_fd_write_count = 0; | |
929 | lli->lli_open_fd_exec_count = 0; | |
930 | mutex_init(&lli->lli_och_mutex); | |
931 | spin_lock_init(&lli->lli_agl_lock); | |
932 | lli->lli_has_smd = false; | |
933 | lli->lli_layout_gen = LL_LAYOUT_GEN_NONE; | |
934 | lli->lli_clob = NULL; | |
935 | ||
936 | LASSERT(lli->lli_vfs_inode.i_mode != 0); | |
937 | if (S_ISDIR(lli->lli_vfs_inode.i_mode)) { | |
938 | mutex_init(&lli->lli_readdir_mutex); | |
939 | lli->lli_opendir_key = NULL; | |
940 | lli->lli_sai = NULL; | |
941 | lli->lli_def_acl = NULL; | |
942 | spin_lock_init(&lli->lli_sa_lock); | |
943 | lli->lli_opendir_pid = 0; | |
944 | } else { | |
945 | sema_init(&lli->lli_size_sem, 1); | |
946 | lli->lli_size_sem_owner = NULL; | |
947 | lli->lli_symlink_name = NULL; | |
948 | init_rwsem(&lli->lli_trunc_sem); | |
949 | mutex_init(&lli->lli_write_mutex); | |
950 | init_rwsem(&lli->lli_glimpse_sem); | |
951 | lli->lli_glimpse_time = 0; | |
952 | INIT_LIST_HEAD(&lli->lli_agl_list); | |
953 | lli->lli_agl_index = 0; | |
954 | lli->lli_async_rc = 0; | |
955 | lli->lli_volatile = false; | |
956 | } | |
957 | mutex_init(&lli->lli_layout_mutex); | |
958 | } | |
959 | ||
960 | static inline int ll_bdi_register(struct backing_dev_info *bdi) | |
961 | { | |
962 | static atomic_t ll_bdi_num = ATOMIC_INIT(0); | |
963 | ||
964 | bdi->name = "lustre"; | |
965 | return bdi_register(bdi, NULL, "lustre-%d", | |
966 | atomic_inc_return(&ll_bdi_num)); | |
967 | } | |
968 | ||
969 | int ll_fill_super(struct super_block *sb, struct vfsmount *mnt) | |
970 | { | |
971 | struct lustre_profile *lprof = NULL; | |
972 | struct lustre_sb_info *lsi = s2lsi(sb); | |
973 | struct ll_sb_info *sbi; | |
974 | char *dt = NULL, *md = NULL; | |
975 | char *profilenm = get_profile_name(sb); | |
976 | struct config_llog_instance *cfg; | |
977 | /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */ | |
978 | const int instlen = sizeof(cfg->cfg_instance) * 2 + 2; | |
979 | int err; | |
980 | ENTRY; | |
981 | ||
982 | CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb); | |
983 | ||
984 | OBD_ALLOC_PTR(cfg); | |
985 | if (cfg == NULL) | |
986 | RETURN(-ENOMEM); | |
987 | ||
988 | try_module_get(THIS_MODULE); | |
989 | ||
990 | /* client additional sb info */ | |
991 | lsi->lsi_llsbi = sbi = ll_init_sbi(); | |
992 | if (!sbi) { | |
993 | module_put(THIS_MODULE); | |
994 | OBD_FREE_PTR(cfg); | |
995 | RETURN(-ENOMEM); | |
996 | } | |
997 | ||
998 | err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags); | |
999 | if (err) | |
1000 | GOTO(out_free, err); | |
1001 | ||
1002 | err = bdi_init(&lsi->lsi_bdi); | |
1003 | if (err) | |
1004 | GOTO(out_free, err); | |
1005 | lsi->lsi_flags |= LSI_BDI_INITIALIZED; | |
1006 | lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY; | |
1007 | err = ll_bdi_register(&lsi->lsi_bdi); | |
1008 | if (err) | |
1009 | GOTO(out_free, err); | |
1010 | ||
1011 | sb->s_bdi = &lsi->lsi_bdi; | |
1012 | ||
1013 | /* Generate a string unique to this super, in case some joker tries | |
1014 | to mount the same fs at two mount points. | |
1015 | Use the address of the super itself.*/ | |
1016 | cfg->cfg_instance = sb; | |
1017 | cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid; | |
1018 | cfg->cfg_callback = class_config_llog_handler; | |
1019 | /* set up client obds */ | |
1020 | err = lustre_process_log(sb, profilenm, cfg); | |
1021 | if (err < 0) { | |
1022 | CERROR("Unable to process log: %d\n", err); | |
1023 | GOTO(out_free, err); | |
1024 | } | |
1025 | ||
1026 | /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */ | |
1027 | lprof = class_get_profile(profilenm); | |
1028 | if (lprof == NULL) { | |
1029 | LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be" | |
1030 | " read from the MGS. Does that filesystem " | |
1031 | "exist?\n", profilenm); | |
1032 | GOTO(out_free, err = -EINVAL); | |
1033 | } | |
1034 | CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm, | |
1035 | lprof->lp_md, lprof->lp_dt); | |
1036 | ||
1037 | OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2); | |
1038 | if (!dt) | |
1039 | GOTO(out_free, err = -ENOMEM); | |
1040 | sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance); | |
1041 | ||
1042 | OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2); | |
1043 | if (!md) | |
1044 | GOTO(out_free, err = -ENOMEM); | |
1045 | sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance); | |
1046 | ||
1047 | /* connections, registrations, sb setup */ | |
1048 | err = client_common_fill_super(sb, md, dt, mnt); | |
1049 | ||
1050 | out_free: | |
1051 | if (md) | |
1052 | OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2); | |
1053 | if (dt) | |
1054 | OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2); | |
1055 | if (err) | |
1056 | ll_put_super(sb); | |
1057 | else if (sbi->ll_flags & LL_SBI_VERBOSE) | |
1058 | LCONSOLE_WARN("Mounted %s\n", profilenm); | |
1059 | ||
1060 | OBD_FREE_PTR(cfg); | |
1061 | RETURN(err); | |
1062 | } /* ll_fill_super */ | |
1063 | ||
d7e09d03 PT |
1064 | void ll_put_super(struct super_block *sb) |
1065 | { | |
1066 | struct config_llog_instance cfg; | |
1067 | struct obd_device *obd; | |
1068 | struct lustre_sb_info *lsi = s2lsi(sb); | |
1069 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
1070 | char *profilenm = get_profile_name(sb); | |
c52f69c5 | 1071 | int next, force = 1; |
d7e09d03 PT |
1072 | ENTRY; |
1073 | ||
1074 | CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm); | |
1075 | ||
1076 | ll_print_capa_stat(sbi); | |
1077 | ||
1078 | cfg.cfg_instance = sb; | |
1079 | lustre_end_log(sb, profilenm, &cfg); | |
1080 | ||
1081 | if (sbi->ll_md_exp) { | |
1082 | obd = class_exp2obd(sbi->ll_md_exp); | |
1083 | if (obd) | |
1084 | force = obd->obd_force; | |
1085 | } | |
1086 | ||
d7e09d03 PT |
1087 | /* We need to set force before the lov_disconnect in |
1088 | lustre_common_put_super, since l_d cleans up osc's as well. */ | |
1089 | if (force) { | |
1090 | next = 0; | |
1091 | while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, | |
1092 | &next)) != NULL) { | |
1093 | obd->obd_force = force; | |
1094 | } | |
1095 | } | |
1096 | ||
1097 | if (sbi->ll_lcq) { | |
1098 | /* Only if client_common_fill_super succeeded */ | |
1099 | client_common_put_super(sb); | |
1100 | } | |
1101 | ||
1102 | next = 0; | |
1103 | while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) { | |
1104 | class_manual_cleanup(obd); | |
1105 | } | |
1106 | ||
1107 | if (sbi->ll_flags & LL_SBI_VERBOSE) | |
1108 | LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : ""); | |
1109 | ||
1110 | if (profilenm) | |
1111 | class_del_profile(profilenm); | |
1112 | ||
1113 | if (lsi->lsi_flags & LSI_BDI_INITIALIZED) { | |
1114 | bdi_destroy(&lsi->lsi_bdi); | |
1115 | lsi->lsi_flags &= ~LSI_BDI_INITIALIZED; | |
1116 | } | |
1117 | ||
1118 | ll_free_sbi(sb); | |
1119 | lsi->lsi_llsbi = NULL; | |
1120 | ||
1121 | lustre_common_put_super(sb); | |
1122 | ||
1123 | module_put(THIS_MODULE); | |
1124 | ||
1125 | EXIT; | |
1126 | } /* client_put_super */ | |
1127 | ||
1128 | struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock) | |
1129 | { | |
1130 | struct inode *inode = NULL; | |
1131 | ||
1132 | /* NOTE: we depend on atomic igrab() -bzzz */ | |
1133 | lock_res_and_lock(lock); | |
1134 | if (lock->l_resource->lr_lvb_inode) { | |
1135 | struct ll_inode_info * lli; | |
1136 | lli = ll_i2info(lock->l_resource->lr_lvb_inode); | |
1137 | if (lli->lli_inode_magic == LLI_INODE_MAGIC) { | |
1138 | inode = igrab(lock->l_resource->lr_lvb_inode); | |
1139 | } else { | |
1140 | inode = lock->l_resource->lr_lvb_inode; | |
1141 | LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO : | |
1142 | D_WARNING, lock, "lr_lvb_inode %p is " | |
1143 | "bogus: magic %08x", | |
1144 | lock->l_resource->lr_lvb_inode, | |
1145 | lli->lli_inode_magic); | |
1146 | inode = NULL; | |
1147 | } | |
1148 | } | |
1149 | unlock_res_and_lock(lock); | |
1150 | return inode; | |
1151 | } | |
1152 | ||
1153 | struct inode *ll_inode_from_lock(struct ldlm_lock *lock) | |
1154 | { | |
1155 | struct inode *inode = NULL; | |
1156 | /* NOTE: we depend on atomic igrab() -bzzz */ | |
1157 | lock_res_and_lock(lock); | |
1158 | if (lock->l_ast_data) { | |
1159 | struct ll_inode_info *lli = ll_i2info(lock->l_ast_data); | |
1160 | if (lli->lli_inode_magic == LLI_INODE_MAGIC) { | |
1161 | inode = igrab(lock->l_ast_data); | |
1162 | } else { | |
1163 | inode = lock->l_ast_data; | |
1164 | LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO : | |
1165 | D_WARNING, lock, "l_ast_data %p is " | |
1166 | "bogus: magic %08x", lock->l_ast_data, | |
1167 | lli->lli_inode_magic); | |
1168 | inode = NULL; | |
1169 | } | |
1170 | } | |
1171 | unlock_res_and_lock(lock); | |
1172 | return inode; | |
1173 | } | |
1174 | ||
1175 | void ll_clear_inode(struct inode *inode) | |
1176 | { | |
1177 | struct ll_inode_info *lli = ll_i2info(inode); | |
1178 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1179 | ENTRY; | |
1180 | ||
1181 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, | |
1182 | inode->i_generation, inode); | |
1183 | ||
1184 | if (S_ISDIR(inode->i_mode)) { | |
1185 | /* these should have been cleared in ll_file_release */ | |
1186 | LASSERT(lli->lli_opendir_key == NULL); | |
1187 | LASSERT(lli->lli_sai == NULL); | |
1188 | LASSERT(lli->lli_opendir_pid == 0); | |
1189 | } | |
1190 | ||
1191 | ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK; | |
1192 | md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode)); | |
1193 | ||
1194 | LASSERT(!lli->lli_open_fd_write_count); | |
1195 | LASSERT(!lli->lli_open_fd_read_count); | |
1196 | LASSERT(!lli->lli_open_fd_exec_count); | |
1197 | ||
1198 | if (lli->lli_mds_write_och) | |
1199 | ll_md_real_close(inode, FMODE_WRITE); | |
1200 | if (lli->lli_mds_exec_och) | |
1201 | ll_md_real_close(inode, FMODE_EXEC); | |
1202 | if (lli->lli_mds_read_och) | |
1203 | ll_md_real_close(inode, FMODE_READ); | |
1204 | ||
1205 | if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) { | |
1206 | OBD_FREE(lli->lli_symlink_name, | |
1207 | strlen(lli->lli_symlink_name) + 1); | |
1208 | lli->lli_symlink_name = NULL; | |
1209 | } | |
1210 | ||
1211 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
1212 | LASSERT(lli->lli_posix_acl == NULL); | |
1213 | if (lli->lli_remote_perms) { | |
1214 | free_rmtperm_hash(lli->lli_remote_perms); | |
1215 | lli->lli_remote_perms = NULL; | |
1216 | } | |
1217 | } | |
1218 | #ifdef CONFIG_FS_POSIX_ACL | |
1219 | else if (lli->lli_posix_acl) { | |
1220 | LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1); | |
1221 | LASSERT(lli->lli_remote_perms == NULL); | |
1222 | posix_acl_release(lli->lli_posix_acl); | |
1223 | lli->lli_posix_acl = NULL; | |
1224 | } | |
1225 | #endif | |
1226 | lli->lli_inode_magic = LLI_INODE_DEAD; | |
1227 | ||
1228 | ll_clear_inode_capas(inode); | |
1229 | if (!S_ISDIR(inode->i_mode)) | |
1230 | LASSERT(list_empty(&lli->lli_agl_list)); | |
1231 | ||
1232 | /* | |
1233 | * XXX This has to be done before lsm is freed below, because | |
1234 | * cl_object still uses inode lsm. | |
1235 | */ | |
1236 | cl_inode_fini(inode); | |
1237 | lli->lli_has_smd = false; | |
1238 | ||
1239 | EXIT; | |
1240 | } | |
1241 | ||
1242 | int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data, | |
1243 | struct md_open_data **mod) | |
1244 | { | |
1245 | struct lustre_md md; | |
1246 | struct inode *inode = dentry->d_inode; | |
1247 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1248 | struct ptlrpc_request *request = NULL; | |
1249 | int rc, ia_valid; | |
1250 | ENTRY; | |
1251 | ||
1252 | op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, | |
1253 | LUSTRE_OPC_ANY, NULL); | |
1254 | if (IS_ERR(op_data)) | |
1255 | RETURN(PTR_ERR(op_data)); | |
1256 | ||
1257 | rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0, | |
1258 | &request, mod); | |
1259 | if (rc) { | |
1260 | ptlrpc_req_finished(request); | |
1261 | if (rc == -ENOENT) { | |
1262 | clear_nlink(inode); | |
1263 | /* Unlinked special device node? Or just a race? | |
1264 | * Pretend we done everything. */ | |
1265 | if (!S_ISREG(inode->i_mode) && | |
1266 | !S_ISDIR(inode->i_mode)) { | |
1267 | ia_valid = op_data->op_attr.ia_valid; | |
1268 | op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS; | |
1269 | rc = simple_setattr(dentry, &op_data->op_attr); | |
1270 | op_data->op_attr.ia_valid = ia_valid; | |
1271 | } | |
1272 | } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) { | |
1273 | CERROR("md_setattr fails: rc = %d\n", rc); | |
1274 | } | |
1275 | RETURN(rc); | |
1276 | } | |
1277 | ||
1278 | rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp, | |
1279 | sbi->ll_md_exp, &md); | |
1280 | if (rc) { | |
1281 | ptlrpc_req_finished(request); | |
1282 | RETURN(rc); | |
1283 | } | |
1284 | ||
251c4317 JH |
1285 | ia_valid = op_data->op_attr.ia_valid; |
1286 | /* inode size will be in ll_setattr_ost, can't do it now since dirty | |
1287 | * cache is not cleared yet. */ | |
1288 | op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE); | |
1289 | rc = simple_setattr(dentry, &op_data->op_attr); | |
1290 | op_data->op_attr.ia_valid = ia_valid; | |
1291 | ||
d7e09d03 PT |
1292 | /* Extract epoch data if obtained. */ |
1293 | op_data->op_handle = md.body->handle; | |
1294 | op_data->op_ioepoch = md.body->ioepoch; | |
1295 | ||
1296 | ll_update_inode(inode, &md); | |
1297 | ptlrpc_req_finished(request); | |
1298 | ||
1299 | RETURN(rc); | |
1300 | } | |
1301 | ||
1302 | /* Close IO epoch and send Size-on-MDS attribute update. */ | |
1303 | static int ll_setattr_done_writing(struct inode *inode, | |
1304 | struct md_op_data *op_data, | |
1305 | struct md_open_data *mod) | |
1306 | { | |
1307 | struct ll_inode_info *lli = ll_i2info(inode); | |
1308 | int rc = 0; | |
1309 | ENTRY; | |
1310 | ||
1311 | LASSERT(op_data != NULL); | |
1312 | if (!S_ISREG(inode->i_mode)) | |
1313 | RETURN(0); | |
1314 | ||
1315 | CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n", | |
1316 | op_data->op_ioepoch, PFID(&lli->lli_fid)); | |
1317 | ||
1318 | op_data->op_flags = MF_EPOCH_CLOSE; | |
1319 | ll_done_writing_attr(inode, op_data); | |
1320 | ll_pack_inode2opdata(inode, op_data, NULL); | |
1321 | ||
1322 | rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod); | |
1323 | if (rc == -EAGAIN) { | |
1324 | /* MDS has instructed us to obtain Size-on-MDS attribute | |
1325 | * from OSTs and send setattr to back to MDS. */ | |
1326 | rc = ll_som_update(inode, op_data); | |
1327 | } else if (rc) { | |
1328 | CERROR("inode %lu mdc truncate failed: rc = %d\n", | |
1329 | inode->i_ino, rc); | |
1330 | } | |
1331 | RETURN(rc); | |
1332 | } | |
1333 | ||
1334 | static int ll_setattr_ost(struct inode *inode, struct iattr *attr) | |
1335 | { | |
1336 | struct obd_capa *capa; | |
1337 | int rc; | |
1338 | ||
1339 | if (attr->ia_valid & ATTR_SIZE) | |
1340 | capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC); | |
1341 | else | |
1342 | capa = ll_mdscapa_get(inode); | |
1343 | ||
1344 | rc = cl_setattr_ost(inode, attr, capa); | |
1345 | ||
1346 | if (attr->ia_valid & ATTR_SIZE) | |
1347 | ll_truncate_free_capa(capa); | |
1348 | else | |
1349 | capa_put(capa); | |
1350 | ||
1351 | return rc; | |
1352 | } | |
1353 | ||
1354 | ||
1355 | /* If this inode has objects allocated to it (lsm != NULL), then the OST | |
1356 | * object(s) determine the file size and mtime. Otherwise, the MDS will | |
1357 | * keep these values until such a time that objects are allocated for it. | |
1358 | * We do the MDS operations first, as it is checking permissions for us. | |
1359 | * We don't to the MDS RPC if there is nothing that we want to store there, | |
1360 | * otherwise there is no harm in updating mtime/atime on the MDS if we are | |
1361 | * going to do an RPC anyways. | |
1362 | * | |
1363 | * If we are doing a truncate, we will send the mtime and ctime updates | |
1364 | * to the OST with the punch RPC, otherwise we do an explicit setattr RPC. | |
1365 | * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE | |
1366 | * at the same time. | |
1367 | */ | |
1368 | int ll_setattr_raw(struct dentry *dentry, struct iattr *attr) | |
1369 | { | |
1370 | struct inode *inode = dentry->d_inode; | |
1371 | struct ll_inode_info *lli = ll_i2info(inode); | |
1372 | struct md_op_data *op_data = NULL; | |
1373 | struct md_open_data *mod = NULL; | |
1374 | int rc = 0, rc1 = 0; | |
1375 | ENTRY; | |
1376 | ||
1377 | CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, " | |
1378 | "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode, | |
1379 | PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size, | |
1380 | attr->ia_valid); | |
1381 | ||
1382 | if (attr->ia_valid & ATTR_SIZE) { | |
1383 | /* Check new size against VFS/VM file size limit and rlimit */ | |
1384 | rc = inode_newsize_ok(inode, attr->ia_size); | |
1385 | if (rc) | |
1386 | RETURN(rc); | |
1387 | ||
1388 | /* The maximum Lustre file size is variable, based on the | |
1389 | * OST maximum object size and number of stripes. This | |
1390 | * needs another check in addition to the VFS check above. */ | |
1391 | if (attr->ia_size > ll_file_maxbytes(inode)) { | |
1392 | CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n", | |
1393 | PFID(&lli->lli_fid), attr->ia_size, | |
1394 | ll_file_maxbytes(inode)); | |
1395 | RETURN(-EFBIG); | |
1396 | } | |
1397 | ||
1398 | attr->ia_valid |= ATTR_MTIME | ATTR_CTIME; | |
1399 | } | |
1400 | ||
1401 | /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */ | |
1402 | if (attr->ia_valid & TIMES_SET_FLAGS) { | |
4b1a25f0 | 1403 | if ((!uid_eq(current_fsuid(), inode->i_uid)) && |
d7e09d03 PT |
1404 | !cfs_capable(CFS_CAP_FOWNER)) |
1405 | RETURN(-EPERM); | |
1406 | } | |
1407 | ||
1408 | /* We mark all of the fields "set" so MDS/OST does not re-set them */ | |
1409 | if (attr->ia_valid & ATTR_CTIME) { | |
1410 | attr->ia_ctime = CFS_CURRENT_TIME; | |
1411 | attr->ia_valid |= ATTR_CTIME_SET; | |
1412 | } | |
1413 | if (!(attr->ia_valid & ATTR_ATIME_SET) && | |
1414 | (attr->ia_valid & ATTR_ATIME)) { | |
1415 | attr->ia_atime = CFS_CURRENT_TIME; | |
1416 | attr->ia_valid |= ATTR_ATIME_SET; | |
1417 | } | |
1418 | if (!(attr->ia_valid & ATTR_MTIME_SET) && | |
1419 | (attr->ia_valid & ATTR_MTIME)) { | |
1420 | attr->ia_mtime = CFS_CURRENT_TIME; | |
1421 | attr->ia_valid |= ATTR_MTIME_SET; | |
1422 | } | |
1423 | ||
1424 | if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME)) | |
1425 | CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n", | |
1426 | LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime), | |
1427 | cfs_time_current_sec()); | |
1428 | ||
1429 | /* If we are changing file size, file content is modified, flag it. */ | |
1430 | if (attr->ia_valid & ATTR_SIZE) { | |
1431 | attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE; | |
1432 | spin_lock(&lli->lli_lock); | |
1433 | lli->lli_flags |= LLIF_DATA_MODIFIED; | |
1434 | spin_unlock(&lli->lli_lock); | |
1435 | } | |
1436 | ||
1437 | /* We always do an MDS RPC, even if we're only changing the size; | |
1438 | * only the MDS knows whether truncate() should fail with -ETXTBUSY */ | |
1439 | ||
1440 | OBD_ALLOC_PTR(op_data); | |
1441 | if (op_data == NULL) | |
1442 | RETURN(-ENOMEM); | |
1443 | ||
1444 | if (!S_ISDIR(inode->i_mode)) { | |
1445 | if (attr->ia_valid & ATTR_SIZE) | |
1446 | inode_dio_write_done(inode); | |
1447 | mutex_unlock(&inode->i_mutex); | |
1448 | down_write(&lli->lli_trunc_sem); | |
1449 | } | |
1450 | ||
1451 | memcpy(&op_data->op_attr, attr, sizeof(*attr)); | |
1452 | ||
1453 | /* Open epoch for truncate. */ | |
1454 | if (exp_connect_som(ll_i2mdexp(inode)) && | |
1455 | (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET))) | |
1456 | op_data->op_flags = MF_EPOCH_OPEN; | |
1457 | ||
1458 | rc = ll_md_setattr(dentry, op_data, &mod); | |
1459 | if (rc) | |
1460 | GOTO(out, rc); | |
1461 | ||
1462 | /* RPC to MDT is sent, cancel data modification flag */ | |
1463 | if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) { | |
1464 | spin_lock(&lli->lli_lock); | |
1465 | lli->lli_flags &= ~LLIF_DATA_MODIFIED; | |
1466 | spin_unlock(&lli->lli_lock); | |
1467 | } | |
1468 | ||
1469 | ll_ioepoch_open(lli, op_data->op_ioepoch); | |
1470 | if (!S_ISREG(inode->i_mode)) | |
1471 | GOTO(out, rc = 0); | |
1472 | ||
1473 | if (attr->ia_valid & (ATTR_SIZE | | |
1474 | ATTR_ATIME | ATTR_ATIME_SET | | |
1475 | ATTR_MTIME | ATTR_MTIME_SET)) | |
1476 | /* For truncate and utimes sending attributes to OSTs, setting | |
1477 | * mtime/atime to the past will be performed under PW [0:EOF] | |
1478 | * extent lock (new_size:EOF for truncate). It may seem | |
1479 | * excessive to send mtime/atime updates to OSTs when not | |
1480 | * setting times to past, but it is necessary due to possible | |
1481 | * time de-synchronization between MDT inode and OST objects */ | |
1482 | rc = ll_setattr_ost(inode, attr); | |
1483 | EXIT; | |
1484 | out: | |
1485 | if (op_data) { | |
1486 | if (op_data->op_ioepoch) { | |
1487 | rc1 = ll_setattr_done_writing(inode, op_data, mod); | |
1488 | if (!rc) | |
1489 | rc = rc1; | |
1490 | } | |
1491 | ll_finish_md_op_data(op_data); | |
1492 | } | |
1493 | if (!S_ISDIR(inode->i_mode)) { | |
1494 | up_write(&lli->lli_trunc_sem); | |
1495 | mutex_lock(&inode->i_mutex); | |
1496 | if (attr->ia_valid & ATTR_SIZE) | |
1497 | inode_dio_wait(inode); | |
1498 | } | |
1499 | ||
1500 | ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ? | |
1501 | LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1); | |
1502 | ||
251c4317 | 1503 | return rc; |
d7e09d03 PT |
1504 | } |
1505 | ||
1506 | int ll_setattr(struct dentry *de, struct iattr *attr) | |
1507 | { | |
1508 | int mode = de->d_inode->i_mode; | |
1509 | ||
1510 | if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) == | |
1511 | (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) | |
1512 | attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE; | |
1513 | ||
1514 | if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) == | |
1515 | (ATTR_SIZE|ATTR_MODE)) && | |
1516 | (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) || | |
1517 | (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) && | |
1518 | !(attr->ia_mode & S_ISGID)))) | |
1519 | attr->ia_valid |= ATTR_FORCE; | |
1520 | ||
1521 | if ((mode & S_ISUID) && | |
1522 | !(attr->ia_mode & S_ISUID) && | |
1523 | !(attr->ia_valid & ATTR_KILL_SUID)) | |
1524 | attr->ia_valid |= ATTR_KILL_SUID; | |
1525 | ||
1526 | if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) && | |
1527 | !(attr->ia_mode & S_ISGID) && | |
1528 | !(attr->ia_valid & ATTR_KILL_SGID)) | |
1529 | attr->ia_valid |= ATTR_KILL_SGID; | |
1530 | ||
1531 | return ll_setattr_raw(de, attr); | |
1532 | } | |
1533 | ||
1534 | int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs, | |
1535 | __u64 max_age, __u32 flags) | |
1536 | { | |
1537 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
1538 | struct obd_statfs obd_osfs; | |
1539 | int rc; | |
1540 | ENTRY; | |
1541 | ||
1542 | rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags); | |
1543 | if (rc) { | |
1544 | CERROR("md_statfs fails: rc = %d\n", rc); | |
1545 | RETURN(rc); | |
1546 | } | |
1547 | ||
1548 | osfs->os_type = sb->s_magic; | |
1549 | ||
1550 | CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n", | |
1551 | osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files); | |
1552 | ||
1553 | if (sbi->ll_flags & LL_SBI_LAZYSTATFS) | |
1554 | flags |= OBD_STATFS_NODELAY; | |
1555 | ||
1556 | rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags); | |
1557 | if (rc) { | |
1558 | CERROR("obd_statfs fails: rc = %d\n", rc); | |
1559 | RETURN(rc); | |
1560 | } | |
1561 | ||
1562 | CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n", | |
1563 | obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree, | |
1564 | obd_osfs.os_files); | |
1565 | ||
1566 | osfs->os_bsize = obd_osfs.os_bsize; | |
1567 | osfs->os_blocks = obd_osfs.os_blocks; | |
1568 | osfs->os_bfree = obd_osfs.os_bfree; | |
1569 | osfs->os_bavail = obd_osfs.os_bavail; | |
1570 | ||
1571 | /* If we don't have as many objects free on the OST as inodes | |
1572 | * on the MDS, we reduce the total number of inodes to | |
1573 | * compensate, so that the "inodes in use" number is correct. | |
1574 | */ | |
1575 | if (obd_osfs.os_ffree < osfs->os_ffree) { | |
1576 | osfs->os_files = (osfs->os_files - osfs->os_ffree) + | |
1577 | obd_osfs.os_ffree; | |
1578 | osfs->os_ffree = obd_osfs.os_ffree; | |
1579 | } | |
1580 | ||
1581 | RETURN(rc); | |
1582 | } | |
1583 | int ll_statfs(struct dentry *de, struct kstatfs *sfs) | |
1584 | { | |
1585 | struct super_block *sb = de->d_sb; | |
1586 | struct obd_statfs osfs; | |
1587 | int rc; | |
1588 | ||
1589 | CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64()); | |
1590 | ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1); | |
1591 | ||
1592 | /* Some amount of caching on the client is allowed */ | |
1593 | rc = ll_statfs_internal(sb, &osfs, | |
1594 | cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), | |
1595 | 0); | |
1596 | if (rc) | |
1597 | return rc; | |
1598 | ||
1599 | statfs_unpack(sfs, &osfs); | |
1600 | ||
1601 | /* We need to downshift for all 32-bit kernels, because we can't | |
1602 | * tell if the kernel is being called via sys_statfs64() or not. | |
1603 | * Stop before overflowing f_bsize - in which case it is better | |
1604 | * to just risk EOVERFLOW if caller is using old sys_statfs(). */ | |
1605 | if (sizeof(long) < 8) { | |
1606 | while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) { | |
1607 | sfs->f_bsize <<= 1; | |
1608 | ||
1609 | osfs.os_blocks >>= 1; | |
1610 | osfs.os_bfree >>= 1; | |
1611 | osfs.os_bavail >>= 1; | |
1612 | } | |
1613 | } | |
1614 | ||
1615 | sfs->f_blocks = osfs.os_blocks; | |
1616 | sfs->f_bfree = osfs.os_bfree; | |
1617 | sfs->f_bavail = osfs.os_bavail; | |
1618 | ||
1619 | return 0; | |
1620 | } | |
1621 | ||
1622 | void ll_inode_size_lock(struct inode *inode) | |
1623 | { | |
1624 | struct ll_inode_info *lli; | |
1625 | ||
1626 | LASSERT(!S_ISDIR(inode->i_mode)); | |
1627 | ||
1628 | lli = ll_i2info(inode); | |
1629 | LASSERT(lli->lli_size_sem_owner != current); | |
1630 | down(&lli->lli_size_sem); | |
1631 | LASSERT(lli->lli_size_sem_owner == NULL); | |
1632 | lli->lli_size_sem_owner = current; | |
1633 | } | |
1634 | ||
1635 | void ll_inode_size_unlock(struct inode *inode) | |
1636 | { | |
1637 | struct ll_inode_info *lli; | |
1638 | ||
1639 | lli = ll_i2info(inode); | |
1640 | LASSERT(lli->lli_size_sem_owner == current); | |
1641 | lli->lli_size_sem_owner = NULL; | |
1642 | up(&lli->lli_size_sem); | |
1643 | } | |
1644 | ||
1645 | void ll_update_inode(struct inode *inode, struct lustre_md *md) | |
1646 | { | |
1647 | struct ll_inode_info *lli = ll_i2info(inode); | |
1648 | struct mdt_body *body = md->body; | |
1649 | struct lov_stripe_md *lsm = md->lsm; | |
1650 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1651 | ||
1652 | LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0)); | |
1653 | if (lsm != NULL) { | |
1654 | if (!lli->lli_has_smd && | |
1655 | !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK)) | |
1656 | cl_file_inode_init(inode, md); | |
1657 | ||
1658 | lli->lli_maxbytes = lsm->lsm_maxbytes; | |
1659 | if (lli->lli_maxbytes > MAX_LFS_FILESIZE) | |
1660 | lli->lli_maxbytes = MAX_LFS_FILESIZE; | |
1661 | } | |
1662 | ||
1663 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT) { | |
1664 | if (body->valid & OBD_MD_FLRMTPERM) | |
1665 | ll_update_remote_perm(inode, md->remote_perm); | |
1666 | } | |
1667 | #ifdef CONFIG_FS_POSIX_ACL | |
1668 | else if (body->valid & OBD_MD_FLACL) { | |
1669 | spin_lock(&lli->lli_lock); | |
1670 | if (lli->lli_posix_acl) | |
1671 | posix_acl_release(lli->lli_posix_acl); | |
1672 | lli->lli_posix_acl = md->posix_acl; | |
1673 | spin_unlock(&lli->lli_lock); | |
1674 | } | |
1675 | #endif | |
c1e2699d | 1676 | inode->i_ino = cl_fid_build_ino(&body->fid1, |
1677 | sbi->ll_flags & LL_SBI_32BIT_API); | |
d7e09d03 PT |
1678 | inode->i_generation = cl_fid_build_gen(&body->fid1); |
1679 | ||
1680 | if (body->valid & OBD_MD_FLATIME) { | |
1681 | if (body->atime > LTIME_S(inode->i_atime)) | |
1682 | LTIME_S(inode->i_atime) = body->atime; | |
1683 | lli->lli_lvb.lvb_atime = body->atime; | |
1684 | } | |
1685 | if (body->valid & OBD_MD_FLMTIME) { | |
1686 | if (body->mtime > LTIME_S(inode->i_mtime)) { | |
1687 | CDEBUG(D_INODE, "setting ino %lu mtime from %lu " | |
1688 | "to "LPU64"\n", inode->i_ino, | |
1689 | LTIME_S(inode->i_mtime), body->mtime); | |
1690 | LTIME_S(inode->i_mtime) = body->mtime; | |
1691 | } | |
1692 | lli->lli_lvb.lvb_mtime = body->mtime; | |
1693 | } | |
1694 | if (body->valid & OBD_MD_FLCTIME) { | |
1695 | if (body->ctime > LTIME_S(inode->i_ctime)) | |
1696 | LTIME_S(inode->i_ctime) = body->ctime; | |
1697 | lli->lli_lvb.lvb_ctime = body->ctime; | |
1698 | } | |
1699 | if (body->valid & OBD_MD_FLMODE) | |
1700 | inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT); | |
1701 | if (body->valid & OBD_MD_FLTYPE) | |
1702 | inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT); | |
1703 | LASSERT(inode->i_mode != 0); | |
1704 | if (S_ISREG(inode->i_mode)) { | |
1705 | inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS); | |
1706 | } else { | |
1707 | inode->i_blkbits = inode->i_sb->s_blocksize_bits; | |
1708 | } | |
1709 | if (body->valid & OBD_MD_FLUID) | |
4b1a25f0 | 1710 | inode->i_uid = make_kuid(&init_user_ns, body->uid); |
d7e09d03 | 1711 | if (body->valid & OBD_MD_FLGID) |
4b1a25f0 | 1712 | inode->i_gid = make_kgid(&init_user_ns, body->gid); |
d7e09d03 PT |
1713 | if (body->valid & OBD_MD_FLFLAGS) |
1714 | inode->i_flags = ll_ext_to_inode_flags(body->flags); | |
1715 | if (body->valid & OBD_MD_FLNLINK) | |
1716 | set_nlink(inode, body->nlink); | |
1717 | if (body->valid & OBD_MD_FLRDEV) | |
1718 | inode->i_rdev = old_decode_dev(body->rdev); | |
1719 | ||
1720 | if (body->valid & OBD_MD_FLID) { | |
1721 | /* FID shouldn't be changed! */ | |
1722 | if (fid_is_sane(&lli->lli_fid)) { | |
1723 | LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1), | |
1724 | "Trying to change FID "DFID | |
1725 | " to the "DFID", inode %lu/%u(%p)\n", | |
1726 | PFID(&lli->lli_fid), PFID(&body->fid1), | |
1727 | inode->i_ino, inode->i_generation, inode); | |
1728 | } else | |
1729 | lli->lli_fid = body->fid1; | |
1730 | } | |
1731 | ||
1732 | LASSERT(fid_seq(&lli->lli_fid) != 0); | |
1733 | ||
1734 | if (body->valid & OBD_MD_FLSIZE) { | |
1735 | if (exp_connect_som(ll_i2mdexp(inode)) && | |
1736 | S_ISREG(inode->i_mode)) { | |
1737 | struct lustre_handle lockh; | |
1738 | ldlm_mode_t mode; | |
1739 | ||
1740 | /* As it is possible a blocking ast has been processed | |
1741 | * by this time, we need to check there is an UPDATE | |
1742 | * lock on the client and set LLIF_MDS_SIZE_LOCK holding | |
1743 | * it. */ | |
1744 | mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE, | |
1745 | &lockh, LDLM_FL_CBPENDING); | |
1746 | if (mode) { | |
1747 | if (lli->lli_flags & (LLIF_DONE_WRITING | | |
1748 | LLIF_EPOCH_PENDING | | |
1749 | LLIF_SOM_DIRTY)) { | |
1750 | CERROR("ino %lu flags %u still has " | |
1751 | "size authority! do not trust " | |
1752 | "the size got from MDS\n", | |
1753 | inode->i_ino, lli->lli_flags); | |
1754 | } else { | |
1755 | /* Use old size assignment to avoid | |
1756 | * deadlock bz14138 & bz14326 */ | |
1757 | i_size_write(inode, body->size); | |
1758 | lli->lli_flags |= LLIF_MDS_SIZE_LOCK; | |
1759 | } | |
1760 | ldlm_lock_decref(&lockh, mode); | |
1761 | } | |
1762 | } else { | |
1763 | /* Use old size assignment to avoid | |
1764 | * deadlock bz14138 & bz14326 */ | |
1765 | i_size_write(inode, body->size); | |
1766 | ||
1767 | CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n", | |
1768 | inode->i_ino, (unsigned long long)body->size); | |
1769 | } | |
1770 | ||
1771 | if (body->valid & OBD_MD_FLBLOCKS) | |
1772 | inode->i_blocks = body->blocks; | |
1773 | } | |
1774 | ||
1775 | if (body->valid & OBD_MD_FLMDSCAPA) { | |
1776 | LASSERT(md->mds_capa); | |
1777 | ll_add_capa(inode, md->mds_capa); | |
1778 | } | |
1779 | if (body->valid & OBD_MD_FLOSSCAPA) { | |
1780 | LASSERT(md->oss_capa); | |
1781 | ll_add_capa(inode, md->oss_capa); | |
1782 | } | |
1783 | } | |
1784 | ||
1785 | void ll_read_inode2(struct inode *inode, void *opaque) | |
1786 | { | |
1787 | struct lustre_md *md = opaque; | |
1788 | struct ll_inode_info *lli = ll_i2info(inode); | |
1789 | ENTRY; | |
1790 | ||
1791 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n", | |
1792 | PFID(&lli->lli_fid), inode); | |
1793 | ||
1794 | LASSERT(!lli->lli_has_smd); | |
1795 | ||
1796 | /* Core attributes from the MDS first. This is a new inode, and | |
1797 | * the VFS doesn't zero times in the core inode so we have to do | |
1798 | * it ourselves. They will be overwritten by either MDS or OST | |
1799 | * attributes - we just need to make sure they aren't newer. */ | |
1800 | LTIME_S(inode->i_mtime) = 0; | |
1801 | LTIME_S(inode->i_atime) = 0; | |
1802 | LTIME_S(inode->i_ctime) = 0; | |
1803 | inode->i_rdev = 0; | |
1804 | ll_update_inode(inode, md); | |
1805 | ||
1806 | /* OIDEBUG(inode); */ | |
1807 | ||
1808 | /* initializing backing dev info. */ | |
1809 | inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi; | |
1810 | ||
1811 | ||
1812 | if (S_ISREG(inode->i_mode)) { | |
1813 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1814 | inode->i_op = &ll_file_inode_operations; | |
1815 | inode->i_fop = sbi->ll_fop; | |
1816 | inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops; | |
1817 | EXIT; | |
1818 | } else if (S_ISDIR(inode->i_mode)) { | |
1819 | inode->i_op = &ll_dir_inode_operations; | |
1820 | inode->i_fop = &ll_dir_operations; | |
1821 | EXIT; | |
1822 | } else if (S_ISLNK(inode->i_mode)) { | |
1823 | inode->i_op = &ll_fast_symlink_inode_operations; | |
1824 | EXIT; | |
1825 | } else { | |
1826 | inode->i_op = &ll_special_inode_operations; | |
1827 | ||
1828 | init_special_inode(inode, inode->i_mode, | |
1829 | inode->i_rdev); | |
1830 | ||
1831 | EXIT; | |
1832 | } | |
1833 | } | |
1834 | ||
1835 | void ll_delete_inode(struct inode *inode) | |
1836 | { | |
1837 | struct cl_inode_info *lli = cl_i2info(inode); | |
1838 | ENTRY; | |
1839 | ||
1840 | if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) | |
1841 | /* discard all dirty pages before truncating them, required by | |
1842 | * osc_extent implementation at LU-1030. */ | |
65fb55d1 NY |
1843 | cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, |
1844 | CL_FSYNC_DISCARD, 1); | |
d7e09d03 PT |
1845 | |
1846 | truncate_inode_pages(&inode->i_data, 0); | |
1847 | ||
1848 | /* Workaround for LU-118 */ | |
1849 | if (inode->i_data.nrpages) { | |
1850 | TREE_READ_LOCK_IRQ(&inode->i_data); | |
1851 | TREE_READ_UNLOCK_IRQ(&inode->i_data); | |
1852 | LASSERTF(inode->i_data.nrpages == 0, | |
1853 | "inode=%lu/%u(%p) nrpages=%lu, see " | |
1854 | "http://jira.whamcloud.com/browse/LU-118\n", | |
1855 | inode->i_ino, inode->i_generation, inode, | |
1856 | inode->i_data.nrpages); | |
1857 | } | |
1858 | /* Workaround end */ | |
1859 | ||
1860 | ll_clear_inode(inode); | |
1861 | clear_inode(inode); | |
1862 | ||
1863 | EXIT; | |
1864 | } | |
1865 | ||
1866 | int ll_iocontrol(struct inode *inode, struct file *file, | |
1867 | unsigned int cmd, unsigned long arg) | |
1868 | { | |
1869 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1870 | struct ptlrpc_request *req = NULL; | |
1871 | int rc, flags = 0; | |
1872 | ENTRY; | |
1873 | ||
1874 | switch(cmd) { | |
1875 | case FSFILT_IOC_GETFLAGS: { | |
1876 | struct mdt_body *body; | |
1877 | struct md_op_data *op_data; | |
1878 | ||
1879 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, | |
1880 | 0, 0, LUSTRE_OPC_ANY, | |
1881 | NULL); | |
1882 | if (IS_ERR(op_data)) | |
1883 | RETURN(PTR_ERR(op_data)); | |
1884 | ||
1885 | op_data->op_valid = OBD_MD_FLFLAGS; | |
1886 | rc = md_getattr(sbi->ll_md_exp, op_data, &req); | |
1887 | ll_finish_md_op_data(op_data); | |
1888 | if (rc) { | |
1889 | CERROR("failure %d inode %lu\n", rc, inode->i_ino); | |
1890 | RETURN(-abs(rc)); | |
1891 | } | |
1892 | ||
1893 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
1894 | ||
1895 | flags = body->flags; | |
1896 | ||
1897 | ptlrpc_req_finished(req); | |
1898 | ||
1899 | RETURN(put_user(flags, (int *)arg)); | |
1900 | } | |
1901 | case FSFILT_IOC_SETFLAGS: { | |
1902 | struct lov_stripe_md *lsm; | |
1903 | struct obd_info oinfo = { { { 0 } } }; | |
1904 | struct md_op_data *op_data; | |
1905 | ||
1906 | if (get_user(flags, (int *)arg)) | |
1907 | RETURN(-EFAULT); | |
1908 | ||
1909 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
1910 | LUSTRE_OPC_ANY, NULL); | |
1911 | if (IS_ERR(op_data)) | |
1912 | RETURN(PTR_ERR(op_data)); | |
1913 | ||
1914 | ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags; | |
1915 | op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG; | |
1916 | rc = md_setattr(sbi->ll_md_exp, op_data, | |
1917 | NULL, 0, NULL, 0, &req, NULL); | |
1918 | ll_finish_md_op_data(op_data); | |
1919 | ptlrpc_req_finished(req); | |
1920 | if (rc) | |
1921 | RETURN(rc); | |
1922 | ||
1923 | inode->i_flags = ll_ext_to_inode_flags(flags); | |
1924 | ||
1925 | lsm = ccc_inode_lsm_get(inode); | |
5dd16419 JX |
1926 | if (!lsm_has_objects(lsm)) { |
1927 | ccc_inode_lsm_put(inode, lsm); | |
d7e09d03 | 1928 | RETURN(0); |
5dd16419 | 1929 | } |
d7e09d03 PT |
1930 | |
1931 | OBDO_ALLOC(oinfo.oi_oa); | |
1932 | if (!oinfo.oi_oa) { | |
1933 | ccc_inode_lsm_put(inode, lsm); | |
1934 | RETURN(-ENOMEM); | |
1935 | } | |
1936 | oinfo.oi_md = lsm; | |
1937 | oinfo.oi_oa->o_oi = lsm->lsm_oi; | |
1938 | oinfo.oi_oa->o_flags = flags; | |
1939 | oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | | |
1940 | OBD_MD_FLGROUP; | |
1941 | oinfo.oi_capa = ll_mdscapa_get(inode); | |
1942 | obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid); | |
1943 | rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL); | |
1944 | capa_put(oinfo.oi_capa); | |
1945 | OBDO_FREE(oinfo.oi_oa); | |
1946 | ccc_inode_lsm_put(inode, lsm); | |
1947 | ||
1948 | if (rc && rc != -EPERM && rc != -EACCES) | |
1949 | CERROR("osc_setattr_async fails: rc = %d\n", rc); | |
1950 | ||
1951 | RETURN(rc); | |
1952 | } | |
1953 | default: | |
1954 | RETURN(-ENOSYS); | |
1955 | } | |
1956 | ||
1957 | RETURN(0); | |
1958 | } | |
1959 | ||
1960 | int ll_flush_ctx(struct inode *inode) | |
1961 | { | |
1962 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1963 | ||
4b1a25f0 PT |
1964 | CDEBUG(D_SEC, "flush context for user %d\n", |
1965 | from_kuid(&init_user_ns, current_uid())); | |
d7e09d03 PT |
1966 | |
1967 | obd_set_info_async(NULL, sbi->ll_md_exp, | |
1968 | sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX, | |
1969 | 0, NULL, NULL); | |
1970 | obd_set_info_async(NULL, sbi->ll_dt_exp, | |
1971 | sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX, | |
1972 | 0, NULL, NULL); | |
1973 | return 0; | |
1974 | } | |
1975 | ||
1976 | /* umount -f client means force down, don't save state */ | |
1977 | void ll_umount_begin(struct super_block *sb) | |
1978 | { | |
1979 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
1980 | struct obd_device *obd; | |
1981 | struct obd_ioctl_data *ioc_data; | |
1982 | ENTRY; | |
1983 | ||
1984 | ||
1985 | CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb, | |
1986 | sb->s_count, atomic_read(&sb->s_active)); | |
1987 | ||
1988 | obd = class_exp2obd(sbi->ll_md_exp); | |
1989 | if (obd == NULL) { | |
1990 | CERROR("Invalid MDC connection handle "LPX64"\n", | |
1991 | sbi->ll_md_exp->exp_handle.h_cookie); | |
1992 | EXIT; | |
1993 | return; | |
1994 | } | |
1995 | obd->obd_force = 1; | |
1996 | ||
1997 | obd = class_exp2obd(sbi->ll_dt_exp); | |
1998 | if (obd == NULL) { | |
1999 | CERROR("Invalid LOV connection handle "LPX64"\n", | |
2000 | sbi->ll_dt_exp->exp_handle.h_cookie); | |
2001 | EXIT; | |
2002 | return; | |
2003 | } | |
2004 | obd->obd_force = 1; | |
2005 | ||
2006 | OBD_ALLOC_PTR(ioc_data); | |
2007 | if (ioc_data) { | |
2008 | obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, | |
2009 | sizeof *ioc_data, ioc_data, NULL); | |
2010 | ||
2011 | obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, | |
2012 | sizeof *ioc_data, ioc_data, NULL); | |
2013 | ||
2014 | OBD_FREE_PTR(ioc_data); | |
2015 | } | |
2016 | ||
d7e09d03 PT |
2017 | /* Really, we'd like to wait until there are no requests outstanding, |
2018 | * and then continue. For now, we just invalidate the requests, | |
2019 | * schedule() and sleep one second if needed, and hope. | |
2020 | */ | |
2021 | schedule(); | |
2022 | ||
2023 | EXIT; | |
2024 | } | |
2025 | ||
2026 | int ll_remount_fs(struct super_block *sb, int *flags, char *data) | |
2027 | { | |
2028 | struct ll_sb_info *sbi = ll_s2sbi(sb); | |
2029 | char *profilenm = get_profile_name(sb); | |
2030 | int err; | |
2031 | __u32 read_only; | |
2032 | ||
2033 | if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) { | |
2034 | read_only = *flags & MS_RDONLY; | |
2035 | err = obd_set_info_async(NULL, sbi->ll_md_exp, | |
2036 | sizeof(KEY_READ_ONLY), | |
2037 | KEY_READ_ONLY, sizeof(read_only), | |
2038 | &read_only, NULL); | |
2039 | if (err) { | |
2040 | LCONSOLE_WARN("Failed to remount %s %s (%d)\n", | |
2041 | profilenm, read_only ? | |
2042 | "read-only" : "read-write", err); | |
2043 | return err; | |
2044 | } | |
2045 | ||
2046 | if (read_only) | |
2047 | sb->s_flags |= MS_RDONLY; | |
2048 | else | |
2049 | sb->s_flags &= ~MS_RDONLY; | |
2050 | ||
2051 | if (sbi->ll_flags & LL_SBI_VERBOSE) | |
2052 | LCONSOLE_WARN("Remounted %s %s\n", profilenm, | |
2053 | read_only ? "read-only" : "read-write"); | |
2054 | } | |
2055 | return 0; | |
2056 | } | |
2057 | ||
2058 | int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req, | |
2059 | struct super_block *sb, struct lookup_intent *it) | |
2060 | { | |
2061 | struct ll_sb_info *sbi = NULL; | |
2062 | struct lustre_md md; | |
2063 | int rc; | |
2064 | ENTRY; | |
2065 | ||
2066 | LASSERT(*inode || sb); | |
2067 | sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode); | |
2068 | rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp, | |
2069 | sbi->ll_md_exp, &md); | |
2070 | if (rc) | |
2071 | RETURN(rc); | |
2072 | ||
2073 | if (*inode) { | |
2074 | ll_update_inode(*inode, &md); | |
2075 | } else { | |
2076 | LASSERT(sb != NULL); | |
2077 | ||
2078 | /* | |
2079 | * At this point server returns to client's same fid as client | |
2080 | * generated for creating. So using ->fid1 is okay here. | |
2081 | */ | |
2082 | LASSERT(fid_is_sane(&md.body->fid1)); | |
2083 | ||
2084 | *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, | |
c1e2699d | 2085 | sbi->ll_flags & LL_SBI_32BIT_API), |
d7e09d03 PT |
2086 | &md); |
2087 | if (*inode == NULL || IS_ERR(*inode)) { | |
2088 | #ifdef CONFIG_FS_POSIX_ACL | |
2089 | if (md.posix_acl) { | |
2090 | posix_acl_release(md.posix_acl); | |
2091 | md.posix_acl = NULL; | |
2092 | } | |
2093 | #endif | |
2094 | rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM; | |
2095 | *inode = NULL; | |
2096 | CERROR("new_inode -fatal: rc %d\n", rc); | |
2097 | GOTO(out, rc); | |
2098 | } | |
2099 | } | |
2100 | ||
2101 | /* Handling piggyback layout lock. | |
2102 | * Layout lock can be piggybacked by getattr and open request. | |
2103 | * The lsm can be applied to inode only if it comes with a layout lock | |
2104 | * otherwise correct layout may be overwritten, for example: | |
2105 | * 1. proc1: mdt returns a lsm but not granting layout | |
2106 | * 2. layout was changed by another client | |
2107 | * 3. proc2: refresh layout and layout lock granted | |
2108 | * 4. proc1: to apply a stale layout */ | |
2109 | if (it != NULL && it->d.lustre.it_lock_mode != 0) { | |
2110 | struct lustre_handle lockh; | |
2111 | struct ldlm_lock *lock; | |
2112 | ||
2113 | lockh.cookie = it->d.lustre.it_lock_handle; | |
2114 | lock = ldlm_handle2lock(&lockh); | |
2115 | LASSERT(lock != NULL); | |
2116 | if (ldlm_has_layout(lock)) { | |
2117 | struct cl_object_conf conf; | |
2118 | ||
2119 | memset(&conf, 0, sizeof(conf)); | |
2120 | conf.coc_opc = OBJECT_CONF_SET; | |
2121 | conf.coc_inode = *inode; | |
2122 | conf.coc_lock = lock; | |
2123 | conf.u.coc_md = &md; | |
2124 | (void)ll_layout_conf(*inode, &conf); | |
2125 | } | |
2126 | LDLM_LOCK_PUT(lock); | |
2127 | } | |
2128 | ||
2129 | out: | |
2130 | if (md.lsm != NULL) | |
2131 | obd_free_memmd(sbi->ll_dt_exp, &md.lsm); | |
2132 | md_free_lustre_md(sbi->ll_md_exp, &md); | |
2133 | RETURN(rc); | |
2134 | } | |
2135 | ||
2136 | int ll_obd_statfs(struct inode *inode, void *arg) | |
2137 | { | |
2138 | struct ll_sb_info *sbi = NULL; | |
2139 | struct obd_export *exp; | |
2140 | char *buf = NULL; | |
2141 | struct obd_ioctl_data *data = NULL; | |
2142 | __u32 type; | |
2143 | __u32 flags; | |
2144 | int len = 0, rc; | |
2145 | ||
2146 | if (!inode || !(sbi = ll_i2sbi(inode))) | |
2147 | GOTO(out_statfs, rc = -EINVAL); | |
2148 | ||
2149 | rc = obd_ioctl_getdata(&buf, &len, arg); | |
2150 | if (rc) | |
2151 | GOTO(out_statfs, rc); | |
2152 | ||
2153 | data = (void*)buf; | |
2154 | if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 || | |
2155 | !data->ioc_pbuf1 || !data->ioc_pbuf2) | |
2156 | GOTO(out_statfs, rc = -EINVAL); | |
2157 | ||
2158 | if (data->ioc_inllen1 != sizeof(__u32) || | |
2159 | data->ioc_inllen2 != sizeof(__u32) || | |
2160 | data->ioc_plen1 != sizeof(struct obd_statfs) || | |
2161 | data->ioc_plen2 != sizeof(struct obd_uuid)) | |
2162 | GOTO(out_statfs, rc = -EINVAL); | |
2163 | ||
2164 | memcpy(&type, data->ioc_inlbuf1, sizeof(__u32)); | |
2165 | if (type & LL_STATFS_LMV) | |
2166 | exp = sbi->ll_md_exp; | |
2167 | else if (type & LL_STATFS_LOV) | |
2168 | exp = sbi->ll_dt_exp; | |
2169 | else | |
2170 | GOTO(out_statfs, rc = -ENODEV); | |
2171 | ||
2172 | flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0; | |
2173 | rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags); | |
2174 | if (rc) | |
2175 | GOTO(out_statfs, rc); | |
2176 | out_statfs: | |
2177 | if (buf) | |
2178 | obd_ioctl_freedata(buf, len); | |
2179 | return rc; | |
2180 | } | |
2181 | ||
2182 | int ll_process_config(struct lustre_cfg *lcfg) | |
2183 | { | |
2184 | char *ptr; | |
2185 | void *sb; | |
2186 | struct lprocfs_static_vars lvars; | |
2187 | unsigned long x; | |
2188 | int rc = 0; | |
2189 | ||
2190 | lprocfs_llite_init_vars(&lvars); | |
2191 | ||
2192 | /* The instance name contains the sb: lustre-client-aacfe000 */ | |
2193 | ptr = strrchr(lustre_cfg_string(lcfg, 0), '-'); | |
2194 | if (!ptr || !*(++ptr)) | |
2195 | return -EINVAL; | |
2196 | if (sscanf(ptr, "%lx", &x) != 1) | |
2197 | return -EINVAL; | |
2198 | sb = (void *)x; | |
2199 | /* This better be a real Lustre superblock! */ | |
2200 | LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC); | |
2201 | ||
2202 | /* Note we have not called client_common_fill_super yet, so | |
2203 | proc fns must be able to handle that! */ | |
2204 | rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars, | |
2205 | lcfg, sb); | |
2206 | if (rc > 0) | |
2207 | rc = 0; | |
2208 | return(rc); | |
2209 | } | |
2210 | ||
2211 | /* this function prepares md_op_data hint for passing ot down to MD stack. */ | |
2212 | struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data, | |
2213 | struct inode *i1, struct inode *i2, | |
2214 | const char *name, int namelen, | |
2215 | int mode, __u32 opc, void *data) | |
2216 | { | |
2217 | LASSERT(i1 != NULL); | |
2218 | ||
2219 | if (namelen > ll_i2sbi(i1)->ll_namelen) | |
2220 | return ERR_PTR(-ENAMETOOLONG); | |
2221 | ||
2222 | if (op_data == NULL) | |
2223 | OBD_ALLOC_PTR(op_data); | |
2224 | ||
2225 | if (op_data == NULL) | |
2226 | return ERR_PTR(-ENOMEM); | |
2227 | ||
2228 | ll_i2gids(op_data->op_suppgids, i1, i2); | |
2229 | op_data->op_fid1 = *ll_inode2fid(i1); | |
2230 | op_data->op_capa1 = ll_mdscapa_get(i1); | |
2231 | ||
2232 | if (i2) { | |
2233 | op_data->op_fid2 = *ll_inode2fid(i2); | |
2234 | op_data->op_capa2 = ll_mdscapa_get(i2); | |
2235 | } else { | |
2236 | fid_zero(&op_data->op_fid2); | |
2237 | op_data->op_capa2 = NULL; | |
2238 | } | |
2239 | ||
2240 | op_data->op_name = name; | |
2241 | op_data->op_namelen = namelen; | |
2242 | op_data->op_mode = mode; | |
2243 | op_data->op_mod_time = cfs_time_current_sec(); | |
4b1a25f0 PT |
2244 | op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid()); |
2245 | op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid()); | |
d7e09d03 PT |
2246 | op_data->op_cap = cfs_curproc_cap_pack(); |
2247 | op_data->op_bias = 0; | |
2248 | op_data->op_cli_flags = 0; | |
2249 | if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) && | |
2250 | filename_is_volatile(name, namelen, NULL)) | |
2251 | op_data->op_bias |= MDS_CREATE_VOLATILE; | |
2252 | op_data->op_opc = opc; | |
2253 | op_data->op_mds = 0; | |
2254 | op_data->op_data = data; | |
2255 | ||
2256 | /* If the file is being opened after mknod() (normally due to NFS) | |
2257 | * try to use the default stripe data from parent directory for | |
2258 | * allocating OST objects. Try to pass the parent FID to MDS. */ | |
2259 | if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) && | |
2260 | !ll_i2info(i2)->lli_has_smd) { | |
2261 | struct ll_inode_info *lli = ll_i2info(i2); | |
2262 | ||
2263 | spin_lock(&lli->lli_lock); | |
2264 | if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid))) | |
2265 | op_data->op_fid1 = lli->lli_pfid; | |
2266 | spin_unlock(&lli->lli_lock); | |
2267 | /** We ignore parent's capability temporary. */ | |
2268 | } | |
2269 | ||
2270 | /* When called by ll_setattr_raw, file is i1. */ | |
2271 | if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags) | |
2272 | op_data->op_bias |= MDS_DATA_MODIFIED; | |
2273 | ||
2274 | return op_data; | |
2275 | } | |
2276 | ||
2277 | void ll_finish_md_op_data(struct md_op_data *op_data) | |
2278 | { | |
2279 | capa_put(op_data->op_capa1); | |
2280 | capa_put(op_data->op_capa2); | |
2281 | OBD_FREE_PTR(op_data); | |
2282 | } | |
2283 | ||
2284 | int ll_show_options(struct seq_file *seq, struct dentry *dentry) | |
2285 | { | |
2286 | struct ll_sb_info *sbi; | |
2287 | ||
2288 | LASSERT((seq != NULL) && (dentry != NULL)); | |
2289 | sbi = ll_s2sbi(dentry->d_sb); | |
2290 | ||
2291 | if (sbi->ll_flags & LL_SBI_NOLCK) | |
2292 | seq_puts(seq, ",nolock"); | |
2293 | ||
2294 | if (sbi->ll_flags & LL_SBI_FLOCK) | |
2295 | seq_puts(seq, ",flock"); | |
2296 | ||
2297 | if (sbi->ll_flags & LL_SBI_LOCALFLOCK) | |
2298 | seq_puts(seq, ",localflock"); | |
2299 | ||
2300 | if (sbi->ll_flags & LL_SBI_USER_XATTR) | |
2301 | seq_puts(seq, ",user_xattr"); | |
2302 | ||
2303 | if (sbi->ll_flags & LL_SBI_LAZYSTATFS) | |
2304 | seq_puts(seq, ",lazystatfs"); | |
2305 | ||
2306 | if (sbi->ll_flags & LL_SBI_USER_FID2PATH) | |
2307 | seq_puts(seq, ",user_fid2path"); | |
2308 | ||
2309 | RETURN(0); | |
2310 | } | |
2311 | ||
2312 | /** | |
2313 | * Get obd name by cmd, and copy out to user space | |
2314 | */ | |
2315 | int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg) | |
2316 | { | |
2317 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
2318 | struct obd_device *obd; | |
2319 | ENTRY; | |
2320 | ||
2321 | if (cmd == OBD_IOC_GETDTNAME) | |
2322 | obd = class_exp2obd(sbi->ll_dt_exp); | |
2323 | else if (cmd == OBD_IOC_GETMDNAME) | |
2324 | obd = class_exp2obd(sbi->ll_md_exp); | |
2325 | else | |
2326 | RETURN(-EINVAL); | |
2327 | ||
2328 | if (!obd) | |
2329 | RETURN(-ENOENT); | |
2330 | ||
2331 | if (copy_to_user((void *)arg, obd->obd_name, | |
2332 | strlen(obd->obd_name) + 1)) | |
2333 | RETURN(-EFAULT); | |
2334 | ||
2335 | RETURN(0); | |
2336 | } | |
2337 | ||
2338 | /** | |
2339 | * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the | |
2340 | * fsname will be returned in this buffer; otherwise, a static buffer will be | |
2341 | * used to store the fsname and returned to caller. | |
2342 | */ | |
2343 | char *ll_get_fsname(struct super_block *sb, char *buf, int buflen) | |
2344 | { | |
2345 | static char fsname_static[MTI_NAME_MAXLEN]; | |
2346 | struct lustre_sb_info *lsi = s2lsi(sb); | |
2347 | char *ptr; | |
2348 | int len; | |
2349 | ||
2350 | if (buf == NULL) { | |
2351 | /* this means the caller wants to use static buffer | |
2352 | * and it doesn't care about race. Usually this is | |
2353 | * in error reporting path */ | |
2354 | buf = fsname_static; | |
2355 | buflen = sizeof(fsname_static); | |
2356 | } | |
2357 | ||
2358 | len = strlen(lsi->lsi_lmd->lmd_profile); | |
2359 | ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-'); | |
2360 | if (ptr && (strcmp(ptr, "-client") == 0)) | |
2361 | len -= 7; | |
2362 | ||
2363 | if (unlikely(len >= buflen)) | |
2364 | len = buflen - 1; | |
2365 | strncpy(buf, lsi->lsi_lmd->lmd_profile, len); | |
2366 | buf[len] = '\0'; | |
2367 | ||
2368 | return buf; | |
2369 | } | |
2370 | ||
2371 | static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize) | |
2372 | { | |
2373 | char *path = NULL; | |
2374 | ||
2375 | struct path p; | |
2376 | ||
2377 | p.dentry = dentry; | |
2378 | p.mnt = current->fs->root.mnt; | |
2379 | path_get(&p); | |
2380 | path = d_path(&p, buf, bufsize); | |
2381 | path_put(&p); | |
2382 | ||
2383 | return path; | |
2384 | } | |
2385 | ||
2386 | void ll_dirty_page_discard_warn(struct page *page, int ioret) | |
2387 | { | |
2388 | char *buf, *path = NULL; | |
2389 | struct dentry *dentry = NULL; | |
2390 | struct ccc_object *obj = cl_inode2ccc(page->mapping->host); | |
2391 | ||
2392 | /* this can be called inside spin lock so use GFP_ATOMIC. */ | |
2393 | buf = (char *)__get_free_page(GFP_ATOMIC); | |
2394 | if (buf != NULL) { | |
2395 | dentry = d_find_alias(page->mapping->host); | |
2396 | if (dentry != NULL) | |
2397 | path = ll_d_path(dentry, buf, PAGE_SIZE); | |
2398 | } | |
2399 | ||
2400 | CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted " | |
2401 | "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0), | |
2402 | s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev, | |
2403 | PFID(&obj->cob_header.coh_lu.loh_fid), | |
2404 | (path && !IS_ERR(path)) ? path : "", ioret); | |
2405 | ||
2406 | if (dentry != NULL) | |
2407 | dput(dentry); | |
2408 | ||
2409 | if (buf != NULL) | |
2410 | free_page((unsigned long)buf); | |
2411 | } |