4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/llite/llite_lib.c
33 * Lustre Light Super operations
36 #define DEBUG_SUBSYSTEM S_LLITE
38 #include <linux/cpu.h>
39 #include <linux/module.h>
40 #include <linux/random.h>
41 #include <linux/statfs.h>
42 #include <linux/time.h>
43 #include <linux/file.h>
44 #include <linux/types.h>
45 #include <libcfs/linux/linux-uuid.h>
46 #include <linux/version.h>
48 #include <linux/user_namespace.h>
49 #include <linux/delay.h>
50 #include <linux/uidgid.h>
51 #include <linux/fs_struct.h>
53 #ifndef HAVE_CPUS_READ_LOCK
54 #include <libcfs/linux/linux-cpu.h>
56 #include <libcfs/linux/linux-misc.h>
57 #include <uapi/linux/lustre/lustre_ioctl.h>
58 #ifdef HAVE_UAPI_LINUX_MOUNT_H
59 #include <uapi/linux/mount.h>
62 #include <lustre_ha.h>
63 #include <lustre_dlm.h>
64 #include <lprocfs_status.h>
65 #include <lustre_disk.h>
66 #include <uapi/linux/lustre/lustre_param.h>
67 #include <lustre_log.h>
68 #include <cl_object.h>
69 #include <obd_cksum.h>
70 #include "llite_internal.h"
72 struct kmem_cache *ll_file_data_slab;
75 #define log2(n) ffz(~(n))
79 * If there is only one number of core visible to Lustre,
80 * async readahead will be disabled, to avoid massive over
81 * subscription, we use 1/2 of active cores as default max
82 * async readahead requests.
84 static inline unsigned int ll_get_ra_async_max_active(void)
86 return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
89 static struct ll_sb_info *ll_init_sbi(void)
91 struct ll_sb_info *sbi = NULL;
93 unsigned long lru_page_max;
101 RETURN(ERR_PTR(-ENOMEM));
103 rc = pcc_super_init(&sbi->ll_pcc_super);
107 spin_lock_init(&sbi->ll_lock);
108 mutex_init(&sbi->ll_lco.lco_lock);
109 spin_lock_init(&sbi->ll_pp_extent_lock);
110 spin_lock_init(&sbi->ll_process_lock);
111 sbi->ll_rw_stats_on = 0;
112 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
115 pages = si.totalram - si.totalhigh;
116 lru_page_max = pages / 2;
118 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
119 sbi->ll_ra_info.ll_readahead_wq =
120 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
122 sbi->ll_ra_info.ra_async_max_active);
123 if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
124 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
126 /* initialize ll_cache data */
127 sbi->ll_cache = cl_cache_init(lru_page_max);
128 if (sbi->ll_cache == NULL)
129 GOTO(out_destroy_ra, rc = -ENOMEM);
131 /* initialize foreign symlink prefix path */
132 OBD_ALLOC(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
133 if (sbi->ll_foreign_symlink_prefix == NULL)
134 GOTO(out_destroy_ra, rc = -ENOMEM);
135 memcpy(sbi->ll_foreign_symlink_prefix, "/mnt/", sizeof("/mnt/"));
136 sbi->ll_foreign_symlink_prefix_size = sizeof("/mnt/");
138 /* initialize foreign symlink upcall path, none by default */
139 OBD_ALLOC(sbi->ll_foreign_symlink_upcall, sizeof("none"));
140 if (sbi->ll_foreign_symlink_upcall == NULL)
141 GOTO(out_destroy_ra, rc = -ENOMEM);
142 memcpy(sbi->ll_foreign_symlink_upcall, "none", sizeof("none"));
143 sbi->ll_foreign_symlink_upcall_items = NULL;
144 sbi->ll_foreign_symlink_upcall_nb_items = 0;
145 init_rwsem(&sbi->ll_foreign_symlink_sem);
146 /* foreign symlink support (LL_SBI_FOREIGN_SYMLINK in ll_flags)
147 * not enabled by default
150 sbi->ll_ra_info.ra_max_pages =
151 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
152 sbi->ll_ra_info.ra_max_pages_per_file =
153 min(sbi->ll_ra_info.ra_max_pages / 4,
154 SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
155 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
156 sbi->ll_ra_info.ra_max_pages_per_file;
157 sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
158 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
159 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
161 set_bit(LL_SBI_VERBOSE, sbi->ll_flags);
162 #ifdef ENABLE_CHECKSUM
163 set_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
166 set_bit(LL_SBI_FLOCK, sbi->ll_flags);
169 #ifdef HAVE_LRU_RESIZE_SUPPORT
170 set_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags);
172 set_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
174 /* metadata statahead is enabled by default */
175 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
176 sbi->ll_sa_max = LL_SA_RPC_DEF;
177 atomic_set(&sbi->ll_sa_total, 0);
178 atomic_set(&sbi->ll_sa_wrong, 0);
179 atomic_set(&sbi->ll_sa_running, 0);
180 atomic_set(&sbi->ll_agl_total, 0);
181 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
182 set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
183 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
184 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
185 ll_sbi_set_encrypt(sbi, true);
188 sbi->ll_squash.rsi_uid = 0;
189 sbi->ll_squash.rsi_gid = 0;
190 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
191 spin_lock_init(&sbi->ll_squash.rsi_lock);
193 /* Per-filesystem file heat */
194 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
195 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
197 /* Per-fs open heat level before requesting open lock */
198 sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
199 sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
200 sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
203 if (sbi->ll_foreign_symlink_prefix)
204 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
206 cl_cache_decref(sbi->ll_cache);
207 sbi->ll_cache = NULL;
209 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
211 pcc_super_fini(&sbi->ll_pcc_super);
217 static void ll_free_sbi(struct super_block *sb)
219 struct ll_sb_info *sbi = ll_s2sbi(sb);
223 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
224 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
225 if (sbi->ll_ra_info.ll_readahead_wq)
226 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
227 if (sbi->ll_cache != NULL) {
228 cl_cache_decref(sbi->ll_cache);
229 sbi->ll_cache = NULL;
231 if (sbi->ll_foreign_symlink_prefix) {
232 OBD_FREE(sbi->ll_foreign_symlink_prefix,
233 sbi->ll_foreign_symlink_prefix_size);
234 sbi->ll_foreign_symlink_prefix = NULL;
236 if (sbi->ll_foreign_symlink_upcall) {
237 OBD_FREE(sbi->ll_foreign_symlink_upcall,
238 strlen(sbi->ll_foreign_symlink_upcall) +
240 sbi->ll_foreign_symlink_upcall = NULL;
242 if (sbi->ll_foreign_symlink_upcall_items) {
244 int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
245 struct ll_foreign_symlink_upcall_item *items =
246 sbi->ll_foreign_symlink_upcall_items;
248 for (i = 0 ; i < nb_items; i++)
249 if (items[i].type == STRING_TYPE)
250 OBD_FREE(items[i].string,
253 OBD_FREE_LARGE(items, nb_items *
254 sizeof(struct ll_foreign_symlink_upcall_item));
255 sbi->ll_foreign_symlink_upcall_items = NULL;
257 ll_free_rw_stats_info(sbi);
258 pcc_super_fini(&sbi->ll_pcc_super);
259 OBD_FREE(sbi, sizeof(*sbi));
264 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
266 struct inode *root = NULL;
267 struct ll_sb_info *sbi = ll_s2sbi(sb);
268 struct obd_statfs *osfs = NULL;
269 struct ptlrpc_request *request = NULL;
270 struct obd_connect_data *data = NULL;
271 struct obd_uuid *uuid;
272 struct md_op_data *op_data;
273 struct lustre_md lmd;
275 int size, err, checksum;
281 sbi->ll_md_obd = class_name2obd(md);
282 if (!sbi->ll_md_obd) {
283 CERROR("MD %s: not setup or attached\n", md);
297 /* pass client page size via ocd_grant_blkbits, the server should report
298 * back its backend blocksize for grant calculation purpose */
299 data->ocd_grant_blkbits = PAGE_SHIFT;
301 /* indicate MDT features supported by this client */
302 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
303 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
304 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
305 OBD_CONNECT_SRVLOCK |
306 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
307 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
308 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
309 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
310 OBD_CONNECT_64BITHASH |
311 OBD_CONNECT_EINPROGRESS |
312 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
313 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
314 OBD_CONNECT_MAX_EASIZE |
315 OBD_CONNECT_FLOCK_DEAD |
316 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
317 OBD_CONNECT_OPEN_BY_FID |
318 OBD_CONNECT_DIR_STRIPE |
319 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
320 OBD_CONNECT_SUBTREE |
321 OBD_CONNECT_MULTIMODRPCS |
322 OBD_CONNECT_GRANT_PARAM |
323 OBD_CONNECT_GRANT_SHRINK |
324 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
326 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
327 OBD_CONNECT2_SUM_STATFS |
328 OBD_CONNECT2_OVERSTRIPING |
330 OBD_CONNECT2_LOCK_CONVERT |
331 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
332 OBD_CONNECT2_INC_XID |
334 OBD_CONNECT2_ASYNC_DISCARD |
336 OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
337 OBD_CONNECT2_GETATTR_PFID |
338 OBD_CONNECT2_DOM_LVB |
339 OBD_CONNECT2_REP_MBITS |
340 OBD_CONNECT2_ATOMIC_OPEN_LOCK;
342 #ifdef HAVE_LRU_RESIZE_SUPPORT
343 if (test_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags))
344 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
346 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
348 data->ocd_cksum_types = obd_cksum_types_supported_client();
350 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
351 /* flag mdc connection as lightweight, only used for test
352 * purpose, use with care */
353 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
355 data->ocd_ibits_known = MDS_INODELOCK_FULL;
356 data->ocd_version = LUSTRE_VERSION_CODE;
358 if (sb->s_flags & SB_RDONLY)
359 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
360 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags))
361 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
364 /* Setting this indicates we correctly support S_NOSEC (See kernel
365 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
367 sb->s_flags |= SB_NOSEC;
369 sbi->ll_fop = ll_select_file_operations(sbi);
371 /* always ping even if server suppress_pings */
372 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
373 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
375 obd_connect_set_secctx(data);
376 if (ll_sbi_has_encrypt(sbi))
377 obd_connect_set_enc(data);
379 #if defined(CONFIG_SECURITY)
380 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
383 data->ocd_brw_size = MD_MAX_BRW_SIZE;
385 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
386 &sbi->ll_sb_uuid, data, sbi->ll_cache);
388 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
389 "recovery, of which this client is not a "
390 "part. Please wait for recovery to complete,"
391 " abort, or time out.\n", md);
394 CERROR("cannot connect to %s: rc = %d\n", md, err);
398 sbi->ll_md_exp->exp_connect_data = *data;
400 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
401 LUSTRE_SEQ_METADATA);
403 CERROR("%s: Can't init metadata layer FID infrastructure, "
404 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
408 /* For mount, we only need fs info from MDT0, and also in DNE, it
409 * can make sure the client can be mounted as long as MDT0 is
411 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
412 ktime_get_seconds() - sbi->ll_statfs_max_age,
413 OBD_STATFS_FOR_MDT0);
415 GOTO(out_md_fid, err);
417 /* This needs to be after statfs to ensure connect has finished.
418 * Note that "data" does NOT contain the valid connect reply.
419 * If connecting to a 1.8 server there will be no LMV device, so
420 * we can access the MDC export directly and exp_connect_flags will
421 * be non-zero, but if accessing an upgraded 2.1 server it will
422 * have the correct flags filled in.
423 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
424 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
425 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
426 valid != CLIENT_CONNECT_MDT_REQD) {
429 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
430 obd_connect_flags2str(buf, PAGE_SIZE,
431 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
432 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
433 "feature(s) needed for correct operation "
434 "of this client (%s). Please upgrade "
435 "server or downgrade client.\n",
436 sbi->ll_md_exp->exp_obd->obd_name, buf);
437 OBD_FREE(buf, PAGE_SIZE);
438 GOTO(out_md_fid, err = -EPROTO);
441 size = sizeof(*data);
442 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
443 KEY_CONN_DATA, &size, data);
445 CERROR("%s: Get connect data failed: rc = %d\n",
446 sbi->ll_md_exp->exp_obd->obd_name, err);
447 GOTO(out_md_fid, err);
450 LASSERT(osfs->os_bsize);
451 sb->s_blocksize = osfs->os_bsize;
452 sb->s_blocksize_bits = log2(osfs->os_bsize);
453 sb->s_magic = LL_SUPER_MAGIC;
454 sb->s_maxbytes = MAX_LFS_FILESIZE;
455 sbi->ll_inode_cache_enabled = 1;
456 sbi->ll_namelen = osfs->os_namelen;
457 sbi->ll_mnt.mnt = current->fs->root.mnt;
458 sbi->ll_mnt_ns = current->nsproxy->mnt_ns;
460 if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags) &&
461 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
462 LCONSOLE_INFO("Disabling user_xattr feature because "
463 "it is not supported on the server\n");
464 clear_bit(LL_SBI_USER_XATTR, sbi->ll_flags);
467 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
469 sb->s_flags |= SB_POSIXACL;
471 set_bit(LL_SBI_ACL, sbi->ll_flags);
473 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
475 sb->s_flags &= ~SB_POSIXACL;
477 clear_bit(LL_SBI_ACL, sbi->ll_flags);
480 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
481 set_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
483 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
484 set_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags);
486 if (obd_connect_has_secctx(data))
487 set_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags);
489 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
490 if (ll_sbi_has_test_dummy_encryption(sbi))
491 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
493 sbi->ll_md_exp->exp_obd->obd_name);
494 ll_sbi_set_encrypt(sbi, false);
497 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
498 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
499 LCONSOLE_INFO("%s: disabling xattr cache due to "
500 "unknown maximum xattr size.\n", dt);
501 } else if (!sbi->ll_xattr_cache_set) {
502 /* If xattr_cache is already set (no matter 0 or 1)
503 * during processing llog, it won't be enabled here. */
504 set_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags);
505 sbi->ll_xattr_cache_enabled = 1;
509 sbi->ll_dt_obd = class_name2obd(dt);
510 if (!sbi->ll_dt_obd) {
511 CERROR("DT %s: not setup or attached\n", dt);
512 GOTO(out_md_fid, err = -ENODEV);
515 /* pass client page size via ocd_grant_blkbits, the server should report
516 * back its backend blocksize for grant calculation purpose */
517 data->ocd_grant_blkbits = PAGE_SHIFT;
519 /* indicate OST features supported by this client */
520 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
521 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
522 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
523 OBD_CONNECT_SRVLOCK |
524 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
525 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
526 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
527 OBD_CONNECT_EINPROGRESS |
528 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
529 OBD_CONNECT_LAYOUTLOCK |
530 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
531 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
532 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
533 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
534 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
535 OBD_CONNECT2_REP_MBITS;
537 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
538 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
540 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
541 * disabled by default, because it can still be enabled on the
542 * fly via /sys. As a consequence, we still need to come to an
543 * agreement on the supported algorithms at connect time
545 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
547 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
548 data->ocd_cksum_types = OBD_CKSUM_ADLER;
550 data->ocd_cksum_types = obd_cksum_types_supported_client();
552 #ifdef HAVE_LRU_RESIZE_SUPPORT
553 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
555 /* always ping even if server suppress_pings */
556 if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
557 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
559 if (ll_sbi_has_encrypt(sbi))
560 obd_connect_set_enc(data);
562 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
563 "ocd_grant: %d\n", data->ocd_connect_flags,
564 data->ocd_version, data->ocd_grant);
566 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
567 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
569 data->ocd_brw_size = DT_MAX_BRW_SIZE;
571 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
572 &sbi->ll_sb_uuid, data, sbi->ll_cache);
574 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
575 "recovery, of which this client is not a "
576 "part. Please wait for recovery to "
577 "complete, abort, or time out.\n", dt);
580 CERROR("%s: Cannot connect to %s: rc = %d\n",
581 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
585 if (ll_sbi_has_encrypt(sbi) &&
586 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
587 if (ll_sbi_has_test_dummy_encryption(sbi))
588 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
590 ll_sbi_set_encrypt(sbi, false);
591 } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
592 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
595 sbi->ll_dt_exp->exp_connect_data = *data;
597 /* Don't change value if it was specified in the config log */
598 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
599 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
600 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
601 (data->ocd_brw_size >> PAGE_SHIFT));
602 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
603 sbi->ll_ra_info.ra_max_pages_per_file)
604 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
605 sbi->ll_ra_info.ra_max_pages_per_file;
608 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
609 LUSTRE_SEQ_METADATA);
611 CERROR("%s: Can't init data layer FID infrastructure, "
612 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
616 mutex_lock(&sbi->ll_lco.lco_lock);
617 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
618 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
619 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
620 mutex_unlock(&sbi->ll_lco.lco_lock);
622 fid_zero(&sbi->ll_root_fid);
623 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
626 CERROR("cannot mds_connect: rc = %d\n", err);
627 GOTO(out_lock_cn_cb, err);
629 if (!fid_is_sane(&sbi->ll_root_fid)) {
630 CERROR("%s: Invalid root fid "DFID" during mount\n",
631 sbi->ll_md_exp->exp_obd->obd_name,
632 PFID(&sbi->ll_root_fid));
633 GOTO(out_lock_cn_cb, err = -EINVAL);
635 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
637 sb->s_op = &lustre_super_operations;
638 sb->s_xattr = ll_xattr_handlers;
639 #if THREAD_SIZE >= 8192 /*b=17630*/
640 sb->s_export_op = &lustre_export_operations;
642 #ifdef HAVE_LUSTRE_CRYPTO
643 llcrypt_set_ops(sb, &lustre_cryptops);
647 * XXX: move this to after cbd setup? */
648 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE |
650 if (test_bit(LL_SBI_ACL, sbi->ll_flags))
651 valid |= OBD_MD_FLACL;
653 OBD_ALLOC_PTR(op_data);
655 GOTO(out_lock_cn_cb, err = -ENOMEM);
657 op_data->op_fid1 = sbi->ll_root_fid;
658 op_data->op_mode = 0;
659 op_data->op_valid = valid;
661 err = md_getattr(sbi->ll_md_exp, op_data, &request);
663 /* We need enc ctx info, so reset it in op_data to
664 * prevent it from being freed.
666 encctx = op_data->op_file_encctx;
667 encctxlen = op_data->op_file_encctx_size;
668 op_data->op_file_encctx = NULL;
669 op_data->op_file_encctx_size = 0;
670 OBD_FREE_PTR(op_data);
672 CERROR("%s: md_getattr failed for root: rc = %d\n",
673 sbi->ll_md_exp->exp_obd->obd_name, err);
674 GOTO(out_lock_cn_cb, err);
677 err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
678 sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
680 CERROR("failed to understand root inode md: rc = %d\n", err);
681 ptlrpc_req_finished(request);
682 GOTO(out_lock_cn_cb, err);
685 LASSERT(fid_is_sane(&sbi->ll_root_fid));
686 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
687 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, api32), &lmd);
688 md_free_lustre_md(sbi->ll_md_exp, &lmd);
692 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
694 CERROR("%s: bad ll_iget() for root: rc = %d\n",
695 sbi->ll_fsname, err);
696 ptlrpc_req_finished(request);
702 "server returned encryption ctx for root inode "DFID"\n",
703 PFID(&sbi->ll_root_fid));
704 err = ll_set_encflags(root, encctx, encctxlen, true);
706 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
708 PFID(&sbi->ll_root_fid), err);
710 ptlrpc_req_finished(request);
712 checksum = test_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
713 if (sbi->ll_checksum_set) {
714 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
715 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
716 sizeof(checksum), &checksum, NULL);
718 CERROR("%s: Set checksum failed: rc = %d\n",
719 sbi->ll_dt_exp->exp_obd->obd_name, err);
725 sb->s_root = d_make_root(root);
726 if (sb->s_root == NULL) {
728 CERROR("%s: can't make root dentry: rc = %d\n",
729 sbi->ll_fsname, err);
733 sbi->ll_sdev_orig = sb->s_dev;
735 /* We set sb->s_dev equal on all lustre clients in order to support
736 * NFS export clustering. NFSD requires that the FSID be the same
738 /* s_dev is also used in lt_compare() to compare two fs, but that is
739 * only a node-local comparison. */
740 uuid = obd_get_uuid(sbi->ll_md_exp);
742 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
749 if (sbi->ll_dt_obd) {
750 err = sysfs_create_link(&sbi->ll_kset.kobj,
751 &sbi->ll_dt_obd->obd_kset.kobj,
752 sbi->ll_dt_obd->obd_type->typ_name);
754 CERROR("%s: could not register %s in llite: rc = %d\n",
755 dt, sbi->ll_fsname, err);
760 if (sbi->ll_md_obd) {
761 err = sysfs_create_link(&sbi->ll_kset.kobj,
762 &sbi->ll_md_obd->obd_kset.kobj,
763 sbi->ll_md_obd->obd_type->typ_name);
765 CERROR("%s: could not register %s in llite: rc = %d\n",
766 md, sbi->ll_fsname, err);
775 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
777 obd_disconnect(sbi->ll_dt_exp);
778 sbi->ll_dt_exp = NULL;
779 sbi->ll_dt_obd = NULL;
781 obd_fid_fini(sbi->ll_md_exp->exp_obd);
783 obd_disconnect(sbi->ll_md_exp);
784 sbi->ll_md_exp = NULL;
785 sbi->ll_md_obd = NULL;
794 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
798 size = sizeof(*lmmsize);
799 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
800 KEY_MAX_EASIZE, &size, lmmsize);
802 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
803 sbi->ll_dt_exp->exp_obd->obd_name, rc);
807 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
810 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
811 KEY_MAX_EASIZE, &size, lmmsize);
813 CERROR("Get max mdsize error rc %d\n", rc);
815 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
821 * Get the value of the default_easize parameter.
823 * \see client_obd::cl_default_mds_easize
825 * \param[in] sbi superblock info for this filesystem
826 * \param[out] lmmsize pointer to storage location for value
828 * \retval 0 on success
829 * \retval negative negated errno on failure
831 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
836 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
837 KEY_DEFAULT_EASIZE, &size, lmmsize);
839 CERROR("Get default mdsize error rc %d\n", rc);
845 * Set the default_easize parameter to the given value.
847 * \see client_obd::cl_default_mds_easize
849 * \param[in] sbi superblock info for this filesystem
850 * \param[in] lmmsize the size to set
852 * \retval 0 on success
853 * \retval negative negated errno on failure
855 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
859 if (lmmsize < sizeof(struct lov_mds_md) ||
860 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
863 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
864 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
865 sizeof(int), &lmmsize, NULL);
870 static void client_common_put_super(struct super_block *sb)
872 struct ll_sb_info *sbi = ll_s2sbi(sb);
877 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
878 obd_disconnect(sbi->ll_dt_exp);
879 sbi->ll_dt_exp = NULL;
881 ll_debugfs_unregister_super(sb);
883 obd_fid_fini(sbi->ll_md_exp->exp_obd);
884 obd_disconnect(sbi->ll_md_exp);
885 sbi->ll_md_exp = NULL;
890 void ll_kill_super(struct super_block *sb)
892 struct ll_sb_info *sbi;
896 if (!(sb->s_flags & SB_ACTIVE))
900 /* we need restore s_dev from changed for clustred NFS before put_super
901 * because new kernels have cached s_dev and change sb->s_dev in
902 * put_super not affected real removing devices */
904 sb->s_dev = sbi->ll_sdev_orig;
906 /* wait running statahead threads to quit */
907 while (atomic_read(&sbi->ll_sa_running) > 0)
908 schedule_timeout_uninterruptible(
909 cfs_time_seconds(1) >> 3);
915 /* Since we use this table for ll_sbi_flags_seq_show make
916 * sure what you want displayed for a specific token that
917 * is listed more than once below be listed first. For
918 * example we want "checksum" displayed, not "nochecksum"
921 static const match_table_t ll_sbi_flags_name = {
922 {LL_SBI_NOLCK, "nolock"},
923 {LL_SBI_CHECKSUM, "checksum"},
924 {LL_SBI_CHECKSUM, "nochecksum"},
925 {LL_SBI_LOCALFLOCK, "localflock"},
926 {LL_SBI_FLOCK, "flock"},
927 {LL_SBI_FLOCK, "noflock"},
928 {LL_SBI_USER_XATTR, "user_xattr"},
929 {LL_SBI_USER_XATTR, "nouser_xattr"},
930 {LL_SBI_LRU_RESIZE, "lruresize"},
931 {LL_SBI_LRU_RESIZE, "nolruresize"},
932 {LL_SBI_LAZYSTATFS, "lazystatfs"},
933 {LL_SBI_LAZYSTATFS, "nolazystatfs"},
934 {LL_SBI_32BIT_API, "32bitapi"},
935 {LL_SBI_USER_FID2PATH, "user_fid2path"},
936 {LL_SBI_USER_FID2PATH, "nouser_fid2path"},
937 {LL_SBI_VERBOSE, "verbose"},
938 {LL_SBI_VERBOSE, "noverbose"},
939 {LL_SBI_ALWAYS_PING, "always_ping"},
940 {LL_SBI_TEST_DUMMY_ENCRYPTION, "test_dummy_encryption"},
941 {LL_SBI_ENCRYPT, "encrypt"},
942 {LL_SBI_ENCRYPT, "noencrypt"},
943 {LL_SBI_FOREIGN_SYMLINK, "foreign_symlink=%s"},
944 {LL_SBI_NUM_MOUNT_OPT, NULL},
947 {LL_SBI_AGL_ENABLED, "agl"},
948 {LL_SBI_64BIT_HASH, "64bit_hash"},
949 {LL_SBI_LAYOUT_LOCK, "layout"},
950 {LL_SBI_XATTR_CACHE, "xattr_cache"},
951 {LL_SBI_NOROOTSQUASH, "norootsquash"},
952 {LL_SBI_FAST_READ, "fast_read"},
953 {LL_SBI_FILE_SECCTX, "file_secctx"},
954 {LL_SBI_TINY_WRITE, "tiny_write"},
955 {LL_SBI_FILE_HEAT, "file_heat"},
956 {LL_SBI_PARALLEL_DIO, "parallel_dio"},
959 int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
961 struct super_block *sb = m->private;
964 for (i = 0; i < LL_SBI_NUM_FLAGS; i++) {
967 if (!test_bit(i, ll_s2sbi(sb)->ll_flags))
970 for (j = 0; j < ARRAY_SIZE(ll_sbi_flags_name); j++) {
971 if (ll_sbi_flags_name[j].token == i &&
972 ll_sbi_flags_name[j].pattern) {
974 ll_sbi_flags_name[j].pattern);
983 /* non-client-specific mount options are parsed in lmd_parse */
984 static int ll_options(char *options, struct super_block *sb)
986 struct ll_sb_info *sbi = ll_s2sbi(sb);
987 char *s2, *s1, *opts;
993 /* Don't stomp on lmd_opts */
994 opts = kstrdup(options, GFP_KERNEL);
1000 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
1002 while ((s1 = strsep(&opts, ",")) != NULL) {
1003 substring_t args[MAX_OPT_ARGS];
1004 bool turn_off = false;
1010 CDEBUG(D_SUPER, "next opt=%s\n", s1);
1012 if (strncmp(s1, "no", 2) == 0)
1016 * Initialize args struct so we know whether arg was
1017 * found; some options take optional arguments.
1020 args[0].from = NULL;
1021 token = match_token(s1, ll_sbi_flags_name, args);
1022 if (token == LL_SBI_NUM_MOUNT_OPT) {
1023 if (match_wildcard("context", s1) ||
1024 match_wildcard("fscontext", s1) ||
1025 match_wildcard("defcontext", s1) ||
1026 match_wildcard("rootcontext",s1))
1029 LCONSOLE_ERROR_MSG(0x152,
1030 "Unknown option '%s', won't mount.\n",
1037 case LL_SBI_32BIT_API:
1038 case LL_SBI_64BIT_HASH:
1039 case LL_SBI_ALWAYS_PING:
1040 set_bit(token, sbi->ll_flags);
1044 clear_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags);
1046 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1048 set_bit(token, sbi->ll_flags);
1051 case LL_SBI_LOCALFLOCK:
1052 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1053 set_bit(token, sbi->ll_flags);
1056 case LL_SBI_CHECKSUM:
1057 sbi->ll_checksum_set = 1;
1059 case LL_SBI_USER_XATTR:
1060 case LL_SBI_USER_FID2PATH:
1061 case LL_SBI_LRU_RESIZE:
1062 case LL_SBI_LAZYSTATFS:
1063 case LL_SBI_VERBOSE:
1065 clear_bit(token, sbi->ll_flags);
1067 set_bit(token, sbi->ll_flags);
1069 case LL_SBI_TEST_DUMMY_ENCRYPTION: {
1070 #ifdef HAVE_LUSTRE_CRYPTO
1071 set_bit(token, sbi->ll_flags);
1073 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1077 case LL_SBI_ENCRYPT:
1078 #ifdef HAVE_LUSTRE_CRYPTO
1080 clear_bit(token, sbi->ll_flags);
1082 set_bit(token, sbi->ll_flags);
1084 LCONSOLE_WARN("noencrypt or encrypt mount option ignored: encryption not supported\n");
1087 case LL_SBI_FOREIGN_SYMLINK:
1088 /* non-default prefix provided ? */
1093 /* path must be absolute */
1094 if (args->from[0] != '/') {
1095 LCONSOLE_ERROR_MSG(0x152,
1096 "foreign prefix '%s' must be an absolute path\n",
1101 old_len = sbi->ll_foreign_symlink_prefix_size;
1102 old = sbi->ll_foreign_symlink_prefix;
1103 /* alloc for path length and '\0' */
1104 sbi->ll_foreign_symlink_prefix = match_strdup(args);
1105 if (!sbi->ll_foreign_symlink_prefix) {
1106 /* restore previous */
1107 sbi->ll_foreign_symlink_prefix = old;
1108 sbi->ll_foreign_symlink_prefix_size =
1112 sbi->ll_foreign_symlink_prefix_size =
1113 args->to - args->from + 1;
1114 OBD_ALLOC_POST(sbi->ll_foreign_symlink_prefix,
1115 sbi->ll_foreign_symlink_prefix_size,
1118 OBD_FREE(old, old_len);
1120 /* enable foreign symlink support */
1121 set_bit(token, sbi->ll_flags);
1123 LCONSOLE_ERROR_MSG(0x152,
1124 "invalid %s option\n", s1);
1135 void ll_lli_init(struct ll_inode_info *lli)
1137 lli->lli_inode_magic = LLI_INODE_MAGIC;
1139 rwlock_init(&lli->lli_lock);
1140 lli->lli_posix_acl = NULL;
1141 /* Do not set lli_fid, it has been initialized already. */
1142 fid_zero(&lli->lli_pfid);
1143 lli->lli_mds_read_och = NULL;
1144 lli->lli_mds_write_och = NULL;
1145 lli->lli_mds_exec_och = NULL;
1146 lli->lli_open_fd_read_count = 0;
1147 lli->lli_open_fd_write_count = 0;
1148 lli->lli_open_fd_exec_count = 0;
1149 mutex_init(&lli->lli_och_mutex);
1150 spin_lock_init(&lli->lli_agl_lock);
1151 spin_lock_init(&lli->lli_layout_lock);
1152 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1153 lli->lli_clob = NULL;
1155 init_rwsem(&lli->lli_xattrs_list_rwsem);
1156 mutex_init(&lli->lli_xattrs_enq_lock);
1158 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1159 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1160 lli->lli_opendir_key = NULL;
1161 lli->lli_sai = NULL;
1162 spin_lock_init(&lli->lli_sa_lock);
1163 lli->lli_opendir_pid = 0;
1164 lli->lli_sa_enabled = 0;
1165 init_rwsem(&lli->lli_lsm_sem);
1167 mutex_init(&lli->lli_size_mutex);
1168 mutex_init(&lli->lli_setattr_mutex);
1169 lli->lli_symlink_name = NULL;
1170 ll_trunc_sem_init(&lli->lli_trunc_sem);
1171 range_lock_tree_init(&lli->lli_write_tree);
1172 init_rwsem(&lli->lli_glimpse_sem);
1173 lli->lli_glimpse_time = ktime_set(0, 0);
1174 INIT_LIST_HEAD(&lli->lli_agl_list);
1175 lli->lli_agl_index = 0;
1176 lli->lli_async_rc = 0;
1177 spin_lock_init(&lli->lli_heat_lock);
1178 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1179 lli->lli_heat_flags = 0;
1180 mutex_init(&lli->lli_pcc_lock);
1181 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1182 lli->lli_pcc_inode = NULL;
1183 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1184 lli->lli_pcc_generation = 0;
1185 mutex_init(&lli->lli_group_mutex);
1186 lli->lli_group_users = 0;
1187 lli->lli_group_gid = 0;
1189 mutex_init(&lli->lli_layout_mutex);
1190 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1191 /* ll_cl_context initialize */
1192 INIT_LIST_HEAD(&lli->lli_lccs);
1195 #define MAX_STRING_SIZE 128
1197 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1198 #ifndef HAVE_BDI_CAP_MAP_COPY
1199 # define BDI_CAP_MAP_COPY 0
1202 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1204 struct lustre_sb_info *lsi = s2lsi(sb);
1205 char buf[MAX_STRING_SIZE];
1209 err = bdi_init(&lsi->lsi_bdi);
1213 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1214 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1215 lsi->lsi_bdi.name = "lustre";
1216 va_start(args, fmt);
1217 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1219 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1222 sb->s_bdi = &lsi->lsi_bdi;
1226 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1228 int ll_fill_super(struct super_block *sb)
1230 struct lustre_profile *lprof = NULL;
1231 struct lustre_sb_info *lsi = s2lsi(sb);
1232 struct ll_sb_info *sbi = NULL;
1233 char *dt = NULL, *md = NULL;
1234 char *profilenm = get_profile_name(sb);
1235 struct config_llog_instance *cfg;
1236 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1237 const int instlen = LUSTRE_MAXINSTANCE + 2;
1238 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1239 char name[MAX_STRING_SIZE];
1248 /* for ASLR, to map between cfg_instance and hashed ptr */
1249 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1250 profilenm, cfg_instance, sb);
1252 OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1256 GOTO(out_free_cfg, err = -ENOMEM);
1258 /* client additional sb info */
1259 lsi->lsi_llsbi = sbi = ll_init_sbi();
1261 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1263 err = ll_options(lsi->lsi_lmd->lmd_opts, sb);
1265 GOTO(out_free_cfg, err);
1267 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1268 sb->s_d_op = &ll_d_ops;
1271 generate_random_uuid(uuid.b);
1272 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1274 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1277 len = strlen(profilenm);
1278 ptr = strrchr(profilenm, '-');
1279 if (ptr && (strcmp(ptr, "-client") == 0))
1282 if (len > LUSTRE_MAXFSNAME) {
1283 if (unlikely(len >= MAX_STRING_SIZE))
1284 len = MAX_STRING_SIZE - 1;
1285 strncpy(name, profilenm, len);
1287 err = -ENAMETOOLONG;
1288 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1289 name, LUSTRE_MAXFSNAME, err);
1290 GOTO(out_free_cfg, err);
1292 strncpy(sbi->ll_fsname, profilenm, len);
1293 sbi->ll_fsname[len] = '\0';
1296 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1297 profilenm, cfg_instance);
1299 err = super_setup_bdi_name(sb, "%s", name);
1301 GOTO(out_free_cfg, err);
1303 /* disable kernel readahead */
1304 sb->s_bdi->ra_pages = 0;
1306 /* Call ll_debugfs_register_super() before lustre_process_log()
1307 * so that "llite.*.*" params can be processed correctly.
1309 err = ll_debugfs_register_super(sb, name);
1311 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1312 sbi->ll_fsname, err);
1316 /* The cfg_instance is a value unique to this super, in case some
1317 * joker tries to mount the same fs at two mount points.
1319 cfg->cfg_instance = cfg_instance;
1320 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1321 cfg->cfg_callback = class_config_llog_handler;
1322 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1323 /* set up client obds */
1324 err = lustre_process_log(sb, profilenm, cfg);
1326 GOTO(out_debugfs, err);
1328 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1329 lprof = class_get_profile(profilenm);
1330 if (lprof == NULL) {
1331 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1332 " read from the MGS. Does that filesystem "
1333 "exist?\n", profilenm);
1334 GOTO(out_debugfs, err = -EINVAL);
1336 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1337 lprof->lp_md, lprof->lp_dt);
1339 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1340 OBD_ALLOC(dt, dt_len);
1342 GOTO(out_profile, err = -ENOMEM);
1343 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1345 md_len = strlen(lprof->lp_md) + instlen + 2;
1346 OBD_ALLOC(md, md_len);
1348 GOTO(out_free_dt, err = -ENOMEM);
1349 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1351 /* connections, registrations, sb setup */
1352 err = client_common_fill_super(sb, md, dt);
1354 GOTO(out_free_md, err);
1356 sbi->ll_client_common_fill_super_succeeded = 1;
1360 OBD_FREE(md, md_len);
1363 OBD_FREE(dt, dt_len);
1366 class_put_profile(lprof);
1369 ll_debugfs_unregister_super(sb);
1376 else if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1377 LCONSOLE_WARN("Mounted %s\n", profilenm);
1379 } /* ll_fill_super */
1381 void ll_put_super(struct super_block *sb)
1383 struct config_llog_instance cfg, params_cfg;
1384 struct obd_device *obd;
1385 struct lustre_sb_info *lsi = s2lsi(sb);
1386 struct ll_sb_info *sbi = ll_s2sbi(sb);
1387 char *profilenm = get_profile_name(sb);
1388 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1390 int next, force = 1, rc = 0;
1394 GOTO(out_no_sbi, 0);
1396 /* Should replace instance_id with something better for ASLR */
1397 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1398 profilenm, cfg_instance, sb);
1400 cfg.cfg_instance = cfg_instance;
1401 lustre_end_log(sb, profilenm, &cfg);
1403 params_cfg.cfg_instance = cfg_instance;
1404 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1406 if (sbi->ll_md_exp) {
1407 obd = class_exp2obd(sbi->ll_md_exp);
1409 force = obd->obd_force;
1412 /* Wait for unstable pages to be committed to stable storage */
1414 rc = l_wait_event_abortable(
1415 sbi->ll_cache->ccc_unstable_waitq,
1416 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1419 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1420 if (force == 0 && rc != -ERESTARTSYS)
1421 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1423 /* We need to set force before the lov_disconnect in
1424 * lustre_common_put_super, since l_d cleans up osc's as well.
1428 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1430 obd->obd_force = force;
1434 if (sbi->ll_client_common_fill_super_succeeded) {
1435 /* Only if client_common_fill_super succeeded */
1436 client_common_put_super(sb);
1440 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1441 class_manual_cleanup(obd);
1443 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1444 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1447 class_del_profile(profilenm);
1449 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1450 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1451 bdi_destroy(&lsi->lsi_bdi);
1452 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1457 lsi->lsi_llsbi = NULL;
1459 lustre_common_put_super(sb);
1461 cl_env_cache_purge(~0);
1464 } /* client_put_super */
1466 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1468 struct inode *inode = NULL;
1470 /* NOTE: we depend on atomic igrab() -bzzz */
1471 lock_res_and_lock(lock);
1472 if (lock->l_resource->lr_lvb_inode) {
1473 struct ll_inode_info * lli;
1474 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1475 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1476 inode = igrab(lock->l_resource->lr_lvb_inode);
1478 inode = lock->l_resource->lr_lvb_inode;
1479 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1480 D_WARNING, lock, "lr_lvb_inode %p is "
1481 "bogus: magic %08x",
1482 lock->l_resource->lr_lvb_inode,
1483 lli->lli_inode_magic);
1487 unlock_res_and_lock(lock);
1491 void ll_dir_clear_lsm_md(struct inode *inode)
1493 struct ll_inode_info *lli = ll_i2info(inode);
1495 LASSERT(S_ISDIR(inode->i_mode));
1497 if (lli->lli_lsm_md) {
1498 lmv_free_memmd(lli->lli_lsm_md);
1499 lli->lli_lsm_md = NULL;
1502 if (lli->lli_default_lsm_md) {
1503 lmv_free_memmd(lli->lli_default_lsm_md);
1504 lli->lli_default_lsm_md = NULL;
1508 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1509 const struct lu_fid *fid,
1510 struct lustre_md *md)
1512 struct ll_sb_info *sbi = ll_s2sbi(sb);
1513 struct ll_inode_info *lli;
1514 struct mdt_body *body = md->body;
1515 struct inode *inode;
1521 ino = cl_fid_build_ino(fid, test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1522 inode = iget_locked(sb, ino);
1523 if (inode == NULL) {
1524 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1525 sbi->ll_fsname, PFID(fid));
1526 RETURN(ERR_PTR(-ENOENT));
1529 lli = ll_i2info(inode);
1530 if (inode->i_state & I_NEW) {
1531 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1532 (body->mbo_mode & S_IFMT);
1533 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1536 inode->i_mtime.tv_sec = 0;
1537 inode->i_atime.tv_sec = 0;
1538 inode->i_ctime.tv_sec = 0;
1541 #ifdef HAVE_BACKING_DEV_INFO
1542 /* initializing backing dev info. */
1543 inode->i_mapping->backing_dev_info =
1544 &s2lsi(inode->i_sb)->lsi_bdi;
1546 inode->i_op = &ll_dir_inode_operations;
1547 inode->i_fop = &ll_dir_operations;
1548 lli->lli_fid = *fid;
1551 /* master object FID */
1552 lli->lli_pfid = body->mbo_fid1;
1553 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1554 lli, PFID(fid), PFID(&lli->lli_pfid));
1555 unlock_new_inode(inode);
1557 /* in directory restripe/auto-split, a directory will be
1558 * transformed to a stripe if it's plain, set its pfid here,
1559 * otherwise ll_lock_cancel_bits() can't find the master inode.
1561 lli->lli_pfid = body->mbo_fid1;
1567 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1570 struct lmv_stripe_md *lsm = md->lmv;
1571 struct ll_inode_info *lli = ll_i2info(inode);
1574 LASSERT(lsm != NULL);
1576 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1577 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1578 lsm_md_dump(D_INODE, lsm);
1580 if (!lmv_dir_striped(lsm))
1583 /* XXX sigh, this lsm_root initialization should be in
1584 * LMV layer, but it needs ll_iget right now, so we
1585 * put this here right now. */
1586 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1587 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1588 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1590 if (!fid_is_sane(fid))
1593 /* Unfortunately ll_iget will call ll_update_inode,
1594 * where the initialization of slave inode is slightly
1595 * different, so it reset lsm_md to NULL to avoid
1596 * initializing lsm for slave inode. */
1597 lsm->lsm_md_oinfo[i].lmo_root =
1598 ll_iget_anon_dir(inode->i_sb, fid, md);
1599 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1600 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1602 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1604 iput(lsm->lsm_md_oinfo[i].lmo_root);
1605 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1611 lli->lli_lsm_md = lsm;
1616 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1618 struct ll_inode_info *lli = ll_i2info(inode);
1622 if (!md->default_lmv) {
1623 /* clear default lsm */
1624 if (lli->lli_default_lsm_md) {
1625 down_write(&lli->lli_lsm_sem);
1626 if (lli->lli_default_lsm_md) {
1627 lmv_free_memmd(lli->lli_default_lsm_md);
1628 lli->lli_default_lsm_md = NULL;
1630 up_write(&lli->lli_lsm_sem);
1635 if (lli->lli_default_lsm_md) {
1636 /* do nonthing if default lsm isn't changed */
1637 down_read(&lli->lli_lsm_sem);
1638 if (lli->lli_default_lsm_md &&
1639 lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1640 up_read(&lli->lli_lsm_sem);
1643 up_read(&lli->lli_lsm_sem);
1646 down_write(&lli->lli_lsm_sem);
1647 if (lli->lli_default_lsm_md)
1648 lmv_free_memmd(lli->lli_default_lsm_md);
1649 lli->lli_default_lsm_md = md->default_lmv;
1650 lsm_md_dump(D_INODE, md->default_lmv);
1651 md->default_lmv = NULL;
1652 up_write(&lli->lli_lsm_sem);
1656 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1658 struct ll_inode_info *lli = ll_i2info(inode);
1659 struct lmv_stripe_md *lsm = md->lmv;
1660 struct cl_attr *attr;
1665 LASSERT(S_ISDIR(inode->i_mode));
1666 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1667 PFID(ll_inode2fid(inode)));
1669 /* update default LMV */
1670 if (md->default_lmv)
1671 ll_update_default_lsm_md(inode, md);
1673 /* after dir migration/restripe, a stripe may be turned into a
1674 * directory, in this case, zero out its lli_pfid.
1676 if (unlikely(fid_is_norm(&lli->lli_pfid)))
1677 fid_zero(&lli->lli_pfid);
1680 * no striped information from request, lustre_md from req does not
1681 * include stripeEA, see ll_md_setattr()
1687 * normally dir layout doesn't change, only take read lock to check
1688 * that to avoid blocking other MD operations.
1690 down_read(&lli->lli_lsm_sem);
1692 /* some current lookup initialized lsm, and unchanged */
1693 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1694 GOTO(unlock, rc = 0);
1696 /* if dir layout doesn't match, check whether version is increased,
1697 * which means layout is changed, this happens in dir split/merge and
1700 * foreign LMV should not change.
1702 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1703 lsm->lsm_md_layout_version <=
1704 lli->lli_lsm_md->lsm_md_layout_version) {
1705 CERROR("%s: "DFID" dir layout mismatch:\n",
1706 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1707 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1708 lsm_md_dump(D_ERROR, lsm);
1709 GOTO(unlock, rc = -EINVAL);
1712 up_read(&lli->lli_lsm_sem);
1713 down_write(&lli->lli_lsm_sem);
1714 /* clear existing lsm */
1715 if (lli->lli_lsm_md) {
1716 lmv_free_memmd(lli->lli_lsm_md);
1717 lli->lli_lsm_md = NULL;
1720 rc = ll_init_lsm_md(inode, md);
1722 up_write(&lli->lli_lsm_sem);
1726 /* md_merge_attr() may take long, since lsm is already set, switch to
1729 downgrade_write(&lli->lli_lsm_sem);
1731 /* set md->lmv to NULL, so the following free lustre_md will not free
1736 if (!lmv_dir_striped(lli->lli_lsm_md))
1737 GOTO(unlock, rc = 0);
1739 OBD_ALLOC_PTR(attr);
1741 GOTO(unlock, rc = -ENOMEM);
1743 /* validate the lsm */
1744 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1745 ll_md_blocking_ast);
1747 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1748 md->body->mbo_nlink = attr->cat_nlink;
1749 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1750 md->body->mbo_size = attr->cat_size;
1751 if (md->body->mbo_valid & OBD_MD_FLATIME)
1752 md->body->mbo_atime = attr->cat_atime;
1753 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1754 md->body->mbo_ctime = attr->cat_ctime;
1755 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1756 md->body->mbo_mtime = attr->cat_mtime;
1762 up_read(&lli->lli_lsm_sem);
1767 void ll_clear_inode(struct inode *inode)
1769 struct ll_inode_info *lli = ll_i2info(inode);
1770 struct ll_sb_info *sbi = ll_i2sbi(inode);
1774 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1775 PFID(ll_inode2fid(inode)), inode);
1777 if (S_ISDIR(inode->i_mode)) {
1778 /* these should have been cleared in ll_file_release */
1779 LASSERT(lli->lli_opendir_key == NULL);
1780 LASSERT(lli->lli_sai == NULL);
1781 LASSERT(lli->lli_opendir_pid == 0);
1783 pcc_inode_free(inode);
1786 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1788 LASSERT(!lli->lli_open_fd_write_count);
1789 LASSERT(!lli->lli_open_fd_read_count);
1790 LASSERT(!lli->lli_open_fd_exec_count);
1792 if (lli->lli_mds_write_och)
1793 ll_md_real_close(inode, FMODE_WRITE);
1794 if (lli->lli_mds_exec_och)
1795 ll_md_real_close(inode, FMODE_EXEC);
1796 if (lli->lli_mds_read_och)
1797 ll_md_real_close(inode, FMODE_READ);
1799 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1800 OBD_FREE(lli->lli_symlink_name,
1801 strlen(lli->lli_symlink_name) + 1);
1802 lli->lli_symlink_name = NULL;
1805 ll_xattr_cache_destroy(inode);
1807 forget_all_cached_acls(inode);
1809 lli->lli_inode_magic = LLI_INODE_DEAD;
1811 if (S_ISDIR(inode->i_mode))
1812 ll_dir_clear_lsm_md(inode);
1813 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1814 LASSERT(list_empty(&lli->lli_agl_list));
1817 * XXX This has to be done before lsm is freed below, because
1818 * cl_object still uses inode lsm.
1820 cl_inode_fini(inode);
1822 llcrypt_put_encryption_info(inode);
1827 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1829 struct lustre_md md;
1830 struct inode *inode = dentry->d_inode;
1831 struct ll_sb_info *sbi = ll_i2sbi(inode);
1832 struct ptlrpc_request *request = NULL;
1837 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1838 LUSTRE_OPC_ANY, NULL);
1839 if (IS_ERR(op_data))
1840 RETURN(PTR_ERR(op_data));
1842 /* If this is a chgrp of a regular file, we want to reserve enough
1843 * quota to cover the entire file size.
1845 if (S_ISREG(inode->i_mode) && op_data->op_attr.ia_valid & ATTR_GID &&
1846 from_kgid(&init_user_ns, op_data->op_attr.ia_gid) !=
1847 from_kgid(&init_user_ns, inode->i_gid)) {
1848 op_data->op_xvalid |= OP_XVALID_BLOCKS;
1849 op_data->op_attr_blocks = inode->i_blocks;
1853 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1855 ptlrpc_req_finished(request);
1856 if (rc == -ENOENT) {
1858 /* Unlinked special device node? Or just a race?
1859 * Pretend we done everything. */
1860 if (!S_ISREG(inode->i_mode) &&
1861 !S_ISDIR(inode->i_mode)) {
1862 ia_valid = op_data->op_attr.ia_valid;
1863 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1864 rc = simple_setattr(&init_user_ns, dentry,
1866 op_data->op_attr.ia_valid = ia_valid;
1868 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1869 CERROR("md_setattr fails: rc = %d\n", rc);
1874 rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1875 sbi->ll_md_exp, &md);
1877 ptlrpc_req_finished(request);
1881 ia_valid = op_data->op_attr.ia_valid;
1882 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1883 * cache is not cleared yet. */
1884 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1885 if (S_ISREG(inode->i_mode))
1887 rc = simple_setattr(&init_user_ns, dentry, &op_data->op_attr);
1888 if (S_ISREG(inode->i_mode))
1889 inode_unlock(inode);
1890 op_data->op_attr.ia_valid = ia_valid;
1892 rc = ll_update_inode(inode, &md);
1893 ptlrpc_req_finished(request);
1899 * Zero portion of page that is part of @inode.
1900 * This implies, if necessary:
1901 * - taking cl_lock on range corresponding to concerned page
1902 * - grabbing vm page
1903 * - associating cl_page
1904 * - proceeding to clio read
1905 * - zeroing range in page
1906 * - proceeding to cl_page flush
1907 * - releasing cl_lock
1909 * \param[in] inode inode
1910 * \param[in] index page index
1911 * \param[in] offset offset in page to start zero from
1912 * \param[in] len len to zero
1914 * \retval 0 on success
1915 * \retval negative errno on failure
1917 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1920 struct ll_inode_info *lli = ll_i2info(inode);
1921 struct cl_object *clob = lli->lli_clob;
1923 struct lu_env *env = NULL;
1924 struct cl_io *io = NULL;
1925 struct cl_page *clpage = NULL;
1926 struct page *vmpage = NULL;
1927 unsigned from = index << PAGE_SHIFT;
1928 struct cl_lock *lock = NULL;
1929 struct cl_lock_descr *descr = NULL;
1930 struct cl_2queue *queue = NULL;
1931 struct cl_sync_io *anchor = NULL;
1932 bool holdinglock = false;
1937 env = cl_env_get(&refcheck);
1939 RETURN(PTR_ERR(env));
1941 io = vvp_env_thread_io(env);
1943 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1947 lock = vvp_env_lock(env);
1948 descr = &lock->cll_descr;
1949 descr->cld_obj = io->ci_obj;
1950 descr->cld_start = cl_index(io->ci_obj, from);
1951 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1952 descr->cld_mode = CLM_WRITE;
1953 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1955 /* request lock for page */
1956 rc = cl_lock_request(env, io, lock);
1957 /* -ECANCELED indicates a matching lock with a different extent
1958 * was already present, and -EEXIST indicates a matching lock
1959 * on exactly the same extent was already present.
1960 * In both cases it means we are covered.
1962 if (rc == -ECANCELED || rc == -EEXIST)
1970 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1972 GOTO(rellock, rc = -EOPNOTSUPP);
1974 if (!PageDirty(vmpage)) {
1975 /* associate cl_page */
1976 clpage = cl_page_find(env, clob, vmpage->index,
1977 vmpage, CPT_CACHEABLE);
1979 GOTO(pagefini, rc = PTR_ERR(clpage));
1981 cl_page_assume(env, io, clpage);
1984 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1985 !PageWriteback(vmpage)) {
1987 /* Set PagePrivate2 to detect special case of empty page
1988 * in osc_brw_fini_request().
1989 * It is also used to tell ll_io_read_page() that we do not
1990 * want the vmpage to be unlocked.
1992 SetPagePrivate2(vmpage);
1993 rc = ll_io_read_page(env, io, clpage, NULL);
1994 if (!PagePrivate2(vmpage)) {
1995 /* PagePrivate2 was cleared in osc_brw_fini_request()
1996 * meaning we read an empty page. In this case, in order
1997 * to avoid allocating unnecessary block in truncated
1998 * file, we must not zero and write as below. Subsequent
1999 * server-side truncate will handle things correctly.
2001 cl_page_unassume(env, io, clpage);
2002 GOTO(clpfini, rc = 0);
2004 ClearPagePrivate2(vmpage);
2009 /* Thanks to PagePrivate2 flag, ll_io_read_page() did not unlock
2010 * the vmpage, so we are good to proceed and zero range in page.
2012 zero_user(vmpage, offset, len);
2014 if (holdinglock && clpage) {
2015 /* explicitly write newly modified page */
2016 queue = &io->ci_queue;
2017 cl_2queue_init(queue);
2018 anchor = &vvp_env_info(env)->vti_anchor;
2019 cl_sync_io_init(anchor, 1);
2020 clpage->cp_sync_io = anchor;
2021 cl_2queue_add(queue, clpage, true);
2022 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
2024 GOTO(queuefini1, rc);
2025 rc = cl_sync_io_wait(env, anchor, 0);
2027 GOTO(queuefini2, rc);
2028 cl_page_assume(env, io, clpage);
2031 cl_2queue_discard(env, io, queue);
2033 cl_2queue_disown(env, io, queue);
2034 cl_2queue_fini(env, queue);
2039 cl_page_put(env, clpage);
2041 unlock_page(vmpage);
2045 cl_lock_release(env, lock);
2047 cl_io_fini(env, io);
2050 cl_env_put(env, &refcheck);
2056 * Get reference file from volatile file name.
2057 * Volatile file name may look like:
2058 * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
2059 * where fd is opened descriptor of reference file.
2061 * \param[in] volatile_name volatile file name
2062 * \param[in] volatile_len volatile file name length
2063 * \param[out] ref_file pointer to struct file of reference file
2065 * \retval 0 on success
2066 * \retval negative errno on failure
2068 int volatile_ref_file(const char *volatile_name, int volatile_len,
2069 struct file **ref_file)
2071 char *p, *q, *fd_str;
2074 p = strnstr(volatile_name, ":fd=", volatile_len);
2075 if (!p || strlen(p + 4) == 0)
2078 q = strchrnul(p + 4, ':');
2079 fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
2082 rc = kstrtouint(fd_str, 10, &fd);
2087 *ref_file = fget(fd);
2093 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2094 * object(s) determine the file size and mtime. Otherwise, the MDS will
2095 * keep these values until such a time that objects are allocated for it.
2096 * We do the MDS operations first, as it is checking permissions for us.
2097 * We don't to the MDS RPC if there is nothing that we want to store there,
2098 * otherwise there is no harm in updating mtime/atime on the MDS if we are
2099 * going to do an RPC anyways.
2101 * If we are doing a truncate, we will send the mtime and ctime updates
2102 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2103 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2106 * In case of HSMimport, we only set attr on MDS.
2108 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2109 enum op_xvalid xvalid, bool hsm_import)
2111 struct inode *inode = dentry->d_inode;
2112 struct ll_inode_info *lli = ll_i2info(inode);
2113 struct md_op_data *op_data = NULL;
2114 ktime_t kstart = ktime_get();
2119 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2120 "valid %x, hsm_import %d\n",
2121 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2122 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2125 if (attr->ia_valid & ATTR_SIZE) {
2126 /* Check new size against VFS/VM file size limit and rlimit */
2127 rc = inode_newsize_ok(inode, attr->ia_size);
2131 /* The maximum Lustre file size is variable, based on the
2132 * OST maximum object size and number of stripes. This
2133 * needs another check in addition to the VFS check above. */
2134 if (attr->ia_size > ll_file_maxbytes(inode)) {
2135 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2136 PFID(&lli->lli_fid), attr->ia_size,
2137 ll_file_maxbytes(inode));
2141 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2144 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2145 if (attr->ia_valid & TIMES_SET_FLAGS) {
2146 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2147 !capable(CAP_FOWNER))
2151 /* We mark all of the fields "set" so MDS/OST does not re-set them */
2152 if (!(xvalid & OP_XVALID_CTIME_SET) &&
2153 (attr->ia_valid & ATTR_CTIME)) {
2154 attr->ia_ctime = current_time(inode);
2155 xvalid |= OP_XVALID_CTIME_SET;
2157 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2158 (attr->ia_valid & ATTR_ATIME)) {
2159 attr->ia_atime = current_time(inode);
2160 attr->ia_valid |= ATTR_ATIME_SET;
2162 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2163 (attr->ia_valid & ATTR_MTIME)) {
2164 attr->ia_mtime = current_time(inode);
2165 attr->ia_valid |= ATTR_MTIME_SET;
2168 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2169 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2170 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2171 ktime_get_real_seconds());
2173 if (S_ISREG(inode->i_mode))
2174 inode_unlock(inode);
2176 /* We always do an MDS RPC, even if we're only changing the size;
2177 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2179 OBD_ALLOC_PTR(op_data);
2180 if (op_data == NULL)
2181 GOTO(out, rc = -ENOMEM);
2183 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2184 /* If we are changing file size, file content is
2185 * modified, flag it.
2187 xvalid |= OP_XVALID_OWNEROVERRIDE;
2188 op_data->op_bias |= MDS_DATA_MODIFIED;
2189 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2192 if (attr->ia_valid & ATTR_FILE) {
2193 struct ll_file_data *fd = attr->ia_file->private_data;
2195 if (fd->fd_lease_och)
2196 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2199 op_data->op_attr = *attr;
2200 op_data->op_xvalid = xvalid;
2202 rc = ll_md_setattr(dentry, op_data);
2206 if (!S_ISREG(inode->i_mode) || hsm_import)
2209 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2210 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2211 xvalid & OP_XVALID_CTIME_SET) {
2212 bool cached = false;
2214 rc = pcc_inode_setattr(inode, attr, &cached);
2217 CERROR("%s: PCC inode "DFID" setattr failed: "
2219 ll_i2sbi(inode)->ll_fsname,
2220 PFID(&lli->lli_fid), rc);
2224 unsigned int flags = 0;
2226 /* For truncate and utimes sending attributes to OSTs,
2227 * setting mtime/atime to the past will be performed
2228 * under PW [0:EOF] extent lock (new_size:EOF for
2229 * truncate). It may seem excessive to send mtime/atime
2230 * updates to OSTs when not setting times to past, but
2231 * it is necessary due to possible time
2232 * de-synchronization between MDT inode and OST objects
2234 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2235 xvalid |= OP_XVALID_FLAGS;
2236 flags = LUSTRE_ENCRYPT_FL;
2237 /* Call to ll_io_zero_page is not necessary if
2238 * truncating on PAGE_SIZE boundary, because
2239 * whole pages will be wiped.
2240 * In case of Direct IO, all we need is to set
2243 if (attr->ia_valid & ATTR_SIZE &&
2244 attr->ia_size & ~PAGE_MASK &&
2245 !(attr->ia_valid & ATTR_FILE &&
2246 attr->ia_file->f_flags & O_DIRECT)) {
2248 attr->ia_size & (PAGE_SIZE - 1);
2250 rc = ll_io_zero_page(inode,
2251 attr->ia_size >> PAGE_SHIFT,
2252 offset, PAGE_SIZE - offset);
2256 /* If encrypted volatile file without the key,
2257 * we need to fetch size from reference file,
2258 * and set it on OST objects. This happens when
2259 * migrating or extending an encrypted file
2262 if (filename_is_volatile(dentry->d_name.name,
2265 llcrypt_require_key(inode) == -ENOKEY) {
2266 struct file *ref_file;
2267 struct inode *ref_inode;
2268 struct ll_inode_info *ref_lli;
2269 struct cl_object *ref_obj;
2270 struct cl_attr ref_attr = { 0 };
2274 rc = volatile_ref_file(
2275 dentry->d_name.name,
2281 ref_inode = file_inode(ref_file);
2284 GOTO(out, rc = -EINVAL);
2287 env = cl_env_get(&refcheck);
2289 GOTO(out, rc = PTR_ERR(env));
2291 ref_lli = ll_i2info(ref_inode);
2292 ref_obj = ref_lli->lli_clob;
2293 cl_object_attr_lock(ref_obj);
2294 rc = cl_object_attr_get(env, ref_obj,
2296 cl_object_attr_unlock(ref_obj);
2297 cl_env_put(env, &refcheck);
2302 attr->ia_valid |= ATTR_SIZE;
2303 attr->ia_size = ref_attr.cat_size;
2306 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2310 /* If the file was restored, it needs to set dirty flag.
2312 * We've already sent MDS_DATA_MODIFIED flag in
2313 * ll_md_setattr() for truncate. However, the MDT refuses to
2314 * set the HS_DIRTY flag on released files, so we have to set
2315 * it again if the file has been restored. Please check how
2316 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2318 * Please notice that if the file is not released, the previous
2319 * MDS_DATA_MODIFIED has taken effect and usually
2320 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2321 * This way we can save an RPC for common open + trunc
2323 if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2324 struct hsm_state_set hss = {
2325 .hss_valid = HSS_SETMASK,
2326 .hss_setmask = HS_DIRTY,
2330 rc2 = ll_hsm_state_set(inode, &hss);
2331 /* truncate and write can happen at the same time, so that
2332 * the file can be set modified even though the file is not
2333 * restored from released state, and ll_hsm_state_set() is
2334 * not applicable for the file, and rc2 < 0 is normal in this
2337 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2338 PFID(ll_inode2fid(inode)), rc2);
2343 if (op_data != NULL)
2344 ll_finish_md_op_data(op_data);
2346 if (S_ISREG(inode->i_mode)) {
2348 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2349 inode_dio_wait(inode);
2350 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2351 * flag. ll_update_inode (called from ll_md_setattr), clears
2352 * inode flags, so there is a gap where S_NOSEC is not set.
2353 * This can cause a writer to take the i_mutex unnecessarily,
2354 * but this is safe to do and should be rare. */
2355 inode_has_no_xattr(inode);
2359 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2360 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2361 ktime_us_delta(ktime_get(), kstart));
2366 int ll_setattr(struct user_namespace *mnt_userns, struct dentry *de,
2369 int mode = de->d_inode->i_mode;
2370 enum op_xvalid xvalid = 0;
2373 rc = llcrypt_prepare_setattr(de, attr);
2377 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2378 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2379 xvalid |= OP_XVALID_OWNEROVERRIDE;
2381 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2382 (ATTR_SIZE|ATTR_MODE)) &&
2383 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2384 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2385 !(attr->ia_mode & S_ISGID))))
2386 attr->ia_valid |= ATTR_FORCE;
2388 if ((attr->ia_valid & ATTR_MODE) &&
2390 !(attr->ia_mode & S_ISUID) &&
2391 !(attr->ia_valid & ATTR_KILL_SUID))
2392 attr->ia_valid |= ATTR_KILL_SUID;
2394 if ((attr->ia_valid & ATTR_MODE) &&
2395 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2396 !(attr->ia_mode & S_ISGID) &&
2397 !(attr->ia_valid & ATTR_KILL_SGID))
2398 attr->ia_valid |= ATTR_KILL_SGID;
2400 return ll_setattr_raw(de, attr, xvalid, false);
2403 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2406 struct obd_statfs obd_osfs = { 0 };
2411 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2413 if (test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags))
2414 flags |= OBD_STATFS_NODELAY;
2416 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2420 osfs->os_type = LL_SUPER_MAGIC;
2422 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2423 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2425 if (osfs->os_state & OS_STATFS_SUM)
2428 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2429 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2432 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2433 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2436 osfs->os_bsize = obd_osfs.os_bsize;
2437 osfs->os_blocks = obd_osfs.os_blocks;
2438 osfs->os_bfree = obd_osfs.os_bfree;
2439 osfs->os_bavail = obd_osfs.os_bavail;
2441 /* If we have _some_ OSTs, but don't have as many free objects on the
2442 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2443 * to compensate, so that the "inodes in use" number is correct.
2444 * This should be kept in sync with lod_statfs() behaviour.
2446 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2447 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2449 osfs->os_ffree = obd_osfs.os_ffree;
2456 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2458 struct if_quotactl qctl = {
2459 .qc_cmd = LUSTRE_Q_GETQUOTA,
2460 .qc_type = PRJQUOTA,
2461 .qc_valid = QC_GENERAL,
2463 u64 limit, curblock;
2466 qctl.qc_id = ll_i2info(inode)->lli_projid;
2467 ret = quotactl_ioctl(inode->i_sb, &qctl);
2469 /* ignore errors if project ID does not have
2470 * a quota limit or feature unsupported.
2472 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2477 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2478 qctl.qc_dqblk.dqb_bsoftlimit :
2479 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2480 if (limit && sfs->f_blocks > limit) {
2481 curblock = (qctl.qc_dqblk.dqb_curspace +
2482 sfs->f_bsize - 1) / sfs->f_bsize;
2483 sfs->f_blocks = limit;
2484 sfs->f_bfree = sfs->f_bavail =
2485 (sfs->f_blocks > curblock) ?
2486 (sfs->f_blocks - curblock) : 0;
2489 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2490 qctl.qc_dqblk.dqb_isoftlimit :
2491 qctl.qc_dqblk.dqb_ihardlimit;
2492 if (limit && sfs->f_files > limit) {
2493 sfs->f_files = limit;
2494 sfs->f_ffree = (sfs->f_files >
2495 qctl.qc_dqblk.dqb_curinodes) ?
2496 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2502 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2504 struct super_block *sb = de->d_sb;
2505 struct obd_statfs osfs;
2506 __u64 fsid = huge_encode_dev(sb->s_dev);
2507 ktime_t kstart = ktime_get();
2510 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2512 /* Some amount of caching on the client is allowed */
2513 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2517 statfs_unpack(sfs, &osfs);
2519 /* We need to downshift for all 32-bit kernels, because we can't
2520 * tell if the kernel is being called via sys_statfs64() or not.
2521 * Stop before overflowing f_bsize - in which case it is better
2522 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2523 if (sizeof(long) < 8) {
2524 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2527 osfs.os_blocks >>= 1;
2528 osfs.os_bfree >>= 1;
2529 osfs.os_bavail >>= 1;
2533 sfs->f_blocks = osfs.os_blocks;
2534 sfs->f_bfree = osfs.os_bfree;
2535 sfs->f_bavail = osfs.os_bavail;
2536 sfs->f_fsid.val[0] = (__u32)fsid;
2537 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2538 if (ll_i2info(de->d_inode)->lli_projid)
2539 return ll_statfs_project(de->d_inode, sfs);
2541 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2542 ktime_us_delta(ktime_get(), kstart));
2547 void ll_inode_size_lock(struct inode *inode)
2549 struct ll_inode_info *lli;
2551 LASSERT(!S_ISDIR(inode->i_mode));
2553 lli = ll_i2info(inode);
2554 mutex_lock(&lli->lli_size_mutex);
2557 void ll_inode_size_unlock(struct inode *inode)
2559 struct ll_inode_info *lli;
2561 lli = ll_i2info(inode);
2562 mutex_unlock(&lli->lli_size_mutex);
2565 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2567 /* do not clear encryption flag */
2568 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2569 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2570 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2571 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2573 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2576 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2578 struct ll_inode_info *lli = ll_i2info(inode);
2579 struct mdt_body *body = md->body;
2580 struct ll_sb_info *sbi = ll_i2sbi(inode);
2584 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2585 rc = cl_file_inode_init(inode, md);
2590 if (S_ISDIR(inode->i_mode)) {
2591 rc = ll_update_lsm_md(inode, md);
2596 if (body->mbo_valid & OBD_MD_FLACL)
2597 lli_replace_acl(lli, md);
2599 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2600 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1, api32);
2601 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2603 if (body->mbo_valid & OBD_MD_FLATIME) {
2604 if (body->mbo_atime > inode->i_atime.tv_sec)
2605 inode->i_atime.tv_sec = body->mbo_atime;
2606 lli->lli_atime = body->mbo_atime;
2609 if (body->mbo_valid & OBD_MD_FLMTIME) {
2610 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2612 "setting ino %lu mtime from %lld to %llu\n",
2613 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2615 inode->i_mtime.tv_sec = body->mbo_mtime;
2617 lli->lli_mtime = body->mbo_mtime;
2620 if (body->mbo_valid & OBD_MD_FLCTIME) {
2621 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2622 inode->i_ctime.tv_sec = body->mbo_ctime;
2623 lli->lli_ctime = body->mbo_ctime;
2626 if (body->mbo_valid & OBD_MD_FLBTIME)
2627 lli->lli_btime = body->mbo_btime;
2629 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2630 if (body->mbo_valid & OBD_MD_FLFLAGS)
2631 ll_update_inode_flags(inode, body->mbo_flags);
2632 if (body->mbo_valid & OBD_MD_FLMODE)
2633 inode->i_mode = (inode->i_mode & S_IFMT) |
2634 (body->mbo_mode & ~S_IFMT);
2636 if (body->mbo_valid & OBD_MD_FLTYPE)
2637 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2638 (body->mbo_mode & S_IFMT);
2640 LASSERT(inode->i_mode != 0);
2641 if (body->mbo_valid & OBD_MD_FLUID)
2642 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2643 if (body->mbo_valid & OBD_MD_FLGID)
2644 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2645 if (body->mbo_valid & OBD_MD_FLPROJID)
2646 lli->lli_projid = body->mbo_projid;
2647 if (body->mbo_valid & OBD_MD_FLNLINK) {
2648 spin_lock(&inode->i_lock);
2649 set_nlink(inode, body->mbo_nlink);
2650 spin_unlock(&inode->i_lock);
2652 if (body->mbo_valid & OBD_MD_FLRDEV)
2653 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2655 if (body->mbo_valid & OBD_MD_FLID) {
2656 /* FID shouldn't be changed! */
2657 if (fid_is_sane(&lli->lli_fid)) {
2658 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2659 "Trying to change FID "DFID
2660 " to the "DFID", inode "DFID"(%p)\n",
2661 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2662 PFID(ll_inode2fid(inode)), inode);
2664 lli->lli_fid = body->mbo_fid1;
2668 LASSERT(fid_seq(&lli->lli_fid) != 0);
2670 /* In case of encrypted file without the key, please do not lose
2671 * clear text size stored into lli_lazysize in ll_merge_attr(),
2672 * we will need it in ll_prepare_close().
2674 if (lli->lli_attr_valid & OBD_MD_FLLAZYSIZE && lli->lli_lazysize &&
2675 llcrypt_require_key(inode) == -ENOKEY)
2676 lli->lli_attr_valid = body->mbo_valid | OBD_MD_FLLAZYSIZE;
2678 lli->lli_attr_valid = body->mbo_valid;
2679 if (body->mbo_valid & OBD_MD_FLSIZE) {
2680 i_size_write(inode, body->mbo_size);
2682 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2683 PFID(ll_inode2fid(inode)),
2684 (unsigned long long)body->mbo_size);
2686 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2687 inode->i_blocks = body->mbo_blocks;
2689 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2690 lli->lli_lazysize = body->mbo_size;
2691 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2692 lli->lli_lazyblocks = body->mbo_blocks;
2695 if (body->mbo_valid & OBD_MD_TSTATE) {
2696 /* Set LLIF_FILE_RESTORING if restore ongoing and
2697 * clear it when done to ensure to start again
2698 * glimpsing updated attrs
2700 if (body->mbo_t_state & MS_RESTORE)
2701 set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2703 clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2709 /* update directory depth to ROOT, called after LOOKUP lock is fetched. */
2710 void ll_update_dir_depth(struct inode *dir, struct inode *inode)
2712 struct ll_inode_info *lli;
2714 if (!S_ISDIR(inode->i_mode))
2720 lli = ll_i2info(inode);
2721 lli->lli_dir_depth = ll_i2info(dir)->lli_dir_depth + 1;
2722 CDEBUG(D_INODE, DFID" depth %hu\n",
2723 PFID(&lli->lli_fid), lli->lli_dir_depth);
2726 void ll_truncate_inode_pages_final(struct inode *inode)
2728 struct address_space *mapping = &inode->i_data;
2729 unsigned long nrpages;
2730 unsigned long flags;
2732 truncate_inode_pages_final(mapping);
2734 /* Workaround for LU-118: Note nrpages may not be totally updated when
2735 * truncate_inode_pages() returns, as there can be a page in the process
2736 * of deletion (inside __delete_from_page_cache()) in the specified
2737 * range. Thus mapping->nrpages can be non-zero when this function
2738 * returns even after truncation of the whole mapping. Only do this if
2739 * npages isn't already zero.
2741 nrpages = mapping->nrpages;
2743 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2744 nrpages = mapping->nrpages;
2745 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2746 } /* Workaround end */
2748 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2749 "see https://jira.whamcloud.com/browse/LU-118\n",
2750 ll_i2sbi(inode)->ll_fsname,
2751 PFID(ll_inode2fid(inode)), inode, nrpages);
2754 int ll_read_inode2(struct inode *inode, void *opaque)
2756 struct lustre_md *md = opaque;
2757 struct ll_inode_info *lli = ll_i2info(inode);
2761 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2762 PFID(&lli->lli_fid), inode);
2764 /* Core attributes from the MDS first. This is a new inode, and
2765 * the VFS doesn't zero times in the core inode so we have to do
2766 * it ourselves. They will be overwritten by either MDS or OST
2767 * attributes - we just need to make sure they aren't newer.
2769 inode->i_mtime.tv_sec = 0;
2770 inode->i_atime.tv_sec = 0;
2771 inode->i_ctime.tv_sec = 0;
2773 rc = ll_update_inode(inode, md);
2777 /* OIDEBUG(inode); */
2779 #ifdef HAVE_BACKING_DEV_INFO
2780 /* initializing backing dev info. */
2781 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2783 if (S_ISREG(inode->i_mode)) {
2784 struct ll_sb_info *sbi = ll_i2sbi(inode);
2785 inode->i_op = &ll_file_inode_operations;
2786 inode->i_fop = sbi->ll_fop;
2787 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2789 } else if (S_ISDIR(inode->i_mode)) {
2790 inode->i_op = &ll_dir_inode_operations;
2791 inode->i_fop = &ll_dir_operations;
2793 } else if (S_ISLNK(inode->i_mode)) {
2794 inode->i_op = &ll_fast_symlink_inode_operations;
2797 inode->i_op = &ll_special_inode_operations;
2799 init_special_inode(inode, inode->i_mode,
2808 void ll_delete_inode(struct inode *inode)
2810 struct ll_inode_info *lli = ll_i2info(inode);
2813 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2814 /* It is last chance to write out dirty pages,
2815 * otherwise we may lose data while umount.
2817 * If i_nlink is 0 then just discard data. This is safe because
2818 * local inode gets i_nlink 0 from server only for the last
2819 * unlink, so that file is not opened somewhere else
2821 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2822 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2825 ll_truncate_inode_pages_final(inode);
2826 ll_clear_inode(inode);
2832 int ll_iocontrol(struct inode *inode, struct file *file,
2833 unsigned int cmd, unsigned long arg)
2835 struct ll_sb_info *sbi = ll_i2sbi(inode);
2836 struct ptlrpc_request *req = NULL;
2841 case FS_IOC_GETFLAGS: {
2842 struct mdt_body *body;
2843 struct md_op_data *op_data;
2845 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2846 0, 0, LUSTRE_OPC_ANY,
2848 if (IS_ERR(op_data))
2849 RETURN(PTR_ERR(op_data));
2851 op_data->op_valid = OBD_MD_FLFLAGS;
2852 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2853 ll_finish_md_op_data(op_data);
2855 CERROR("%s: failure inode "DFID": rc = %d\n",
2856 sbi->ll_md_exp->exp_obd->obd_name,
2857 PFID(ll_inode2fid(inode)), rc);
2861 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2863 flags = body->mbo_flags;
2865 ptlrpc_req_finished(req);
2867 RETURN(put_user(flags, (int __user *)arg));
2869 case FS_IOC_SETFLAGS: {
2871 struct md_op_data *op_data;
2872 struct cl_object *obj;
2873 struct fsxattr fa = { 0 };
2875 if (get_user(flags, (int __user *)arg))
2878 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2879 if (flags & LUSTRE_PROJINHERIT_FL)
2880 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2882 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2887 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2888 LUSTRE_OPC_ANY, NULL);
2889 if (IS_ERR(op_data))
2890 RETURN(PTR_ERR(op_data));
2892 op_data->op_attr_flags = flags;
2893 op_data->op_xvalid |= OP_XVALID_FLAGS;
2894 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2895 ll_finish_md_op_data(op_data);
2896 ptlrpc_req_finished(req);
2900 ll_update_inode_flags(inode, flags);
2902 obj = ll_i2info(inode)->lli_clob;
2906 OBD_ALLOC_PTR(attr);
2910 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2922 int ll_flush_ctx(struct inode *inode)
2924 struct ll_sb_info *sbi = ll_i2sbi(inode);
2926 CDEBUG(D_SEC, "flush context for user %d\n",
2927 from_kuid(&init_user_ns, current_uid()));
2929 obd_set_info_async(NULL, sbi->ll_md_exp,
2930 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2932 obd_set_info_async(NULL, sbi->ll_dt_exp,
2933 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2938 /* umount -f client means force down, don't save state */
2939 void ll_umount_begin(struct super_block *sb)
2941 struct ll_sb_info *sbi = ll_s2sbi(sb);
2942 struct obd_device *obd;
2943 struct obd_ioctl_data *ioc_data;
2947 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2948 sb->s_count, atomic_read(&sb->s_active));
2950 obd = class_exp2obd(sbi->ll_md_exp);
2952 CERROR("Invalid MDC connection handle %#llx\n",
2953 sbi->ll_md_exp->exp_handle.h_cookie);
2959 obd = class_exp2obd(sbi->ll_dt_exp);
2961 CERROR("Invalid LOV connection handle %#llx\n",
2962 sbi->ll_dt_exp->exp_handle.h_cookie);
2968 OBD_ALLOC_PTR(ioc_data);
2970 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2971 sizeof *ioc_data, ioc_data, NULL);
2973 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2974 sizeof *ioc_data, ioc_data, NULL);
2976 OBD_FREE_PTR(ioc_data);
2979 /* Really, we'd like to wait until there are no requests outstanding,
2980 * and then continue. For now, we just periodically checking for vfs
2981 * to decrement mnt_cnt and hope to finish it within 10sec.
2985 !may_umount(sbi->ll_mnt.mnt)) {
2993 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2995 struct ll_sb_info *sbi = ll_s2sbi(sb);
2996 char *profilenm = get_profile_name(sb);
3000 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
3001 read_only = *flags & MS_RDONLY;
3002 err = obd_set_info_async(NULL, sbi->ll_md_exp,
3003 sizeof(KEY_READ_ONLY),
3004 KEY_READ_ONLY, sizeof(read_only),
3007 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
3008 profilenm, read_only ?
3009 "read-only" : "read-write", err);
3014 sb->s_flags |= SB_RDONLY;
3016 sb->s_flags &= ~SB_RDONLY;
3018 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
3019 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
3020 read_only ? "read-only" : "read-write");
3026 * Cleanup the open handle that is cached on MDT-side.
3028 * For open case, the client side open handling thread may hit error
3029 * after the MDT grant the open. Under such case, the client should
3030 * send close RPC to the MDT as cleanup; otherwise, the open handle
3031 * on the MDT will be leaked there until the client umount or evicted.
3033 * In further, if someone unlinked the file, because the open handle
3034 * holds the reference on such file/object, then it will block the
3035 * subsequent threads that want to locate such object via FID.
3037 * \param[in] sb super block for this file-system
3038 * \param[in] open_req pointer to the original open request
3040 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
3042 struct mdt_body *body;
3043 struct md_op_data *op_data;
3044 struct ptlrpc_request *close_req = NULL;
3045 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
3048 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
3049 OBD_ALLOC_PTR(op_data);
3050 if (op_data == NULL) {
3051 CWARN("%s: cannot allocate op_data to release open handle for "
3052 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
3057 op_data->op_fid1 = body->mbo_fid1;
3058 op_data->op_open_handle = body->mbo_open_handle;
3059 op_data->op_mod_time = ktime_get_real_seconds();
3060 md_close(exp, op_data, NULL, &close_req);
3061 ptlrpc_req_finished(close_req);
3062 ll_finish_md_op_data(op_data);
3067 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
3068 struct super_block *sb, struct lookup_intent *it)
3070 struct ll_sb_info *sbi = NULL;
3071 struct lustre_md md = { NULL };
3072 bool default_lmv_deleted = false;
3077 LASSERT(*inode || sb);
3078 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
3079 rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
3080 sbi->ll_md_exp, &md);
3085 * clear default_lmv only if intent_getattr reply doesn't contain it.
3086 * but it needs to be done after iget, check this early because
3087 * ll_update_lsm_md() may change md.
3089 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
3090 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
3091 default_lmv_deleted = true;
3094 rc = ll_update_inode(*inode, &md);
3098 bool api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
3099 struct lu_fid *fid1 = &md.body->mbo_fid1;
3101 LASSERT(sb != NULL);
3104 * At this point server returns to client's same fid as client
3105 * generated for creating. So using ->fid1 is okay here.
3107 if (!fid_is_sane(fid1)) {
3108 CERROR("%s: Fid is insane "DFID"\n",
3109 sbi->ll_fsname, PFID(fid1));
3110 GOTO(out, rc = -EINVAL);
3113 *inode = ll_iget(sb, cl_fid_build_ino(fid1, api32), &md);
3114 if (IS_ERR(*inode)) {
3116 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
3118 CERROR("new_inode -fatal: rc %d\n", rc);
3123 /* Handling piggyback layout lock.
3124 * Layout lock can be piggybacked by getattr and open request.
3125 * The lsm can be applied to inode only if it comes with a layout lock
3126 * otherwise correct layout may be overwritten, for example:
3127 * 1. proc1: mdt returns a lsm but not granting layout
3128 * 2. layout was changed by another client
3129 * 3. proc2: refresh layout and layout lock granted
3130 * 4. proc1: to apply a stale layout */
3131 if (it != NULL && it->it_lock_mode != 0) {
3132 struct lustre_handle lockh;
3133 struct ldlm_lock *lock;
3135 lockh.cookie = it->it_lock_handle;
3136 lock = ldlm_handle2lock(&lockh);
3137 LASSERT(lock != NULL);
3138 if (ldlm_has_layout(lock)) {
3139 struct cl_object_conf conf;
3141 memset(&conf, 0, sizeof(conf));
3142 conf.coc_opc = OBJECT_CONF_SET;
3143 conf.coc_inode = *inode;
3144 conf.coc_lock = lock;
3145 conf.u.coc_layout = md.layout;
3146 (void)ll_layout_conf(*inode, &conf);
3148 LDLM_LOCK_PUT(lock);
3151 if (default_lmv_deleted)
3152 ll_update_default_lsm_md(*inode, &md);
3154 /* we may want to apply some policy for foreign file/dir */
3155 if (ll_sbi_has_foreign_symlink(sbi)) {
3156 rc = ll_manage_foreign(*inode, &md);
3164 /* cleanup will be done if necessary */
3165 md_free_lustre_md(sbi->ll_md_exp, &md);
3167 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3168 ll_intent_drop_lock(it);
3169 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3175 int ll_obd_statfs(struct inode *inode, void __user *arg)
3177 struct ll_sb_info *sbi = NULL;
3178 struct obd_export *exp;
3179 struct obd_ioctl_data *data = NULL;
3184 sbi = ll_i2sbi(inode);
3186 GOTO(out_statfs, rc = -EINVAL);
3188 rc = obd_ioctl_getdata(&data, &len, arg);
3190 GOTO(out_statfs, rc);
3192 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3193 !data->ioc_pbuf1 || !data->ioc_pbuf2)
3194 GOTO(out_statfs, rc = -EINVAL);
3196 if (data->ioc_inllen1 != sizeof(__u32) ||
3197 data->ioc_inllen2 != sizeof(__u32) ||
3198 data->ioc_plen1 != sizeof(struct obd_statfs) ||
3199 data->ioc_plen2 != sizeof(struct obd_uuid))
3200 GOTO(out_statfs, rc = -EINVAL);
3202 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3203 if (type & LL_STATFS_LMV)
3204 exp = sbi->ll_md_exp;
3205 else if (type & LL_STATFS_LOV)
3206 exp = sbi->ll_dt_exp;
3208 GOTO(out_statfs, rc = -ENODEV);
3210 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3212 GOTO(out_statfs, rc);
3214 OBD_FREE_LARGE(data, len);
3219 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3220 * be called early to avoid deadlock.
3222 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3224 if (op_data->op_mea2_sem) {
3225 up_read_non_owner(op_data->op_mea2_sem);
3226 op_data->op_mea2_sem = NULL;
3229 if (op_data->op_mea1_sem) {
3230 up_read_non_owner(op_data->op_mea1_sem);
3231 op_data->op_mea1_sem = NULL;
3235 /* this function prepares md_op_data hint for passing it down to MD stack. */
3236 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3237 struct inode *i1, struct inode *i2,
3238 const char *name, size_t namelen,
3239 __u32 mode, enum md_op_code opc,
3242 struct llcrypt_name fname = { 0 };
3245 LASSERT(i1 != NULL);
3248 /* Do not reuse namelen for something else. */
3250 return ERR_PTR(-EINVAL);
3252 if ((!IS_ENCRYPTED(i1) ||
3253 (opc != LUSTRE_OPC_LOOKUP && opc != LUSTRE_OPC_CREATE)) &&
3254 namelen > ll_i2sbi(i1)->ll_namelen)
3255 return ERR_PTR(-ENAMETOOLONG);
3257 /* "/" is not valid name, but it's allowed */
3258 if (!lu_name_is_valid_2(name, namelen) &&
3259 strncmp("/", name, namelen) != 0)
3260 return ERR_PTR(-EINVAL);
3263 if (op_data == NULL)
3264 OBD_ALLOC_PTR(op_data);
3266 if (op_data == NULL)
3267 return ERR_PTR(-ENOMEM);
3269 ll_i2gids(op_data->op_suppgids, i1, i2);
3270 /* If the client is using a subdir mount and looks at what it sees as
3271 * /.fscrypt, interpret it as the .fscrypt dir at the root of the fs.
3273 if (unlikely(i1->i_sb && i1->i_sb->s_root && is_root_inode(i1) &&
3274 !fid_is_root(ll_inode2fid(i1)) &&
3275 name && namelen == strlen(dot_fscrypt_name) &&
3276 strncmp(name, dot_fscrypt_name, namelen) == 0))
3277 lu_root_fid(&op_data->op_fid1);
3279 op_data->op_fid1 = *ll_inode2fid(i1);
3281 if (S_ISDIR(i1->i_mode)) {
3282 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3283 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3284 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3285 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3289 op_data->op_fid2 = *ll_inode2fid(i2);
3290 if (S_ISDIR(i2->i_mode)) {
3292 /* i2 is typically a child of i1, and MUST be
3293 * further from the root to avoid deadlocks.
3295 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3296 op_data->op_mea2_sem =
3297 &ll_i2info(i2)->lli_lsm_sem;
3299 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3302 fid_zero(&op_data->op_fid2);
3305 if (test_bit(LL_SBI_64BIT_HASH, ll_i2sbi(i1)->ll_flags))
3306 op_data->op_cli_flags |= CLI_HASH64;
3308 if (ll_need_32bit_api(ll_i2sbi(i1)))
3309 op_data->op_cli_flags |= CLI_API32;
3311 if ((i2 && is_root_inode(i2)) ||
3312 opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_CREATE) {
3313 /* In case of lookup, ll_setup_filename() has already been
3314 * called in ll_lookup_it(), so just take provided name.
3315 * Also take provided name if we are dealing with root inode.
3317 fname.disk_name.name = (unsigned char *)name;
3318 fname.disk_name.len = namelen;
3319 } else if (name && namelen) {
3320 struct qstr dname = QSTR_INIT(name, namelen);
3322 struct lu_fid *pfid = NULL;
3326 if (!S_ISDIR(i1->i_mode) && i2 && S_ISDIR(i2->i_mode)) {
3327 /* special case when called from ll_link() */
3332 lookup = (int)(opc == LUSTRE_OPC_ANY);
3334 if (opc == LUSTRE_OPC_ANY && lookup)
3336 rc = ll_setup_filename(dir, &dname, lookup, &fname, pfid);
3338 ll_finish_md_op_data(op_data);
3341 if (pfid && !fid_is_zero(pfid)) {
3343 op_data->op_fid2 = fid;
3344 op_data->op_bias = MDS_FID_OP;
3346 if (fname.disk_name.name &&
3347 fname.disk_name.name != (unsigned char *)name)
3348 /* op_data->op_name must be freed after use */
3349 op_data->op_flags |= MF_OPNAME_KMALLOCED;
3352 /* In fact LUSTRE_OPC_LOOKUP, LUSTRE_OPC_OPEN
3353 * are LUSTRE_OPC_ANY
3355 if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_OPEN)
3356 op_data->op_code = LUSTRE_OPC_ANY;
3358 op_data->op_code = opc;
3359 op_data->op_name = fname.disk_name.name;
3360 op_data->op_namelen = fname.disk_name.len;
3361 op_data->op_mode = mode;
3362 op_data->op_mod_time = ktime_get_real_seconds();
3363 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3364 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3365 op_data->op_cap = current_cap();
3366 op_data->op_mds = 0;
3367 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3368 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3369 op_data->op_bias |= MDS_CREATE_VOLATILE;
3371 op_data->op_data = data;
3376 void ll_finish_md_op_data(struct md_op_data *op_data)
3378 ll_unlock_md_op_lsm(op_data);
3379 ll_security_release_secctx(op_data->op_file_secctx,
3380 op_data->op_file_secctx_size);
3381 if (op_data->op_flags & MF_OPNAME_KMALLOCED)
3382 /* allocated via ll_setup_filename called
3383 * from ll_prep_md_op_data
3385 kfree(op_data->op_name);
3386 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3387 OBD_FREE_PTR(op_data);
3390 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3392 struct ll_sb_info *sbi;
3395 LASSERT(seq && dentry);
3396 sbi = ll_s2sbi(dentry->d_sb);
3398 if (test_bit(LL_SBI_NOLCK, sbi->ll_flags))
3399 seq_puts(seq, "nolock");
3401 for (i = 1; ll_sbi_flags_name[i].token != LL_SBI_NUM_MOUNT_OPT; i++) {
3402 /* match_table in some cases has patterns for both enabled and
3403 * disabled cases. Ignore 'no'xxx versions if bit is set.
3405 if (test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3406 strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3407 if (ll_sbi_flags_name[i].token ==
3408 LL_SBI_FOREIGN_SYMLINK) {
3409 seq_show_option(seq, "foreign_symlink",
3410 sbi->ll_foreign_symlink_prefix);
3412 seq_printf(seq, ",%s",
3413 ll_sbi_flags_name[i].pattern);
3416 /* You can have either localflock or flock but not
3417 * both. If localflock is set don't print flock or
3420 if (ll_sbi_flags_name[i].token == LL_SBI_LOCALFLOCK)
3422 } else if (!test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3423 !strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3424 seq_printf(seq, ",%s",
3425 ll_sbi_flags_name[i].pattern);
3433 * Get obd name by cmd, and copy out to user space
3435 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3437 struct ll_sb_info *sbi = ll_i2sbi(inode);
3438 struct obd_device *obd;
3441 if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3442 obd = class_exp2obd(sbi->ll_dt_exp);
3443 else if (cmd == OBD_IOC_GETMDNAME)
3444 obd = class_exp2obd(sbi->ll_md_exp);
3451 if (copy_to_user((void __user *)arg, obd->obd_name,
3452 strlen(obd->obd_name) + 1))
3459 struct work_struct db_work;
3460 struct dentry *db_dentry;
3461 /* Let's hope the path is not too long, 32 bytes for the work struct
3464 char buf[PAGE_SIZE - sizeof(struct work_struct) - sizeof(void *)];
3467 static void ll_dput_later(struct work_struct *work)
3469 struct dname_buf *db = container_of(work, struct dname_buf, db_work);
3471 dput(db->db_dentry);
3472 free_page((unsigned long)db);
3475 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3482 p.mnt = current->fs->root.mnt;
3484 path = d_path(&p, buf, bufsize);
3489 void ll_dirty_page_discard_warn(struct inode *inode, int ioret)
3491 struct dname_buf *db;
3493 struct dentry *dentry = NULL;
3495 /* this can be called inside spin lock so use GFP_ATOMIC. */
3496 db = (struct dname_buf *)__get_free_page(GFP_ATOMIC);
3499 dentry = d_find_alias(inode);
3501 path = ll_d_path(dentry, db->buf, sizeof(db->buf));
3504 /* The below message is checked in recovery-small.sh test_24b */
3506 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3507 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3508 s2lsi(inode->i_sb)->lsi_lmd->lmd_dev,
3509 PFID(ll_inode2fid(inode)),
3510 (path && !IS_ERR(path)) ? path : "", ioret);
3512 if (dentry != NULL) {
3513 /* We cannot dput here since if we happen to be the last holder
3514 * then we can end up waiting for page evictions that
3515 * in turn wait for RPCs that need this instance of ptlrpcd
3516 * (callng brw_interpret->*page_completion*->vmpage_error->here)
3519 INIT_WORK(&db->db_work, ll_dput_later);
3520 db->db_dentry = dentry;
3521 schedule_work(&db->db_work);
3524 free_page((unsigned long)db);
3528 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3529 struct lov_user_md **kbuf)
3531 struct lov_user_md lum;
3535 if (copy_from_user(&lum, md, sizeof(lum)))
3538 lum_size = ll_lov_user_md_size(&lum);
3542 OBD_ALLOC_LARGE(*kbuf, lum_size);
3546 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3547 OBD_FREE_LARGE(*kbuf, lum_size);
3555 * Compute llite root squash state after a change of root squash
3556 * configuration setting or add/remove of a lnet nid
3558 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3560 struct root_squash_info *squash = &sbi->ll_squash;
3563 struct lnet_processid id;
3565 /* Update norootsquash flag */
3566 spin_lock(&squash->rsi_lock);
3567 if (list_empty(&squash->rsi_nosquash_nids))
3568 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3570 /* Do not apply root squash as soon as one of our NIDs is
3571 * in the nosquash_nids list */
3574 while (LNetGetId(i++, &id) != -ENOENT) {
3575 if (nid_is_lo0(&id.nid))
3577 if (cfs_match_nid(lnet_nid_to_nid4(&id.nid),
3578 &squash->rsi_nosquash_nids)) {
3584 set_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3586 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3588 spin_unlock(&squash->rsi_lock);
3592 * Parse linkea content to extract information about a given hardlink
3594 * \param[in] ldata - Initialized linkea data
3595 * \param[in] linkno - Link identifier
3596 * \param[out] parent_fid - The entry's parent FID
3597 * \param[out] ln - Entry name destination buffer
3599 * \retval 0 on success
3600 * \retval Appropriate negative error code on failure
3602 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3603 struct lu_fid *parent_fid, struct lu_name *ln)
3609 rc = linkea_init_with_rec(ldata);
3613 if (linkno >= ldata->ld_leh->leh_reccount)
3614 /* beyond last link */
3617 linkea_first_entry(ldata);
3618 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3619 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3624 linkea_next_entry(ldata);
3634 * Get parent FID and name of an identified link. Operation is performed for
3635 * a given link number, letting the caller iterate over linkno to list one or
3636 * all links of an entry.
3638 * \param[in] file - File descriptor against which to perform the operation
3639 * \param[in,out] arg - User-filled structure containing the linkno to operate
3640 * on and the available size. It is eventually filled with
3641 * the requested information or left untouched on error
3643 * \retval - 0 on success
3644 * \retval - Appropriate negative error code on failure
3646 int ll_getparent(struct file *file, struct getparent __user *arg)
3648 struct inode *inode = file_inode(file);
3649 struct linkea_data *ldata;
3650 struct lu_buf buf = LU_BUF_NULL;
3652 struct lu_fid parent_fid;
3659 if (!capable(CAP_DAC_READ_SEARCH) &&
3660 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3663 if (get_user(name_size, &arg->gp_name_size))
3666 if (get_user(linkno, &arg->gp_linkno))
3669 if (name_size > PATH_MAX)
3672 OBD_ALLOC(ldata, sizeof(*ldata));
3676 rc = linkea_data_new(ldata, &buf);
3678 GOTO(ldata_free, rc);
3680 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3681 buf.lb_len, OBD_MD_FLXATTR);
3685 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3689 if (ln.ln_namelen >= name_size)
3690 GOTO(lb_free, rc = -EOVERFLOW);
3692 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3693 GOTO(lb_free, rc = -EFAULT);
3695 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3696 GOTO(lb_free, rc = -EFAULT);
3698 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3699 GOTO(lb_free, rc = -EFAULT);
3704 OBD_FREE(ldata, sizeof(*ldata));