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/types.h>
44 #include <libcfs/linux/linux-uuid.h>
45 #include <linux/version.h>
47 #include <linux/user_namespace.h>
48 #include <linux/delay.h>
49 #include <linux/uidgid.h>
50 #include <linux/fs_struct.h>
52 #ifndef HAVE_CPUS_READ_LOCK
53 #include <libcfs/linux/linux-cpu.h>
55 #include <uapi/linux/lustre/lustre_ioctl.h>
56 #ifdef HAVE_UAPI_LINUX_MOUNT_H
57 #include <uapi/linux/mount.h>
60 #include <lustre_ha.h>
61 #include <lustre_dlm.h>
62 #include <lprocfs_status.h>
63 #include <lustre_disk.h>
64 #include <uapi/linux/lustre/lustre_param.h>
65 #include <lustre_log.h>
66 #include <cl_object.h>
67 #include <obd_cksum.h>
68 #include "llite_internal.h"
70 struct kmem_cache *ll_file_data_slab;
73 #define log2(n) ffz(~(n))
77 * If there is only one number of core visible to Lustre,
78 * async readahead will be disabled, to avoid massive over
79 * subscription, we use 1/2 of active cores as default max
80 * async readahead requests.
82 static inline unsigned int ll_get_ra_async_max_active(void)
84 return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
87 static struct ll_sb_info *ll_init_sbi(void)
89 struct ll_sb_info *sbi = NULL;
91 unsigned long lru_page_max;
100 RETURN(ERR_PTR(-ENOMEM));
102 rc = pcc_super_init(&sbi->ll_pcc_super);
106 spin_lock_init(&sbi->ll_lock);
107 mutex_init(&sbi->ll_lco.lco_lock);
108 spin_lock_init(&sbi->ll_pp_extent_lock);
109 spin_lock_init(&sbi->ll_process_lock);
110 sbi->ll_rw_stats_on = 0;
111 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
114 pages = si.totalram - si.totalhigh;
115 lru_page_max = pages / 2;
117 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
118 sbi->ll_ra_info.ll_readahead_wq =
119 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
121 sbi->ll_ra_info.ra_async_max_active);
122 if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
123 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
125 /* initialize ll_cache data */
126 sbi->ll_cache = cl_cache_init(lru_page_max);
127 if (sbi->ll_cache == NULL)
128 GOTO(out_destroy_ra, rc = -ENOMEM);
130 /* initialize foreign symlink prefix path */
131 OBD_ALLOC(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
132 if (sbi->ll_foreign_symlink_prefix == NULL)
133 GOTO(out_destroy_ra, rc = -ENOMEM);
134 memcpy(sbi->ll_foreign_symlink_prefix, "/mnt/", sizeof("/mnt/"));
135 sbi->ll_foreign_symlink_prefix_size = sizeof("/mnt/");
137 /* initialize foreign symlink upcall path, none by default */
138 OBD_ALLOC(sbi->ll_foreign_symlink_upcall, sizeof("none"));
139 if (sbi->ll_foreign_symlink_upcall == NULL)
140 GOTO(out_destroy_ra, rc = -ENOMEM);
141 memcpy(sbi->ll_foreign_symlink_upcall, "none", sizeof("none"));
142 sbi->ll_foreign_symlink_upcall_items = NULL;
143 sbi->ll_foreign_symlink_upcall_nb_items = 0;
144 init_rwsem(&sbi->ll_foreign_symlink_sem);
145 /* foreign symlink support (LL_SBI_FOREIGN_SYMLINK in ll_flags)
146 * not enabled by default
149 sbi->ll_ra_info.ra_max_pages =
150 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
151 sbi->ll_ra_info.ra_max_pages_per_file =
152 min(sbi->ll_ra_info.ra_max_pages / 4,
153 SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
154 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
155 sbi->ll_ra_info.ra_max_pages_per_file;
156 sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
157 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
158 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
160 sbi->ll_flags |= LL_SBI_VERBOSE;
161 #ifdef ENABLE_CHECKSUM
162 sbi->ll_flags |= LL_SBI_CHECKSUM;
165 sbi->ll_flags |= LL_SBI_FLOCK;
168 #ifdef HAVE_LRU_RESIZE_SUPPORT
169 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
171 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
173 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
174 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
176 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
180 /* metadata statahead is enabled by default */
181 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
182 sbi->ll_sa_max = LL_SA_RPC_DEF;
183 atomic_set(&sbi->ll_sa_total, 0);
184 atomic_set(&sbi->ll_sa_wrong, 0);
185 atomic_set(&sbi->ll_sa_running, 0);
186 atomic_set(&sbi->ll_agl_total, 0);
187 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
188 sbi->ll_flags |= LL_SBI_FAST_READ;
189 sbi->ll_flags |= LL_SBI_TINY_WRITE;
190 ll_sbi_set_encrypt(sbi, true);
193 sbi->ll_squash.rsi_uid = 0;
194 sbi->ll_squash.rsi_gid = 0;
195 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
196 spin_lock_init(&sbi->ll_squash.rsi_lock);
198 /* Per-filesystem file heat */
199 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
200 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
202 /* Per-fs open heat level before requesting open lock */
203 sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
204 sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
205 sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
208 if (sbi->ll_foreign_symlink_prefix)
209 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
211 cl_cache_decref(sbi->ll_cache);
212 sbi->ll_cache = NULL;
214 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
216 pcc_super_fini(&sbi->ll_pcc_super);
222 static void ll_free_sbi(struct super_block *sb)
224 struct ll_sb_info *sbi = ll_s2sbi(sb);
228 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
229 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
230 if (sbi->ll_ra_info.ll_readahead_wq)
231 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
232 if (sbi->ll_cache != NULL) {
233 cl_cache_decref(sbi->ll_cache);
234 sbi->ll_cache = NULL;
236 if (sbi->ll_foreign_symlink_prefix) {
237 OBD_FREE(sbi->ll_foreign_symlink_prefix,
238 sbi->ll_foreign_symlink_prefix_size);
239 sbi->ll_foreign_symlink_prefix = NULL;
241 if (sbi->ll_foreign_symlink_upcall) {
242 OBD_FREE(sbi->ll_foreign_symlink_upcall,
243 strlen(sbi->ll_foreign_symlink_upcall) +
245 sbi->ll_foreign_symlink_upcall = NULL;
247 if (sbi->ll_foreign_symlink_upcall_items) {
249 int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
250 struct ll_foreign_symlink_upcall_item *items =
251 sbi->ll_foreign_symlink_upcall_items;
253 for (i = 0 ; i < nb_items; i++)
254 if (items[i].type == STRING_TYPE)
255 OBD_FREE(items[i].string,
258 OBD_FREE_LARGE(items, nb_items *
259 sizeof(struct ll_foreign_symlink_upcall_item));
260 sbi->ll_foreign_symlink_upcall_items = NULL;
262 pcc_super_fini(&sbi->ll_pcc_super);
263 OBD_FREE(sbi, sizeof(*sbi));
268 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
270 struct inode *root = NULL;
271 struct ll_sb_info *sbi = ll_s2sbi(sb);
272 struct obd_statfs *osfs = NULL;
273 struct ptlrpc_request *request = NULL;
274 struct obd_connect_data *data = NULL;
275 struct obd_uuid *uuid;
276 struct md_op_data *op_data;
277 struct lustre_md lmd;
279 int size, err, checksum;
282 sbi->ll_md_obd = class_name2obd(md);
283 if (!sbi->ll_md_obd) {
284 CERROR("MD %s: not setup or attached\n", md);
298 /* pass client page size via ocd_grant_blkbits, the server should report
299 * back its backend blocksize for grant calculation purpose */
300 data->ocd_grant_blkbits = PAGE_SHIFT;
302 /* indicate MDT features supported by this client */
303 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
304 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
305 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
306 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
307 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
308 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
309 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
310 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
311 OBD_CONNECT_64BITHASH |
312 OBD_CONNECT_EINPROGRESS |
313 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
314 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
315 OBD_CONNECT_MAX_EASIZE |
316 OBD_CONNECT_FLOCK_DEAD |
317 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
318 OBD_CONNECT_OPEN_BY_FID |
319 OBD_CONNECT_DIR_STRIPE |
320 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
321 OBD_CONNECT_SUBTREE |
322 OBD_CONNECT_MULTIMODRPCS |
323 OBD_CONNECT_GRANT_PARAM |
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 (sbi->ll_flags & LL_SBI_LRU_RESIZE)
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 (sbi->ll_flags & LL_SBI_USER_XATTR)
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 (sbi->ll_flags & LL_SBI_ALWAYS_PING)
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_namelen = osfs->os_namelen;
456 sbi->ll_mnt.mnt = current->fs->root.mnt;
458 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
459 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
460 LCONSOLE_INFO("Disabling user_xattr feature because "
461 "it is not supported on the server\n");
462 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
465 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
467 sb->s_flags |= SB_POSIXACL;
469 sbi->ll_flags |= LL_SBI_ACL;
471 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
473 sb->s_flags &= ~SB_POSIXACL;
475 sbi->ll_flags &= ~LL_SBI_ACL;
478 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
479 sbi->ll_flags |= LL_SBI_64BIT_HASH;
481 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
482 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
484 if (obd_connect_has_secctx(data))
485 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
487 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
488 if (ll_sbi_has_test_dummy_encryption(sbi))
489 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
491 sbi->ll_md_exp->exp_obd->obd_name);
492 ll_sbi_set_encrypt(sbi, false);
495 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
496 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
497 LCONSOLE_INFO("%s: disabling xattr cache due to "
498 "unknown maximum xattr size.\n", dt);
499 } else if (!sbi->ll_xattr_cache_set) {
500 /* If xattr_cache is already set (no matter 0 or 1)
501 * during processing llog, it won't be enabled here. */
502 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
503 sbi->ll_xattr_cache_enabled = 1;
507 sbi->ll_dt_obd = class_name2obd(dt);
508 if (!sbi->ll_dt_obd) {
509 CERROR("DT %s: not setup or attached\n", dt);
510 GOTO(out_md_fid, err = -ENODEV);
513 /* pass client page size via ocd_grant_blkbits, the server should report
514 * back its backend blocksize for grant calculation purpose */
515 data->ocd_grant_blkbits = PAGE_SHIFT;
517 /* indicate OST features supported by this client */
518 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
519 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
520 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
521 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
522 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
523 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
524 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
525 OBD_CONNECT_EINPROGRESS |
526 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
527 OBD_CONNECT_LAYOUTLOCK |
528 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
529 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
530 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
531 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
532 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
533 OBD_CONNECT2_REP_MBITS;
535 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
536 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
538 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
539 * disabled by default, because it can still be enabled on the
540 * fly via /sys. As a consequence, we still need to come to an
541 * agreement on the supported algorithms at connect time
543 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
545 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
546 data->ocd_cksum_types = OBD_CKSUM_ADLER;
548 data->ocd_cksum_types = obd_cksum_types_supported_client();
550 #ifdef HAVE_LRU_RESIZE_SUPPORT
551 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
553 /* always ping even if server suppress_pings */
554 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
555 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
557 if (ll_sbi_has_encrypt(sbi))
558 obd_connect_set_enc(data);
560 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
561 "ocd_grant: %d\n", data->ocd_connect_flags,
562 data->ocd_version, data->ocd_grant);
564 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
565 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
567 data->ocd_brw_size = DT_MAX_BRW_SIZE;
569 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
570 &sbi->ll_sb_uuid, data, sbi->ll_cache);
572 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
573 "recovery, of which this client is not a "
574 "part. Please wait for recovery to "
575 "complete, abort, or time out.\n", dt);
578 CERROR("%s: Cannot connect to %s: rc = %d\n",
579 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
583 if (ll_sbi_has_encrypt(sbi) &&
584 !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
585 if (ll_sbi_has_test_dummy_encryption(sbi))
586 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
588 ll_sbi_set_encrypt(sbi, false);
589 } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
590 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
593 sbi->ll_dt_exp->exp_connect_data = *data;
595 /* Don't change value if it was specified in the config log */
596 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
597 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
598 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
599 (data->ocd_brw_size >> PAGE_SHIFT));
600 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
601 sbi->ll_ra_info.ra_max_pages_per_file)
602 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
603 sbi->ll_ra_info.ra_max_pages_per_file;
606 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
607 LUSTRE_SEQ_METADATA);
609 CERROR("%s: Can't init data layer FID infrastructure, "
610 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
614 mutex_lock(&sbi->ll_lco.lco_lock);
615 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
616 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
617 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
618 mutex_unlock(&sbi->ll_lco.lco_lock);
620 fid_zero(&sbi->ll_root_fid);
621 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
624 CERROR("cannot mds_connect: rc = %d\n", err);
625 GOTO(out_lock_cn_cb, err);
627 if (!fid_is_sane(&sbi->ll_root_fid)) {
628 CERROR("%s: Invalid root fid "DFID" during mount\n",
629 sbi->ll_md_exp->exp_obd->obd_name,
630 PFID(&sbi->ll_root_fid));
631 GOTO(out_lock_cn_cb, err = -EINVAL);
633 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
635 sb->s_op = &lustre_super_operations;
636 sb->s_xattr = ll_xattr_handlers;
637 #if THREAD_SIZE >= 8192 /*b=17630*/
638 sb->s_export_op = &lustre_export_operations;
640 #ifdef HAVE_LUSTRE_CRYPTO
641 llcrypt_set_ops(sb, &lustre_cryptops);
645 * XXX: move this to after cbd setup? */
646 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
647 if (sbi->ll_flags & LL_SBI_ACL)
648 valid |= OBD_MD_FLACL;
650 OBD_ALLOC_PTR(op_data);
652 GOTO(out_lock_cn_cb, err = -ENOMEM);
654 op_data->op_fid1 = sbi->ll_root_fid;
655 op_data->op_mode = 0;
656 op_data->op_valid = valid;
658 err = md_getattr(sbi->ll_md_exp, op_data, &request);
660 OBD_FREE_PTR(op_data);
662 CERROR("%s: md_getattr failed for root: rc = %d\n",
663 sbi->ll_md_exp->exp_obd->obd_name, err);
664 GOTO(out_lock_cn_cb, err);
667 err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
668 sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
670 CERROR("failed to understand root inode md: rc = %d\n", err);
671 ptlrpc_req_finished(request);
672 GOTO(out_lock_cn_cb, err);
675 LASSERT(fid_is_sane(&sbi->ll_root_fid));
676 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
677 sbi->ll_flags & LL_SBI_32BIT_API),
679 md_free_lustre_md(sbi->ll_md_exp, &lmd);
680 ptlrpc_req_finished(request);
684 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
686 CERROR("%s: bad ll_iget() for root: rc = %d\n",
687 sbi->ll_fsname, err);
691 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
692 if (sbi->ll_checksum_set) {
693 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
694 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
695 sizeof(checksum), &checksum, NULL);
697 CERROR("%s: Set checksum failed: rc = %d\n",
698 sbi->ll_dt_exp->exp_obd->obd_name, err);
704 sb->s_root = d_make_root(root);
705 if (sb->s_root == NULL) {
707 CERROR("%s: can't make root dentry: rc = %d\n",
708 sbi->ll_fsname, err);
712 sbi->ll_sdev_orig = sb->s_dev;
714 /* We set sb->s_dev equal on all lustre clients in order to support
715 * NFS export clustering. NFSD requires that the FSID be the same
717 /* s_dev is also used in lt_compare() to compare two fs, but that is
718 * only a node-local comparison. */
719 uuid = obd_get_uuid(sbi->ll_md_exp);
721 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
728 if (sbi->ll_dt_obd) {
729 err = sysfs_create_link(&sbi->ll_kset.kobj,
730 &sbi->ll_dt_obd->obd_kset.kobj,
731 sbi->ll_dt_obd->obd_type->typ_name);
733 CERROR("%s: could not register %s in llite: rc = %d\n",
734 dt, sbi->ll_fsname, err);
739 if (sbi->ll_md_obd) {
740 err = sysfs_create_link(&sbi->ll_kset.kobj,
741 &sbi->ll_md_obd->obd_kset.kobj,
742 sbi->ll_md_obd->obd_type->typ_name);
744 CERROR("%s: could not register %s in llite: rc = %d\n",
745 md, sbi->ll_fsname, err);
754 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
756 obd_disconnect(sbi->ll_dt_exp);
757 sbi->ll_dt_exp = NULL;
758 sbi->ll_dt_obd = NULL;
760 obd_fid_fini(sbi->ll_md_exp->exp_obd);
762 obd_disconnect(sbi->ll_md_exp);
763 sbi->ll_md_exp = NULL;
764 sbi->ll_md_obd = NULL;
773 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
777 size = sizeof(*lmmsize);
778 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
779 KEY_MAX_EASIZE, &size, lmmsize);
781 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
782 sbi->ll_dt_exp->exp_obd->obd_name, rc);
786 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
789 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
790 KEY_MAX_EASIZE, &size, lmmsize);
792 CERROR("Get max mdsize error rc %d\n", rc);
794 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
800 * Get the value of the default_easize parameter.
802 * \see client_obd::cl_default_mds_easize
804 * \param[in] sbi superblock info for this filesystem
805 * \param[out] lmmsize pointer to storage location for value
807 * \retval 0 on success
808 * \retval negative negated errno on failure
810 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
815 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
816 KEY_DEFAULT_EASIZE, &size, lmmsize);
818 CERROR("Get default mdsize error rc %d\n", rc);
824 * Set the default_easize parameter to the given value.
826 * \see client_obd::cl_default_mds_easize
828 * \param[in] sbi superblock info for this filesystem
829 * \param[in] lmmsize the size to set
831 * \retval 0 on success
832 * \retval negative negated errno on failure
834 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
838 if (lmmsize < sizeof(struct lov_mds_md) ||
839 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
842 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
843 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
844 sizeof(int), &lmmsize, NULL);
849 static void client_common_put_super(struct super_block *sb)
851 struct ll_sb_info *sbi = ll_s2sbi(sb);
856 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
857 obd_disconnect(sbi->ll_dt_exp);
858 sbi->ll_dt_exp = NULL;
860 ll_debugfs_unregister_super(sb);
862 obd_fid_fini(sbi->ll_md_exp->exp_obd);
863 obd_disconnect(sbi->ll_md_exp);
864 sbi->ll_md_exp = NULL;
869 void ll_kill_super(struct super_block *sb)
871 struct ll_sb_info *sbi;
875 if (!(sb->s_flags & SB_ACTIVE))
879 /* we need restore s_dev from changed for clustred NFS before put_super
880 * because new kernels have cached s_dev and change sb->s_dev in
881 * put_super not affected real removing devices */
883 sb->s_dev = sbi->ll_sdev_orig;
885 /* wait running statahead threads to quit */
886 while (atomic_read(&sbi->ll_sa_running) > 0)
887 schedule_timeout_uninterruptible(
888 cfs_time_seconds(1) >> 3);
894 static inline int ll_set_opt(const char *opt, char *data, int fl)
896 if (strncmp(opt, data, strlen(opt)) != 0)
902 /* non-client-specific mount options are parsed in lmd_parse */
903 static int ll_options(char *options, struct ll_sb_info *sbi)
906 char *s1 = options, *s2;
907 int *flags = &sbi->ll_flags;
913 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
916 CDEBUG(D_SUPER, "next opt=%s\n", s1);
917 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
922 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
924 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
927 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
929 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
932 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
937 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
942 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
947 tmp = ll_set_opt("context", s1, 1);
950 tmp = ll_set_opt("fscontext", s1, 1);
953 tmp = ll_set_opt("defcontext", s1, 1);
956 tmp = ll_set_opt("rootcontext", s1, 1);
959 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
964 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
970 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
973 sbi->ll_checksum_set = 1;
976 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
979 sbi->ll_checksum_set = 1;
982 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
987 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
992 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
997 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
1002 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
1007 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
1012 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
1017 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
1022 tmp = ll_set_opt("test_dummy_encryption", s1,
1023 LL_SBI_TEST_DUMMY_ENCRYPTION);
1025 #ifdef HAVE_LUSTRE_CRYPTO
1028 LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1032 tmp = ll_set_opt("noencrypt", s1, LL_SBI_ENCRYPT);
1034 #ifdef HAVE_LUSTRE_CRYPTO
1037 LCONSOLE_WARN("noencrypt mount option ignored: encryption not supported\n");
1041 tmp = ll_set_opt("foreign_symlink", s1, LL_SBI_FOREIGN_SYMLINK);
1043 int prefix_pos = sizeof("foreign_symlink=") - 1;
1044 int equal_pos = sizeof("foreign_symlink=") - 2;
1046 /* non-default prefix provided ? */
1047 if (strlen(s1) >= sizeof("foreign_symlink=") &&
1048 *(s1 + equal_pos) == '=') {
1049 char *old = sbi->ll_foreign_symlink_prefix;
1051 sbi->ll_foreign_symlink_prefix_size;
1053 /* path must be absolute */
1054 if (*(s1 + sizeof("foreign_symlink=")
1056 LCONSOLE_ERROR_MSG(0x152,
1057 "foreign prefix '%s' must be an absolute path\n",
1062 s2 = strchrnul(s1 + prefix_pos, ',');
1064 if (sbi->ll_foreign_symlink_prefix) {
1065 sbi->ll_foreign_symlink_prefix = NULL;
1066 sbi->ll_foreign_symlink_prefix_size = 0;
1068 /* alloc for path length and '\0' */
1069 OBD_ALLOC(sbi->ll_foreign_symlink_prefix,
1070 s2 - (s1 + prefix_pos) + 1);
1071 if (!sbi->ll_foreign_symlink_prefix) {
1072 /* restore previous */
1073 sbi->ll_foreign_symlink_prefix = old;
1074 sbi->ll_foreign_symlink_prefix_size =
1079 OBD_FREE(old, old_len);
1080 strncpy(sbi->ll_foreign_symlink_prefix,
1082 s2 - (s1 + prefix_pos));
1083 sbi->ll_foreign_symlink_prefix_size =
1084 s2 - (s1 + prefix_pos) + 1;
1086 LCONSOLE_ERROR_MSG(0x152,
1087 "invalid %s option\n", s1);
1089 /* enable foreign symlink support */
1093 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1099 s2 = strchr(s1, ',');
1107 void ll_lli_init(struct ll_inode_info *lli)
1109 lli->lli_inode_magic = LLI_INODE_MAGIC;
1111 spin_lock_init(&lli->lli_lock);
1112 lli->lli_posix_acl = NULL;
1113 /* Do not set lli_fid, it has been initialized already. */
1114 fid_zero(&lli->lli_pfid);
1115 lli->lli_mds_read_och = NULL;
1116 lli->lli_mds_write_och = NULL;
1117 lli->lli_mds_exec_och = NULL;
1118 lli->lli_open_fd_read_count = 0;
1119 lli->lli_open_fd_write_count = 0;
1120 lli->lli_open_fd_exec_count = 0;
1121 mutex_init(&lli->lli_och_mutex);
1122 spin_lock_init(&lli->lli_agl_lock);
1123 spin_lock_init(&lli->lli_layout_lock);
1124 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1125 lli->lli_clob = NULL;
1127 init_rwsem(&lli->lli_xattrs_list_rwsem);
1128 mutex_init(&lli->lli_xattrs_enq_lock);
1130 LASSERT(lli->lli_vfs_inode.i_mode != 0);
1131 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1132 lli->lli_opendir_key = NULL;
1133 lli->lli_sai = NULL;
1134 spin_lock_init(&lli->lli_sa_lock);
1135 lli->lli_opendir_pid = 0;
1136 lli->lli_sa_enabled = 0;
1137 init_rwsem(&lli->lli_lsm_sem);
1139 mutex_init(&lli->lli_size_mutex);
1140 mutex_init(&lli->lli_setattr_mutex);
1141 lli->lli_symlink_name = NULL;
1142 ll_trunc_sem_init(&lli->lli_trunc_sem);
1143 range_lock_tree_init(&lli->lli_write_tree);
1144 init_rwsem(&lli->lli_glimpse_sem);
1145 lli->lli_glimpse_time = ktime_set(0, 0);
1146 INIT_LIST_HEAD(&lli->lli_agl_list);
1147 lli->lli_agl_index = 0;
1148 lli->lli_async_rc = 0;
1149 spin_lock_init(&lli->lli_heat_lock);
1150 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1151 lli->lli_heat_flags = 0;
1152 mutex_init(&lli->lli_pcc_lock);
1153 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1154 lli->lli_pcc_inode = NULL;
1155 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1156 lli->lli_pcc_generation = 0;
1157 mutex_init(&lli->lli_group_mutex);
1158 lli->lli_group_users = 0;
1159 lli->lli_group_gid = 0;
1161 mutex_init(&lli->lli_layout_mutex);
1162 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1165 #define MAX_STRING_SIZE 128
1167 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1169 #define LSI_BDI_INITIALIZED 0x00400000
1171 #ifndef HAVE_BDI_CAP_MAP_COPY
1172 # define BDI_CAP_MAP_COPY 0
1175 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1177 struct lustre_sb_info *lsi = s2lsi(sb);
1178 char buf[MAX_STRING_SIZE];
1182 err = bdi_init(&lsi->lsi_bdi);
1186 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1187 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1188 lsi->lsi_bdi.name = "lustre";
1189 va_start(args, fmt);
1190 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1192 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1195 sb->s_bdi = &lsi->lsi_bdi;
1199 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1201 int ll_fill_super(struct super_block *sb)
1203 struct lustre_profile *lprof = NULL;
1204 struct lustre_sb_info *lsi = s2lsi(sb);
1205 struct ll_sb_info *sbi = NULL;
1206 char *dt = NULL, *md = NULL;
1207 char *profilenm = get_profile_name(sb);
1208 struct config_llog_instance *cfg;
1209 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1210 const int instlen = LUSTRE_MAXINSTANCE + 2;
1211 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1212 char name[MAX_STRING_SIZE];
1221 /* for ASLR, to map between cfg_instance and hashed ptr */
1222 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1223 profilenm, cfg_instance, sb);
1225 OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1229 GOTO(out_free_cfg, err = -ENOMEM);
1231 /* client additional sb info */
1232 lsi->lsi_llsbi = sbi = ll_init_sbi();
1234 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1236 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1238 GOTO(out_free_cfg, err);
1240 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1241 sb->s_d_op = &ll_d_ops;
1244 generate_random_uuid(uuid.b);
1245 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1247 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1250 len = strlen(profilenm);
1251 ptr = strrchr(profilenm, '-');
1252 if (ptr && (strcmp(ptr, "-client") == 0))
1255 if (len > LUSTRE_MAXFSNAME) {
1256 if (unlikely(len >= MAX_STRING_SIZE))
1257 len = MAX_STRING_SIZE - 1;
1258 strncpy(name, profilenm, len);
1260 err = -ENAMETOOLONG;
1261 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1262 name, LUSTRE_MAXFSNAME, err);
1263 GOTO(out_free_cfg, err);
1265 strncpy(sbi->ll_fsname, profilenm, len);
1266 sbi->ll_fsname[len] = '\0';
1269 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1270 profilenm, cfg_instance);
1272 err = super_setup_bdi_name(sb, "%s", name);
1274 GOTO(out_free_cfg, err);
1276 /* Call ll_debugfs_register_super() before lustre_process_log()
1277 * so that "llite.*.*" params can be processed correctly.
1279 err = ll_debugfs_register_super(sb, name);
1281 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1282 sbi->ll_fsname, err);
1286 /* The cfg_instance is a value unique to this super, in case some
1287 * joker tries to mount the same fs at two mount points.
1289 cfg->cfg_instance = cfg_instance;
1290 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1291 cfg->cfg_callback = class_config_llog_handler;
1292 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1293 /* set up client obds */
1294 err = lustre_process_log(sb, profilenm, cfg);
1296 GOTO(out_debugfs, err);
1298 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1299 lprof = class_get_profile(profilenm);
1300 if (lprof == NULL) {
1301 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1302 " read from the MGS. Does that filesystem "
1303 "exist?\n", profilenm);
1304 GOTO(out_debugfs, err = -EINVAL);
1306 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1307 lprof->lp_md, lprof->lp_dt);
1309 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1310 OBD_ALLOC(dt, dt_len);
1312 GOTO(out_profile, err = -ENOMEM);
1313 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1315 md_len = strlen(lprof->lp_md) + instlen + 2;
1316 OBD_ALLOC(md, md_len);
1318 GOTO(out_free_dt, err = -ENOMEM);
1319 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1321 /* connections, registrations, sb setup */
1322 err = client_common_fill_super(sb, md, dt);
1324 GOTO(out_free_md, err);
1326 sbi->ll_client_common_fill_super_succeeded = 1;
1330 OBD_FREE(md, md_len);
1333 OBD_FREE(dt, dt_len);
1336 class_put_profile(lprof);
1339 ll_debugfs_unregister_super(sb);
1346 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1347 LCONSOLE_WARN("Mounted %s\n", profilenm);
1349 } /* ll_fill_super */
1351 void ll_put_super(struct super_block *sb)
1353 struct config_llog_instance cfg, params_cfg;
1354 struct obd_device *obd;
1355 struct lustre_sb_info *lsi = s2lsi(sb);
1356 struct ll_sb_info *sbi = ll_s2sbi(sb);
1357 char *profilenm = get_profile_name(sb);
1358 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1360 int next, force = 1, rc = 0;
1364 GOTO(out_no_sbi, 0);
1366 /* Should replace instance_id with something better for ASLR */
1367 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1368 profilenm, cfg_instance, sb);
1370 cfg.cfg_instance = cfg_instance;
1371 lustre_end_log(sb, profilenm, &cfg);
1373 params_cfg.cfg_instance = cfg_instance;
1374 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1376 if (sbi->ll_md_exp) {
1377 obd = class_exp2obd(sbi->ll_md_exp);
1379 force = obd->obd_force;
1382 /* Wait for unstable pages to be committed to stable storage */
1384 rc = l_wait_event_abortable(
1385 sbi->ll_cache->ccc_unstable_waitq,
1386 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1389 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1390 if (force == 0 && rc != -ERESTARTSYS)
1391 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1393 /* We need to set force before the lov_disconnect in
1394 * lustre_common_put_super, since l_d cleans up osc's as well.
1398 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1400 obd->obd_force = force;
1404 if (sbi->ll_client_common_fill_super_succeeded) {
1405 /* Only if client_common_fill_super succeeded */
1406 client_common_put_super(sb);
1410 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1411 class_manual_cleanup(obd);
1413 if (sbi->ll_flags & LL_SBI_VERBOSE)
1414 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1417 class_del_profile(profilenm);
1419 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1420 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1421 bdi_destroy(&lsi->lsi_bdi);
1422 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1427 lsi->lsi_llsbi = NULL;
1429 lustre_common_put_super(sb);
1431 cl_env_cache_purge(~0);
1434 } /* client_put_super */
1436 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1438 struct inode *inode = NULL;
1440 /* NOTE: we depend on atomic igrab() -bzzz */
1441 lock_res_and_lock(lock);
1442 if (lock->l_resource->lr_lvb_inode) {
1443 struct ll_inode_info * lli;
1444 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1445 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1446 inode = igrab(lock->l_resource->lr_lvb_inode);
1448 inode = lock->l_resource->lr_lvb_inode;
1449 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1450 D_WARNING, lock, "lr_lvb_inode %p is "
1451 "bogus: magic %08x",
1452 lock->l_resource->lr_lvb_inode,
1453 lli->lli_inode_magic);
1457 unlock_res_and_lock(lock);
1461 void ll_dir_clear_lsm_md(struct inode *inode)
1463 struct ll_inode_info *lli = ll_i2info(inode);
1465 LASSERT(S_ISDIR(inode->i_mode));
1467 if (lli->lli_lsm_md) {
1468 lmv_free_memmd(lli->lli_lsm_md);
1469 lli->lli_lsm_md = NULL;
1472 if (lli->lli_default_lsm_md) {
1473 lmv_free_memmd(lli->lli_default_lsm_md);
1474 lli->lli_default_lsm_md = NULL;
1478 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1479 const struct lu_fid *fid,
1480 struct lustre_md *md)
1482 struct ll_sb_info *sbi = ll_s2sbi(sb);
1483 struct mdt_body *body = md->body;
1484 struct inode *inode;
1488 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1489 inode = iget_locked(sb, ino);
1490 if (inode == NULL) {
1491 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1492 sbi->ll_fsname, PFID(fid));
1493 RETURN(ERR_PTR(-ENOENT));
1496 if (inode->i_state & I_NEW) {
1497 struct ll_inode_info *lli = ll_i2info(inode);
1498 struct lmv_stripe_md *lsm = md->lmv;
1500 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1501 (body->mbo_mode & S_IFMT);
1502 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1505 inode->i_mtime.tv_sec = 0;
1506 inode->i_atime.tv_sec = 0;
1507 inode->i_ctime.tv_sec = 0;
1510 #ifdef HAVE_BACKING_DEV_INFO
1511 /* initializing backing dev info. */
1512 inode->i_mapping->backing_dev_info =
1513 &s2lsi(inode->i_sb)->lsi_bdi;
1515 inode->i_op = &ll_dir_inode_operations;
1516 inode->i_fop = &ll_dir_operations;
1517 lli->lli_fid = *fid;
1520 LASSERT(lsm != NULL);
1521 /* master object FID */
1522 lli->lli_pfid = body->mbo_fid1;
1523 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1524 lli, PFID(fid), PFID(&lli->lli_pfid));
1525 unlock_new_inode(inode);
1531 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1534 struct lmv_stripe_md *lsm = md->lmv;
1535 struct ll_inode_info *lli = ll_i2info(inode);
1538 LASSERT(lsm != NULL);
1540 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1541 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1542 lsm_md_dump(D_INODE, lsm);
1544 if (!lmv_dir_striped(lsm))
1547 /* XXX sigh, this lsm_root initialization should be in
1548 * LMV layer, but it needs ll_iget right now, so we
1549 * put this here right now. */
1550 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1551 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1552 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1554 if (!fid_is_sane(fid))
1557 /* Unfortunately ll_iget will call ll_update_inode,
1558 * where the initialization of slave inode is slightly
1559 * different, so it reset lsm_md to NULL to avoid
1560 * initializing lsm for slave inode. */
1561 lsm->lsm_md_oinfo[i].lmo_root =
1562 ll_iget_anon_dir(inode->i_sb, fid, md);
1563 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1564 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1566 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1568 iput(lsm->lsm_md_oinfo[i].lmo_root);
1569 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1575 lli->lli_lsm_md = lsm;
1580 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1582 struct ll_inode_info *lli = ll_i2info(inode);
1584 if (!md->default_lmv) {
1585 /* clear default lsm */
1586 if (lli->lli_default_lsm_md) {
1587 down_write(&lli->lli_lsm_sem);
1588 if (lli->lli_default_lsm_md) {
1589 lmv_free_memmd(lli->lli_default_lsm_md);
1590 lli->lli_default_lsm_md = NULL;
1592 up_write(&lli->lli_lsm_sem);
1597 if (lli->lli_default_lsm_md) {
1598 /* do nonthing if default lsm isn't changed */
1599 down_read(&lli->lli_lsm_sem);
1600 if (lli->lli_default_lsm_md &&
1601 lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1602 up_read(&lli->lli_lsm_sem);
1605 up_read(&lli->lli_lsm_sem);
1608 down_write(&lli->lli_lsm_sem);
1609 if (lli->lli_default_lsm_md)
1610 lmv_free_memmd(lli->lli_default_lsm_md);
1611 lli->lli_default_lsm_md = md->default_lmv;
1612 lsm_md_dump(D_INODE, md->default_lmv);
1613 md->default_lmv = NULL;
1614 up_write(&lli->lli_lsm_sem);
1617 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1619 struct ll_inode_info *lli = ll_i2info(inode);
1620 struct lmv_stripe_md *lsm = md->lmv;
1621 struct cl_attr *attr;
1626 LASSERT(S_ISDIR(inode->i_mode));
1627 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1628 PFID(ll_inode2fid(inode)));
1630 /* update default LMV */
1631 if (md->default_lmv)
1632 ll_update_default_lsm_md(inode, md);
1635 * no striped information from request, lustre_md from req does not
1636 * include stripeEA, see ll_md_setattr()
1642 * normally dir layout doesn't change, only take read lock to check
1643 * that to avoid blocking other MD operations.
1645 down_read(&lli->lli_lsm_sem);
1647 /* some current lookup initialized lsm, and unchanged */
1648 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1649 GOTO(unlock, rc = 0);
1651 /* if dir layout doesn't match, check whether version is increased,
1652 * which means layout is changed, this happens in dir split/merge and
1655 * foreign LMV should not change.
1657 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1658 lsm->lsm_md_layout_version <=
1659 lli->lli_lsm_md->lsm_md_layout_version) {
1660 CERROR("%s: "DFID" dir layout mismatch:\n",
1661 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1662 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1663 lsm_md_dump(D_ERROR, lsm);
1664 GOTO(unlock, rc = -EINVAL);
1667 up_read(&lli->lli_lsm_sem);
1668 down_write(&lli->lli_lsm_sem);
1669 /* clear existing lsm */
1670 if (lli->lli_lsm_md) {
1671 lmv_free_memmd(lli->lli_lsm_md);
1672 lli->lli_lsm_md = NULL;
1675 rc = ll_init_lsm_md(inode, md);
1676 up_write(&lli->lli_lsm_sem);
1681 /* set md->lmv to NULL, so the following free lustre_md will not free
1686 /* md_merge_attr() may take long, since lsm is already set, switch to
1689 down_read(&lli->lli_lsm_sem);
1691 if (!lmv_dir_striped(lli->lli_lsm_md))
1692 GOTO(unlock, rc = 0);
1694 OBD_ALLOC_PTR(attr);
1696 GOTO(unlock, rc = -ENOMEM);
1698 /* validate the lsm */
1699 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1700 ll_md_blocking_ast);
1702 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1703 md->body->mbo_nlink = attr->cat_nlink;
1704 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1705 md->body->mbo_size = attr->cat_size;
1706 if (md->body->mbo_valid & OBD_MD_FLATIME)
1707 md->body->mbo_atime = attr->cat_atime;
1708 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1709 md->body->mbo_ctime = attr->cat_ctime;
1710 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1711 md->body->mbo_mtime = attr->cat_mtime;
1717 up_read(&lli->lli_lsm_sem);
1722 void ll_clear_inode(struct inode *inode)
1724 struct ll_inode_info *lli = ll_i2info(inode);
1725 struct ll_sb_info *sbi = ll_i2sbi(inode);
1729 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1730 PFID(ll_inode2fid(inode)), inode);
1732 if (S_ISDIR(inode->i_mode)) {
1733 /* these should have been cleared in ll_file_release */
1734 LASSERT(lli->lli_opendir_key == NULL);
1735 LASSERT(lli->lli_sai == NULL);
1736 LASSERT(lli->lli_opendir_pid == 0);
1738 pcc_inode_free(inode);
1741 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1743 LASSERT(!lli->lli_open_fd_write_count);
1744 LASSERT(!lli->lli_open_fd_read_count);
1745 LASSERT(!lli->lli_open_fd_exec_count);
1747 if (lli->lli_mds_write_och)
1748 ll_md_real_close(inode, FMODE_WRITE);
1749 if (lli->lli_mds_exec_och)
1750 ll_md_real_close(inode, FMODE_EXEC);
1751 if (lli->lli_mds_read_och)
1752 ll_md_real_close(inode, FMODE_READ);
1754 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1755 OBD_FREE(lli->lli_symlink_name,
1756 strlen(lli->lli_symlink_name) + 1);
1757 lli->lli_symlink_name = NULL;
1760 ll_xattr_cache_destroy(inode);
1762 forget_all_cached_acls(inode);
1764 lli->lli_inode_magic = LLI_INODE_DEAD;
1766 if (S_ISDIR(inode->i_mode))
1767 ll_dir_clear_lsm_md(inode);
1768 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1769 LASSERT(list_empty(&lli->lli_agl_list));
1772 * XXX This has to be done before lsm is freed below, because
1773 * cl_object still uses inode lsm.
1775 cl_inode_fini(inode);
1777 llcrypt_put_encryption_info(inode);
1782 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1784 struct lustre_md md;
1785 struct inode *inode = dentry->d_inode;
1786 struct ll_sb_info *sbi = ll_i2sbi(inode);
1787 struct ptlrpc_request *request = NULL;
1791 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1792 LUSTRE_OPC_ANY, NULL);
1793 if (IS_ERR(op_data))
1794 RETURN(PTR_ERR(op_data));
1796 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1798 ptlrpc_req_finished(request);
1799 if (rc == -ENOENT) {
1801 /* Unlinked special device node? Or just a race?
1802 * Pretend we done everything. */
1803 if (!S_ISREG(inode->i_mode) &&
1804 !S_ISDIR(inode->i_mode)) {
1805 ia_valid = op_data->op_attr.ia_valid;
1806 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1807 rc = simple_setattr(dentry, &op_data->op_attr);
1808 op_data->op_attr.ia_valid = ia_valid;
1810 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1811 CERROR("md_setattr fails: rc = %d\n", rc);
1816 rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1817 sbi->ll_md_exp, &md);
1819 ptlrpc_req_finished(request);
1823 ia_valid = op_data->op_attr.ia_valid;
1824 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1825 * cache is not cleared yet. */
1826 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1827 if (S_ISREG(inode->i_mode))
1829 rc = simple_setattr(dentry, &op_data->op_attr);
1830 if (S_ISREG(inode->i_mode))
1831 inode_unlock(inode);
1832 op_data->op_attr.ia_valid = ia_valid;
1834 rc = ll_update_inode(inode, &md);
1835 ptlrpc_req_finished(request);
1841 * Zero portion of page that is part of @inode.
1842 * This implies, if necessary:
1843 * - taking cl_lock on range corresponding to concerned page
1844 * - grabbing vm page
1845 * - associating cl_page
1846 * - proceeding to clio read
1847 * - zeroing range in page
1848 * - proceeding to cl_page flush
1849 * - releasing cl_lock
1851 * \param[in] inode inode
1852 * \param[in] index page index
1853 * \param[in] offset offset in page to start zero from
1854 * \param[in] len len to zero
1856 * \retval 0 on success
1857 * \retval negative errno on failure
1859 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1862 struct ll_inode_info *lli = ll_i2info(inode);
1863 struct cl_object *clob = lli->lli_clob;
1865 struct lu_env *env = NULL;
1866 struct cl_io *io = NULL;
1867 struct cl_page *clpage = NULL;
1868 struct page *vmpage = NULL;
1869 unsigned from = index << PAGE_SHIFT;
1870 struct cl_lock *lock = NULL;
1871 struct cl_lock_descr *descr = NULL;
1872 struct cl_2queue *queue = NULL;
1873 struct cl_sync_io *anchor = NULL;
1874 bool holdinglock = false;
1875 bool lockedbymyself = true;
1880 env = cl_env_get(&refcheck);
1882 RETURN(PTR_ERR(env));
1884 io = vvp_env_thread_io(env);
1886 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1890 lock = vvp_env_lock(env);
1891 descr = &lock->cll_descr;
1892 descr->cld_obj = io->ci_obj;
1893 descr->cld_start = cl_index(io->ci_obj, from);
1894 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1895 descr->cld_mode = CLM_WRITE;
1896 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1898 /* request lock for page */
1899 rc = cl_lock_request(env, io, lock);
1900 /* -ECANCELED indicates a matching lock with a different extent
1901 * was already present, and -EEXIST indicates a matching lock
1902 * on exactly the same extent was already present.
1903 * In both cases it means we are covered.
1905 if (rc == -ECANCELED || rc == -EEXIST)
1913 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1915 GOTO(rellock, rc = -EOPNOTSUPP);
1917 if (!PageDirty(vmpage)) {
1918 /* associate cl_page */
1919 clpage = cl_page_find(env, clob, vmpage->index,
1920 vmpage, CPT_CACHEABLE);
1922 GOTO(pagefini, rc = PTR_ERR(clpage));
1924 cl_page_assume(env, io, clpage);
1927 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1928 !PageWriteback(vmpage)) {
1930 /* set PagePrivate2 to detect special case of empty page
1931 * in osc_brw_fini_request()
1933 SetPagePrivate2(vmpage);
1934 rc = ll_io_read_page(env, io, clpage, NULL);
1935 if (!PagePrivate2(vmpage))
1936 /* PagePrivate2 was cleared in osc_brw_fini_request()
1937 * meaning we read an empty page. In this case, in order
1938 * to avoid allocating unnecessary block in truncated
1939 * file, we must not zero and write as below. Subsequent
1940 * server-side truncate will handle things correctly.
1942 GOTO(clpfini, rc = 0);
1943 ClearPagePrivate2(vmpage);
1946 lockedbymyself = trylock_page(vmpage);
1947 cl_page_assume(env, io, clpage);
1950 /* zero range in page */
1951 zero_user(vmpage, offset, len);
1953 if (holdinglock && clpage) {
1954 /* explicitly write newly modified page */
1955 queue = &io->ci_queue;
1956 cl_2queue_init(queue);
1957 anchor = &vvp_env_info(env)->vti_anchor;
1958 cl_sync_io_init(anchor, 1);
1959 clpage->cp_sync_io = anchor;
1960 cl_2queue_add(queue, clpage);
1961 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1963 GOTO(queuefini1, rc);
1964 rc = cl_sync_io_wait(env, anchor, 0);
1966 GOTO(queuefini2, rc);
1967 cl_page_assume(env, io, clpage);
1970 cl_2queue_discard(env, io, queue);
1972 cl_2queue_disown(env, io, queue);
1973 cl_2queue_fini(env, queue);
1978 cl_page_put(env, clpage);
1980 if (lockedbymyself) {
1981 unlock_page(vmpage);
1986 cl_lock_release(env, lock);
1988 cl_io_fini(env, io);
1991 cl_env_put(env, &refcheck);
1996 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1997 * object(s) determine the file size and mtime. Otherwise, the MDS will
1998 * keep these values until such a time that objects are allocated for it.
1999 * We do the MDS operations first, as it is checking permissions for us.
2000 * We don't to the MDS RPC if there is nothing that we want to store there,
2001 * otherwise there is no harm in updating mtime/atime on the MDS if we are
2002 * going to do an RPC anyways.
2004 * If we are doing a truncate, we will send the mtime and ctime updates
2005 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2006 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2009 * In case of HSMimport, we only set attr on MDS.
2011 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2012 enum op_xvalid xvalid, bool hsm_import)
2014 struct inode *inode = dentry->d_inode;
2015 struct ll_inode_info *lli = ll_i2info(inode);
2016 struct md_op_data *op_data = NULL;
2017 ktime_t kstart = ktime_get();
2022 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2023 "valid %x, hsm_import %d\n",
2024 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2025 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2028 if (attr->ia_valid & ATTR_SIZE) {
2029 /* Check new size against VFS/VM file size limit and rlimit */
2030 rc = inode_newsize_ok(inode, attr->ia_size);
2034 /* The maximum Lustre file size is variable, based on the
2035 * OST maximum object size and number of stripes. This
2036 * needs another check in addition to the VFS check above. */
2037 if (attr->ia_size > ll_file_maxbytes(inode)) {
2038 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2039 PFID(&lli->lli_fid), attr->ia_size,
2040 ll_file_maxbytes(inode));
2044 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2047 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2048 if (attr->ia_valid & TIMES_SET_FLAGS) {
2049 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2050 !capable(CAP_FOWNER))
2054 /* We mark all of the fields "set" so MDS/OST does not re-set them */
2055 if (!(xvalid & OP_XVALID_CTIME_SET) &&
2056 (attr->ia_valid & ATTR_CTIME)) {
2057 attr->ia_ctime = current_time(inode);
2058 xvalid |= OP_XVALID_CTIME_SET;
2060 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2061 (attr->ia_valid & ATTR_ATIME)) {
2062 attr->ia_atime = current_time(inode);
2063 attr->ia_valid |= ATTR_ATIME_SET;
2065 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2066 (attr->ia_valid & ATTR_MTIME)) {
2067 attr->ia_mtime = current_time(inode);
2068 attr->ia_valid |= ATTR_MTIME_SET;
2071 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2072 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2073 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2074 ktime_get_real_seconds());
2076 if (S_ISREG(inode->i_mode))
2077 inode_unlock(inode);
2079 /* We always do an MDS RPC, even if we're only changing the size;
2080 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2082 OBD_ALLOC_PTR(op_data);
2083 if (op_data == NULL)
2084 GOTO(out, rc = -ENOMEM);
2086 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2087 /* If we are changing file size, file content is
2088 * modified, flag it.
2090 xvalid |= OP_XVALID_OWNEROVERRIDE;
2091 op_data->op_bias |= MDS_DATA_MODIFIED;
2092 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2095 if (attr->ia_valid & ATTR_FILE) {
2096 struct ll_file_data *fd = attr->ia_file->private_data;
2098 if (fd->fd_lease_och)
2099 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2102 op_data->op_attr = *attr;
2103 op_data->op_xvalid = xvalid;
2105 rc = ll_md_setattr(dentry, op_data);
2109 if (!S_ISREG(inode->i_mode) || hsm_import)
2112 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2113 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2114 xvalid & OP_XVALID_CTIME_SET) {
2115 bool cached = false;
2117 rc = pcc_inode_setattr(inode, attr, &cached);
2120 CERROR("%s: PCC inode "DFID" setattr failed: "
2122 ll_i2sbi(inode)->ll_fsname,
2123 PFID(&lli->lli_fid), rc);
2127 unsigned int flags = 0;
2129 /* For truncate and utimes sending attributes to OSTs,
2130 * setting mtime/atime to the past will be performed
2131 * under PW [0:EOF] extent lock (new_size:EOF for
2132 * truncate). It may seem excessive to send mtime/atime
2133 * updates to OSTs when not setting times to past, but
2134 * it is necessary due to possible time
2135 * de-synchronization between MDT inode and OST objects
2137 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2138 attr->ia_valid & ATTR_SIZE) {
2139 xvalid |= OP_XVALID_FLAGS;
2140 flags = LUSTRE_ENCRYPT_FL;
2141 /* Call to ll_io_zero_page is not necessary if
2142 * truncating on PAGE_SIZE boundary, because
2143 * whole pages will be wiped.
2144 * In case of Direct IO, all we need is to set
2147 if (attr->ia_size & ~PAGE_MASK &&
2148 !(attr->ia_valid & ATTR_FILE &&
2149 attr->ia_file->f_flags & O_DIRECT)) {
2151 attr->ia_size & (PAGE_SIZE - 1);
2153 rc = ll_io_zero_page(inode,
2154 attr->ia_size >> PAGE_SHIFT,
2155 offset, PAGE_SIZE - offset);
2160 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2164 /* If the file was restored, it needs to set dirty flag.
2166 * We've already sent MDS_DATA_MODIFIED flag in
2167 * ll_md_setattr() for truncate. However, the MDT refuses to
2168 * set the HS_DIRTY flag on released files, so we have to set
2169 * it again if the file has been restored. Please check how
2170 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2172 * Please notice that if the file is not released, the previous
2173 * MDS_DATA_MODIFIED has taken effect and usually
2174 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2175 * This way we can save an RPC for common open + trunc
2177 if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2178 struct hsm_state_set hss = {
2179 .hss_valid = HSS_SETMASK,
2180 .hss_setmask = HS_DIRTY,
2184 rc2 = ll_hsm_state_set(inode, &hss);
2185 /* truncate and write can happen at the same time, so that
2186 * the file can be set modified even though the file is not
2187 * restored from released state, and ll_hsm_state_set() is
2188 * not applicable for the file, and rc2 < 0 is normal in this
2191 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2192 PFID(ll_inode2fid(inode)), rc2);
2197 if (op_data != NULL)
2198 ll_finish_md_op_data(op_data);
2200 if (S_ISREG(inode->i_mode)) {
2202 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2203 inode_dio_wait(inode);
2204 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2205 * flag. ll_update_inode (called from ll_md_setattr), clears
2206 * inode flags, so there is a gap where S_NOSEC is not set.
2207 * This can cause a writer to take the i_mutex unnecessarily,
2208 * but this is safe to do and should be rare. */
2209 inode_has_no_xattr(inode);
2213 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2214 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2215 ktime_us_delta(ktime_get(), kstart));
2220 int ll_setattr(struct dentry *de, struct iattr *attr)
2222 int mode = de->d_inode->i_mode;
2223 enum op_xvalid xvalid = 0;
2226 rc = llcrypt_prepare_setattr(de, attr);
2230 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2231 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2232 xvalid |= OP_XVALID_OWNEROVERRIDE;
2234 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2235 (ATTR_SIZE|ATTR_MODE)) &&
2236 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2237 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2238 !(attr->ia_mode & S_ISGID))))
2239 attr->ia_valid |= ATTR_FORCE;
2241 if ((attr->ia_valid & ATTR_MODE) &&
2243 !(attr->ia_mode & S_ISUID) &&
2244 !(attr->ia_valid & ATTR_KILL_SUID))
2245 attr->ia_valid |= ATTR_KILL_SUID;
2247 if ((attr->ia_valid & ATTR_MODE) &&
2248 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2249 !(attr->ia_mode & S_ISGID) &&
2250 !(attr->ia_valid & ATTR_KILL_SGID))
2251 attr->ia_valid |= ATTR_KILL_SGID;
2253 return ll_setattr_raw(de, attr, xvalid, false);
2256 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2259 struct obd_statfs obd_osfs = { 0 };
2264 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2266 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2267 flags |= OBD_STATFS_NODELAY;
2269 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2273 osfs->os_type = LL_SUPER_MAGIC;
2275 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2276 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2278 if (osfs->os_state & OS_STATFS_SUM)
2281 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2282 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2285 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2286 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2289 osfs->os_bsize = obd_osfs.os_bsize;
2290 osfs->os_blocks = obd_osfs.os_blocks;
2291 osfs->os_bfree = obd_osfs.os_bfree;
2292 osfs->os_bavail = obd_osfs.os_bavail;
2294 /* If we have _some_ OSTs, but don't have as many free objects on the
2295 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2296 * to compensate, so that the "inodes in use" number is correct.
2297 * This should be kept in sync with lod_statfs() behaviour.
2299 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2300 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2302 osfs->os_ffree = obd_osfs.os_ffree;
2309 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2311 struct if_quotactl qctl = {
2312 .qc_cmd = LUSTRE_Q_GETQUOTA,
2313 .qc_type = PRJQUOTA,
2314 .qc_valid = QC_GENERAL,
2316 u64 limit, curblock;
2319 qctl.qc_id = ll_i2info(inode)->lli_projid;
2320 ret = quotactl_ioctl(ll_i2sbi(inode), &qctl);
2322 /* ignore errors if project ID does not have
2323 * a quota limit or feature unsupported.
2325 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2330 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2331 qctl.qc_dqblk.dqb_bsoftlimit :
2332 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2333 if (limit && sfs->f_blocks > limit) {
2334 curblock = (qctl.qc_dqblk.dqb_curspace +
2335 sfs->f_bsize - 1) / sfs->f_bsize;
2336 sfs->f_blocks = limit;
2337 sfs->f_bfree = sfs->f_bavail =
2338 (sfs->f_blocks > curblock) ?
2339 (sfs->f_blocks - curblock) : 0;
2342 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2343 qctl.qc_dqblk.dqb_isoftlimit :
2344 qctl.qc_dqblk.dqb_ihardlimit;
2345 if (limit && sfs->f_files > limit) {
2346 sfs->f_files = limit;
2347 sfs->f_ffree = (sfs->f_files >
2348 qctl.qc_dqblk.dqb_curinodes) ?
2349 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2355 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2357 struct super_block *sb = de->d_sb;
2358 struct obd_statfs osfs;
2359 __u64 fsid = huge_encode_dev(sb->s_dev);
2360 ktime_t kstart = ktime_get();
2363 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2365 /* Some amount of caching on the client is allowed */
2366 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2370 statfs_unpack(sfs, &osfs);
2372 /* We need to downshift for all 32-bit kernels, because we can't
2373 * tell if the kernel is being called via sys_statfs64() or not.
2374 * Stop before overflowing f_bsize - in which case it is better
2375 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2376 if (sizeof(long) < 8) {
2377 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2380 osfs.os_blocks >>= 1;
2381 osfs.os_bfree >>= 1;
2382 osfs.os_bavail >>= 1;
2386 sfs->f_blocks = osfs.os_blocks;
2387 sfs->f_bfree = osfs.os_bfree;
2388 sfs->f_bavail = osfs.os_bavail;
2389 sfs->f_fsid.val[0] = (__u32)fsid;
2390 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2391 if (ll_i2info(de->d_inode)->lli_projid)
2392 return ll_statfs_project(de->d_inode, sfs);
2394 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2395 ktime_us_delta(ktime_get(), kstart));
2400 void ll_inode_size_lock(struct inode *inode)
2402 struct ll_inode_info *lli;
2404 LASSERT(!S_ISDIR(inode->i_mode));
2406 lli = ll_i2info(inode);
2407 mutex_lock(&lli->lli_size_mutex);
2410 void ll_inode_size_unlock(struct inode *inode)
2412 struct ll_inode_info *lli;
2414 lli = ll_i2info(inode);
2415 mutex_unlock(&lli->lli_size_mutex);
2418 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2420 /* do not clear encryption flag */
2421 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2422 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2423 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2424 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2426 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2429 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2431 struct ll_inode_info *lli = ll_i2info(inode);
2432 struct mdt_body *body = md->body;
2433 struct ll_sb_info *sbi = ll_i2sbi(inode);
2436 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2437 rc = cl_file_inode_init(inode, md);
2442 if (S_ISDIR(inode->i_mode)) {
2443 rc = ll_update_lsm_md(inode, md);
2448 if (body->mbo_valid & OBD_MD_FLACL)
2449 lli_replace_acl(lli, md);
2451 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2452 sbi->ll_flags & LL_SBI_32BIT_API);
2453 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2455 if (body->mbo_valid & OBD_MD_FLATIME) {
2456 if (body->mbo_atime > inode->i_atime.tv_sec)
2457 inode->i_atime.tv_sec = body->mbo_atime;
2458 lli->lli_atime = body->mbo_atime;
2461 if (body->mbo_valid & OBD_MD_FLMTIME) {
2462 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2464 "setting ino %lu mtime from %lld to %llu\n",
2465 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2467 inode->i_mtime.tv_sec = body->mbo_mtime;
2469 lli->lli_mtime = body->mbo_mtime;
2472 if (body->mbo_valid & OBD_MD_FLCTIME) {
2473 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2474 inode->i_ctime.tv_sec = body->mbo_ctime;
2475 lli->lli_ctime = body->mbo_ctime;
2478 if (body->mbo_valid & OBD_MD_FLBTIME)
2479 lli->lli_btime = body->mbo_btime;
2481 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2482 if (body->mbo_valid & OBD_MD_FLFLAGS)
2483 ll_update_inode_flags(inode, body->mbo_flags);
2484 if (body->mbo_valid & OBD_MD_FLMODE)
2485 inode->i_mode = (inode->i_mode & S_IFMT) |
2486 (body->mbo_mode & ~S_IFMT);
2488 if (body->mbo_valid & OBD_MD_FLTYPE)
2489 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2490 (body->mbo_mode & S_IFMT);
2492 LASSERT(inode->i_mode != 0);
2493 if (body->mbo_valid & OBD_MD_FLUID)
2494 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2495 if (body->mbo_valid & OBD_MD_FLGID)
2496 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2497 if (body->mbo_valid & OBD_MD_FLPROJID)
2498 lli->lli_projid = body->mbo_projid;
2499 if (body->mbo_valid & OBD_MD_FLNLINK)
2500 set_nlink(inode, body->mbo_nlink);
2501 if (body->mbo_valid & OBD_MD_FLRDEV)
2502 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2504 if (body->mbo_valid & OBD_MD_FLID) {
2505 /* FID shouldn't be changed! */
2506 if (fid_is_sane(&lli->lli_fid)) {
2507 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2508 "Trying to change FID "DFID
2509 " to the "DFID", inode "DFID"(%p)\n",
2510 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2511 PFID(ll_inode2fid(inode)), inode);
2513 lli->lli_fid = body->mbo_fid1;
2517 LASSERT(fid_seq(&lli->lli_fid) != 0);
2519 lli->lli_attr_valid = body->mbo_valid;
2520 if (body->mbo_valid & OBD_MD_FLSIZE) {
2521 i_size_write(inode, body->mbo_size);
2523 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2524 PFID(ll_inode2fid(inode)),
2525 (unsigned long long)body->mbo_size);
2527 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2528 inode->i_blocks = body->mbo_blocks;
2530 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2531 lli->lli_lazysize = body->mbo_size;
2532 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2533 lli->lli_lazyblocks = body->mbo_blocks;
2536 if (body->mbo_valid & OBD_MD_TSTATE) {
2537 /* Set LLIF_FILE_RESTORING if restore ongoing and
2538 * clear it when done to ensure to start again
2539 * glimpsing updated attrs
2541 if (body->mbo_t_state & MS_RESTORE)
2542 set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2544 clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2550 void ll_truncate_inode_pages_final(struct inode *inode)
2552 struct address_space *mapping = &inode->i_data;
2553 unsigned long nrpages;
2554 unsigned long flags;
2556 truncate_inode_pages_final(mapping);
2558 /* Workaround for LU-118: Note nrpages may not be totally updated when
2559 * truncate_inode_pages() returns, as there can be a page in the process
2560 * of deletion (inside __delete_from_page_cache()) in the specified
2561 * range. Thus mapping->nrpages can be non-zero when this function
2562 * returns even after truncation of the whole mapping. Only do this if
2563 * npages isn't already zero.
2565 nrpages = mapping->nrpages;
2567 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2568 nrpages = mapping->nrpages;
2569 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2570 } /* Workaround end */
2572 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2573 "see https://jira.whamcloud.com/browse/LU-118\n",
2574 ll_i2sbi(inode)->ll_fsname,
2575 PFID(ll_inode2fid(inode)), inode, nrpages);
2578 int ll_read_inode2(struct inode *inode, void *opaque)
2580 struct lustre_md *md = opaque;
2581 struct ll_inode_info *lli = ll_i2info(inode);
2585 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2586 PFID(&lli->lli_fid), inode);
2588 /* Core attributes from the MDS first. This is a new inode, and
2589 * the VFS doesn't zero times in the core inode so we have to do
2590 * it ourselves. They will be overwritten by either MDS or OST
2591 * attributes - we just need to make sure they aren't newer.
2593 inode->i_mtime.tv_sec = 0;
2594 inode->i_atime.tv_sec = 0;
2595 inode->i_ctime.tv_sec = 0;
2597 rc = ll_update_inode(inode, md);
2601 /* OIDEBUG(inode); */
2603 #ifdef HAVE_BACKING_DEV_INFO
2604 /* initializing backing dev info. */
2605 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2607 if (S_ISREG(inode->i_mode)) {
2608 struct ll_sb_info *sbi = ll_i2sbi(inode);
2609 inode->i_op = &ll_file_inode_operations;
2610 inode->i_fop = sbi->ll_fop;
2611 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2613 } else if (S_ISDIR(inode->i_mode)) {
2614 inode->i_op = &ll_dir_inode_operations;
2615 inode->i_fop = &ll_dir_operations;
2617 } else if (S_ISLNK(inode->i_mode)) {
2618 inode->i_op = &ll_fast_symlink_inode_operations;
2621 inode->i_op = &ll_special_inode_operations;
2623 init_special_inode(inode, inode->i_mode,
2632 void ll_delete_inode(struct inode *inode)
2634 struct ll_inode_info *lli = ll_i2info(inode);
2637 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2638 /* It is last chance to write out dirty pages,
2639 * otherwise we may lose data while umount.
2641 * If i_nlink is 0 then just discard data. This is safe because
2642 * local inode gets i_nlink 0 from server only for the last
2643 * unlink, so that file is not opened somewhere else
2645 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2646 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2649 ll_truncate_inode_pages_final(inode);
2650 ll_clear_inode(inode);
2656 int ll_iocontrol(struct inode *inode, struct file *file,
2657 unsigned int cmd, unsigned long arg)
2659 struct ll_sb_info *sbi = ll_i2sbi(inode);
2660 struct ptlrpc_request *req = NULL;
2665 case FS_IOC_GETFLAGS: {
2666 struct mdt_body *body;
2667 struct md_op_data *op_data;
2669 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2670 0, 0, LUSTRE_OPC_ANY,
2672 if (IS_ERR(op_data))
2673 RETURN(PTR_ERR(op_data));
2675 op_data->op_valid = OBD_MD_FLFLAGS;
2676 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2677 ll_finish_md_op_data(op_data);
2679 CERROR("%s: failure inode "DFID": rc = %d\n",
2680 sbi->ll_md_exp->exp_obd->obd_name,
2681 PFID(ll_inode2fid(inode)), rc);
2685 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2687 flags = body->mbo_flags;
2689 ptlrpc_req_finished(req);
2691 RETURN(put_user(flags, (int __user *)arg));
2693 case FS_IOC_SETFLAGS: {
2695 struct md_op_data *op_data;
2696 struct cl_object *obj;
2697 struct fsxattr fa = { 0 };
2699 if (get_user(flags, (int __user *)arg))
2702 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2703 if (flags & LUSTRE_PROJINHERIT_FL)
2704 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2706 rc = ll_ioctl_check_project(inode, &fa);
2710 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2711 LUSTRE_OPC_ANY, NULL);
2712 if (IS_ERR(op_data))
2713 RETURN(PTR_ERR(op_data));
2715 op_data->op_attr_flags = flags;
2716 op_data->op_xvalid |= OP_XVALID_FLAGS;
2717 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2718 ll_finish_md_op_data(op_data);
2719 ptlrpc_req_finished(req);
2723 ll_update_inode_flags(inode, flags);
2725 obj = ll_i2info(inode)->lli_clob;
2729 OBD_ALLOC_PTR(attr);
2733 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2745 int ll_flush_ctx(struct inode *inode)
2747 struct ll_sb_info *sbi = ll_i2sbi(inode);
2749 CDEBUG(D_SEC, "flush context for user %d\n",
2750 from_kuid(&init_user_ns, current_uid()));
2752 obd_set_info_async(NULL, sbi->ll_md_exp,
2753 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2755 obd_set_info_async(NULL, sbi->ll_dt_exp,
2756 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2761 /* umount -f client means force down, don't save state */
2762 void ll_umount_begin(struct super_block *sb)
2764 struct ll_sb_info *sbi = ll_s2sbi(sb);
2765 struct obd_device *obd;
2766 struct obd_ioctl_data *ioc_data;
2770 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2771 sb->s_count, atomic_read(&sb->s_active));
2773 obd = class_exp2obd(sbi->ll_md_exp);
2775 CERROR("Invalid MDC connection handle %#llx\n",
2776 sbi->ll_md_exp->exp_handle.h_cookie);
2782 obd = class_exp2obd(sbi->ll_dt_exp);
2784 CERROR("Invalid LOV connection handle %#llx\n",
2785 sbi->ll_dt_exp->exp_handle.h_cookie);
2791 OBD_ALLOC_PTR(ioc_data);
2793 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2794 sizeof *ioc_data, ioc_data, NULL);
2796 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2797 sizeof *ioc_data, ioc_data, NULL);
2799 OBD_FREE_PTR(ioc_data);
2802 /* Really, we'd like to wait until there are no requests outstanding,
2803 * and then continue. For now, we just periodically checking for vfs
2804 * to decrement mnt_cnt and hope to finish it within 10sec.
2808 !may_umount(sbi->ll_mnt.mnt)) {
2816 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2818 struct ll_sb_info *sbi = ll_s2sbi(sb);
2819 char *profilenm = get_profile_name(sb);
2823 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2824 read_only = *flags & MS_RDONLY;
2825 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2826 sizeof(KEY_READ_ONLY),
2827 KEY_READ_ONLY, sizeof(read_only),
2830 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2831 profilenm, read_only ?
2832 "read-only" : "read-write", err);
2837 sb->s_flags |= SB_RDONLY;
2839 sb->s_flags &= ~SB_RDONLY;
2841 if (sbi->ll_flags & LL_SBI_VERBOSE)
2842 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2843 read_only ? "read-only" : "read-write");
2849 * Cleanup the open handle that is cached on MDT-side.
2851 * For open case, the client side open handling thread may hit error
2852 * after the MDT grant the open. Under such case, the client should
2853 * send close RPC to the MDT as cleanup; otherwise, the open handle
2854 * on the MDT will be leaked there until the client umount or evicted.
2856 * In further, if someone unlinked the file, because the open handle
2857 * holds the reference on such file/object, then it will block the
2858 * subsequent threads that want to locate such object via FID.
2860 * \param[in] sb super block for this file-system
2861 * \param[in] open_req pointer to the original open request
2863 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
2865 struct mdt_body *body;
2866 struct md_op_data *op_data;
2867 struct ptlrpc_request *close_req = NULL;
2868 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2871 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
2872 OBD_ALLOC_PTR(op_data);
2873 if (op_data == NULL) {
2874 CWARN("%s: cannot allocate op_data to release open handle for "
2875 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2880 op_data->op_fid1 = body->mbo_fid1;
2881 op_data->op_open_handle = body->mbo_open_handle;
2882 op_data->op_mod_time = ktime_get_real_seconds();
2883 md_close(exp, op_data, NULL, &close_req);
2884 ptlrpc_req_finished(close_req);
2885 ll_finish_md_op_data(op_data);
2890 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
2891 struct super_block *sb, struct lookup_intent *it)
2893 struct ll_sb_info *sbi = NULL;
2894 struct lustre_md md = { NULL };
2895 bool default_lmv_deleted = false;
2900 LASSERT(*inode || sb);
2901 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2902 rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
2903 sbi->ll_md_exp, &md);
2908 * clear default_lmv only if intent_getattr reply doesn't contain it.
2909 * but it needs to be done after iget, check this early because
2910 * ll_update_lsm_md() may change md.
2912 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2913 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2914 default_lmv_deleted = true;
2917 rc = ll_update_inode(*inode, &md);
2921 LASSERT(sb != NULL);
2924 * At this point server returns to client's same fid as client
2925 * generated for creating. So using ->fid1 is okay here.
2927 if (!fid_is_sane(&md.body->mbo_fid1)) {
2928 CERROR("%s: Fid is insane "DFID"\n",
2930 PFID(&md.body->mbo_fid1));
2931 GOTO(out, rc = -EINVAL);
2934 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2935 sbi->ll_flags & LL_SBI_32BIT_API),
2937 if (IS_ERR(*inode)) {
2939 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2941 CERROR("new_inode -fatal: rc %d\n", rc);
2946 /* Handling piggyback layout lock.
2947 * Layout lock can be piggybacked by getattr and open request.
2948 * The lsm can be applied to inode only if it comes with a layout lock
2949 * otherwise correct layout may be overwritten, for example:
2950 * 1. proc1: mdt returns a lsm but not granting layout
2951 * 2. layout was changed by another client
2952 * 3. proc2: refresh layout and layout lock granted
2953 * 4. proc1: to apply a stale layout */
2954 if (it != NULL && it->it_lock_mode != 0) {
2955 struct lustre_handle lockh;
2956 struct ldlm_lock *lock;
2958 lockh.cookie = it->it_lock_handle;
2959 lock = ldlm_handle2lock(&lockh);
2960 LASSERT(lock != NULL);
2961 if (ldlm_has_layout(lock)) {
2962 struct cl_object_conf conf;
2964 memset(&conf, 0, sizeof(conf));
2965 conf.coc_opc = OBJECT_CONF_SET;
2966 conf.coc_inode = *inode;
2967 conf.coc_lock = lock;
2968 conf.u.coc_layout = md.layout;
2969 (void)ll_layout_conf(*inode, &conf);
2971 LDLM_LOCK_PUT(lock);
2974 if (default_lmv_deleted)
2975 ll_update_default_lsm_md(*inode, &md);
2977 /* we may want to apply some policy for foreign file/dir */
2978 if (ll_sbi_has_foreign_symlink(sbi)) {
2979 rc = ll_manage_foreign(*inode, &md);
2987 /* cleanup will be done if necessary */
2988 md_free_lustre_md(sbi->ll_md_exp, &md);
2990 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2991 ll_intent_drop_lock(it);
2992 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
2998 int ll_obd_statfs(struct inode *inode, void __user *arg)
3000 struct ll_sb_info *sbi = NULL;
3001 struct obd_export *exp;
3002 struct obd_ioctl_data *data = NULL;
3007 sbi = ll_i2sbi(inode);
3009 GOTO(out_statfs, rc = -EINVAL);
3011 rc = obd_ioctl_getdata(&data, &len, arg);
3013 GOTO(out_statfs, rc);
3015 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3016 !data->ioc_pbuf1 || !data->ioc_pbuf2)
3017 GOTO(out_statfs, rc = -EINVAL);
3019 if (data->ioc_inllen1 != sizeof(__u32) ||
3020 data->ioc_inllen2 != sizeof(__u32) ||
3021 data->ioc_plen1 != sizeof(struct obd_statfs) ||
3022 data->ioc_plen2 != sizeof(struct obd_uuid))
3023 GOTO(out_statfs, rc = -EINVAL);
3025 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3026 if (type & LL_STATFS_LMV)
3027 exp = sbi->ll_md_exp;
3028 else if (type & LL_STATFS_LOV)
3029 exp = sbi->ll_dt_exp;
3031 GOTO(out_statfs, rc = -ENODEV);
3033 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3035 GOTO(out_statfs, rc);
3037 OBD_FREE_LARGE(data, len);
3042 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3043 * be called early to avoid deadlock.
3045 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3047 if (op_data->op_mea2_sem) {
3048 up_read_non_owner(op_data->op_mea2_sem);
3049 op_data->op_mea2_sem = NULL;
3052 if (op_data->op_mea1_sem) {
3053 up_read_non_owner(op_data->op_mea1_sem);
3054 op_data->op_mea1_sem = NULL;
3058 /* this function prepares md_op_data hint for passing it down to MD stack. */
3059 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3060 struct inode *i1, struct inode *i2,
3061 const char *name, size_t namelen,
3062 __u32 mode, enum md_op_code opc,
3065 LASSERT(i1 != NULL);
3068 /* Do not reuse namelen for something else. */
3070 return ERR_PTR(-EINVAL);
3072 if (namelen > ll_i2sbi(i1)->ll_namelen)
3073 return ERR_PTR(-ENAMETOOLONG);
3075 /* "/" is not valid name, but it's allowed */
3076 if (!lu_name_is_valid_2(name, namelen) &&
3077 strncmp("/", name, namelen) != 0)
3078 return ERR_PTR(-EINVAL);
3081 if (op_data == NULL)
3082 OBD_ALLOC_PTR(op_data);
3084 if (op_data == NULL)
3085 return ERR_PTR(-ENOMEM);
3087 ll_i2gids(op_data->op_suppgids, i1, i2);
3088 op_data->op_fid1 = *ll_inode2fid(i1);
3089 op_data->op_code = opc;
3091 if (S_ISDIR(i1->i_mode)) {
3092 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3093 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3094 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3095 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3099 op_data->op_fid2 = *ll_inode2fid(i2);
3100 if (S_ISDIR(i2->i_mode)) {
3102 /* i2 is typically a child of i1, and MUST be
3103 * further from the root to avoid deadlocks.
3105 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3106 op_data->op_mea2_sem =
3107 &ll_i2info(i2)->lli_lsm_sem;
3109 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3112 fid_zero(&op_data->op_fid2);
3115 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
3116 op_data->op_cli_flags |= CLI_HASH64;
3118 if (ll_need_32bit_api(ll_i2sbi(i1)))
3119 op_data->op_cli_flags |= CLI_API32;
3121 op_data->op_name = name;
3122 op_data->op_namelen = namelen;
3123 op_data->op_mode = mode;
3124 op_data->op_mod_time = ktime_get_real_seconds();
3125 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3126 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3127 op_data->op_cap = cfs_curproc_cap_pack();
3128 op_data->op_mds = 0;
3129 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3130 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3131 op_data->op_bias |= MDS_CREATE_VOLATILE;
3133 op_data->op_data = data;
3138 void ll_finish_md_op_data(struct md_op_data *op_data)
3140 ll_unlock_md_op_lsm(op_data);
3141 ll_security_release_secctx(op_data->op_file_secctx,
3142 op_data->op_file_secctx_size);
3143 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3144 OBD_FREE_PTR(op_data);
3147 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3149 struct ll_sb_info *sbi;
3151 LASSERT(seq && dentry);
3152 sbi = ll_s2sbi(dentry->d_sb);
3154 if (sbi->ll_flags & LL_SBI_NOLCK)
3155 seq_puts(seq, ",nolock");
3157 /* "flock" is the default since 2.13, but it wasn't for many years,
3158 * so it is still useful to print this to show it is enabled.
3159 * Start to print "noflock" so it is now clear when flock is disabled.
3161 if (sbi->ll_flags & LL_SBI_FLOCK)
3162 seq_puts(seq, ",flock");
3163 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
3164 seq_puts(seq, ",localflock");
3166 seq_puts(seq, ",noflock");
3168 if (sbi->ll_flags & LL_SBI_USER_XATTR)
3169 seq_puts(seq, ",user_xattr");
3171 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
3172 seq_puts(seq, ",lazystatfs");
3174 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3175 seq_puts(seq, ",user_fid2path");
3177 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3178 seq_puts(seq, ",always_ping");
3180 if (ll_sbi_has_test_dummy_encryption(sbi))
3181 seq_puts(seq, ",test_dummy_encryption");
3183 if (ll_sbi_has_encrypt(sbi))
3184 seq_puts(seq, ",encrypt");
3186 seq_puts(seq, ",noencrypt");
3188 if (sbi->ll_flags & LL_SBI_FOREIGN_SYMLINK) {
3189 seq_puts(seq, ",foreign_symlink=");
3190 seq_puts(seq, sbi->ll_foreign_symlink_prefix);
3197 * Get obd name by cmd, and copy out to user space
3199 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3201 struct ll_sb_info *sbi = ll_i2sbi(inode);
3202 struct obd_device *obd;
3205 if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3206 obd = class_exp2obd(sbi->ll_dt_exp);
3207 else if (cmd == OBD_IOC_GETMDNAME)
3208 obd = class_exp2obd(sbi->ll_md_exp);
3215 if (copy_to_user((void __user *)arg, obd->obd_name,
3216 strlen(obd->obd_name) + 1))
3222 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3229 p.mnt = current->fs->root.mnt;
3231 path = d_path(&p, buf, bufsize);
3236 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3238 char *buf, *path = NULL;
3239 struct dentry *dentry = NULL;
3240 struct inode *inode = page->mapping->host;
3242 /* this can be called inside spin lock so use GFP_ATOMIC. */
3243 buf = (char *)__get_free_page(GFP_ATOMIC);
3245 dentry = d_find_alias(page->mapping->host);
3247 path = ll_d_path(dentry, buf, PAGE_SIZE);
3250 /* The below message is checked in recovery-small.sh test_24b */
3252 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3253 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3254 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3255 PFID(ll_inode2fid(inode)),
3256 (path && !IS_ERR(path)) ? path : "", ioret);
3262 free_page((unsigned long)buf);
3265 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3266 struct lov_user_md **kbuf)
3268 struct lov_user_md lum;
3272 if (copy_from_user(&lum, md, sizeof(lum)))
3275 lum_size = ll_lov_user_md_size(&lum);
3279 OBD_ALLOC_LARGE(*kbuf, lum_size);
3283 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3284 OBD_FREE_LARGE(*kbuf, lum_size);
3292 * Compute llite root squash state after a change of root squash
3293 * configuration setting or add/remove of a lnet nid
3295 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3297 struct root_squash_info *squash = &sbi->ll_squash;
3300 struct lnet_process_id id;
3302 /* Update norootsquash flag */
3303 spin_lock(&squash->rsi_lock);
3304 if (list_empty(&squash->rsi_nosquash_nids))
3305 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3307 /* Do not apply root squash as soon as one of our NIDs is
3308 * in the nosquash_nids list */
3311 while (LNetGetId(i++, &id) != -ENOENT) {
3312 if (id.nid == LNET_NID_LO_0)
3314 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3320 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
3322 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3324 spin_unlock(&squash->rsi_lock);
3328 * Parse linkea content to extract information about a given hardlink
3330 * \param[in] ldata - Initialized linkea data
3331 * \param[in] linkno - Link identifier
3332 * \param[out] parent_fid - The entry's parent FID
3333 * \param[out] ln - Entry name destination buffer
3335 * \retval 0 on success
3336 * \retval Appropriate negative error code on failure
3338 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3339 struct lu_fid *parent_fid, struct lu_name *ln)
3345 rc = linkea_init_with_rec(ldata);
3349 if (linkno >= ldata->ld_leh->leh_reccount)
3350 /* beyond last link */
3353 linkea_first_entry(ldata);
3354 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3355 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3360 linkea_next_entry(ldata);
3370 * Get parent FID and name of an identified link. Operation is performed for
3371 * a given link number, letting the caller iterate over linkno to list one or
3372 * all links of an entry.
3374 * \param[in] file - File descriptor against which to perform the operation
3375 * \param[in,out] arg - User-filled structure containing the linkno to operate
3376 * on and the available size. It is eventually filled with
3377 * the requested information or left untouched on error
3379 * \retval - 0 on success
3380 * \retval - Appropriate negative error code on failure
3382 int ll_getparent(struct file *file, struct getparent __user *arg)
3384 struct inode *inode = file_inode(file);
3385 struct linkea_data *ldata;
3386 struct lu_buf buf = LU_BUF_NULL;
3388 struct lu_fid parent_fid;
3395 if (!capable(CAP_DAC_READ_SEARCH) &&
3396 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3399 if (get_user(name_size, &arg->gp_name_size))
3402 if (get_user(linkno, &arg->gp_linkno))
3405 if (name_size > PATH_MAX)
3408 OBD_ALLOC(ldata, sizeof(*ldata));
3412 rc = linkea_data_new(ldata, &buf);
3414 GOTO(ldata_free, rc);
3416 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3417 buf.lb_len, OBD_MD_FLXATTR);
3421 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3425 if (ln.ln_namelen >= name_size)
3426 GOTO(lb_free, rc = -EOVERFLOW);
3428 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3429 GOTO(lb_free, rc = -EFAULT);
3431 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3432 GOTO(lb_free, rc = -EFAULT);
3434 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3435 GOTO(lb_free, rc = -EFAULT);
3440 OBD_FREE(ldata, sizeof(*ldata));