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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/cpu.h>
40 #include <linux/module.h>
41 #include <linux/random.h>
42 #include <linux/statfs.h>
43 #include <linux/time.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/security.h>
52 #include <linux/fs_struct.h>
54 #ifndef HAVE_CPUS_READ_LOCK
55 #include <libcfs/linux/linux-cpu.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;
102 RETURN(ERR_PTR(-ENOMEM));
104 rc = pcc_super_init(&sbi->ll_pcc_super);
108 spin_lock_init(&sbi->ll_lock);
109 mutex_init(&sbi->ll_lco.lco_lock);
110 spin_lock_init(&sbi->ll_pp_extent_lock);
111 spin_lock_init(&sbi->ll_process_lock);
112 sbi->ll_rw_stats_on = 0;
113 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
116 pages = si.totalram - si.totalhigh;
117 lru_page_max = pages / 2;
119 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
120 sbi->ll_ra_info.ll_readahead_wq =
121 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
123 sbi->ll_ra_info.ra_async_max_active);
124 if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
125 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
127 /* initialize ll_cache data */
128 sbi->ll_cache = cl_cache_init(lru_page_max);
129 if (sbi->ll_cache == NULL)
130 GOTO(out_destroy_ra, rc = -ENOMEM);
132 /* initialize foreign symlink prefix path */
133 OBD_ALLOC(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
134 if (sbi->ll_foreign_symlink_prefix == NULL)
135 GOTO(out_destroy_ra, rc = -ENOMEM);
136 memcpy(sbi->ll_foreign_symlink_prefix, "/mnt/", sizeof("/mnt/"));
137 sbi->ll_foreign_symlink_prefix_size = sizeof("/mnt/");
139 /* initialize foreign symlink upcall path, none by default */
140 OBD_ALLOC(sbi->ll_foreign_symlink_upcall, sizeof("none"));
141 if (sbi->ll_foreign_symlink_upcall == NULL)
142 GOTO(out_destroy_ra, rc = -ENOMEM);
143 memcpy(sbi->ll_foreign_symlink_upcall, "none", sizeof("none"));
144 sbi->ll_foreign_symlink_upcall_items = NULL;
145 sbi->ll_foreign_symlink_upcall_nb_items = 0;
146 init_rwsem(&sbi->ll_foreign_symlink_sem);
147 /* foreign symlink support (LL_SBI_FOREIGN_SYMLINK in ll_flags)
148 * not enabled by default
151 sbi->ll_ra_info.ra_max_pages =
152 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
153 sbi->ll_ra_info.ra_max_pages_per_file =
154 min(sbi->ll_ra_info.ra_max_pages / 4,
155 SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
156 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
157 sbi->ll_ra_info.ra_max_pages_per_file;
158 sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
159 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
160 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
162 sbi->ll_flags |= LL_SBI_VERBOSE;
163 #ifdef ENABLE_CHECKSUM
164 sbi->ll_flags |= LL_SBI_CHECKSUM;
167 sbi->ll_flags |= LL_SBI_FLOCK;
170 #ifdef HAVE_LRU_RESIZE_SUPPORT
171 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
173 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
175 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
176 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
178 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
182 /* metadata statahead is enabled by default */
183 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
184 sbi->ll_sa_max = LL_SA_RPC_DEF;
185 atomic_set(&sbi->ll_sa_total, 0);
186 atomic_set(&sbi->ll_sa_wrong, 0);
187 atomic_set(&sbi->ll_sa_running, 0);
188 atomic_set(&sbi->ll_agl_total, 0);
189 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
190 sbi->ll_flags |= LL_SBI_FAST_READ;
191 sbi->ll_flags |= LL_SBI_TINY_WRITE;
192 ll_sbi_set_encrypt(sbi, true);
195 sbi->ll_squash.rsi_uid = 0;
196 sbi->ll_squash.rsi_gid = 0;
197 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
198 spin_lock_init(&sbi->ll_squash.rsi_lock);
200 /* Per-filesystem file heat */
201 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
202 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
205 if (sbi->ll_foreign_symlink_prefix)
206 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
208 cl_cache_decref(sbi->ll_cache);
209 sbi->ll_cache = NULL;
211 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
213 pcc_super_fini(&sbi->ll_pcc_super);
219 static void ll_free_sbi(struct super_block *sb)
221 struct ll_sb_info *sbi = ll_s2sbi(sb);
225 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
226 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
227 if (sbi->ll_ra_info.ll_readahead_wq)
228 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
229 if (sbi->ll_cache != NULL) {
230 cl_cache_decref(sbi->ll_cache);
231 sbi->ll_cache = NULL;
233 if (sbi->ll_foreign_symlink_prefix) {
234 OBD_FREE(sbi->ll_foreign_symlink_prefix,
235 sbi->ll_foreign_symlink_prefix_size);
236 sbi->ll_foreign_symlink_prefix = NULL;
238 if (sbi->ll_foreign_symlink_upcall) {
239 OBD_FREE(sbi->ll_foreign_symlink_upcall,
240 strlen(sbi->ll_foreign_symlink_upcall) +
242 sbi->ll_foreign_symlink_upcall = NULL;
244 if (sbi->ll_foreign_symlink_upcall_items) {
246 int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
247 struct ll_foreign_symlink_upcall_item *items =
248 sbi->ll_foreign_symlink_upcall_items;
250 for (i = 0 ; i < nb_items; i++)
251 if (items[i].type == STRING_TYPE)
252 OBD_FREE(items[i].string,
255 OBD_FREE_LARGE(items, nb_items *
256 sizeof(struct ll_foreign_symlink_upcall_item));
257 sbi->ll_foreign_symlink_upcall_items = NULL;
259 pcc_super_fini(&sbi->ll_pcc_super);
260 OBD_FREE(sbi, sizeof(*sbi));
265 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
267 struct inode *root = NULL;
268 struct ll_sb_info *sbi = ll_s2sbi(sb);
269 struct obd_statfs *osfs = NULL;
270 struct ptlrpc_request *request = NULL;
271 struct obd_connect_data *data = NULL;
272 struct obd_uuid *uuid;
273 struct md_op_data *op_data;
274 struct lustre_md lmd;
276 int size, err, checksum;
279 sbi->ll_md_obd = class_name2obd(md);
280 if (!sbi->ll_md_obd) {
281 CERROR("MD %s: not setup or attached\n", md);
295 /* pass client page size via ocd_grant_blkbits, the server should report
296 * back its backend blocksize for grant calculation purpose */
297 data->ocd_grant_blkbits = PAGE_SHIFT;
299 /* indicate MDT features supported by this client */
300 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
301 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
302 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
303 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
304 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
305 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
306 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
307 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
308 OBD_CONNECT_64BITHASH |
309 OBD_CONNECT_EINPROGRESS |
310 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
311 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
312 OBD_CONNECT_MAX_EASIZE |
313 OBD_CONNECT_FLOCK_DEAD |
314 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
315 OBD_CONNECT_OPEN_BY_FID |
316 OBD_CONNECT_DIR_STRIPE |
317 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
318 OBD_CONNECT_SUBTREE |
319 OBD_CONNECT_MULTIMODRPCS |
320 OBD_CONNECT_GRANT_PARAM |
321 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
323 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
324 OBD_CONNECT2_SUM_STATFS |
325 OBD_CONNECT2_OVERSTRIPING |
327 OBD_CONNECT2_LOCK_CONVERT |
328 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
329 OBD_CONNECT2_INC_XID |
331 OBD_CONNECT2_ASYNC_DISCARD |
333 OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
334 OBD_CONNECT2_GETATTR_PFID |
335 OBD_CONNECT2_DOM_LVB;
337 #ifdef HAVE_LRU_RESIZE_SUPPORT
338 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
339 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
341 data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
343 data->ocd_cksum_types = obd_cksum_types_supported_client();
345 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
346 /* flag mdc connection as lightweight, only used for test
347 * purpose, use with care */
348 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
350 data->ocd_ibits_known = MDS_INODELOCK_FULL;
351 data->ocd_version = LUSTRE_VERSION_CODE;
353 if (sb->s_flags & SB_RDONLY)
354 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
355 if (sbi->ll_flags & LL_SBI_USER_XATTR)
356 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
359 /* Setting this indicates we correctly support S_NOSEC (See kernel
360 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
362 sb->s_flags |= SB_NOSEC;
365 if (sbi->ll_flags & LL_SBI_FLOCK)
366 sbi->ll_fop = &ll_file_operations_flock;
367 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
368 sbi->ll_fop = &ll_file_operations;
370 sbi->ll_fop = &ll_file_operations_noflock;
372 /* always ping even if server suppress_pings */
373 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
374 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
376 obd_connect_set_secctx(data);
377 if (ll_sbi_has_encrypt(sbi))
378 obd_connect_set_enc(data);
380 #if defined(CONFIG_SECURITY)
381 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
384 data->ocd_brw_size = MD_MAX_BRW_SIZE;
386 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
387 &sbi->ll_sb_uuid, data, sbi->ll_cache);
389 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
390 "recovery, of which this client is not a "
391 "part. Please wait for recovery to complete,"
392 " abort, or time out.\n", md);
395 CERROR("cannot connect to %s: rc = %d\n", md, err);
399 sbi->ll_md_exp->exp_connect_data = *data;
401 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
402 LUSTRE_SEQ_METADATA);
404 CERROR("%s: Can't init metadata layer FID infrastructure, "
405 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
409 /* For mount, we only need fs info from MDT0, and also in DNE, it
410 * can make sure the client can be mounted as long as MDT0 is
412 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
413 ktime_get_seconds() - sbi->ll_statfs_max_age,
414 OBD_STATFS_FOR_MDT0);
416 GOTO(out_md_fid, err);
418 /* This needs to be after statfs to ensure connect has finished.
419 * Note that "data" does NOT contain the valid connect reply.
420 * If connecting to a 1.8 server there will be no LMV device, so
421 * we can access the MDC export directly and exp_connect_flags will
422 * be non-zero, but if accessing an upgraded 2.1 server it will
423 * have the correct flags filled in.
424 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
425 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
426 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
427 valid != CLIENT_CONNECT_MDT_REQD) {
430 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
431 obd_connect_flags2str(buf, PAGE_SIZE,
432 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
433 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
434 "feature(s) needed for correct operation "
435 "of this client (%s). Please upgrade "
436 "server or downgrade client.\n",
437 sbi->ll_md_exp->exp_obd->obd_name, buf);
438 OBD_FREE(buf, PAGE_SIZE);
439 GOTO(out_md_fid, err = -EPROTO);
442 size = sizeof(*data);
443 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
444 KEY_CONN_DATA, &size, data);
446 CERROR("%s: Get connect data failed: rc = %d\n",
447 sbi->ll_md_exp->exp_obd->obd_name, err);
448 GOTO(out_md_fid, err);
451 LASSERT(osfs->os_bsize);
452 sb->s_blocksize = osfs->os_bsize;
453 sb->s_blocksize_bits = log2(osfs->os_bsize);
454 sb->s_magic = LL_SUPER_MAGIC;
455 sb->s_maxbytes = MAX_LFS_FILESIZE;
456 sbi->ll_namelen = osfs->os_namelen;
457 sbi->ll_mnt.mnt = current->fs->root.mnt;
459 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
460 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
461 LCONSOLE_INFO("Disabling user_xattr feature because "
462 "it is not supported on the server\n");
463 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
466 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
468 sb->s_flags |= SB_POSIXACL;
470 sbi->ll_flags |= LL_SBI_ACL;
472 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
474 sb->s_flags &= ~SB_POSIXACL;
476 sbi->ll_flags &= ~LL_SBI_ACL;
479 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
480 sbi->ll_flags |= LL_SBI_64BIT_HASH;
482 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
483 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
485 if (obd_connect_has_secctx(data))
486 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
488 if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
489 if (ll_sbi_has_test_dummy_encryption(sbi))
490 LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
492 sbi->ll_md_exp->exp_obd->obd_name);
493 ll_sbi_set_encrypt(sbi, false);
496 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
497 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
498 LCONSOLE_INFO("%s: disabling xattr cache due to "
499 "unknown maximum xattr size.\n", dt);
500 } else if (!sbi->ll_xattr_cache_set) {
501 /* If xattr_cache is already set (no matter 0 or 1)
502 * during processing llog, it won't be enabled here. */
503 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
504 sbi->ll_xattr_cache_enabled = 1;
508 sbi->ll_dt_obd = class_name2obd(dt);
509 if (!sbi->ll_dt_obd) {
510 CERROR("DT %s: not setup or attached\n", dt);
511 GOTO(out_md_fid, err = -ENODEV);
514 /* pass client page size via ocd_grant_blkbits, the server should report
515 * back its backend blocksize for grant calculation purpose */
516 data->ocd_grant_blkbits = PAGE_SHIFT;
518 /* indicate OST features supported by this client */
519 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
520 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
521 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
522 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
523 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
524 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
525 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
526 OBD_CONNECT_EINPROGRESS |
527 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
528 OBD_CONNECT_LAYOUTLOCK |
529 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
530 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
531 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
532 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
533 OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK;
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, sbi->ll_dt_exp,
668 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);
1227 GOTO(out_free_cfg, err = -ENOMEM);
1229 /* client additional sb info */
1230 lsi->lsi_llsbi = sbi = ll_init_sbi();
1232 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1234 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1236 GOTO(out_free_cfg, err);
1238 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1239 sb->s_d_op = &ll_d_ops;
1242 generate_random_uuid(uuid.b);
1243 snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1245 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1248 len = strlen(profilenm);
1249 ptr = strrchr(profilenm, '-');
1250 if (ptr && (strcmp(ptr, "-client") == 0))
1253 if (len > LUSTRE_MAXFSNAME) {
1254 if (unlikely(len >= MAX_STRING_SIZE))
1255 len = MAX_STRING_SIZE - 1;
1256 strncpy(name, profilenm, len);
1258 err = -ENAMETOOLONG;
1259 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1260 name, LUSTRE_MAXFSNAME, err);
1261 GOTO(out_free_cfg, err);
1263 strncpy(sbi->ll_fsname, profilenm, len);
1264 sbi->ll_fsname[len] = '\0';
1267 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1268 profilenm, cfg_instance);
1270 err = super_setup_bdi_name(sb, "%s", name);
1272 GOTO(out_free_cfg, err);
1274 /* Call ll_debugfs_register_super() before lustre_process_log()
1275 * so that "llite.*.*" params can be processed correctly.
1277 err = ll_debugfs_register_super(sb, name);
1279 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1280 sbi->ll_fsname, err);
1284 /* The cfg_instance is a value unique to this super, in case some
1285 * joker tries to mount the same fs at two mount points.
1287 cfg->cfg_instance = cfg_instance;
1288 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1289 cfg->cfg_callback = class_config_llog_handler;
1290 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1291 /* set up client obds */
1292 err = lustre_process_log(sb, profilenm, cfg);
1294 GOTO(out_debugfs, err);
1296 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1297 lprof = class_get_profile(profilenm);
1298 if (lprof == NULL) {
1299 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1300 " read from the MGS. Does that filesystem "
1301 "exist?\n", profilenm);
1302 GOTO(out_debugfs, err = -EINVAL);
1304 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1305 lprof->lp_md, lprof->lp_dt);
1307 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1308 OBD_ALLOC(dt, dt_len);
1310 GOTO(out_profile, err = -ENOMEM);
1311 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1313 md_len = strlen(lprof->lp_md) + instlen + 2;
1314 OBD_ALLOC(md, md_len);
1316 GOTO(out_free_dt, err = -ENOMEM);
1317 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1319 /* connections, registrations, sb setup */
1320 err = client_common_fill_super(sb, md, dt);
1322 GOTO(out_free_md, err);
1324 sbi->ll_client_common_fill_super_succeeded = 1;
1328 OBD_FREE(md, md_len);
1331 OBD_FREE(dt, dt_len);
1334 class_put_profile(lprof);
1337 ll_debugfs_unregister_super(sb);
1344 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1345 LCONSOLE_WARN("Mounted %s\n", profilenm);
1347 } /* ll_fill_super */
1349 void ll_put_super(struct super_block *sb)
1351 struct config_llog_instance cfg, params_cfg;
1352 struct obd_device *obd;
1353 struct lustre_sb_info *lsi = s2lsi(sb);
1354 struct ll_sb_info *sbi = ll_s2sbi(sb);
1355 char *profilenm = get_profile_name(sb);
1356 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1358 int next, force = 1, rc = 0;
1362 GOTO(out_no_sbi, 0);
1364 /* Should replace instance_id with something better for ASLR */
1365 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1366 profilenm, cfg_instance, sb);
1368 cfg.cfg_instance = cfg_instance;
1369 lustre_end_log(sb, profilenm, &cfg);
1371 params_cfg.cfg_instance = cfg_instance;
1372 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1374 if (sbi->ll_md_exp) {
1375 obd = class_exp2obd(sbi->ll_md_exp);
1377 force = obd->obd_force;
1380 /* Wait for unstable pages to be committed to stable storage */
1382 rc = l_wait_event_abortable(
1383 sbi->ll_cache->ccc_unstable_waitq,
1384 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1387 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1388 if (force == 0 && rc != -ERESTARTSYS)
1389 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1391 /* We need to set force before the lov_disconnect in
1392 * lustre_common_put_super, since l_d cleans up osc's as well.
1396 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1398 obd->obd_force = force;
1402 if (sbi->ll_client_common_fill_super_succeeded) {
1403 /* Only if client_common_fill_super succeeded */
1404 client_common_put_super(sb);
1408 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1409 class_manual_cleanup(obd);
1411 if (sbi->ll_flags & LL_SBI_VERBOSE)
1412 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1415 class_del_profile(profilenm);
1417 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1418 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1419 bdi_destroy(&lsi->lsi_bdi);
1420 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1425 lsi->lsi_llsbi = NULL;
1427 lustre_common_put_super(sb);
1429 cl_env_cache_purge(~0);
1432 } /* client_put_super */
1434 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1436 struct inode *inode = NULL;
1438 /* NOTE: we depend on atomic igrab() -bzzz */
1439 lock_res_and_lock(lock);
1440 if (lock->l_resource->lr_lvb_inode) {
1441 struct ll_inode_info * lli;
1442 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1443 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1444 inode = igrab(lock->l_resource->lr_lvb_inode);
1446 inode = lock->l_resource->lr_lvb_inode;
1447 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1448 D_WARNING, lock, "lr_lvb_inode %p is "
1449 "bogus: magic %08x",
1450 lock->l_resource->lr_lvb_inode,
1451 lli->lli_inode_magic);
1455 unlock_res_and_lock(lock);
1459 void ll_dir_clear_lsm_md(struct inode *inode)
1461 struct ll_inode_info *lli = ll_i2info(inode);
1463 LASSERT(S_ISDIR(inode->i_mode));
1465 if (lli->lli_lsm_md) {
1466 lmv_free_memmd(lli->lli_lsm_md);
1467 lli->lli_lsm_md = NULL;
1470 if (lli->lli_default_lsm_md) {
1471 lmv_free_memmd(lli->lli_default_lsm_md);
1472 lli->lli_default_lsm_md = NULL;
1476 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1477 const struct lu_fid *fid,
1478 struct lustre_md *md)
1480 struct ll_sb_info *sbi = ll_s2sbi(sb);
1481 struct mdt_body *body = md->body;
1482 struct inode *inode;
1486 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1487 inode = iget_locked(sb, ino);
1488 if (inode == NULL) {
1489 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1490 sbi->ll_fsname, PFID(fid));
1491 RETURN(ERR_PTR(-ENOENT));
1494 if (inode->i_state & I_NEW) {
1495 struct ll_inode_info *lli = ll_i2info(inode);
1496 struct lmv_stripe_md *lsm = md->lmv;
1498 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1499 (body->mbo_mode & S_IFMT);
1500 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1503 inode->i_mtime.tv_sec = 0;
1504 inode->i_atime.tv_sec = 0;
1505 inode->i_ctime.tv_sec = 0;
1508 #ifdef HAVE_BACKING_DEV_INFO
1509 /* initializing backing dev info. */
1510 inode->i_mapping->backing_dev_info =
1511 &s2lsi(inode->i_sb)->lsi_bdi;
1513 inode->i_op = &ll_dir_inode_operations;
1514 inode->i_fop = &ll_dir_operations;
1515 lli->lli_fid = *fid;
1518 LASSERT(lsm != NULL);
1519 /* master object FID */
1520 lli->lli_pfid = body->mbo_fid1;
1521 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1522 lli, PFID(fid), PFID(&lli->lli_pfid));
1523 unlock_new_inode(inode);
1529 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1532 struct lmv_stripe_md *lsm = md->lmv;
1533 struct ll_inode_info *lli = ll_i2info(inode);
1536 LASSERT(lsm != NULL);
1538 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1539 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1540 lsm_md_dump(D_INODE, lsm);
1542 if (!lmv_dir_striped(lsm))
1545 /* XXX sigh, this lsm_root initialization should be in
1546 * LMV layer, but it needs ll_iget right now, so we
1547 * put this here right now. */
1548 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1549 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1550 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1552 if (!fid_is_sane(fid))
1555 /* Unfortunately ll_iget will call ll_update_inode,
1556 * where the initialization of slave inode is slightly
1557 * different, so it reset lsm_md to NULL to avoid
1558 * initializing lsm for slave inode. */
1559 lsm->lsm_md_oinfo[i].lmo_root =
1560 ll_iget_anon_dir(inode->i_sb, fid, md);
1561 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1562 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1564 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1566 iput(lsm->lsm_md_oinfo[i].lmo_root);
1567 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1573 lli->lli_lsm_md = lsm;
1578 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1580 struct ll_inode_info *lli = ll_i2info(inode);
1582 if (!md->default_lmv) {
1583 /* clear default lsm */
1584 if (lli->lli_default_lsm_md) {
1585 down_write(&lli->lli_lsm_sem);
1586 if (lli->lli_default_lsm_md) {
1587 lmv_free_memmd(lli->lli_default_lsm_md);
1588 lli->lli_default_lsm_md = NULL;
1590 up_write(&lli->lli_lsm_sem);
1592 } else if (lli->lli_default_lsm_md) {
1593 /* update default lsm if it changes */
1594 down_read(&lli->lli_lsm_sem);
1595 if (lli->lli_default_lsm_md &&
1596 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1597 up_read(&lli->lli_lsm_sem);
1598 down_write(&lli->lli_lsm_sem);
1599 if (lli->lli_default_lsm_md)
1600 lmv_free_memmd(lli->lli_default_lsm_md);
1601 lli->lli_default_lsm_md = md->default_lmv;
1602 lsm_md_dump(D_INODE, md->default_lmv);
1603 md->default_lmv = NULL;
1604 up_write(&lli->lli_lsm_sem);
1606 up_read(&lli->lli_lsm_sem);
1609 /* init default lsm */
1610 down_write(&lli->lli_lsm_sem);
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);
1618 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1620 struct ll_inode_info *lli = ll_i2info(inode);
1621 struct lmv_stripe_md *lsm = md->lmv;
1622 struct cl_attr *attr;
1627 LASSERT(S_ISDIR(inode->i_mode));
1628 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1629 PFID(ll_inode2fid(inode)));
1631 /* update default LMV */
1632 if (md->default_lmv)
1633 ll_update_default_lsm_md(inode, md);
1636 * no striped information from request, lustre_md from req does not
1637 * include stripeEA, see ll_md_setattr()
1643 * normally dir layout doesn't change, only take read lock to check
1644 * that to avoid blocking other MD operations.
1646 down_read(&lli->lli_lsm_sem);
1648 /* some current lookup initialized lsm, and unchanged */
1649 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1650 GOTO(unlock, rc = 0);
1652 /* if dir layout doesn't match, check whether version is increased,
1653 * which means layout is changed, this happens in dir split/merge and
1656 * foreign LMV should not change.
1658 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1659 lsm->lsm_md_layout_version <=
1660 lli->lli_lsm_md->lsm_md_layout_version) {
1661 CERROR("%s: "DFID" dir layout mismatch:\n",
1662 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1663 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1664 lsm_md_dump(D_ERROR, lsm);
1665 GOTO(unlock, rc = -EINVAL);
1668 up_read(&lli->lli_lsm_sem);
1669 down_write(&lli->lli_lsm_sem);
1670 /* clear existing lsm */
1671 if (lli->lli_lsm_md) {
1672 lmv_free_memmd(lli->lli_lsm_md);
1673 lli->lli_lsm_md = NULL;
1676 rc = ll_init_lsm_md(inode, md);
1677 up_write(&lli->lli_lsm_sem);
1682 /* set md->lmv to NULL, so the following free lustre_md will not free
1687 /* md_merge_attr() may take long, since lsm is already set, switch to
1690 down_read(&lli->lli_lsm_sem);
1692 if (!lmv_dir_striped(lli->lli_lsm_md))
1693 GOTO(unlock, rc = 0);
1695 OBD_ALLOC_PTR(attr);
1697 GOTO(unlock, rc = -ENOMEM);
1699 /* validate the lsm */
1700 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1701 ll_md_blocking_ast);
1703 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1704 md->body->mbo_nlink = attr->cat_nlink;
1705 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1706 md->body->mbo_size = attr->cat_size;
1707 if (md->body->mbo_valid & OBD_MD_FLATIME)
1708 md->body->mbo_atime = attr->cat_atime;
1709 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1710 md->body->mbo_ctime = attr->cat_ctime;
1711 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1712 md->body->mbo_mtime = attr->cat_mtime;
1718 up_read(&lli->lli_lsm_sem);
1723 void ll_clear_inode(struct inode *inode)
1725 struct ll_inode_info *lli = ll_i2info(inode);
1726 struct ll_sb_info *sbi = ll_i2sbi(inode);
1730 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1731 PFID(ll_inode2fid(inode)), inode);
1733 if (S_ISDIR(inode->i_mode)) {
1734 /* these should have been cleared in ll_file_release */
1735 LASSERT(lli->lli_opendir_key == NULL);
1736 LASSERT(lli->lli_sai == NULL);
1737 LASSERT(lli->lli_opendir_pid == 0);
1739 pcc_inode_free(inode);
1742 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1744 LASSERT(!lli->lli_open_fd_write_count);
1745 LASSERT(!lli->lli_open_fd_read_count);
1746 LASSERT(!lli->lli_open_fd_exec_count);
1748 if (lli->lli_mds_write_och)
1749 ll_md_real_close(inode, FMODE_WRITE);
1750 if (lli->lli_mds_exec_och)
1751 ll_md_real_close(inode, FMODE_EXEC);
1752 if (lli->lli_mds_read_och)
1753 ll_md_real_close(inode, FMODE_READ);
1755 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1756 OBD_FREE(lli->lli_symlink_name,
1757 strlen(lli->lli_symlink_name) + 1);
1758 lli->lli_symlink_name = NULL;
1761 ll_xattr_cache_destroy(inode);
1763 forget_all_cached_acls(inode);
1765 lli->lli_inode_magic = LLI_INODE_DEAD;
1767 if (S_ISDIR(inode->i_mode))
1768 ll_dir_clear_lsm_md(inode);
1769 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1770 LASSERT(list_empty(&lli->lli_agl_list));
1773 * XXX This has to be done before lsm is freed below, because
1774 * cl_object still uses inode lsm.
1776 cl_inode_fini(inode);
1778 llcrypt_put_encryption_info(inode);
1783 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1785 struct lustre_md md;
1786 struct inode *inode = dentry->d_inode;
1787 struct ll_sb_info *sbi = ll_i2sbi(inode);
1788 struct ptlrpc_request *request = NULL;
1792 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1793 LUSTRE_OPC_ANY, NULL);
1794 if (IS_ERR(op_data))
1795 RETURN(PTR_ERR(op_data));
1797 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1799 ptlrpc_req_finished(request);
1800 if (rc == -ENOENT) {
1802 /* Unlinked special device node? Or just a race?
1803 * Pretend we done everything. */
1804 if (!S_ISREG(inode->i_mode) &&
1805 !S_ISDIR(inode->i_mode)) {
1806 ia_valid = op_data->op_attr.ia_valid;
1807 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1808 rc = simple_setattr(dentry, &op_data->op_attr);
1809 op_data->op_attr.ia_valid = ia_valid;
1811 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1812 CERROR("md_setattr fails: rc = %d\n", rc);
1817 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1818 sbi->ll_md_exp, &md);
1820 ptlrpc_req_finished(request);
1824 ia_valid = op_data->op_attr.ia_valid;
1825 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1826 * cache is not cleared yet. */
1827 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1828 if (S_ISREG(inode->i_mode))
1830 rc = simple_setattr(dentry, &op_data->op_attr);
1831 if (S_ISREG(inode->i_mode))
1832 inode_unlock(inode);
1833 op_data->op_attr.ia_valid = ia_valid;
1835 rc = ll_update_inode(inode, &md);
1836 ptlrpc_req_finished(request);
1842 * Zero portion of page that is part of @inode.
1843 * This implies, if necessary:
1844 * - taking cl_lock on range corresponding to concerned page
1845 * - grabbing vm page
1846 * - associating cl_page
1847 * - proceeding to clio read
1848 * - zeroing range in page
1849 * - proceeding to cl_page flush
1850 * - releasing cl_lock
1852 * \param[in] inode inode
1853 * \param[in] index page index
1854 * \param[in] offset offset in page to start zero from
1855 * \param[in] len len to zero
1857 * \retval 0 on success
1858 * \retval negative errno on failure
1860 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1863 struct ll_inode_info *lli = ll_i2info(inode);
1864 struct cl_object *clob = lli->lli_clob;
1866 struct lu_env *env = NULL;
1867 struct cl_io *io = NULL;
1868 struct cl_page *clpage = NULL;
1869 struct page *vmpage = NULL;
1870 unsigned from = index << PAGE_SHIFT;
1871 struct cl_lock *lock = NULL;
1872 struct cl_lock_descr *descr = NULL;
1873 struct cl_2queue *queue = NULL;
1874 struct cl_sync_io *anchor = NULL;
1875 bool holdinglock = false;
1876 bool lockedbymyself = true;
1881 env = cl_env_get(&refcheck);
1883 RETURN(PTR_ERR(env));
1885 io = vvp_env_thread_io(env);
1887 rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1891 lock = vvp_env_lock(env);
1892 descr = &lock->cll_descr;
1893 descr->cld_obj = io->ci_obj;
1894 descr->cld_start = cl_index(io->ci_obj, from);
1895 descr->cld_end = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1896 descr->cld_mode = CLM_WRITE;
1897 descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1899 /* request lock for page */
1900 rc = cl_lock_request(env, io, lock);
1901 /* -ECANCELED indicates a matching lock with a different extent
1902 * was already present, and -EEXIST indicates a matching lock
1903 * on exactly the same extent was already present.
1904 * In both cases it means we are covered.
1906 if (rc == -ECANCELED || rc == -EEXIST)
1914 vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1916 GOTO(rellock, rc = -EOPNOTSUPP);
1918 if (!PageDirty(vmpage)) {
1919 /* associate cl_page */
1920 clpage = cl_page_find(env, clob, vmpage->index,
1921 vmpage, CPT_CACHEABLE);
1923 GOTO(pagefini, rc = PTR_ERR(clpage));
1925 cl_page_assume(env, io, clpage);
1928 if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1929 !PageWriteback(vmpage)) {
1931 /* set PagePrivate2 to detect special case of empty page
1932 * in osc_brw_fini_request()
1934 SetPagePrivate2(vmpage);
1935 rc = ll_io_read_page(env, io, clpage, NULL);
1936 if (!PagePrivate2(vmpage))
1937 /* PagePrivate2 was cleared in osc_brw_fini_request()
1938 * meaning we read an empty page. In this case, in order
1939 * to avoid allocating unnecessary block in truncated
1940 * file, we must not zero and write as below. Subsequent
1941 * server-side truncate will handle things correctly.
1943 GOTO(clpfini, rc = 0);
1944 ClearPagePrivate2(vmpage);
1947 lockedbymyself = trylock_page(vmpage);
1948 cl_page_assume(env, io, clpage);
1951 /* zero range in page */
1952 zero_user(vmpage, offset, len);
1954 if (holdinglock && clpage) {
1955 /* explicitly write newly modified page */
1956 queue = &io->ci_queue;
1957 cl_2queue_init(queue);
1958 anchor = &vvp_env_info(env)->vti_anchor;
1959 cl_sync_io_init(anchor, 1);
1960 clpage->cp_sync_io = anchor;
1961 cl_2queue_add(queue, clpage);
1962 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1964 GOTO(queuefini1, rc);
1965 rc = cl_sync_io_wait(env, anchor, 0);
1967 GOTO(queuefini2, rc);
1968 cl_page_assume(env, io, clpage);
1971 cl_2queue_discard(env, io, queue);
1973 cl_2queue_disown(env, io, queue);
1974 cl_2queue_fini(env, queue);
1979 cl_page_put(env, clpage);
1981 if (lockedbymyself) {
1982 unlock_page(vmpage);
1987 cl_lock_release(env, lock);
1989 cl_io_fini(env, io);
1992 cl_env_put(env, &refcheck);
1997 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1998 * object(s) determine the file size and mtime. Otherwise, the MDS will
1999 * keep these values until such a time that objects are allocated for it.
2000 * We do the MDS operations first, as it is checking permissions for us.
2001 * We don't to the MDS RPC if there is nothing that we want to store there,
2002 * otherwise there is no harm in updating mtime/atime on the MDS if we are
2003 * going to do an RPC anyways.
2005 * If we are doing a truncate, we will send the mtime and ctime updates
2006 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2007 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2010 * In case of HSMimport, we only set attr on MDS.
2012 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2013 enum op_xvalid xvalid, bool hsm_import)
2015 struct inode *inode = dentry->d_inode;
2016 struct ll_inode_info *lli = ll_i2info(inode);
2017 struct md_op_data *op_data = NULL;
2018 ktime_t kstart = ktime_get();
2023 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2024 "valid %x, hsm_import %d\n",
2025 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2026 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2029 if (attr->ia_valid & ATTR_SIZE) {
2030 /* Check new size against VFS/VM file size limit and rlimit */
2031 rc = inode_newsize_ok(inode, attr->ia_size);
2035 /* The maximum Lustre file size is variable, based on the
2036 * OST maximum object size and number of stripes. This
2037 * needs another check in addition to the VFS check above. */
2038 if (attr->ia_size > ll_file_maxbytes(inode)) {
2039 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2040 PFID(&lli->lli_fid), attr->ia_size,
2041 ll_file_maxbytes(inode));
2045 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2048 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2049 if (attr->ia_valid & TIMES_SET_FLAGS) {
2050 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2051 !cfs_capable(CAP_FOWNER))
2055 /* We mark all of the fields "set" so MDS/OST does not re-set them */
2056 if (!(xvalid & OP_XVALID_CTIME_SET) &&
2057 (attr->ia_valid & ATTR_CTIME)) {
2058 attr->ia_ctime = current_time(inode);
2059 xvalid |= OP_XVALID_CTIME_SET;
2061 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2062 (attr->ia_valid & ATTR_ATIME)) {
2063 attr->ia_atime = current_time(inode);
2064 attr->ia_valid |= ATTR_ATIME_SET;
2066 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2067 (attr->ia_valid & ATTR_MTIME)) {
2068 attr->ia_mtime = current_time(inode);
2069 attr->ia_valid |= ATTR_MTIME_SET;
2072 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2073 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2074 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2075 ktime_get_real_seconds());
2077 if (S_ISREG(inode->i_mode))
2078 inode_unlock(inode);
2080 /* We always do an MDS RPC, even if we're only changing the size;
2081 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2083 OBD_ALLOC_PTR(op_data);
2084 if (op_data == NULL)
2085 GOTO(out, rc = -ENOMEM);
2087 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2088 /* If we are changing file size, file content is
2089 * modified, flag it.
2091 xvalid |= OP_XVALID_OWNEROVERRIDE;
2092 op_data->op_bias |= MDS_DATA_MODIFIED;
2093 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
2096 if (attr->ia_valid & ATTR_FILE) {
2097 struct ll_file_data *fd = attr->ia_file->private_data;
2099 if (fd->fd_lease_och)
2100 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2103 op_data->op_attr = *attr;
2104 op_data->op_xvalid = xvalid;
2106 rc = ll_md_setattr(dentry, op_data);
2110 if (!S_ISREG(inode->i_mode) || hsm_import)
2113 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2114 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2115 xvalid & OP_XVALID_CTIME_SET) {
2116 bool cached = false;
2118 rc = pcc_inode_setattr(inode, attr, &cached);
2121 CERROR("%s: PCC inode "DFID" setattr failed: "
2123 ll_i2sbi(inode)->ll_fsname,
2124 PFID(&lli->lli_fid), rc);
2128 unsigned int flags = 0;
2130 /* For truncate and utimes sending attributes to OSTs,
2131 * setting mtime/atime to the past will be performed
2132 * under PW [0:EOF] extent lock (new_size:EOF for
2133 * truncate). It may seem excessive to send mtime/atime
2134 * updates to OSTs when not setting times to past, but
2135 * it is necessary due to possible time
2136 * de-synchronization between MDT inode and OST objects
2138 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2139 attr->ia_valid & ATTR_SIZE) {
2140 xvalid |= OP_XVALID_FLAGS;
2141 flags = LUSTRE_ENCRYPT_FL;
2142 /* Call to ll_io_zero_page is not necessary if
2143 * truncating on PAGE_SIZE boundary, because
2144 * whole pages will be wiped.
2145 * In case of Direct IO, all we need is to set
2148 if (attr->ia_size & ~PAGE_MASK &&
2149 !(attr->ia_valid & ATTR_FILE &&
2150 attr->ia_file->f_flags & O_DIRECT)) {
2152 attr->ia_size & (PAGE_SIZE - 1);
2154 rc = ll_io_zero_page(inode,
2155 attr->ia_size >> PAGE_SHIFT,
2156 offset, PAGE_SIZE - offset);
2161 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2165 /* If the file was restored, it needs to set dirty flag.
2167 * We've already sent MDS_DATA_MODIFIED flag in
2168 * ll_md_setattr() for truncate. However, the MDT refuses to
2169 * set the HS_DIRTY flag on released files, so we have to set
2170 * it again if the file has been restored. Please check how
2171 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2173 * Please notice that if the file is not released, the previous
2174 * MDS_DATA_MODIFIED has taken effect and usually
2175 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2176 * This way we can save an RPC for common open + trunc
2178 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
2179 struct hsm_state_set hss = {
2180 .hss_valid = HSS_SETMASK,
2181 .hss_setmask = HS_DIRTY,
2185 rc2 = ll_hsm_state_set(inode, &hss);
2186 /* truncate and write can happen at the same time, so that
2187 * the file can be set modified even though the file is not
2188 * restored from released state, and ll_hsm_state_set() is
2189 * not applicable for the file, and rc2 < 0 is normal in this
2192 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2193 PFID(ll_inode2fid(inode)), rc2);
2198 if (op_data != NULL)
2199 ll_finish_md_op_data(op_data);
2201 if (S_ISREG(inode->i_mode)) {
2203 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2204 inode_dio_wait(inode);
2205 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2206 * flag. ll_update_inode (called from ll_md_setattr), clears
2207 * inode flags, so there is a gap where S_NOSEC is not set.
2208 * This can cause a writer to take the i_mutex unnecessarily,
2209 * but this is safe to do and should be rare. */
2210 inode_has_no_xattr(inode);
2214 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2215 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2216 ktime_us_delta(ktime_get(), kstart));
2221 int ll_setattr(struct dentry *de, struct iattr *attr)
2223 int mode = de->d_inode->i_mode;
2224 enum op_xvalid xvalid = 0;
2227 rc = llcrypt_prepare_setattr(de, attr);
2231 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2232 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2233 xvalid |= OP_XVALID_OWNEROVERRIDE;
2235 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2236 (ATTR_SIZE|ATTR_MODE)) &&
2237 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2238 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2239 !(attr->ia_mode & S_ISGID))))
2240 attr->ia_valid |= ATTR_FORCE;
2242 if ((attr->ia_valid & ATTR_MODE) &&
2244 !(attr->ia_mode & S_ISUID) &&
2245 !(attr->ia_valid & ATTR_KILL_SUID))
2246 attr->ia_valid |= ATTR_KILL_SUID;
2248 if ((attr->ia_valid & ATTR_MODE) &&
2249 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2250 !(attr->ia_mode & S_ISGID) &&
2251 !(attr->ia_valid & ATTR_KILL_SGID))
2252 attr->ia_valid |= ATTR_KILL_SGID;
2254 return ll_setattr_raw(de, attr, xvalid, false);
2257 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2260 struct obd_statfs obd_osfs = { 0 };
2265 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2267 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2268 flags |= OBD_STATFS_NODELAY;
2270 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2274 osfs->os_type = LL_SUPER_MAGIC;
2276 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2277 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2279 if (osfs->os_state & OS_STATFS_SUM)
2282 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2283 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
2286 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2287 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2290 osfs->os_bsize = obd_osfs.os_bsize;
2291 osfs->os_blocks = obd_osfs.os_blocks;
2292 osfs->os_bfree = obd_osfs.os_bfree;
2293 osfs->os_bavail = obd_osfs.os_bavail;
2295 /* If we have _some_ OSTs, but don't have as many free objects on the
2296 * OSTs as inodes on the MDTs, reduce the reported number of inodes
2297 * to compensate, so that the "inodes in use" number is correct.
2298 * This should be kept in sync with lod_statfs() behaviour.
2300 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2301 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2303 osfs->os_ffree = obd_osfs.os_ffree;
2310 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2312 struct if_quotactl qctl = {
2313 .qc_cmd = LUSTRE_Q_GETQUOTA,
2314 .qc_type = PRJQUOTA,
2315 .qc_valid = QC_GENERAL,
2317 u64 limit, curblock;
2320 qctl.qc_id = ll_i2info(inode)->lli_projid;
2321 ret = quotactl_ioctl(ll_i2sbi(inode), &qctl);
2323 /* ignore errors if project ID does not have
2324 * a quota limit or feature unsupported.
2326 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2331 limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2332 qctl.qc_dqblk.dqb_bsoftlimit :
2333 qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2334 if (limit && sfs->f_blocks > limit) {
2335 curblock = (qctl.qc_dqblk.dqb_curspace +
2336 sfs->f_bsize - 1) / sfs->f_bsize;
2337 sfs->f_blocks = limit;
2338 sfs->f_bfree = sfs->f_bavail =
2339 (sfs->f_blocks > curblock) ?
2340 (sfs->f_blocks - curblock) : 0;
2343 limit = qctl.qc_dqblk.dqb_isoftlimit ?
2344 qctl.qc_dqblk.dqb_isoftlimit :
2345 qctl.qc_dqblk.dqb_ihardlimit;
2346 if (limit && sfs->f_files > limit) {
2347 sfs->f_files = limit;
2348 sfs->f_ffree = (sfs->f_files >
2349 qctl.qc_dqblk.dqb_curinodes) ?
2350 (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2356 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2358 struct super_block *sb = de->d_sb;
2359 struct obd_statfs osfs;
2360 __u64 fsid = huge_encode_dev(sb->s_dev);
2361 ktime_t kstart = ktime_get();
2364 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2366 /* Some amount of caching on the client is allowed */
2367 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2371 statfs_unpack(sfs, &osfs);
2373 /* We need to downshift for all 32-bit kernels, because we can't
2374 * tell if the kernel is being called via sys_statfs64() or not.
2375 * Stop before overflowing f_bsize - in which case it is better
2376 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2377 if (sizeof(long) < 8) {
2378 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2381 osfs.os_blocks >>= 1;
2382 osfs.os_bfree >>= 1;
2383 osfs.os_bavail >>= 1;
2387 sfs->f_blocks = osfs.os_blocks;
2388 sfs->f_bfree = osfs.os_bfree;
2389 sfs->f_bavail = osfs.os_bavail;
2390 sfs->f_fsid.val[0] = (__u32)fsid;
2391 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2392 if (ll_i2info(de->d_inode)->lli_projid)
2393 return ll_statfs_project(de->d_inode, sfs);
2395 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2396 ktime_us_delta(ktime_get(), kstart));
2401 void ll_inode_size_lock(struct inode *inode)
2403 struct ll_inode_info *lli;
2405 LASSERT(!S_ISDIR(inode->i_mode));
2407 lli = ll_i2info(inode);
2408 mutex_lock(&lli->lli_size_mutex);
2411 void ll_inode_size_unlock(struct inode *inode)
2413 struct ll_inode_info *lli;
2415 lli = ll_i2info(inode);
2416 mutex_unlock(&lli->lli_size_mutex);
2419 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2421 /* do not clear encryption flag */
2422 ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2423 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2424 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2425 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2427 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2430 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2432 struct ll_inode_info *lli = ll_i2info(inode);
2433 struct mdt_body *body = md->body;
2434 struct ll_sb_info *sbi = ll_i2sbi(inode);
2437 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2438 rc = cl_file_inode_init(inode, md);
2443 if (S_ISDIR(inode->i_mode)) {
2444 rc = ll_update_lsm_md(inode, md);
2449 if (body->mbo_valid & OBD_MD_FLACL)
2450 lli_replace_acl(lli, md);
2452 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2453 sbi->ll_flags & LL_SBI_32BIT_API);
2454 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2456 if (body->mbo_valid & OBD_MD_FLATIME) {
2457 if (body->mbo_atime > inode->i_atime.tv_sec)
2458 inode->i_atime.tv_sec = body->mbo_atime;
2459 lli->lli_atime = body->mbo_atime;
2462 if (body->mbo_valid & OBD_MD_FLMTIME) {
2463 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2465 "setting ino %lu mtime from %lld to %llu\n",
2466 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2468 inode->i_mtime.tv_sec = body->mbo_mtime;
2470 lli->lli_mtime = body->mbo_mtime;
2473 if (body->mbo_valid & OBD_MD_FLCTIME) {
2474 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2475 inode->i_ctime.tv_sec = body->mbo_ctime;
2476 lli->lli_ctime = body->mbo_ctime;
2479 if (body->mbo_valid & OBD_MD_FLBTIME)
2480 lli->lli_btime = body->mbo_btime;
2482 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2483 if (body->mbo_valid & OBD_MD_FLFLAGS)
2484 ll_update_inode_flags(inode, body->mbo_flags);
2485 if (body->mbo_valid & OBD_MD_FLMODE)
2486 inode->i_mode = (inode->i_mode & S_IFMT) |
2487 (body->mbo_mode & ~S_IFMT);
2489 if (body->mbo_valid & OBD_MD_FLTYPE)
2490 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2491 (body->mbo_mode & S_IFMT);
2493 LASSERT(inode->i_mode != 0);
2494 if (body->mbo_valid & OBD_MD_FLUID)
2495 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2496 if (body->mbo_valid & OBD_MD_FLGID)
2497 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2498 if (body->mbo_valid & OBD_MD_FLPROJID)
2499 lli->lli_projid = body->mbo_projid;
2500 if (body->mbo_valid & OBD_MD_FLNLINK)
2501 set_nlink(inode, body->mbo_nlink);
2502 if (body->mbo_valid & OBD_MD_FLRDEV)
2503 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2505 if (body->mbo_valid & OBD_MD_FLID) {
2506 /* FID shouldn't be changed! */
2507 if (fid_is_sane(&lli->lli_fid)) {
2508 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2509 "Trying to change FID "DFID
2510 " to the "DFID", inode "DFID"(%p)\n",
2511 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2512 PFID(ll_inode2fid(inode)), inode);
2514 lli->lli_fid = body->mbo_fid1;
2518 LASSERT(fid_seq(&lli->lli_fid) != 0);
2520 lli->lli_attr_valid = body->mbo_valid;
2521 if (body->mbo_valid & OBD_MD_FLSIZE) {
2522 i_size_write(inode, body->mbo_size);
2524 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2525 PFID(ll_inode2fid(inode)),
2526 (unsigned long long)body->mbo_size);
2528 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2529 inode->i_blocks = body->mbo_blocks;
2531 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2532 lli->lli_lazysize = body->mbo_size;
2533 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2534 lli->lli_lazyblocks = body->mbo_blocks;
2537 if (body->mbo_valid & OBD_MD_TSTATE) {
2538 /* Set LLIF_FILE_RESTORING if restore ongoing and
2539 * clear it when done to ensure to start again
2540 * glimpsing updated attrs
2542 if (body->mbo_t_state & MS_RESTORE)
2543 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2545 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2551 int ll_read_inode2(struct inode *inode, void *opaque)
2553 struct lustre_md *md = opaque;
2554 struct ll_inode_info *lli = ll_i2info(inode);
2558 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2559 PFID(&lli->lli_fid), inode);
2561 /* Core attributes from the MDS first. This is a new inode, and
2562 * the VFS doesn't zero times in the core inode so we have to do
2563 * it ourselves. They will be overwritten by either MDS or OST
2564 * attributes - we just need to make sure they aren't newer.
2566 inode->i_mtime.tv_sec = 0;
2567 inode->i_atime.tv_sec = 0;
2568 inode->i_ctime.tv_sec = 0;
2570 rc = ll_update_inode(inode, md);
2574 /* OIDEBUG(inode); */
2576 #ifdef HAVE_BACKING_DEV_INFO
2577 /* initializing backing dev info. */
2578 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2580 if (S_ISREG(inode->i_mode)) {
2581 struct ll_sb_info *sbi = ll_i2sbi(inode);
2582 inode->i_op = &ll_file_inode_operations;
2583 inode->i_fop = sbi->ll_fop;
2584 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2586 } else if (S_ISDIR(inode->i_mode)) {
2587 inode->i_op = &ll_dir_inode_operations;
2588 inode->i_fop = &ll_dir_operations;
2590 } else if (S_ISLNK(inode->i_mode)) {
2591 inode->i_op = &ll_fast_symlink_inode_operations;
2594 inode->i_op = &ll_special_inode_operations;
2596 init_special_inode(inode, inode->i_mode,
2605 void ll_delete_inode(struct inode *inode)
2607 struct ll_inode_info *lli = ll_i2info(inode);
2608 struct address_space *mapping = &inode->i_data;
2609 unsigned long nrpages;
2610 unsigned long flags;
2614 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2615 /* It is last chance to write out dirty pages,
2616 * otherwise we may lose data while umount.
2618 * If i_nlink is 0 then just discard data. This is safe because
2619 * local inode gets i_nlink 0 from server only for the last
2620 * unlink, so that file is not opened somewhere else
2622 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2623 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2625 truncate_inode_pages_final(mapping);
2627 /* Workaround for LU-118: Note nrpages may not be totally updated when
2628 * truncate_inode_pages() returns, as there can be a page in the process
2629 * of deletion (inside __delete_from_page_cache()) in the specified
2630 * range. Thus mapping->nrpages can be non-zero when this function
2631 * returns even after truncation of the whole mapping. Only do this if
2632 * npages isn't already zero.
2634 nrpages = mapping->nrpages;
2636 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2637 nrpages = mapping->nrpages;
2638 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2639 } /* Workaround end */
2641 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2642 "see https://jira.whamcloud.com/browse/LU-118\n",
2643 ll_i2sbi(inode)->ll_fsname,
2644 PFID(ll_inode2fid(inode)), inode, nrpages);
2646 ll_clear_inode(inode);
2652 int ll_iocontrol(struct inode *inode, struct file *file,
2653 unsigned int cmd, unsigned long arg)
2655 struct ll_sb_info *sbi = ll_i2sbi(inode);
2656 struct ptlrpc_request *req = NULL;
2661 case FS_IOC_GETFLAGS: {
2662 struct mdt_body *body;
2663 struct md_op_data *op_data;
2665 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2666 0, 0, LUSTRE_OPC_ANY,
2668 if (IS_ERR(op_data))
2669 RETURN(PTR_ERR(op_data));
2671 op_data->op_valid = OBD_MD_FLFLAGS;
2672 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2673 ll_finish_md_op_data(op_data);
2675 CERROR("%s: failure inode "DFID": rc = %d\n",
2676 sbi->ll_md_exp->exp_obd->obd_name,
2677 PFID(ll_inode2fid(inode)), rc);
2681 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2683 flags = body->mbo_flags;
2685 ptlrpc_req_finished(req);
2687 RETURN(put_user(flags, (int __user *)arg));
2689 case FS_IOC_SETFLAGS: {
2691 struct md_op_data *op_data;
2692 struct cl_object *obj;
2693 struct fsxattr fa = { 0 };
2695 if (get_user(flags, (int __user *)arg))
2698 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2699 if (flags & LUSTRE_PROJINHERIT_FL)
2700 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2702 rc = ll_ioctl_check_project(inode, &fa);
2706 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2707 LUSTRE_OPC_ANY, NULL);
2708 if (IS_ERR(op_data))
2709 RETURN(PTR_ERR(op_data));
2711 op_data->op_attr_flags = flags;
2712 op_data->op_xvalid |= OP_XVALID_FLAGS;
2713 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2714 ll_finish_md_op_data(op_data);
2715 ptlrpc_req_finished(req);
2719 ll_update_inode_flags(inode, flags);
2721 obj = ll_i2info(inode)->lli_clob;
2725 OBD_ALLOC_PTR(attr);
2729 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2741 int ll_flush_ctx(struct inode *inode)
2743 struct ll_sb_info *sbi = ll_i2sbi(inode);
2745 CDEBUG(D_SEC, "flush context for user %d\n",
2746 from_kuid(&init_user_ns, current_uid()));
2748 obd_set_info_async(NULL, sbi->ll_md_exp,
2749 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2751 obd_set_info_async(NULL, sbi->ll_dt_exp,
2752 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2757 /* umount -f client means force down, don't save state */
2758 void ll_umount_begin(struct super_block *sb)
2760 struct ll_sb_info *sbi = ll_s2sbi(sb);
2761 struct obd_device *obd;
2762 struct obd_ioctl_data *ioc_data;
2766 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2767 sb->s_count, atomic_read(&sb->s_active));
2769 obd = class_exp2obd(sbi->ll_md_exp);
2771 CERROR("Invalid MDC connection handle %#llx\n",
2772 sbi->ll_md_exp->exp_handle.h_cookie);
2778 obd = class_exp2obd(sbi->ll_dt_exp);
2780 CERROR("Invalid LOV connection handle %#llx\n",
2781 sbi->ll_dt_exp->exp_handle.h_cookie);
2787 OBD_ALLOC_PTR(ioc_data);
2789 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2790 sizeof *ioc_data, ioc_data, NULL);
2792 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2793 sizeof *ioc_data, ioc_data, NULL);
2795 OBD_FREE_PTR(ioc_data);
2798 /* Really, we'd like to wait until there are no requests outstanding,
2799 * and then continue. For now, we just periodically checking for vfs
2800 * to decrement mnt_cnt and hope to finish it within 10sec.
2804 !may_umount(sbi->ll_mnt.mnt)) {
2812 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2814 struct ll_sb_info *sbi = ll_s2sbi(sb);
2815 char *profilenm = get_profile_name(sb);
2819 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2820 read_only = *flags & MS_RDONLY;
2821 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2822 sizeof(KEY_READ_ONLY),
2823 KEY_READ_ONLY, sizeof(read_only),
2826 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2827 profilenm, read_only ?
2828 "read-only" : "read-write", err);
2833 sb->s_flags |= SB_RDONLY;
2835 sb->s_flags &= ~SB_RDONLY;
2837 if (sbi->ll_flags & LL_SBI_VERBOSE)
2838 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2839 read_only ? "read-only" : "read-write");
2845 * Cleanup the open handle that is cached on MDT-side.
2847 * For open case, the client side open handling thread may hit error
2848 * after the MDT grant the open. Under such case, the client should
2849 * send close RPC to the MDT as cleanup; otherwise, the open handle
2850 * on the MDT will be leaked there until the client umount or evicted.
2852 * In further, if someone unlinked the file, because the open handle
2853 * holds the reference on such file/object, then it will block the
2854 * subsequent threads that want to locate such object via FID.
2856 * \param[in] sb super block for this file-system
2857 * \param[in] open_req pointer to the original open request
2859 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2861 struct mdt_body *body;
2862 struct md_op_data *op_data;
2863 struct ptlrpc_request *close_req = NULL;
2864 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2867 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2868 OBD_ALLOC_PTR(op_data);
2869 if (op_data == NULL) {
2870 CWARN("%s: cannot allocate op_data to release open handle for "
2871 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2876 op_data->op_fid1 = body->mbo_fid1;
2877 op_data->op_open_handle = body->mbo_open_handle;
2878 op_data->op_mod_time = ktime_get_real_seconds();
2879 md_close(exp, op_data, NULL, &close_req);
2880 ptlrpc_req_finished(close_req);
2881 ll_finish_md_op_data(op_data);
2886 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2887 struct super_block *sb, struct lookup_intent *it)
2889 struct ll_sb_info *sbi = NULL;
2890 struct lustre_md md = { NULL };
2891 bool default_lmv_deleted = false;
2896 LASSERT(*inode || sb);
2897 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2898 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2899 sbi->ll_md_exp, &md);
2904 * clear default_lmv only if intent_getattr reply doesn't contain it.
2905 * but it needs to be done after iget, check this early because
2906 * ll_update_lsm_md() may change md.
2908 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2909 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2910 default_lmv_deleted = true;
2913 rc = ll_update_inode(*inode, &md);
2917 LASSERT(sb != NULL);
2920 * At this point server returns to client's same fid as client
2921 * generated for creating. So using ->fid1 is okay here.
2923 if (!fid_is_sane(&md.body->mbo_fid1)) {
2924 CERROR("%s: Fid is insane "DFID"\n",
2926 PFID(&md.body->mbo_fid1));
2927 GOTO(out, rc = -EINVAL);
2930 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2931 sbi->ll_flags & LL_SBI_32BIT_API),
2933 if (IS_ERR(*inode)) {
2935 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2937 CERROR("new_inode -fatal: rc %d\n", rc);
2942 /* Handling piggyback layout lock.
2943 * Layout lock can be piggybacked by getattr and open request.
2944 * The lsm can be applied to inode only if it comes with a layout lock
2945 * otherwise correct layout may be overwritten, for example:
2946 * 1. proc1: mdt returns a lsm but not granting layout
2947 * 2. layout was changed by another client
2948 * 3. proc2: refresh layout and layout lock granted
2949 * 4. proc1: to apply a stale layout */
2950 if (it != NULL && it->it_lock_mode != 0) {
2951 struct lustre_handle lockh;
2952 struct ldlm_lock *lock;
2954 lockh.cookie = it->it_lock_handle;
2955 lock = ldlm_handle2lock(&lockh);
2956 LASSERT(lock != NULL);
2957 if (ldlm_has_layout(lock)) {
2958 struct cl_object_conf conf;
2960 memset(&conf, 0, sizeof(conf));
2961 conf.coc_opc = OBJECT_CONF_SET;
2962 conf.coc_inode = *inode;
2963 conf.coc_lock = lock;
2964 conf.u.coc_layout = md.layout;
2965 (void)ll_layout_conf(*inode, &conf);
2967 LDLM_LOCK_PUT(lock);
2970 if (default_lmv_deleted)
2971 ll_update_default_lsm_md(*inode, &md);
2973 /* we may want to apply some policy for foreign file/dir */
2974 if (ll_sbi_has_foreign_symlink(sbi)) {
2975 rc = ll_manage_foreign(*inode, &md);
2983 /* cleanup will be done if necessary */
2984 md_free_lustre_md(sbi->ll_md_exp, &md);
2986 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2987 ll_intent_drop_lock(it);
2988 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2994 int ll_obd_statfs(struct inode *inode, void __user *arg)
2996 struct ll_sb_info *sbi = NULL;
2997 struct obd_export *exp;
2998 struct obd_ioctl_data *data = NULL;
3003 sbi = ll_i2sbi(inode);
3005 GOTO(out_statfs, rc = -EINVAL);
3007 rc = obd_ioctl_getdata(&data, &len, arg);
3009 GOTO(out_statfs, rc);
3011 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3012 !data->ioc_pbuf1 || !data->ioc_pbuf2)
3013 GOTO(out_statfs, rc = -EINVAL);
3015 if (data->ioc_inllen1 != sizeof(__u32) ||
3016 data->ioc_inllen2 != sizeof(__u32) ||
3017 data->ioc_plen1 != sizeof(struct obd_statfs) ||
3018 data->ioc_plen2 != sizeof(struct obd_uuid))
3019 GOTO(out_statfs, rc = -EINVAL);
3021 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3022 if (type & LL_STATFS_LMV)
3023 exp = sbi->ll_md_exp;
3024 else if (type & LL_STATFS_LOV)
3025 exp = sbi->ll_dt_exp;
3027 GOTO(out_statfs, rc = -ENODEV);
3029 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3031 GOTO(out_statfs, rc);
3033 OBD_FREE_LARGE(data, len);
3038 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3039 * be called early to avoid deadlock.
3041 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3043 if (op_data->op_mea2_sem) {
3044 up_read_non_owner(op_data->op_mea2_sem);
3045 op_data->op_mea2_sem = NULL;
3048 if (op_data->op_mea1_sem) {
3049 up_read_non_owner(op_data->op_mea1_sem);
3050 op_data->op_mea1_sem = NULL;
3054 /* this function prepares md_op_data hint for passing it down to MD stack. */
3055 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3056 struct inode *i1, struct inode *i2,
3057 const char *name, size_t namelen,
3058 __u32 mode, enum md_op_code opc,
3061 LASSERT(i1 != NULL);
3064 /* Do not reuse namelen for something else. */
3066 return ERR_PTR(-EINVAL);
3068 if (namelen > ll_i2sbi(i1)->ll_namelen)
3069 return ERR_PTR(-ENAMETOOLONG);
3071 /* "/" is not valid name, but it's allowed */
3072 if (!lu_name_is_valid_2(name, namelen) &&
3073 strncmp("/", name, namelen) != 0)
3074 return ERR_PTR(-EINVAL);
3077 if (op_data == NULL)
3078 OBD_ALLOC_PTR(op_data);
3080 if (op_data == NULL)
3081 return ERR_PTR(-ENOMEM);
3083 ll_i2gids(op_data->op_suppgids, i1, i2);
3084 op_data->op_fid1 = *ll_inode2fid(i1);
3085 op_data->op_code = opc;
3087 if (S_ISDIR(i1->i_mode)) {
3088 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3089 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3090 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3091 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3095 op_data->op_fid2 = *ll_inode2fid(i2);
3096 if (S_ISDIR(i2->i_mode)) {
3098 /* i2 is typically a child of i1, and MUST be
3099 * further from the root to avoid deadlocks.
3101 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3102 op_data->op_mea2_sem =
3103 &ll_i2info(i2)->lli_lsm_sem;
3105 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3108 fid_zero(&op_data->op_fid2);
3111 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
3112 op_data->op_cli_flags |= CLI_HASH64;
3114 if (ll_need_32bit_api(ll_i2sbi(i1)))
3115 op_data->op_cli_flags |= CLI_API32;
3117 op_data->op_name = name;
3118 op_data->op_namelen = namelen;
3119 op_data->op_mode = mode;
3120 op_data->op_mod_time = ktime_get_real_seconds();
3121 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3122 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3123 op_data->op_cap = cfs_curproc_cap_pack();
3124 op_data->op_mds = 0;
3125 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3126 filename_is_volatile(name, namelen, &op_data->op_mds)) {
3127 op_data->op_bias |= MDS_CREATE_VOLATILE;
3129 op_data->op_data = data;
3134 void ll_finish_md_op_data(struct md_op_data *op_data)
3136 ll_unlock_md_op_lsm(op_data);
3137 security_release_secctx(op_data->op_file_secctx,
3138 op_data->op_file_secctx_size);
3139 llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3140 OBD_FREE_PTR(op_data);
3143 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3145 struct ll_sb_info *sbi;
3147 LASSERT(seq && dentry);
3148 sbi = ll_s2sbi(dentry->d_sb);
3150 if (sbi->ll_flags & LL_SBI_NOLCK)
3151 seq_puts(seq, ",nolock");
3153 /* "flock" is the default since 2.13, but it wasn't for many years,
3154 * so it is still useful to print this to show it is enabled.
3155 * Start to print "noflock" so it is now clear when flock is disabled.
3157 if (sbi->ll_flags & LL_SBI_FLOCK)
3158 seq_puts(seq, ",flock");
3159 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
3160 seq_puts(seq, ",localflock");
3162 seq_puts(seq, ",noflock");
3164 if (sbi->ll_flags & LL_SBI_USER_XATTR)
3165 seq_puts(seq, ",user_xattr");
3167 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
3168 seq_puts(seq, ",lazystatfs");
3170 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3171 seq_puts(seq, ",user_fid2path");
3173 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3174 seq_puts(seq, ",always_ping");
3176 if (ll_sbi_has_test_dummy_encryption(sbi))
3177 seq_puts(seq, ",test_dummy_encryption");
3179 if (ll_sbi_has_encrypt(sbi))
3180 seq_puts(seq, ",encrypt");
3182 seq_puts(seq, ",noencrypt");
3184 if (sbi->ll_flags & LL_SBI_FOREIGN_SYMLINK) {
3185 seq_puts(seq, ",foreign_symlink=");
3186 seq_puts(seq, sbi->ll_foreign_symlink_prefix);
3193 * Get obd name by cmd, and copy out to user space
3195 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3197 struct ll_sb_info *sbi = ll_i2sbi(inode);
3198 struct obd_device *obd;
3201 if (cmd == OBD_IOC_GETDTNAME)
3202 obd = class_exp2obd(sbi->ll_dt_exp);
3203 else if (cmd == OBD_IOC_GETMDNAME)
3204 obd = class_exp2obd(sbi->ll_md_exp);
3211 if (copy_to_user((void __user *)arg, obd->obd_name,
3212 strlen(obd->obd_name) + 1))
3218 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3225 p.mnt = current->fs->root.mnt;
3227 path = d_path(&p, buf, bufsize);
3232 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3234 char *buf, *path = NULL;
3235 struct dentry *dentry = NULL;
3236 struct inode *inode = page->mapping->host;
3238 /* this can be called inside spin lock so use GFP_ATOMIC. */
3239 buf = (char *)__get_free_page(GFP_ATOMIC);
3241 dentry = d_find_alias(page->mapping->host);
3243 path = ll_d_path(dentry, buf, PAGE_SIZE);
3246 /* The below message is checked in recovery-small.sh test_24b */
3248 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3249 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3250 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3251 PFID(ll_inode2fid(inode)),
3252 (path && !IS_ERR(path)) ? path : "", ioret);
3258 free_page((unsigned long)buf);
3261 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3262 struct lov_user_md **kbuf)
3264 struct lov_user_md lum;
3268 if (copy_from_user(&lum, md, sizeof(lum)))
3271 lum_size = ll_lov_user_md_size(&lum);
3275 OBD_ALLOC_LARGE(*kbuf, lum_size);
3279 if (copy_from_user(*kbuf, md, lum_size) != 0) {
3280 OBD_FREE_LARGE(*kbuf, lum_size);
3288 * Compute llite root squash state after a change of root squash
3289 * configuration setting or add/remove of a lnet nid
3291 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3293 struct root_squash_info *squash = &sbi->ll_squash;
3296 struct lnet_process_id id;
3298 /* Update norootsquash flag */
3299 spin_lock(&squash->rsi_lock);
3300 if (list_empty(&squash->rsi_nosquash_nids))
3301 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3303 /* Do not apply root squash as soon as one of our NIDs is
3304 * in the nosquash_nids list */
3307 while (LNetGetId(i++, &id) != -ENOENT) {
3308 if (id.nid == LNET_NID_LO_0)
3310 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3316 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
3318 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3320 spin_unlock(&squash->rsi_lock);
3324 * Parse linkea content to extract information about a given hardlink
3326 * \param[in] ldata - Initialized linkea data
3327 * \param[in] linkno - Link identifier
3328 * \param[out] parent_fid - The entry's parent FID
3329 * \param[out] ln - Entry name destination buffer
3331 * \retval 0 on success
3332 * \retval Appropriate negative error code on failure
3334 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3335 struct lu_fid *parent_fid, struct lu_name *ln)
3341 rc = linkea_init_with_rec(ldata);
3345 if (linkno >= ldata->ld_leh->leh_reccount)
3346 /* beyond last link */
3349 linkea_first_entry(ldata);
3350 for (idx = 0; ldata->ld_lee != NULL; idx++) {
3351 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3356 linkea_next_entry(ldata);
3366 * Get parent FID and name of an identified link. Operation is performed for
3367 * a given link number, letting the caller iterate over linkno to list one or
3368 * all links of an entry.
3370 * \param[in] file - File descriptor against which to perform the operation
3371 * \param[in,out] arg - User-filled structure containing the linkno to operate
3372 * on and the available size. It is eventually filled with
3373 * the requested information or left untouched on error
3375 * \retval - 0 on success
3376 * \retval - Appropriate negative error code on failure
3378 int ll_getparent(struct file *file, struct getparent __user *arg)
3380 struct inode *inode = file_inode(file);
3381 struct linkea_data *ldata;
3382 struct lu_buf buf = LU_BUF_NULL;
3384 struct lu_fid parent_fid;
3391 if (!cfs_capable(CAP_DAC_READ_SEARCH) &&
3392 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3395 if (get_user(name_size, &arg->gp_name_size))
3398 if (get_user(linkno, &arg->gp_linkno))
3401 if (name_size > PATH_MAX)
3404 OBD_ALLOC(ldata, sizeof(*ldata));
3408 rc = linkea_data_new(ldata, &buf);
3410 GOTO(ldata_free, rc);
3412 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3413 buf.lb_len, OBD_MD_FLXATTR);
3417 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3421 if (ln.ln_namelen >= name_size)
3422 GOTO(lb_free, rc = -EOVERFLOW);
3424 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3425 GOTO(lb_free, rc = -EFAULT);
3427 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3428 GOTO(lb_free, rc = -EFAULT);
3430 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3431 GOTO(lb_free, rc = -EFAULT);
3436 OBD_FREE(ldata, sizeof(*ldata));