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/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/security.h>
52 #include <uapi/linux/lustre/lustre_ioctl.h>
53 #ifdef HAVE_UAPI_LINUX_MOUNT_H
54 #include <uapi/linux/mount.h>
57 #include <lustre_ha.h>
58 #include <lustre_dlm.h>
59 #include <lprocfs_status.h>
60 #include <lustre_disk.h>
61 #include <uapi/linux/lustre/lustre_param.h>
62 #include <lustre_log.h>
63 #include <cl_object.h>
64 #include <obd_cksum.h>
65 #include "llite_internal.h"
67 struct kmem_cache *ll_file_data_slab;
70 #define log2(n) ffz(~(n))
73 static struct ll_sb_info *ll_init_sbi(void)
75 struct ll_sb_info *sbi = NULL;
77 unsigned long lru_page_max;
86 RETURN(ERR_PTR(-ENOMEM));
88 rc = pcc_super_init(&sbi->ll_pcc_super);
92 spin_lock_init(&sbi->ll_lock);
93 mutex_init(&sbi->ll_lco.lco_lock);
94 spin_lock_init(&sbi->ll_pp_extent_lock);
95 spin_lock_init(&sbi->ll_process_lock);
96 sbi->ll_rw_stats_on = 0;
97 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
100 pages = si.totalram - si.totalhigh;
101 lru_page_max = pages / 2;
103 sbi->ll_ra_info.ra_async_max_active = 0;
104 sbi->ll_ra_info.ll_readahead_wq =
105 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
106 sbi->ll_ra_info.ra_async_max_active);
107 if (!sbi->ll_ra_info.ll_readahead_wq)
108 GOTO(out_pcc, rc = -ENOMEM);
110 /* initialize ll_cache data */
111 sbi->ll_cache = cl_cache_init(lru_page_max);
112 if (sbi->ll_cache == NULL)
113 GOTO(out_destroy_ra, rc = -ENOMEM);
115 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
116 SBI_DEFAULT_READAHEAD_MAX);
117 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
118 sbi->ll_ra_info.ra_max_pages_per_file;
119 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
120 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
122 sbi->ll_flags |= LL_SBI_VERBOSE;
123 #ifdef ENABLE_CHECKSUM
124 sbi->ll_flags |= LL_SBI_CHECKSUM;
127 sbi->ll_flags |= LL_SBI_FLOCK;
130 #ifdef HAVE_LRU_RESIZE_SUPPORT
131 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
133 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
135 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
136 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
138 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
142 /* metadata statahead is enabled by default */
143 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
144 sbi->ll_sa_max = LL_SA_RPC_DEF;
145 atomic_set(&sbi->ll_sa_total, 0);
146 atomic_set(&sbi->ll_sa_wrong, 0);
147 atomic_set(&sbi->ll_sa_running, 0);
148 atomic_set(&sbi->ll_agl_total, 0);
149 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
150 sbi->ll_flags |= LL_SBI_FAST_READ;
151 sbi->ll_flags |= LL_SBI_TINY_WRITE;
154 sbi->ll_squash.rsi_uid = 0;
155 sbi->ll_squash.rsi_gid = 0;
156 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
157 spin_lock_init(&sbi->ll_squash.rsi_lock);
159 /* Per-filesystem file heat */
160 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
161 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
164 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
166 pcc_super_fini(&sbi->ll_pcc_super);
172 static void ll_free_sbi(struct super_block *sb)
174 struct ll_sb_info *sbi = ll_s2sbi(sb);
178 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
179 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
180 if (sbi->ll_ra_info.ll_readahead_wq)
181 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
182 if (sbi->ll_cache != NULL) {
183 cl_cache_decref(sbi->ll_cache);
184 sbi->ll_cache = NULL;
186 pcc_super_fini(&sbi->ll_pcc_super);
187 OBD_FREE(sbi, sizeof(*sbi));
192 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
194 struct inode *root = NULL;
195 struct ll_sb_info *sbi = ll_s2sbi(sb);
196 struct obd_statfs *osfs = NULL;
197 struct ptlrpc_request *request = NULL;
198 struct obd_connect_data *data = NULL;
199 struct obd_uuid *uuid;
200 struct md_op_data *op_data;
201 struct lustre_md lmd;
203 int size, err, checksum;
206 sbi->ll_md_obd = class_name2obd(md);
207 if (!sbi->ll_md_obd) {
208 CERROR("MD %s: not setup or attached\n", md);
222 /* pass client page size via ocd_grant_blkbits, the server should report
223 * back its backend blocksize for grant calculation purpose */
224 data->ocd_grant_blkbits = PAGE_SHIFT;
226 /* indicate MDT features supported by this client */
227 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
228 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
229 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
230 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
231 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
232 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
233 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
234 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
235 OBD_CONNECT_64BITHASH |
236 OBD_CONNECT_EINPROGRESS |
237 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
238 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
239 OBD_CONNECT_MAX_EASIZE |
240 OBD_CONNECT_FLOCK_DEAD |
241 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
242 OBD_CONNECT_OPEN_BY_FID |
243 OBD_CONNECT_DIR_STRIPE |
244 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
245 OBD_CONNECT_SUBTREE |
246 OBD_CONNECT_MULTIMODRPCS |
247 OBD_CONNECT_GRANT_PARAM |
248 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
250 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
251 OBD_CONNECT2_SUM_STATFS |
252 OBD_CONNECT2_OVERSTRIPING |
254 OBD_CONNECT2_LOCK_CONVERT |
255 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
256 OBD_CONNECT2_INC_XID |
258 OBD_CONNECT2_ASYNC_DISCARD |
261 #ifdef HAVE_LRU_RESIZE_SUPPORT
262 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
263 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
265 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
266 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
267 OBD_CONNECT_LARGE_ACL;
270 data->ocd_cksum_types = obd_cksum_types_supported_client();
272 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
273 /* flag mdc connection as lightweight, only used for test
274 * purpose, use with care */
275 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
277 data->ocd_ibits_known = MDS_INODELOCK_FULL;
278 data->ocd_version = LUSTRE_VERSION_CODE;
280 if (sb->s_flags & SB_RDONLY)
281 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
282 if (sbi->ll_flags & LL_SBI_USER_XATTR)
283 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
286 /* Setting this indicates we correctly support S_NOSEC (See kernel
287 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
289 sb->s_flags |= SB_NOSEC;
292 if (sbi->ll_flags & LL_SBI_FLOCK)
293 sbi->ll_fop = &ll_file_operations_flock;
294 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
295 sbi->ll_fop = &ll_file_operations;
297 sbi->ll_fop = &ll_file_operations_noflock;
299 /* always ping even if server suppress_pings */
300 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
301 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
303 obd_connect_set_secctx(data);
305 #if defined(CONFIG_SECURITY)
306 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
309 data->ocd_brw_size = MD_MAX_BRW_SIZE;
311 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
312 &sbi->ll_sb_uuid, data, sbi->ll_cache);
314 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
315 "recovery, of which this client is not a "
316 "part. Please wait for recovery to complete,"
317 " abort, or time out.\n", md);
320 CERROR("cannot connect to %s: rc = %d\n", md, err);
324 sbi->ll_md_exp->exp_connect_data = *data;
326 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
327 LUSTRE_SEQ_METADATA);
329 CERROR("%s: Can't init metadata layer FID infrastructure, "
330 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
334 /* For mount, we only need fs info from MDT0, and also in DNE, it
335 * can make sure the client can be mounted as long as MDT0 is
337 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
338 ktime_get_seconds() - sbi->ll_statfs_max_age,
339 OBD_STATFS_FOR_MDT0);
341 GOTO(out_md_fid, err);
343 /* This needs to be after statfs to ensure connect has finished.
344 * Note that "data" does NOT contain the valid connect reply.
345 * If connecting to a 1.8 server there will be no LMV device, so
346 * we can access the MDC export directly and exp_connect_flags will
347 * be non-zero, but if accessing an upgraded 2.1 server it will
348 * have the correct flags filled in.
349 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
350 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
351 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
352 valid != CLIENT_CONNECT_MDT_REQD) {
355 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
356 obd_connect_flags2str(buf, PAGE_SIZE,
357 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
358 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
359 "feature(s) needed for correct operation "
360 "of this client (%s). Please upgrade "
361 "server or downgrade client.\n",
362 sbi->ll_md_exp->exp_obd->obd_name, buf);
363 OBD_FREE(buf, PAGE_SIZE);
364 GOTO(out_md_fid, err = -EPROTO);
367 size = sizeof(*data);
368 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
369 KEY_CONN_DATA, &size, data);
371 CERROR("%s: Get connect data failed: rc = %d\n",
372 sbi->ll_md_exp->exp_obd->obd_name, err);
373 GOTO(out_md_fid, err);
376 LASSERT(osfs->os_bsize);
377 sb->s_blocksize = osfs->os_bsize;
378 sb->s_blocksize_bits = log2(osfs->os_bsize);
379 sb->s_magic = LL_SUPER_MAGIC;
380 sb->s_maxbytes = MAX_LFS_FILESIZE;
381 sbi->ll_namelen = osfs->os_namelen;
382 sbi->ll_mnt.mnt = current->fs->root.mnt;
384 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
385 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
386 LCONSOLE_INFO("Disabling user_xattr feature because "
387 "it is not supported on the server\n");
388 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
391 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
393 sb->s_flags |= SB_POSIXACL;
395 sbi->ll_flags |= LL_SBI_ACL;
397 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
399 sb->s_flags &= ~SB_POSIXACL;
401 sbi->ll_flags &= ~LL_SBI_ACL;
404 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
405 sbi->ll_flags |= LL_SBI_64BIT_HASH;
407 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
408 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
410 if (obd_connect_has_secctx(data))
411 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
413 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
414 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
415 LCONSOLE_INFO("%s: disabling xattr cache due to "
416 "unknown maximum xattr size.\n", dt);
417 } else if (!sbi->ll_xattr_cache_set) {
418 /* If xattr_cache is already set (no matter 0 or 1)
419 * during processing llog, it won't be enabled here. */
420 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
421 sbi->ll_xattr_cache_enabled = 1;
425 sbi->ll_dt_obd = class_name2obd(dt);
426 if (!sbi->ll_dt_obd) {
427 CERROR("DT %s: not setup or attached\n", dt);
428 GOTO(out_md_fid, err = -ENODEV);
431 /* pass client page size via ocd_grant_blkbits, the server should report
432 * back its backend blocksize for grant calculation purpose */
433 data->ocd_grant_blkbits = PAGE_SHIFT;
435 /* indicate OST features supported by this client */
436 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
437 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
438 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
439 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
440 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
441 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
442 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
443 OBD_CONNECT_EINPROGRESS |
444 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
445 OBD_CONNECT_LAYOUTLOCK |
446 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
447 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
448 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
450 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
451 * can interoperate with an older version of lockahead which was released prior
452 * to landing in master. This support will be dropped when 2.13 development
453 * starts. At the point, we should not just drop the connect flag (below), we
454 * should also remove the support in the code.
456 * Removing it means a few things:
457 * 1. Remove this section here
458 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
459 * 3. Remove function exp_connect_lockahead_old
460 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
462 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
463 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
466 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
467 OBD_CONNECT2_INC_XID;
469 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
470 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
472 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
473 * disabled by default, because it can still be enabled on the
474 * fly via /sys. As a consequence, we still need to come to an
475 * agreement on the supported algorithms at connect time
477 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
479 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
480 data->ocd_cksum_types = OBD_CKSUM_ADLER;
482 data->ocd_cksum_types = obd_cksum_types_supported_client();
484 #ifdef HAVE_LRU_RESIZE_SUPPORT
485 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
487 /* always ping even if server suppress_pings */
488 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
489 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
491 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
492 "ocd_grant: %d\n", data->ocd_connect_flags,
493 data->ocd_version, data->ocd_grant);
495 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
496 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
498 data->ocd_brw_size = DT_MAX_BRW_SIZE;
500 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
501 &sbi->ll_sb_uuid, data, sbi->ll_cache);
503 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
504 "recovery, of which this client is not a "
505 "part. Please wait for recovery to "
506 "complete, abort, or time out.\n", dt);
509 CERROR("%s: Cannot connect to %s: rc = %d\n",
510 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
514 sbi->ll_dt_exp->exp_connect_data = *data;
516 /* Don't change value if it was specified in the config log */
517 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
518 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
519 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
520 (data->ocd_brw_size >> PAGE_SHIFT));
521 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
522 sbi->ll_ra_info.ra_max_pages_per_file)
523 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
524 sbi->ll_ra_info.ra_max_pages_per_file;
527 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
528 LUSTRE_SEQ_METADATA);
530 CERROR("%s: Can't init data layer FID infrastructure, "
531 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
535 mutex_lock(&sbi->ll_lco.lco_lock);
536 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
537 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
538 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
539 mutex_unlock(&sbi->ll_lco.lco_lock);
541 fid_zero(&sbi->ll_root_fid);
542 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
545 CERROR("cannot mds_connect: rc = %d\n", err);
546 GOTO(out_lock_cn_cb, err);
548 if (!fid_is_sane(&sbi->ll_root_fid)) {
549 CERROR("%s: Invalid root fid "DFID" during mount\n",
550 sbi->ll_md_exp->exp_obd->obd_name,
551 PFID(&sbi->ll_root_fid));
552 GOTO(out_lock_cn_cb, err = -EINVAL);
554 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
556 sb->s_op = &lustre_super_operations;
557 sb->s_xattr = ll_xattr_handlers;
558 #if THREAD_SIZE >= 8192 /*b=17630*/
559 sb->s_export_op = &lustre_export_operations;
563 * XXX: move this to after cbd setup? */
564 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
565 if (sbi->ll_flags & LL_SBI_ACL)
566 valid |= OBD_MD_FLACL;
568 OBD_ALLOC_PTR(op_data);
570 GOTO(out_lock_cn_cb, err = -ENOMEM);
572 op_data->op_fid1 = sbi->ll_root_fid;
573 op_data->op_mode = 0;
574 op_data->op_valid = valid;
576 err = md_getattr(sbi->ll_md_exp, op_data, &request);
578 OBD_FREE_PTR(op_data);
580 CERROR("%s: md_getattr failed for root: rc = %d\n",
581 sbi->ll_md_exp->exp_obd->obd_name, err);
582 GOTO(out_lock_cn_cb, err);
585 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
586 sbi->ll_md_exp, &lmd);
588 CERROR("failed to understand root inode md: rc = %d\n", err);
589 ptlrpc_req_finished(request);
590 GOTO(out_lock_cn_cb, err);
593 LASSERT(fid_is_sane(&sbi->ll_root_fid));
594 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
595 sbi->ll_flags & LL_SBI_32BIT_API),
597 md_free_lustre_md(sbi->ll_md_exp, &lmd);
598 ptlrpc_req_finished(request);
601 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
603 posix_acl_release(lmd.posix_acl);
604 lmd.posix_acl = NULL;
607 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
609 CERROR("%s: bad ll_iget() for root: rc = %d\n",
610 sbi->ll_fsname, err);
614 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
615 if (sbi->ll_checksum_set) {
616 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
617 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
618 sizeof(checksum), &checksum, NULL);
620 CERROR("%s: Set checksum failed: rc = %d\n",
621 sbi->ll_dt_exp->exp_obd->obd_name, err);
627 sb->s_root = d_make_root(root);
628 if (sb->s_root == NULL) {
630 CERROR("%s: can't make root dentry: rc = %d\n",
631 sbi->ll_fsname, err);
635 sbi->ll_sdev_orig = sb->s_dev;
637 /* We set sb->s_dev equal on all lustre clients in order to support
638 * NFS export clustering. NFSD requires that the FSID be the same
640 /* s_dev is also used in lt_compare() to compare two fs, but that is
641 * only a node-local comparison. */
642 uuid = obd_get_uuid(sbi->ll_md_exp);
644 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
651 if (sbi->ll_dt_obd) {
652 err = sysfs_create_link(&sbi->ll_kset.kobj,
653 &sbi->ll_dt_obd->obd_kset.kobj,
654 sbi->ll_dt_obd->obd_type->typ_name);
656 CERROR("%s: could not register %s in llite: rc = %d\n",
657 dt, sbi->ll_fsname, err);
662 if (sbi->ll_md_obd) {
663 err = sysfs_create_link(&sbi->ll_kset.kobj,
664 &sbi->ll_md_obd->obd_kset.kobj,
665 sbi->ll_md_obd->obd_type->typ_name);
667 CERROR("%s: could not register %s in llite: rc = %d\n",
668 md, sbi->ll_fsname, err);
678 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
680 obd_disconnect(sbi->ll_dt_exp);
681 sbi->ll_dt_exp = NULL;
682 sbi->ll_dt_obd = NULL;
684 obd_fid_fini(sbi->ll_md_exp->exp_obd);
686 obd_disconnect(sbi->ll_md_exp);
687 sbi->ll_md_exp = NULL;
688 sbi->ll_md_obd = NULL;
697 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
701 size = sizeof(*lmmsize);
702 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
703 KEY_MAX_EASIZE, &size, lmmsize);
705 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
706 sbi->ll_dt_exp->exp_obd->obd_name, rc);
710 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
713 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
714 KEY_MAX_EASIZE, &size, lmmsize);
716 CERROR("Get max mdsize error rc %d\n", rc);
718 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
724 * Get the value of the default_easize parameter.
726 * \see client_obd::cl_default_mds_easize
728 * \param[in] sbi superblock info for this filesystem
729 * \param[out] lmmsize pointer to storage location for value
731 * \retval 0 on success
732 * \retval negative negated errno on failure
734 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
739 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
740 KEY_DEFAULT_EASIZE, &size, lmmsize);
742 CERROR("Get default mdsize error rc %d\n", rc);
748 * Set the default_easize parameter to the given value.
750 * \see client_obd::cl_default_mds_easize
752 * \param[in] sbi superblock info for this filesystem
753 * \param[in] lmmsize the size to set
755 * \retval 0 on success
756 * \retval negative negated errno on failure
758 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
762 if (lmmsize < sizeof(struct lov_mds_md) ||
763 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
766 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
767 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
768 sizeof(int), &lmmsize, NULL);
773 static void client_common_put_super(struct super_block *sb)
775 struct ll_sb_info *sbi = ll_s2sbi(sb);
780 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
781 obd_disconnect(sbi->ll_dt_exp);
782 sbi->ll_dt_exp = NULL;
784 ll_debugfs_unregister_super(sb);
786 obd_fid_fini(sbi->ll_md_exp->exp_obd);
787 obd_disconnect(sbi->ll_md_exp);
788 sbi->ll_md_exp = NULL;
793 void ll_kill_super(struct super_block *sb)
795 struct ll_sb_info *sbi;
799 if (!(sb->s_flags & SB_ACTIVE))
803 /* we need restore s_dev from changed for clustred NFS before put_super
804 * because new kernels have cached s_dev and change sb->s_dev in
805 * put_super not affected real removing devices */
807 sb->s_dev = sbi->ll_sdev_orig;
809 /* wait running statahead threads to quit */
810 while (atomic_read(&sbi->ll_sa_running) > 0) {
811 set_current_state(TASK_UNINTERRUPTIBLE);
812 schedule_timeout(cfs_time_seconds(1) >> 3);
819 static inline int ll_set_opt(const char *opt, char *data, int fl)
821 if (strncmp(opt, data, strlen(opt)) != 0)
827 /* non-client-specific mount options are parsed in lmd_parse */
828 static int ll_options(char *options, struct ll_sb_info *sbi)
831 char *s1 = options, *s2;
832 int *flags = &sbi->ll_flags;
838 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
841 CDEBUG(D_SUPER, "next opt=%s\n", s1);
842 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
847 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
849 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
852 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
854 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
857 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
862 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
867 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
872 tmp = ll_set_opt("context", s1, 1);
875 tmp = ll_set_opt("fscontext", s1, 1);
878 tmp = ll_set_opt("defcontext", s1, 1);
881 tmp = ll_set_opt("rootcontext", s1, 1);
884 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
889 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
895 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
898 sbi->ll_checksum_set = 1;
901 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
904 sbi->ll_checksum_set = 1;
907 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
912 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
917 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
922 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
927 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
932 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
937 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
942 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
947 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
953 s2 = strchr(s1, ',');
961 void ll_lli_init(struct ll_inode_info *lli)
963 lli->lli_inode_magic = LLI_INODE_MAGIC;
965 spin_lock_init(&lli->lli_lock);
966 lli->lli_posix_acl = NULL;
967 /* Do not set lli_fid, it has been initialized already. */
968 fid_zero(&lli->lli_pfid);
969 lli->lli_mds_read_och = NULL;
970 lli->lli_mds_write_och = NULL;
971 lli->lli_mds_exec_och = NULL;
972 lli->lli_open_fd_read_count = 0;
973 lli->lli_open_fd_write_count = 0;
974 lli->lli_open_fd_exec_count = 0;
975 mutex_init(&lli->lli_och_mutex);
976 spin_lock_init(&lli->lli_agl_lock);
977 spin_lock_init(&lli->lli_layout_lock);
978 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
979 lli->lli_clob = NULL;
981 init_rwsem(&lli->lli_xattrs_list_rwsem);
982 mutex_init(&lli->lli_xattrs_enq_lock);
984 LASSERT(lli->lli_vfs_inode.i_mode != 0);
985 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
986 lli->lli_opendir_key = NULL;
988 spin_lock_init(&lli->lli_sa_lock);
989 lli->lli_opendir_pid = 0;
990 lli->lli_sa_enabled = 0;
991 init_rwsem(&lli->lli_lsm_sem);
993 mutex_init(&lli->lli_size_mutex);
994 lli->lli_symlink_name = NULL;
995 ll_trunc_sem_init(&lli->lli_trunc_sem);
996 range_lock_tree_init(&lli->lli_write_tree);
997 init_rwsem(&lli->lli_glimpse_sem);
998 lli->lli_glimpse_time = ktime_set(0, 0);
999 INIT_LIST_HEAD(&lli->lli_agl_list);
1000 lli->lli_agl_index = 0;
1001 lli->lli_async_rc = 0;
1002 spin_lock_init(&lli->lli_heat_lock);
1003 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1004 lli->lli_heat_flags = 0;
1005 mutex_init(&lli->lli_pcc_lock);
1006 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1007 lli->lli_pcc_inode = NULL;
1008 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1009 lli->lli_pcc_generation = 0;
1010 mutex_init(&lli->lli_group_mutex);
1011 lli->lli_group_users = 0;
1012 lli->lli_group_gid = 0;
1014 mutex_init(&lli->lli_layout_mutex);
1015 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1018 #define MAX_STRING_SIZE 128
1020 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1022 #define LSI_BDI_INITIALIZED 0x00400000
1024 #ifndef HAVE_BDI_CAP_MAP_COPY
1025 # define BDI_CAP_MAP_COPY 0
1028 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1030 struct lustre_sb_info *lsi = s2lsi(sb);
1031 char buf[MAX_STRING_SIZE];
1035 err = bdi_init(&lsi->lsi_bdi);
1039 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1040 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1041 lsi->lsi_bdi.name = "lustre";
1042 va_start(args, fmt);
1043 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1045 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1048 sb->s_bdi = &lsi->lsi_bdi;
1052 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1054 int ll_fill_super(struct super_block *sb)
1056 struct lustre_profile *lprof = NULL;
1057 struct lustre_sb_info *lsi = s2lsi(sb);
1058 struct ll_sb_info *sbi = NULL;
1059 char *dt = NULL, *md = NULL;
1060 char *profilenm = get_profile_name(sb);
1061 struct config_llog_instance *cfg;
1062 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1063 const int instlen = LUSTRE_MAXINSTANCE + 2;
1064 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1065 char name[MAX_STRING_SIZE];
1074 /* for ASLR, to map between cfg_instance and hashed ptr */
1075 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1076 profilenm, cfg_instance, sb);
1080 GOTO(out_free_cfg, err = -ENOMEM);
1082 /* client additional sb info */
1083 lsi->lsi_llsbi = sbi = ll_init_sbi();
1085 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1087 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1089 GOTO(out_free_cfg, err);
1091 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1092 sb->s_d_op = &ll_d_ops;
1095 generate_random_uuid(uuid.b);
1096 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1098 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1101 len = strlen(profilenm);
1102 ptr = strrchr(profilenm, '-');
1103 if (ptr && (strcmp(ptr, "-client") == 0))
1106 if (len > LUSTRE_MAXFSNAME) {
1107 if (unlikely(len >= MAX_STRING_SIZE))
1108 len = MAX_STRING_SIZE - 1;
1109 strncpy(name, profilenm, len);
1111 err = -ENAMETOOLONG;
1112 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1113 name, LUSTRE_MAXFSNAME, err);
1114 GOTO(out_free_cfg, err);
1116 strncpy(sbi->ll_fsname, profilenm, len);
1117 sbi->ll_fsname[len] = '\0';
1120 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1121 profilenm, cfg_instance);
1123 err = super_setup_bdi_name(sb, "%s", name);
1125 GOTO(out_free_cfg, err);
1127 /* Call ll_debugfs_register_super() before lustre_process_log()
1128 * so that "llite.*.*" params can be processed correctly.
1130 err = ll_debugfs_register_super(sb, name);
1132 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1133 sbi->ll_fsname, err);
1137 /* The cfg_instance is a value unique to this super, in case some
1138 * joker tries to mount the same fs at two mount points.
1140 cfg->cfg_instance = cfg_instance;
1141 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1142 cfg->cfg_callback = class_config_llog_handler;
1143 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1144 /* set up client obds */
1145 err = lustre_process_log(sb, profilenm, cfg);
1147 GOTO(out_debugfs, err);
1149 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1150 lprof = class_get_profile(profilenm);
1151 if (lprof == NULL) {
1152 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1153 " read from the MGS. Does that filesystem "
1154 "exist?\n", profilenm);
1155 GOTO(out_debugfs, err = -EINVAL);
1157 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1158 lprof->lp_md, lprof->lp_dt);
1160 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1161 OBD_ALLOC(dt, dt_len);
1163 GOTO(out_profile, err = -ENOMEM);
1164 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1166 md_len = strlen(lprof->lp_md) + instlen + 2;
1167 OBD_ALLOC(md, md_len);
1169 GOTO(out_free_dt, err = -ENOMEM);
1170 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1172 /* connections, registrations, sb setup */
1173 err = client_common_fill_super(sb, md, dt);
1175 GOTO(out_free_md, err);
1177 sbi->ll_client_common_fill_super_succeeded = 1;
1181 OBD_FREE(md, md_len);
1184 OBD_FREE(dt, dt_len);
1187 class_put_profile(lprof);
1190 ll_debugfs_unregister_super(sb);
1197 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1198 LCONSOLE_WARN("Mounted %s\n", profilenm);
1200 } /* ll_fill_super */
1202 void ll_put_super(struct super_block *sb)
1204 struct config_llog_instance cfg, params_cfg;
1205 struct obd_device *obd;
1206 struct lustre_sb_info *lsi = s2lsi(sb);
1207 struct ll_sb_info *sbi = ll_s2sbi(sb);
1208 char *profilenm = get_profile_name(sb);
1209 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1211 int next, force = 1, rc = 0;
1215 GOTO(out_no_sbi, 0);
1217 /* Should replace instance_id with something better for ASLR */
1218 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1219 profilenm, cfg_instance, sb);
1221 cfg.cfg_instance = cfg_instance;
1222 lustre_end_log(sb, profilenm, &cfg);
1224 params_cfg.cfg_instance = cfg_instance;
1225 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1227 if (sbi->ll_md_exp) {
1228 obd = class_exp2obd(sbi->ll_md_exp);
1230 force = obd->obd_force;
1233 /* Wait for unstable pages to be committed to stable storage */
1235 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1236 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1237 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1241 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1242 if (force == 0 && rc != -EINTR)
1243 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1245 /* We need to set force before the lov_disconnect in
1246 * lustre_common_put_super, since l_d cleans up osc's as well.
1250 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1252 obd->obd_force = force;
1256 if (sbi->ll_client_common_fill_super_succeeded) {
1257 /* Only if client_common_fill_super succeeded */
1258 client_common_put_super(sb);
1262 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1263 class_manual_cleanup(obd);
1265 if (sbi->ll_flags & LL_SBI_VERBOSE)
1266 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1269 class_del_profile(profilenm);
1271 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1272 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1273 bdi_destroy(&lsi->lsi_bdi);
1274 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1279 lsi->lsi_llsbi = NULL;
1281 lustre_common_put_super(sb);
1283 cl_env_cache_purge(~0);
1285 module_put(THIS_MODULE);
1288 } /* client_put_super */
1290 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1292 struct inode *inode = NULL;
1294 /* NOTE: we depend on atomic igrab() -bzzz */
1295 lock_res_and_lock(lock);
1296 if (lock->l_resource->lr_lvb_inode) {
1297 struct ll_inode_info * lli;
1298 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1299 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1300 inode = igrab(lock->l_resource->lr_lvb_inode);
1302 inode = lock->l_resource->lr_lvb_inode;
1303 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1304 D_WARNING, lock, "lr_lvb_inode %p is "
1305 "bogus: magic %08x",
1306 lock->l_resource->lr_lvb_inode,
1307 lli->lli_inode_magic);
1311 unlock_res_and_lock(lock);
1315 void ll_dir_clear_lsm_md(struct inode *inode)
1317 struct ll_inode_info *lli = ll_i2info(inode);
1319 LASSERT(S_ISDIR(inode->i_mode));
1321 if (lli->lli_lsm_md) {
1322 lmv_free_memmd(lli->lli_lsm_md);
1323 lli->lli_lsm_md = NULL;
1326 if (lli->lli_default_lsm_md) {
1327 lmv_free_memmd(lli->lli_default_lsm_md);
1328 lli->lli_default_lsm_md = NULL;
1332 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1333 const struct lu_fid *fid,
1334 struct lustre_md *md)
1336 struct ll_sb_info *sbi = ll_s2sbi(sb);
1337 struct mdt_body *body = md->body;
1338 struct inode *inode;
1342 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1343 inode = iget_locked(sb, ino);
1344 if (inode == NULL) {
1345 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1346 sbi->ll_fsname, PFID(fid));
1347 RETURN(ERR_PTR(-ENOENT));
1350 if (inode->i_state & I_NEW) {
1351 struct ll_inode_info *lli = ll_i2info(inode);
1352 struct lmv_stripe_md *lsm = md->lmv;
1354 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1355 (body->mbo_mode & S_IFMT);
1356 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1359 inode->i_mtime.tv_sec = 0;
1360 inode->i_atime.tv_sec = 0;
1361 inode->i_ctime.tv_sec = 0;
1364 #ifdef HAVE_BACKING_DEV_INFO
1365 /* initializing backing dev info. */
1366 inode->i_mapping->backing_dev_info =
1367 &s2lsi(inode->i_sb)->lsi_bdi;
1369 inode->i_op = &ll_dir_inode_operations;
1370 inode->i_fop = &ll_dir_operations;
1371 lli->lli_fid = *fid;
1374 LASSERT(lsm != NULL);
1375 /* master object FID */
1376 lli->lli_pfid = body->mbo_fid1;
1377 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1378 lli, PFID(fid), PFID(&lli->lli_pfid));
1379 unlock_new_inode(inode);
1385 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1388 struct lmv_stripe_md *lsm = md->lmv;
1389 struct ll_inode_info *lli = ll_i2info(inode);
1392 LASSERT(lsm != NULL);
1394 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1395 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1396 lsm_md_dump(D_INODE, lsm);
1398 if (!lmv_dir_striped(lsm))
1401 /* XXX sigh, this lsm_root initialization should be in
1402 * LMV layer, but it needs ll_iget right now, so we
1403 * put this here right now. */
1404 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1405 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1406 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1408 if (!fid_is_sane(fid))
1411 /* Unfortunately ll_iget will call ll_update_inode,
1412 * where the initialization of slave inode is slightly
1413 * different, so it reset lsm_md to NULL to avoid
1414 * initializing lsm for slave inode. */
1415 lsm->lsm_md_oinfo[i].lmo_root =
1416 ll_iget_anon_dir(inode->i_sb, fid, md);
1417 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1418 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1420 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1422 iput(lsm->lsm_md_oinfo[i].lmo_root);
1423 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1429 lli->lli_lsm_md = lsm;
1434 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1436 struct ll_inode_info *lli = ll_i2info(inode);
1438 if (!md->default_lmv) {
1439 /* clear default lsm */
1440 if (lli->lli_default_lsm_md) {
1441 down_write(&lli->lli_lsm_sem);
1442 if (lli->lli_default_lsm_md) {
1443 lmv_free_memmd(lli->lli_default_lsm_md);
1444 lli->lli_default_lsm_md = NULL;
1446 up_write(&lli->lli_lsm_sem);
1448 } else if (lli->lli_default_lsm_md) {
1449 /* update default lsm if it changes */
1450 down_read(&lli->lli_lsm_sem);
1451 if (lli->lli_default_lsm_md &&
1452 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1453 up_read(&lli->lli_lsm_sem);
1454 down_write(&lli->lli_lsm_sem);
1455 if (lli->lli_default_lsm_md)
1456 lmv_free_memmd(lli->lli_default_lsm_md);
1457 lli->lli_default_lsm_md = md->default_lmv;
1458 lsm_md_dump(D_INODE, md->default_lmv);
1459 md->default_lmv = NULL;
1460 up_write(&lli->lli_lsm_sem);
1462 up_read(&lli->lli_lsm_sem);
1465 /* init default lsm */
1466 down_write(&lli->lli_lsm_sem);
1467 lli->lli_default_lsm_md = md->default_lmv;
1468 lsm_md_dump(D_INODE, md->default_lmv);
1469 md->default_lmv = NULL;
1470 up_write(&lli->lli_lsm_sem);
1474 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1476 struct ll_inode_info *lli = ll_i2info(inode);
1477 struct lmv_stripe_md *lsm = md->lmv;
1478 struct cl_attr *attr;
1483 LASSERT(S_ISDIR(inode->i_mode));
1484 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1485 PFID(ll_inode2fid(inode)));
1487 /* update default LMV */
1488 if (md->default_lmv)
1489 ll_update_default_lsm_md(inode, md);
1492 * no striped information from request, lustre_md from req does not
1493 * include stripeEA, see ll_md_setattr()
1499 * normally dir layout doesn't change, only take read lock to check
1500 * that to avoid blocking other MD operations.
1502 down_read(&lli->lli_lsm_sem);
1504 /* some current lookup initialized lsm, and unchanged */
1505 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1506 GOTO(unlock, rc = 0);
1508 /* if dir layout doesn't match, check whether version is increased,
1509 * which means layout is changed, this happens in dir split/merge and
1512 * foreign LMV should not change.
1514 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1515 lsm->lsm_md_layout_version <=
1516 lli->lli_lsm_md->lsm_md_layout_version) {
1517 CERROR("%s: "DFID" dir layout mismatch:\n",
1518 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1519 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1520 lsm_md_dump(D_ERROR, lsm);
1521 GOTO(unlock, rc = -EINVAL);
1524 up_read(&lli->lli_lsm_sem);
1525 down_write(&lli->lli_lsm_sem);
1526 /* clear existing lsm */
1527 if (lli->lli_lsm_md) {
1528 lmv_free_memmd(lli->lli_lsm_md);
1529 lli->lli_lsm_md = NULL;
1532 rc = ll_init_lsm_md(inode, md);
1533 up_write(&lli->lli_lsm_sem);
1538 /* set md->lmv to NULL, so the following free lustre_md will not free
1543 /* md_merge_attr() may take long, since lsm is already set, switch to
1546 down_read(&lli->lli_lsm_sem);
1548 if (!lmv_dir_striped(lli->lli_lsm_md))
1549 GOTO(unlock, rc = 0);
1551 OBD_ALLOC_PTR(attr);
1553 GOTO(unlock, rc = -ENOMEM);
1555 /* validate the lsm */
1556 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1557 ll_md_blocking_ast);
1559 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1560 md->body->mbo_nlink = attr->cat_nlink;
1561 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1562 md->body->mbo_size = attr->cat_size;
1563 if (md->body->mbo_valid & OBD_MD_FLATIME)
1564 md->body->mbo_atime = attr->cat_atime;
1565 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1566 md->body->mbo_ctime = attr->cat_ctime;
1567 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1568 md->body->mbo_mtime = attr->cat_mtime;
1574 up_read(&lli->lli_lsm_sem);
1579 void ll_clear_inode(struct inode *inode)
1581 struct ll_inode_info *lli = ll_i2info(inode);
1582 struct ll_sb_info *sbi = ll_i2sbi(inode);
1586 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1587 PFID(ll_inode2fid(inode)), inode);
1589 if (S_ISDIR(inode->i_mode)) {
1590 /* these should have been cleared in ll_file_release */
1591 LASSERT(lli->lli_opendir_key == NULL);
1592 LASSERT(lli->lli_sai == NULL);
1593 LASSERT(lli->lli_opendir_pid == 0);
1595 pcc_inode_free(inode);
1598 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1600 LASSERT(!lli->lli_open_fd_write_count);
1601 LASSERT(!lli->lli_open_fd_read_count);
1602 LASSERT(!lli->lli_open_fd_exec_count);
1604 if (lli->lli_mds_write_och)
1605 ll_md_real_close(inode, FMODE_WRITE);
1606 if (lli->lli_mds_exec_och)
1607 ll_md_real_close(inode, FMODE_EXEC);
1608 if (lli->lli_mds_read_och)
1609 ll_md_real_close(inode, FMODE_READ);
1611 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1612 OBD_FREE(lli->lli_symlink_name,
1613 strlen(lli->lli_symlink_name) + 1);
1614 lli->lli_symlink_name = NULL;
1617 ll_xattr_cache_destroy(inode);
1619 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1620 forget_all_cached_acls(inode);
1621 if (lli->lli_posix_acl) {
1622 posix_acl_release(lli->lli_posix_acl);
1623 lli->lli_posix_acl = NULL;
1626 lli->lli_inode_magic = LLI_INODE_DEAD;
1628 if (S_ISDIR(inode->i_mode))
1629 ll_dir_clear_lsm_md(inode);
1630 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1631 LASSERT(list_empty(&lli->lli_agl_list));
1634 * XXX This has to be done before lsm is freed below, because
1635 * cl_object still uses inode lsm.
1637 cl_inode_fini(inode);
1642 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1644 struct lustre_md md;
1645 struct inode *inode = dentry->d_inode;
1646 struct ll_sb_info *sbi = ll_i2sbi(inode);
1647 struct ptlrpc_request *request = NULL;
1651 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1652 LUSTRE_OPC_ANY, NULL);
1653 if (IS_ERR(op_data))
1654 RETURN(PTR_ERR(op_data));
1656 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1658 ptlrpc_req_finished(request);
1659 if (rc == -ENOENT) {
1661 /* Unlinked special device node? Or just a race?
1662 * Pretend we done everything. */
1663 if (!S_ISREG(inode->i_mode) &&
1664 !S_ISDIR(inode->i_mode)) {
1665 ia_valid = op_data->op_attr.ia_valid;
1666 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1667 rc = simple_setattr(dentry, &op_data->op_attr);
1668 op_data->op_attr.ia_valid = ia_valid;
1670 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1671 CERROR("md_setattr fails: rc = %d\n", rc);
1676 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1677 sbi->ll_md_exp, &md);
1679 ptlrpc_req_finished(request);
1683 ia_valid = op_data->op_attr.ia_valid;
1684 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1685 * cache is not cleared yet. */
1686 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1687 if (S_ISREG(inode->i_mode))
1689 rc = simple_setattr(dentry, &op_data->op_attr);
1690 if (S_ISREG(inode->i_mode))
1691 inode_unlock(inode);
1692 op_data->op_attr.ia_valid = ia_valid;
1694 rc = ll_update_inode(inode, &md);
1695 ptlrpc_req_finished(request);
1700 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1701 * object(s) determine the file size and mtime. Otherwise, the MDS will
1702 * keep these values until such a time that objects are allocated for it.
1703 * We do the MDS operations first, as it is checking permissions for us.
1704 * We don't to the MDS RPC if there is nothing that we want to store there,
1705 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1706 * going to do an RPC anyways.
1708 * If we are doing a truncate, we will send the mtime and ctime updates
1709 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1710 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1713 * In case of HSMimport, we only set attr on MDS.
1715 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1716 enum op_xvalid xvalid, bool hsm_import)
1718 struct inode *inode = dentry->d_inode;
1719 struct ll_inode_info *lli = ll_i2info(inode);
1720 struct md_op_data *op_data = NULL;
1721 ktime_t kstart = ktime_get();
1726 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1727 "valid %x, hsm_import %d\n",
1728 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1729 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1732 if (attr->ia_valid & ATTR_SIZE) {
1733 /* Check new size against VFS/VM file size limit and rlimit */
1734 rc = inode_newsize_ok(inode, attr->ia_size);
1738 /* The maximum Lustre file size is variable, based on the
1739 * OST maximum object size and number of stripes. This
1740 * needs another check in addition to the VFS check above. */
1741 if (attr->ia_size > ll_file_maxbytes(inode)) {
1742 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1743 PFID(&lli->lli_fid), attr->ia_size,
1744 ll_file_maxbytes(inode));
1748 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1751 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1752 if (attr->ia_valid & TIMES_SET_FLAGS) {
1753 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1754 !cfs_capable(CFS_CAP_FOWNER))
1758 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1759 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1760 (attr->ia_valid & ATTR_CTIME)) {
1761 attr->ia_ctime = current_time(inode);
1762 xvalid |= OP_XVALID_CTIME_SET;
1764 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1765 (attr->ia_valid & ATTR_ATIME)) {
1766 attr->ia_atime = current_time(inode);
1767 attr->ia_valid |= ATTR_ATIME_SET;
1769 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1770 (attr->ia_valid & ATTR_MTIME)) {
1771 attr->ia_mtime = current_time(inode);
1772 attr->ia_valid |= ATTR_MTIME_SET;
1775 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1776 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1777 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1778 ktime_get_real_seconds());
1780 if (S_ISREG(inode->i_mode)) {
1781 if (attr->ia_valid & ATTR_SIZE)
1782 inode_dio_write_done(inode);
1783 inode_unlock(inode);
1786 /* We always do an MDS RPC, even if we're only changing the size;
1787 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1789 OBD_ALLOC_PTR(op_data);
1790 if (op_data == NULL)
1791 GOTO(out, rc = -ENOMEM);
1793 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1794 /* If we are changing file size, file content is
1795 * modified, flag it.
1797 xvalid |= OP_XVALID_OWNEROVERRIDE;
1798 op_data->op_bias |= MDS_DATA_MODIFIED;
1799 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1802 if (attr->ia_valid & ATTR_FILE) {
1803 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1805 if (fd->fd_lease_och)
1806 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1809 op_data->op_attr = *attr;
1810 op_data->op_xvalid = xvalid;
1812 rc = ll_md_setattr(dentry, op_data);
1816 if (!S_ISREG(inode->i_mode) || hsm_import)
1819 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1820 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1821 xvalid & OP_XVALID_CTIME_SET) {
1822 bool cached = false;
1824 rc = pcc_inode_setattr(inode, attr, &cached);
1827 CERROR("%s: PCC inode "DFID" setattr failed: "
1829 ll_i2sbi(inode)->ll_fsname,
1830 PFID(&lli->lli_fid), rc);
1834 /* For truncate and utimes sending attributes to OSTs,
1835 * setting mtime/atime to the past will be performed
1836 * under PW [0:EOF] extent lock (new_size:EOF for
1837 * truncate). It may seem excessive to send mtime/atime
1838 * updates to OSTs when not setting times to past, but
1839 * it is necessary due to possible time
1840 * de-synchronization between MDT inode and OST objects
1842 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1846 /* If the file was restored, it needs to set dirty flag.
1848 * We've already sent MDS_DATA_MODIFIED flag in
1849 * ll_md_setattr() for truncate. However, the MDT refuses to
1850 * set the HS_DIRTY flag on released files, so we have to set
1851 * it again if the file has been restored. Please check how
1852 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1854 * Please notice that if the file is not released, the previous
1855 * MDS_DATA_MODIFIED has taken effect and usually
1856 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1857 * This way we can save an RPC for common open + trunc
1859 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1860 struct hsm_state_set hss = {
1861 .hss_valid = HSS_SETMASK,
1862 .hss_setmask = HS_DIRTY,
1866 rc2 = ll_hsm_state_set(inode, &hss);
1867 /* truncate and write can happen at the same time, so that
1868 * the file can be set modified even though the file is not
1869 * restored from released state, and ll_hsm_state_set() is
1870 * not applicable for the file, and rc2 < 0 is normal in this
1873 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1874 PFID(ll_inode2fid(inode)), rc2);
1879 if (op_data != NULL)
1880 ll_finish_md_op_data(op_data);
1882 if (S_ISREG(inode->i_mode)) {
1884 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1885 inode_dio_wait(inode);
1886 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1887 * flag. ll_update_inode (called from ll_md_setattr), clears
1888 * inode flags, so there is a gap where S_NOSEC is not set.
1889 * This can cause a writer to take the i_mutex unnecessarily,
1890 * but this is safe to do and should be rare. */
1891 inode_has_no_xattr(inode);
1895 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
1896 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
1897 ktime_us_delta(ktime_get(), kstart));
1902 int ll_setattr(struct dentry *de, struct iattr *attr)
1904 int mode = de->d_inode->i_mode;
1905 enum op_xvalid xvalid = 0;
1907 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1908 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1909 xvalid |= OP_XVALID_OWNEROVERRIDE;
1911 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1912 (ATTR_SIZE|ATTR_MODE)) &&
1913 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1914 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1915 !(attr->ia_mode & S_ISGID))))
1916 attr->ia_valid |= ATTR_FORCE;
1918 if ((attr->ia_valid & ATTR_MODE) &&
1920 !(attr->ia_mode & S_ISUID) &&
1921 !(attr->ia_valid & ATTR_KILL_SUID))
1922 attr->ia_valid |= ATTR_KILL_SUID;
1924 if ((attr->ia_valid & ATTR_MODE) &&
1925 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1926 !(attr->ia_mode & S_ISGID) &&
1927 !(attr->ia_valid & ATTR_KILL_SGID))
1928 attr->ia_valid |= ATTR_KILL_SGID;
1930 return ll_setattr_raw(de, attr, xvalid, false);
1933 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1936 struct obd_statfs obd_osfs = { 0 };
1941 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1943 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1944 flags |= OBD_STATFS_NODELAY;
1946 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1950 osfs->os_type = LL_SUPER_MAGIC;
1952 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1953 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1955 if (osfs->os_state & OS_STATE_SUM)
1958 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1959 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1962 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1963 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1966 osfs->os_bsize = obd_osfs.os_bsize;
1967 osfs->os_blocks = obd_osfs.os_blocks;
1968 osfs->os_bfree = obd_osfs.os_bfree;
1969 osfs->os_bavail = obd_osfs.os_bavail;
1971 /* If we have _some_ OSTs, but don't have as many free objects on the
1972 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1973 * to compensate, so that the "inodes in use" number is correct.
1974 * This should be kept in sync with lod_statfs() behaviour.
1976 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1977 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1979 osfs->os_ffree = obd_osfs.os_ffree;
1986 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1988 struct super_block *sb = de->d_sb;
1989 struct obd_statfs osfs;
1990 __u64 fsid = huge_encode_dev(sb->s_dev);
1991 ktime_t kstart = ktime_get();
1994 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
1996 /* Some amount of caching on the client is allowed */
1997 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2001 statfs_unpack(sfs, &osfs);
2003 /* We need to downshift for all 32-bit kernels, because we can't
2004 * tell if the kernel is being called via sys_statfs64() or not.
2005 * Stop before overflowing f_bsize - in which case it is better
2006 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2007 if (sizeof(long) < 8) {
2008 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2011 osfs.os_blocks >>= 1;
2012 osfs.os_bfree >>= 1;
2013 osfs.os_bavail >>= 1;
2017 sfs->f_blocks = osfs.os_blocks;
2018 sfs->f_bfree = osfs.os_bfree;
2019 sfs->f_bavail = osfs.os_bavail;
2020 sfs->f_fsid.val[0] = (__u32)fsid;
2021 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2023 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2024 ktime_us_delta(ktime_get(), kstart));
2029 void ll_inode_size_lock(struct inode *inode)
2031 struct ll_inode_info *lli;
2033 LASSERT(!S_ISDIR(inode->i_mode));
2035 lli = ll_i2info(inode);
2036 mutex_lock(&lli->lli_size_mutex);
2039 void ll_inode_size_unlock(struct inode *inode)
2041 struct ll_inode_info *lli;
2043 lli = ll_i2info(inode);
2044 mutex_unlock(&lli->lli_size_mutex);
2047 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2049 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2050 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2051 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2053 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2056 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2058 struct ll_inode_info *lli = ll_i2info(inode);
2059 struct mdt_body *body = md->body;
2060 struct ll_sb_info *sbi = ll_i2sbi(inode);
2063 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2064 rc = cl_file_inode_init(inode, md);
2069 if (S_ISDIR(inode->i_mode)) {
2070 rc = ll_update_lsm_md(inode, md);
2075 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2076 if (body->mbo_valid & OBD_MD_FLACL) {
2077 spin_lock(&lli->lli_lock);
2078 if (lli->lli_posix_acl)
2079 posix_acl_release(lli->lli_posix_acl);
2080 lli->lli_posix_acl = md->posix_acl;
2081 spin_unlock(&lli->lli_lock);
2084 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2085 sbi->ll_flags & LL_SBI_32BIT_API);
2086 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2088 if (body->mbo_valid & OBD_MD_FLATIME) {
2089 if (body->mbo_atime > inode->i_atime.tv_sec)
2090 inode->i_atime.tv_sec = body->mbo_atime;
2091 lli->lli_atime = body->mbo_atime;
2094 if (body->mbo_valid & OBD_MD_FLMTIME) {
2095 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2097 "setting ino %lu mtime from %lld to %llu\n",
2098 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2100 inode->i_mtime.tv_sec = body->mbo_mtime;
2102 lli->lli_mtime = body->mbo_mtime;
2105 if (body->mbo_valid & OBD_MD_FLCTIME) {
2106 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2107 inode->i_ctime.tv_sec = body->mbo_ctime;
2108 lli->lli_ctime = body->mbo_ctime;
2111 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2112 if (body->mbo_valid & OBD_MD_FLFLAGS)
2113 ll_update_inode_flags(inode, body->mbo_flags);
2114 if (body->mbo_valid & OBD_MD_FLMODE)
2115 inode->i_mode = (inode->i_mode & S_IFMT) |
2116 (body->mbo_mode & ~S_IFMT);
2118 if (body->mbo_valid & OBD_MD_FLTYPE)
2119 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2120 (body->mbo_mode & S_IFMT);
2122 LASSERT(inode->i_mode != 0);
2123 if (S_ISREG(inode->i_mode))
2124 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2125 LL_MAX_BLKSIZE_BITS);
2127 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2129 if (body->mbo_valid & OBD_MD_FLUID)
2130 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2131 if (body->mbo_valid & OBD_MD_FLGID)
2132 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2133 if (body->mbo_valid & OBD_MD_FLPROJID)
2134 lli->lli_projid = body->mbo_projid;
2135 if (body->mbo_valid & OBD_MD_FLNLINK)
2136 set_nlink(inode, body->mbo_nlink);
2137 if (body->mbo_valid & OBD_MD_FLRDEV)
2138 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2140 if (body->mbo_valid & OBD_MD_FLID) {
2141 /* FID shouldn't be changed! */
2142 if (fid_is_sane(&lli->lli_fid)) {
2143 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2144 "Trying to change FID "DFID
2145 " to the "DFID", inode "DFID"(%p)\n",
2146 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2147 PFID(ll_inode2fid(inode)), inode);
2149 lli->lli_fid = body->mbo_fid1;
2153 LASSERT(fid_seq(&lli->lli_fid) != 0);
2155 if (body->mbo_valid & OBD_MD_FLSIZE) {
2156 i_size_write(inode, body->mbo_size);
2158 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2159 PFID(ll_inode2fid(inode)),
2160 (unsigned long long)body->mbo_size);
2162 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2163 inode->i_blocks = body->mbo_blocks;
2166 if (body->mbo_valid & OBD_MD_TSTATE) {
2167 /* Set LLIF_FILE_RESTORING if restore ongoing and
2168 * clear it when done to ensure to start again
2169 * glimpsing updated attrs
2171 if (body->mbo_t_state & MS_RESTORE)
2172 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2174 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2180 int ll_read_inode2(struct inode *inode, void *opaque)
2182 struct lustre_md *md = opaque;
2183 struct ll_inode_info *lli = ll_i2info(inode);
2187 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2188 PFID(&lli->lli_fid), inode);
2190 /* Core attributes from the MDS first. This is a new inode, and
2191 * the VFS doesn't zero times in the core inode so we have to do
2192 * it ourselves. They will be overwritten by either MDS or OST
2193 * attributes - we just need to make sure they aren't newer.
2195 inode->i_mtime.tv_sec = 0;
2196 inode->i_atime.tv_sec = 0;
2197 inode->i_ctime.tv_sec = 0;
2199 rc = ll_update_inode(inode, md);
2203 /* OIDEBUG(inode); */
2205 #ifdef HAVE_BACKING_DEV_INFO
2206 /* initializing backing dev info. */
2207 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2209 if (S_ISREG(inode->i_mode)) {
2210 struct ll_sb_info *sbi = ll_i2sbi(inode);
2211 inode->i_op = &ll_file_inode_operations;
2212 inode->i_fop = sbi->ll_fop;
2213 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2215 } else if (S_ISDIR(inode->i_mode)) {
2216 inode->i_op = &ll_dir_inode_operations;
2217 inode->i_fop = &ll_dir_operations;
2219 } else if (S_ISLNK(inode->i_mode)) {
2220 inode->i_op = &ll_fast_symlink_inode_operations;
2223 inode->i_op = &ll_special_inode_operations;
2225 init_special_inode(inode, inode->i_mode,
2234 void ll_delete_inode(struct inode *inode)
2236 struct ll_inode_info *lli = ll_i2info(inode);
2237 struct address_space *mapping = &inode->i_data;
2238 unsigned long nrpages;
2239 unsigned long flags;
2243 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2244 /* It is last chance to write out dirty pages,
2245 * otherwise we may lose data while umount.
2247 * If i_nlink is 0 then just discard data. This is safe because
2248 * local inode gets i_nlink 0 from server only for the last
2249 * unlink, so that file is not opened somewhere else
2251 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2252 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2254 truncate_inode_pages_final(mapping);
2256 /* Workaround for LU-118: Note nrpages may not be totally updated when
2257 * truncate_inode_pages() returns, as there can be a page in the process
2258 * of deletion (inside __delete_from_page_cache()) in the specified
2259 * range. Thus mapping->nrpages can be non-zero when this function
2260 * returns even after truncation of the whole mapping. Only do this if
2261 * npages isn't already zero.
2263 nrpages = mapping->nrpages;
2265 xa_lock_irqsave(&mapping->i_pages, flags);
2266 nrpages = mapping->nrpages;
2267 xa_unlock_irqrestore(&mapping->i_pages, flags);
2268 } /* Workaround end */
2270 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2271 "see https://jira.whamcloud.com/browse/LU-118\n",
2272 ll_i2sbi(inode)->ll_fsname,
2273 PFID(ll_inode2fid(inode)), inode, nrpages);
2275 ll_clear_inode(inode);
2281 int ll_iocontrol(struct inode *inode, struct file *file,
2282 unsigned int cmd, unsigned long arg)
2284 struct ll_sb_info *sbi = ll_i2sbi(inode);
2285 struct ptlrpc_request *req = NULL;
2290 case FS_IOC_GETFLAGS: {
2291 struct mdt_body *body;
2292 struct md_op_data *op_data;
2294 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2295 0, 0, LUSTRE_OPC_ANY,
2297 if (IS_ERR(op_data))
2298 RETURN(PTR_ERR(op_data));
2300 op_data->op_valid = OBD_MD_FLFLAGS;
2301 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2302 ll_finish_md_op_data(op_data);
2304 CERROR("%s: failure inode "DFID": rc = %d\n",
2305 sbi->ll_md_exp->exp_obd->obd_name,
2306 PFID(ll_inode2fid(inode)), rc);
2310 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2312 flags = body->mbo_flags;
2314 ptlrpc_req_finished(req);
2316 RETURN(put_user(flags, (int __user *)arg));
2318 case FS_IOC_SETFLAGS: {
2320 struct md_op_data *op_data;
2321 struct cl_object *obj;
2322 struct fsxattr fa = { 0 };
2324 if (get_user(flags, (int __user *)arg))
2327 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2328 if (flags & LUSTRE_PROJINHERIT_FL)
2329 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2331 rc = ll_ioctl_check_project(inode, &fa);
2335 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2336 LUSTRE_OPC_ANY, NULL);
2337 if (IS_ERR(op_data))
2338 RETURN(PTR_ERR(op_data));
2340 op_data->op_attr_flags = flags;
2341 op_data->op_xvalid |= OP_XVALID_FLAGS;
2342 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2343 ll_finish_md_op_data(op_data);
2344 ptlrpc_req_finished(req);
2348 ll_update_inode_flags(inode, flags);
2350 obj = ll_i2info(inode)->lli_clob;
2354 OBD_ALLOC_PTR(attr);
2358 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2370 int ll_flush_ctx(struct inode *inode)
2372 struct ll_sb_info *sbi = ll_i2sbi(inode);
2374 CDEBUG(D_SEC, "flush context for user %d\n",
2375 from_kuid(&init_user_ns, current_uid()));
2377 obd_set_info_async(NULL, sbi->ll_md_exp,
2378 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2380 obd_set_info_async(NULL, sbi->ll_dt_exp,
2381 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2386 /* umount -f client means force down, don't save state */
2387 void ll_umount_begin(struct super_block *sb)
2389 struct ll_sb_info *sbi = ll_s2sbi(sb);
2390 struct obd_device *obd;
2391 struct obd_ioctl_data *ioc_data;
2395 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2396 sb->s_count, atomic_read(&sb->s_active));
2398 obd = class_exp2obd(sbi->ll_md_exp);
2400 CERROR("Invalid MDC connection handle %#llx\n",
2401 sbi->ll_md_exp->exp_handle.h_cookie);
2407 obd = class_exp2obd(sbi->ll_dt_exp);
2409 CERROR("Invalid LOV connection handle %#llx\n",
2410 sbi->ll_dt_exp->exp_handle.h_cookie);
2416 OBD_ALLOC_PTR(ioc_data);
2418 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2419 sizeof *ioc_data, ioc_data, NULL);
2421 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2422 sizeof *ioc_data, ioc_data, NULL);
2424 OBD_FREE_PTR(ioc_data);
2427 /* Really, we'd like to wait until there are no requests outstanding,
2428 * and then continue. For now, we just periodically checking for vfs
2429 * to decrement mnt_cnt and hope to finish it within 10sec.
2433 !may_umount(sbi->ll_mnt.mnt)) {
2441 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2443 struct ll_sb_info *sbi = ll_s2sbi(sb);
2444 char *profilenm = get_profile_name(sb);
2448 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2449 read_only = *flags & MS_RDONLY;
2450 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2451 sizeof(KEY_READ_ONLY),
2452 KEY_READ_ONLY, sizeof(read_only),
2455 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2456 profilenm, read_only ?
2457 "read-only" : "read-write", err);
2462 sb->s_flags |= SB_RDONLY;
2464 sb->s_flags &= ~SB_RDONLY;
2466 if (sbi->ll_flags & LL_SBI_VERBOSE)
2467 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2468 read_only ? "read-only" : "read-write");
2474 * Cleanup the open handle that is cached on MDT-side.
2476 * For open case, the client side open handling thread may hit error
2477 * after the MDT grant the open. Under such case, the client should
2478 * send close RPC to the MDT as cleanup; otherwise, the open handle
2479 * on the MDT will be leaked there until the client umount or evicted.
2481 * In further, if someone unlinked the file, because the open handle
2482 * holds the reference on such file/object, then it will block the
2483 * subsequent threads that want to locate such object via FID.
2485 * \param[in] sb super block for this file-system
2486 * \param[in] open_req pointer to the original open request
2488 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2490 struct mdt_body *body;
2491 struct md_op_data *op_data;
2492 struct ptlrpc_request *close_req = NULL;
2493 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2496 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2497 OBD_ALLOC_PTR(op_data);
2498 if (op_data == NULL) {
2499 CWARN("%s: cannot allocate op_data to release open handle for "
2500 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2505 op_data->op_fid1 = body->mbo_fid1;
2506 op_data->op_open_handle = body->mbo_open_handle;
2507 op_data->op_mod_time = ktime_get_real_seconds();
2508 md_close(exp, op_data, NULL, &close_req);
2509 ptlrpc_req_finished(close_req);
2510 ll_finish_md_op_data(op_data);
2515 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2516 struct super_block *sb, struct lookup_intent *it)
2518 struct ll_sb_info *sbi = NULL;
2519 struct lustre_md md = { NULL };
2520 bool default_lmv_deleted = false;
2525 LASSERT(*inode || sb);
2526 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2527 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2528 sbi->ll_md_exp, &md);
2533 * clear default_lmv only if intent_getattr reply doesn't contain it.
2534 * but it needs to be done after iget, check this early because
2535 * ll_update_lsm_md() may change md.
2537 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2538 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2539 default_lmv_deleted = true;
2542 rc = ll_update_inode(*inode, &md);
2546 LASSERT(sb != NULL);
2549 * At this point server returns to client's same fid as client
2550 * generated for creating. So using ->fid1 is okay here.
2552 if (!fid_is_sane(&md.body->mbo_fid1)) {
2553 CERROR("%s: Fid is insane "DFID"\n",
2555 PFID(&md.body->mbo_fid1));
2556 GOTO(out, rc = -EINVAL);
2559 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2560 sbi->ll_flags & LL_SBI_32BIT_API),
2562 if (IS_ERR(*inode)) {
2563 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2565 posix_acl_release(md.posix_acl);
2566 md.posix_acl = NULL;
2569 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2571 CERROR("new_inode -fatal: rc %d\n", rc);
2576 /* Handling piggyback layout lock.
2577 * Layout lock can be piggybacked by getattr and open request.
2578 * The lsm can be applied to inode only if it comes with a layout lock
2579 * otherwise correct layout may be overwritten, for example:
2580 * 1. proc1: mdt returns a lsm but not granting layout
2581 * 2. layout was changed by another client
2582 * 3. proc2: refresh layout and layout lock granted
2583 * 4. proc1: to apply a stale layout */
2584 if (it != NULL && it->it_lock_mode != 0) {
2585 struct lustre_handle lockh;
2586 struct ldlm_lock *lock;
2588 lockh.cookie = it->it_lock_handle;
2589 lock = ldlm_handle2lock(&lockh);
2590 LASSERT(lock != NULL);
2591 if (ldlm_has_layout(lock)) {
2592 struct cl_object_conf conf;
2594 memset(&conf, 0, sizeof(conf));
2595 conf.coc_opc = OBJECT_CONF_SET;
2596 conf.coc_inode = *inode;
2597 conf.coc_lock = lock;
2598 conf.u.coc_layout = md.layout;
2599 (void)ll_layout_conf(*inode, &conf);
2601 LDLM_LOCK_PUT(lock);
2604 if (default_lmv_deleted)
2605 ll_update_default_lsm_md(*inode, &md);
2610 /* cleanup will be done if necessary */
2611 md_free_lustre_md(sbi->ll_md_exp, &md);
2613 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2614 ll_intent_drop_lock(it);
2615 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2621 int ll_obd_statfs(struct inode *inode, void __user *arg)
2623 struct ll_sb_info *sbi = NULL;
2624 struct obd_export *exp;
2626 struct obd_ioctl_data *data = NULL;
2630 if (!inode || !(sbi = ll_i2sbi(inode)))
2631 GOTO(out_statfs, rc = -EINVAL);
2633 rc = obd_ioctl_getdata(&buf, &len, arg);
2635 GOTO(out_statfs, rc);
2638 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2639 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2640 GOTO(out_statfs, rc = -EINVAL);
2642 if (data->ioc_inllen1 != sizeof(__u32) ||
2643 data->ioc_inllen2 != sizeof(__u32) ||
2644 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2645 data->ioc_plen2 != sizeof(struct obd_uuid))
2646 GOTO(out_statfs, rc = -EINVAL);
2648 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2649 if (type & LL_STATFS_LMV)
2650 exp = sbi->ll_md_exp;
2651 else if (type & LL_STATFS_LOV)
2652 exp = sbi->ll_dt_exp;
2654 GOTO(out_statfs, rc = -ENODEV);
2656 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2658 GOTO(out_statfs, rc);
2660 OBD_FREE_LARGE(buf, len);
2665 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2666 * be called early to avoid deadlock.
2668 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2670 if (op_data->op_mea2_sem) {
2671 up_read(op_data->op_mea2_sem);
2672 op_data->op_mea2_sem = NULL;
2675 if (op_data->op_mea1_sem) {
2676 up_read(op_data->op_mea1_sem);
2677 op_data->op_mea1_sem = NULL;
2681 /* this function prepares md_op_data hint for passing it down to MD stack. */
2682 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2683 struct inode *i1, struct inode *i2,
2684 const char *name, size_t namelen,
2685 __u32 mode, enum md_op_code opc,
2688 LASSERT(i1 != NULL);
2691 /* Do not reuse namelen for something else. */
2693 return ERR_PTR(-EINVAL);
2695 if (namelen > ll_i2sbi(i1)->ll_namelen)
2696 return ERR_PTR(-ENAMETOOLONG);
2698 if (!lu_name_is_valid_2(name, namelen))
2699 return ERR_PTR(-EINVAL);
2702 if (op_data == NULL)
2703 OBD_ALLOC_PTR(op_data);
2705 if (op_data == NULL)
2706 return ERR_PTR(-ENOMEM);
2708 ll_i2gids(op_data->op_suppgids, i1, i2);
2709 op_data->op_fid1 = *ll_inode2fid(i1);
2710 op_data->op_code = opc;
2712 if (S_ISDIR(i1->i_mode)) {
2713 down_read(&ll_i2info(i1)->lli_lsm_sem);
2714 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2715 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2716 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2720 op_data->op_fid2 = *ll_inode2fid(i2);
2721 if (S_ISDIR(i2->i_mode)) {
2723 down_read(&ll_i2info(i2)->lli_lsm_sem);
2724 op_data->op_mea2_sem =
2725 &ll_i2info(i2)->lli_lsm_sem;
2727 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2730 fid_zero(&op_data->op_fid2);
2733 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2734 op_data->op_cli_flags |= CLI_HASH64;
2736 if (ll_need_32bit_api(ll_i2sbi(i1)))
2737 op_data->op_cli_flags |= CLI_API32;
2739 op_data->op_name = name;
2740 op_data->op_namelen = namelen;
2741 op_data->op_mode = mode;
2742 op_data->op_mod_time = ktime_get_real_seconds();
2743 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2744 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2745 op_data->op_cap = cfs_curproc_cap_pack();
2746 op_data->op_mds = 0;
2747 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2748 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2749 op_data->op_bias |= MDS_CREATE_VOLATILE;
2751 op_data->op_data = data;
2756 void ll_finish_md_op_data(struct md_op_data *op_data)
2758 ll_unlock_md_op_lsm(op_data);
2759 security_release_secctx(op_data->op_file_secctx,
2760 op_data->op_file_secctx_size);
2761 OBD_FREE_PTR(op_data);
2764 #ifdef HAVE_SUPEROPS_USE_DENTRY
2765 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2767 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2770 struct ll_sb_info *sbi;
2772 #ifdef HAVE_SUPEROPS_USE_DENTRY
2773 LASSERT((seq != NULL) && (dentry != NULL));
2774 sbi = ll_s2sbi(dentry->d_sb);
2776 LASSERT((seq != NULL) && (vfs != NULL));
2777 sbi = ll_s2sbi(vfs->mnt_sb);
2780 if (sbi->ll_flags & LL_SBI_NOLCK)
2781 seq_puts(seq, ",nolock");
2783 /* "flock" is the default since 2.13, but it wasn't for many years,
2784 * so it is still useful to print this to show it is enabled.
2785 * Start to print "noflock" so it is now clear when flock is disabled.
2787 if (sbi->ll_flags & LL_SBI_FLOCK)
2788 seq_puts(seq, ",flock");
2789 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2790 seq_puts(seq, ",localflock");
2792 seq_puts(seq, ",noflock");
2794 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2795 seq_puts(seq, ",user_xattr");
2797 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2798 seq_puts(seq, ",lazystatfs");
2800 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2801 seq_puts(seq, ",user_fid2path");
2803 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2804 seq_puts(seq, ",always_ping");
2810 * Get obd name by cmd, and copy out to user space
2812 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2814 struct ll_sb_info *sbi = ll_i2sbi(inode);
2815 struct obd_device *obd;
2818 if (cmd == OBD_IOC_GETDTNAME)
2819 obd = class_exp2obd(sbi->ll_dt_exp);
2820 else if (cmd == OBD_IOC_GETMDNAME)
2821 obd = class_exp2obd(sbi->ll_md_exp);
2828 if (copy_to_user((void __user *)arg, obd->obd_name,
2829 strlen(obd->obd_name) + 1))
2835 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2842 p.mnt = current->fs->root.mnt;
2844 path = d_path(&p, buf, bufsize);
2849 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2851 char *buf, *path = NULL;
2852 struct dentry *dentry = NULL;
2853 struct inode *inode = page->mapping->host;
2855 /* this can be called inside spin lock so use GFP_ATOMIC. */
2856 buf = (char *)__get_free_page(GFP_ATOMIC);
2858 dentry = d_find_alias(page->mapping->host);
2860 path = ll_d_path(dentry, buf, PAGE_SIZE);
2863 /* The below message is checked in recovery-small.sh test_24b */
2865 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2866 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2867 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2868 PFID(ll_inode2fid(inode)),
2869 (path && !IS_ERR(path)) ? path : "", ioret);
2875 free_page((unsigned long)buf);
2878 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2879 struct lov_user_md **kbuf)
2881 struct lov_user_md lum;
2885 if (copy_from_user(&lum, md, sizeof(lum)))
2888 lum_size = ll_lov_user_md_size(&lum);
2892 OBD_ALLOC_LARGE(*kbuf, lum_size);
2896 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2897 OBD_FREE_LARGE(*kbuf, lum_size);
2905 * Compute llite root squash state after a change of root squash
2906 * configuration setting or add/remove of a lnet nid
2908 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2910 struct root_squash_info *squash = &sbi->ll_squash;
2913 struct lnet_process_id id;
2915 /* Update norootsquash flag */
2916 spin_lock(&squash->rsi_lock);
2917 if (list_empty(&squash->rsi_nosquash_nids))
2918 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2920 /* Do not apply root squash as soon as one of our NIDs is
2921 * in the nosquash_nids list */
2924 while (LNetGetId(i++, &id) != -ENOENT) {
2925 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2927 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2933 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2935 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2937 spin_unlock(&squash->rsi_lock);
2941 * Parse linkea content to extract information about a given hardlink
2943 * \param[in] ldata - Initialized linkea data
2944 * \param[in] linkno - Link identifier
2945 * \param[out] parent_fid - The entry's parent FID
2946 * \param[out] ln - Entry name destination buffer
2948 * \retval 0 on success
2949 * \retval Appropriate negative error code on failure
2951 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2952 struct lu_fid *parent_fid, struct lu_name *ln)
2958 rc = linkea_init_with_rec(ldata);
2962 if (linkno >= ldata->ld_leh->leh_reccount)
2963 /* beyond last link */
2966 linkea_first_entry(ldata);
2967 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2968 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2973 linkea_next_entry(ldata);
2983 * Get parent FID and name of an identified link. Operation is performed for
2984 * a given link number, letting the caller iterate over linkno to list one or
2985 * all links of an entry.
2987 * \param[in] file - File descriptor against which to perform the operation
2988 * \param[in,out] arg - User-filled structure containing the linkno to operate
2989 * on and the available size. It is eventually filled with
2990 * the requested information or left untouched on error
2992 * \retval - 0 on success
2993 * \retval - Appropriate negative error code on failure
2995 int ll_getparent(struct file *file, struct getparent __user *arg)
2997 struct inode *inode = file_inode(file);
2998 struct linkea_data *ldata;
2999 struct lu_buf buf = LU_BUF_NULL;
3001 struct lu_fid parent_fid;
3008 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3009 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3012 if (get_user(name_size, &arg->gp_name_size))
3015 if (get_user(linkno, &arg->gp_linkno))
3018 if (name_size > PATH_MAX)
3021 OBD_ALLOC(ldata, sizeof(*ldata));
3025 rc = linkea_data_new(ldata, &buf);
3027 GOTO(ldata_free, rc);
3029 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3030 buf.lb_len, OBD_MD_FLXATTR);
3034 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3038 if (ln.ln_namelen >= name_size)
3039 GOTO(lb_free, rc = -EOVERFLOW);
3041 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3042 GOTO(lb_free, rc = -EFAULT);
3044 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3045 GOTO(lb_free, rc = -EFAULT);
3047 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3048 GOTO(lb_free, rc = -EFAULT);
3053 OBD_FREE(ldata, sizeof(*ldata));