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>
53 #ifndef HAVE_CPUS_READ_LOCK
54 #include <libcfs/linux/linux-cpu.h>
56 #include <uapi/linux/lustre/lustre_ioctl.h>
57 #ifdef HAVE_UAPI_LINUX_MOUNT_H
58 #include <uapi/linux/mount.h>
61 #include <lustre_ha.h>
62 #include <lustre_dlm.h>
63 #include <lprocfs_status.h>
64 #include <lustre_disk.h>
65 #include <uapi/linux/lustre/lustre_param.h>
66 #include <lustre_log.h>
67 #include <cl_object.h>
68 #include <obd_cksum.h>
69 #include "llite_internal.h"
71 struct kmem_cache *ll_file_data_slab;
74 #define log2(n) ffz(~(n))
78 * If there is only one number of core visible to Lustre,
79 * async readahead will be disabled, to avoid massive over
80 * subscription, we use 1/2 of active cores as default max
81 * async readahead requests.
83 static inline unsigned int ll_get_ra_async_max_active(void)
85 return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
88 static struct ll_sb_info *ll_init_sbi(void)
90 struct ll_sb_info *sbi = NULL;
92 unsigned long lru_page_max;
101 RETURN(ERR_PTR(-ENOMEM));
103 rc = pcc_super_init(&sbi->ll_pcc_super);
107 spin_lock_init(&sbi->ll_lock);
108 mutex_init(&sbi->ll_lco.lco_lock);
109 spin_lock_init(&sbi->ll_pp_extent_lock);
110 spin_lock_init(&sbi->ll_process_lock);
111 sbi->ll_rw_stats_on = 0;
112 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
115 pages = si.totalram - si.totalhigh;
116 lru_page_max = pages / 2;
118 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
119 sbi->ll_ra_info.ll_readahead_wq =
120 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
122 sbi->ll_ra_info.ra_async_max_active);
123 if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
124 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
126 /* initialize ll_cache data */
127 sbi->ll_cache = cl_cache_init(lru_page_max);
128 if (sbi->ll_cache == NULL)
129 GOTO(out_destroy_ra, rc = -ENOMEM);
131 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
132 SBI_DEFAULT_READ_AHEAD_MAX);
133 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
134 sbi->ll_ra_info.ra_max_pages_per_file;
135 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
136 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
137 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
139 sbi->ll_flags |= LL_SBI_VERBOSE;
140 #ifdef ENABLE_CHECKSUM
141 sbi->ll_flags |= LL_SBI_CHECKSUM;
144 sbi->ll_flags |= LL_SBI_FLOCK;
147 #ifdef HAVE_LRU_RESIZE_SUPPORT
148 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
150 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
152 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
153 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
155 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
159 /* metadata statahead is enabled by default */
160 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
161 sbi->ll_sa_max = LL_SA_RPC_DEF;
162 atomic_set(&sbi->ll_sa_total, 0);
163 atomic_set(&sbi->ll_sa_wrong, 0);
164 atomic_set(&sbi->ll_sa_running, 0);
165 atomic_set(&sbi->ll_agl_total, 0);
166 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
167 sbi->ll_flags |= LL_SBI_FAST_READ;
168 sbi->ll_flags |= LL_SBI_TINY_WRITE;
171 sbi->ll_squash.rsi_uid = 0;
172 sbi->ll_squash.rsi_gid = 0;
173 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
174 spin_lock_init(&sbi->ll_squash.rsi_lock);
176 /* Per-filesystem file heat */
177 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
178 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
181 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
183 pcc_super_fini(&sbi->ll_pcc_super);
189 static void ll_free_sbi(struct super_block *sb)
191 struct ll_sb_info *sbi = ll_s2sbi(sb);
195 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
196 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
197 if (sbi->ll_ra_info.ll_readahead_wq)
198 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
199 if (sbi->ll_cache != NULL) {
200 cl_cache_decref(sbi->ll_cache);
201 sbi->ll_cache = NULL;
203 pcc_super_fini(&sbi->ll_pcc_super);
204 OBD_FREE(sbi, sizeof(*sbi));
209 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
211 struct inode *root = NULL;
212 struct ll_sb_info *sbi = ll_s2sbi(sb);
213 struct obd_statfs *osfs = NULL;
214 struct ptlrpc_request *request = NULL;
215 struct obd_connect_data *data = NULL;
216 struct obd_uuid *uuid;
217 struct md_op_data *op_data;
218 struct lustre_md lmd;
220 int size, err, checksum;
223 sbi->ll_md_obd = class_name2obd(md);
224 if (!sbi->ll_md_obd) {
225 CERROR("MD %s: not setup or attached\n", md);
239 /* pass client page size via ocd_grant_blkbits, the server should report
240 * back its backend blocksize for grant calculation purpose */
241 data->ocd_grant_blkbits = PAGE_SHIFT;
243 /* indicate MDT features supported by this client */
244 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
245 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
246 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
247 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
248 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
249 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
250 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
251 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
252 OBD_CONNECT_64BITHASH |
253 OBD_CONNECT_EINPROGRESS |
254 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
255 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
256 OBD_CONNECT_MAX_EASIZE |
257 OBD_CONNECT_FLOCK_DEAD |
258 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
259 OBD_CONNECT_OPEN_BY_FID |
260 OBD_CONNECT_DIR_STRIPE |
261 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
262 OBD_CONNECT_SUBTREE |
263 OBD_CONNECT_MULTIMODRPCS |
264 OBD_CONNECT_GRANT_PARAM |
265 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
267 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
268 OBD_CONNECT2_SUM_STATFS |
269 OBD_CONNECT2_OVERSTRIPING |
271 OBD_CONNECT2_LOCK_CONVERT |
272 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
273 OBD_CONNECT2_INC_XID |
275 OBD_CONNECT2_ASYNC_DISCARD |
279 #ifdef HAVE_LRU_RESIZE_SUPPORT
280 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
281 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
283 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
284 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
285 OBD_CONNECT_LARGE_ACL;
288 data->ocd_cksum_types = obd_cksum_types_supported_client();
290 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
291 /* flag mdc connection as lightweight, only used for test
292 * purpose, use with care */
293 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
295 data->ocd_ibits_known = MDS_INODELOCK_FULL;
296 data->ocd_version = LUSTRE_VERSION_CODE;
298 if (sb->s_flags & SB_RDONLY)
299 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
300 if (sbi->ll_flags & LL_SBI_USER_XATTR)
301 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
304 /* Setting this indicates we correctly support S_NOSEC (See kernel
305 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
307 sb->s_flags |= SB_NOSEC;
310 if (sbi->ll_flags & LL_SBI_FLOCK)
311 sbi->ll_fop = &ll_file_operations_flock;
312 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
313 sbi->ll_fop = &ll_file_operations;
315 sbi->ll_fop = &ll_file_operations_noflock;
317 /* always ping even if server suppress_pings */
318 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
319 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
321 obd_connect_set_secctx(data);
323 #if defined(CONFIG_SECURITY)
324 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
327 data->ocd_brw_size = MD_MAX_BRW_SIZE;
329 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
330 &sbi->ll_sb_uuid, data, sbi->ll_cache);
332 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
333 "recovery, of which this client is not a "
334 "part. Please wait for recovery to complete,"
335 " abort, or time out.\n", md);
338 CERROR("cannot connect to %s: rc = %d\n", md, err);
342 sbi->ll_md_exp->exp_connect_data = *data;
344 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
345 LUSTRE_SEQ_METADATA);
347 CERROR("%s: Can't init metadata layer FID infrastructure, "
348 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
352 /* For mount, we only need fs info from MDT0, and also in DNE, it
353 * can make sure the client can be mounted as long as MDT0 is
355 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
356 ktime_get_seconds() - sbi->ll_statfs_max_age,
357 OBD_STATFS_FOR_MDT0);
359 GOTO(out_md_fid, err);
361 /* This needs to be after statfs to ensure connect has finished.
362 * Note that "data" does NOT contain the valid connect reply.
363 * If connecting to a 1.8 server there will be no LMV device, so
364 * we can access the MDC export directly and exp_connect_flags will
365 * be non-zero, but if accessing an upgraded 2.1 server it will
366 * have the correct flags filled in.
367 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
368 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
369 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
370 valid != CLIENT_CONNECT_MDT_REQD) {
373 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
374 obd_connect_flags2str(buf, PAGE_SIZE,
375 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
376 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
377 "feature(s) needed for correct operation "
378 "of this client (%s). Please upgrade "
379 "server or downgrade client.\n",
380 sbi->ll_md_exp->exp_obd->obd_name, buf);
381 OBD_FREE(buf, PAGE_SIZE);
382 GOTO(out_md_fid, err = -EPROTO);
385 size = sizeof(*data);
386 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
387 KEY_CONN_DATA, &size, data);
389 CERROR("%s: Get connect data failed: rc = %d\n",
390 sbi->ll_md_exp->exp_obd->obd_name, err);
391 GOTO(out_md_fid, err);
394 LASSERT(osfs->os_bsize);
395 sb->s_blocksize = osfs->os_bsize;
396 sb->s_blocksize_bits = log2(osfs->os_bsize);
397 sb->s_magic = LL_SUPER_MAGIC;
398 sb->s_maxbytes = MAX_LFS_FILESIZE;
399 sbi->ll_namelen = osfs->os_namelen;
400 sbi->ll_mnt.mnt = current->fs->root.mnt;
402 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
403 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
404 LCONSOLE_INFO("Disabling user_xattr feature because "
405 "it is not supported on the server\n");
406 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
409 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
411 sb->s_flags |= SB_POSIXACL;
413 sbi->ll_flags |= LL_SBI_ACL;
415 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
417 sb->s_flags &= ~SB_POSIXACL;
419 sbi->ll_flags &= ~LL_SBI_ACL;
422 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
423 sbi->ll_flags |= LL_SBI_64BIT_HASH;
425 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
426 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
428 if (obd_connect_has_secctx(data))
429 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
431 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
432 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
433 LCONSOLE_INFO("%s: disabling xattr cache due to "
434 "unknown maximum xattr size.\n", dt);
435 } else if (!sbi->ll_xattr_cache_set) {
436 /* If xattr_cache is already set (no matter 0 or 1)
437 * during processing llog, it won't be enabled here. */
438 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
439 sbi->ll_xattr_cache_enabled = 1;
443 sbi->ll_dt_obd = class_name2obd(dt);
444 if (!sbi->ll_dt_obd) {
445 CERROR("DT %s: not setup or attached\n", dt);
446 GOTO(out_md_fid, err = -ENODEV);
449 /* pass client page size via ocd_grant_blkbits, the server should report
450 * back its backend blocksize for grant calculation purpose */
451 data->ocd_grant_blkbits = PAGE_SHIFT;
453 /* indicate OST features supported by this client */
454 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
455 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
456 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
457 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
458 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
459 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
460 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
461 OBD_CONNECT_EINPROGRESS |
462 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
463 OBD_CONNECT_LAYOUTLOCK |
464 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
465 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
466 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
467 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
468 OBD_CONNECT2_INC_XID;
470 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
471 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
473 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
474 * disabled by default, because it can still be enabled on the
475 * fly via /sys. As a consequence, we still need to come to an
476 * agreement on the supported algorithms at connect time
478 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
480 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
481 data->ocd_cksum_types = OBD_CKSUM_ADLER;
483 data->ocd_cksum_types = obd_cksum_types_supported_client();
485 #ifdef HAVE_LRU_RESIZE_SUPPORT
486 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
488 /* always ping even if server suppress_pings */
489 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
490 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
492 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
493 "ocd_grant: %d\n", data->ocd_connect_flags,
494 data->ocd_version, data->ocd_grant);
496 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
497 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
499 data->ocd_brw_size = DT_MAX_BRW_SIZE;
501 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
502 &sbi->ll_sb_uuid, data, sbi->ll_cache);
504 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
505 "recovery, of which this client is not a "
506 "part. Please wait for recovery to "
507 "complete, abort, or time out.\n", dt);
510 CERROR("%s: Cannot connect to %s: rc = %d\n",
511 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
515 sbi->ll_dt_exp->exp_connect_data = *data;
517 /* Don't change value if it was specified in the config log */
518 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
519 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
520 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
521 (data->ocd_brw_size >> PAGE_SHIFT));
522 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
523 sbi->ll_ra_info.ra_max_pages_per_file)
524 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
525 sbi->ll_ra_info.ra_max_pages_per_file;
528 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
529 LUSTRE_SEQ_METADATA);
531 CERROR("%s: Can't init data layer FID infrastructure, "
532 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
536 mutex_lock(&sbi->ll_lco.lco_lock);
537 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
538 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
539 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
540 mutex_unlock(&sbi->ll_lco.lco_lock);
542 fid_zero(&sbi->ll_root_fid);
543 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
546 CERROR("cannot mds_connect: rc = %d\n", err);
547 GOTO(out_lock_cn_cb, err);
549 if (!fid_is_sane(&sbi->ll_root_fid)) {
550 CERROR("%s: Invalid root fid "DFID" during mount\n",
551 sbi->ll_md_exp->exp_obd->obd_name,
552 PFID(&sbi->ll_root_fid));
553 GOTO(out_lock_cn_cb, err = -EINVAL);
555 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
557 sb->s_op = &lustre_super_operations;
558 sb->s_xattr = ll_xattr_handlers;
559 #if THREAD_SIZE >= 8192 /*b=17630*/
560 sb->s_export_op = &lustre_export_operations;
564 * XXX: move this to after cbd setup? */
565 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
566 if (sbi->ll_flags & LL_SBI_ACL)
567 valid |= OBD_MD_FLACL;
569 OBD_ALLOC_PTR(op_data);
571 GOTO(out_lock_cn_cb, err = -ENOMEM);
573 op_data->op_fid1 = sbi->ll_root_fid;
574 op_data->op_mode = 0;
575 op_data->op_valid = valid;
577 err = md_getattr(sbi->ll_md_exp, op_data, &request);
579 OBD_FREE_PTR(op_data);
581 CERROR("%s: md_getattr failed for root: rc = %d\n",
582 sbi->ll_md_exp->exp_obd->obd_name, err);
583 GOTO(out_lock_cn_cb, err);
586 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
587 sbi->ll_md_exp, &lmd);
589 CERROR("failed to understand root inode md: rc = %d\n", err);
590 ptlrpc_req_finished(request);
591 GOTO(out_lock_cn_cb, err);
594 LASSERT(fid_is_sane(&sbi->ll_root_fid));
595 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
596 sbi->ll_flags & LL_SBI_32BIT_API),
598 md_free_lustre_md(sbi->ll_md_exp, &lmd);
599 ptlrpc_req_finished(request);
602 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
604 posix_acl_release(lmd.posix_acl);
605 lmd.posix_acl = NULL;
608 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
610 CERROR("%s: bad ll_iget() for root: rc = %d\n",
611 sbi->ll_fsname, err);
615 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
616 if (sbi->ll_checksum_set) {
617 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
618 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
619 sizeof(checksum), &checksum, NULL);
621 CERROR("%s: Set checksum failed: rc = %d\n",
622 sbi->ll_dt_exp->exp_obd->obd_name, err);
628 sb->s_root = d_make_root(root);
629 if (sb->s_root == NULL) {
631 CERROR("%s: can't make root dentry: rc = %d\n",
632 sbi->ll_fsname, err);
636 sbi->ll_sdev_orig = sb->s_dev;
638 /* We set sb->s_dev equal on all lustre clients in order to support
639 * NFS export clustering. NFSD requires that the FSID be the same
641 /* s_dev is also used in lt_compare() to compare two fs, but that is
642 * only a node-local comparison. */
643 uuid = obd_get_uuid(sbi->ll_md_exp);
645 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
652 if (sbi->ll_dt_obd) {
653 err = sysfs_create_link(&sbi->ll_kset.kobj,
654 &sbi->ll_dt_obd->obd_kset.kobj,
655 sbi->ll_dt_obd->obd_type->typ_name);
657 CERROR("%s: could not register %s in llite: rc = %d\n",
658 dt, sbi->ll_fsname, err);
663 if (sbi->ll_md_obd) {
664 err = sysfs_create_link(&sbi->ll_kset.kobj,
665 &sbi->ll_md_obd->obd_kset.kobj,
666 sbi->ll_md_obd->obd_type->typ_name);
668 CERROR("%s: could not register %s in llite: rc = %d\n",
669 md, sbi->ll_fsname, err);
679 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
681 obd_disconnect(sbi->ll_dt_exp);
682 sbi->ll_dt_exp = NULL;
683 sbi->ll_dt_obd = NULL;
685 obd_fid_fini(sbi->ll_md_exp->exp_obd);
687 obd_disconnect(sbi->ll_md_exp);
688 sbi->ll_md_exp = NULL;
689 sbi->ll_md_obd = NULL;
698 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
702 size = sizeof(*lmmsize);
703 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
704 KEY_MAX_EASIZE, &size, lmmsize);
706 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
707 sbi->ll_dt_exp->exp_obd->obd_name, rc);
711 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
714 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
715 KEY_MAX_EASIZE, &size, lmmsize);
717 CERROR("Get max mdsize error rc %d\n", rc);
719 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
725 * Get the value of the default_easize parameter.
727 * \see client_obd::cl_default_mds_easize
729 * \param[in] sbi superblock info for this filesystem
730 * \param[out] lmmsize pointer to storage location for value
732 * \retval 0 on success
733 * \retval negative negated errno on failure
735 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
740 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
741 KEY_DEFAULT_EASIZE, &size, lmmsize);
743 CERROR("Get default mdsize error rc %d\n", rc);
749 * Set the default_easize parameter to the given value.
751 * \see client_obd::cl_default_mds_easize
753 * \param[in] sbi superblock info for this filesystem
754 * \param[in] lmmsize the size to set
756 * \retval 0 on success
757 * \retval negative negated errno on failure
759 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
763 if (lmmsize < sizeof(struct lov_mds_md) ||
764 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
767 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
768 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
769 sizeof(int), &lmmsize, NULL);
774 static void client_common_put_super(struct super_block *sb)
776 struct ll_sb_info *sbi = ll_s2sbi(sb);
781 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
782 obd_disconnect(sbi->ll_dt_exp);
783 sbi->ll_dt_exp = NULL;
785 ll_debugfs_unregister_super(sb);
787 obd_fid_fini(sbi->ll_md_exp->exp_obd);
788 obd_disconnect(sbi->ll_md_exp);
789 sbi->ll_md_exp = NULL;
794 void ll_kill_super(struct super_block *sb)
796 struct ll_sb_info *sbi;
800 if (!(sb->s_flags & SB_ACTIVE))
804 /* we need restore s_dev from changed for clustred NFS before put_super
805 * because new kernels have cached s_dev and change sb->s_dev in
806 * put_super not affected real removing devices */
808 sb->s_dev = sbi->ll_sdev_orig;
810 /* wait running statahead threads to quit */
811 while (atomic_read(&sbi->ll_sa_running) > 0)
812 schedule_timeout_uninterruptible(
813 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, sizeof(sbi->ll_sb_uuid), "%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 rc = l_wait_event_abortable(
1236 sbi->ll_cache->ccc_unstable_waitq,
1237 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1240 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1241 if (force == 0 && rc != -ERESTARTSYS)
1242 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1244 /* We need to set force before the lov_disconnect in
1245 * lustre_common_put_super, since l_d cleans up osc's as well.
1249 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1251 obd->obd_force = force;
1255 if (sbi->ll_client_common_fill_super_succeeded) {
1256 /* Only if client_common_fill_super succeeded */
1257 client_common_put_super(sb);
1261 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1262 class_manual_cleanup(obd);
1264 if (sbi->ll_flags & LL_SBI_VERBOSE)
1265 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1268 class_del_profile(profilenm);
1270 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1271 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1272 bdi_destroy(&lsi->lsi_bdi);
1273 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1278 lsi->lsi_llsbi = NULL;
1280 lustre_common_put_super(sb);
1282 cl_env_cache_purge(~0);
1284 module_put(THIS_MODULE);
1287 } /* client_put_super */
1289 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1291 struct inode *inode = NULL;
1293 /* NOTE: we depend on atomic igrab() -bzzz */
1294 lock_res_and_lock(lock);
1295 if (lock->l_resource->lr_lvb_inode) {
1296 struct ll_inode_info * lli;
1297 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1298 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1299 inode = igrab(lock->l_resource->lr_lvb_inode);
1301 inode = lock->l_resource->lr_lvb_inode;
1302 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1303 D_WARNING, lock, "lr_lvb_inode %p is "
1304 "bogus: magic %08x",
1305 lock->l_resource->lr_lvb_inode,
1306 lli->lli_inode_magic);
1310 unlock_res_and_lock(lock);
1314 void ll_dir_clear_lsm_md(struct inode *inode)
1316 struct ll_inode_info *lli = ll_i2info(inode);
1318 LASSERT(S_ISDIR(inode->i_mode));
1320 if (lli->lli_lsm_md) {
1321 lmv_free_memmd(lli->lli_lsm_md);
1322 lli->lli_lsm_md = NULL;
1325 if (lli->lli_default_lsm_md) {
1326 lmv_free_memmd(lli->lli_default_lsm_md);
1327 lli->lli_default_lsm_md = NULL;
1331 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1332 const struct lu_fid *fid,
1333 struct lustre_md *md)
1335 struct ll_sb_info *sbi = ll_s2sbi(sb);
1336 struct mdt_body *body = md->body;
1337 struct inode *inode;
1341 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1342 inode = iget_locked(sb, ino);
1343 if (inode == NULL) {
1344 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1345 sbi->ll_fsname, PFID(fid));
1346 RETURN(ERR_PTR(-ENOENT));
1349 if (inode->i_state & I_NEW) {
1350 struct ll_inode_info *lli = ll_i2info(inode);
1351 struct lmv_stripe_md *lsm = md->lmv;
1353 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1354 (body->mbo_mode & S_IFMT);
1355 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1358 inode->i_mtime.tv_sec = 0;
1359 inode->i_atime.tv_sec = 0;
1360 inode->i_ctime.tv_sec = 0;
1363 #ifdef HAVE_BACKING_DEV_INFO
1364 /* initializing backing dev info. */
1365 inode->i_mapping->backing_dev_info =
1366 &s2lsi(inode->i_sb)->lsi_bdi;
1368 inode->i_op = &ll_dir_inode_operations;
1369 inode->i_fop = &ll_dir_operations;
1370 lli->lli_fid = *fid;
1373 LASSERT(lsm != NULL);
1374 /* master object FID */
1375 lli->lli_pfid = body->mbo_fid1;
1376 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1377 lli, PFID(fid), PFID(&lli->lli_pfid));
1378 unlock_new_inode(inode);
1384 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1387 struct lmv_stripe_md *lsm = md->lmv;
1388 struct ll_inode_info *lli = ll_i2info(inode);
1391 LASSERT(lsm != NULL);
1393 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1394 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1395 lsm_md_dump(D_INODE, lsm);
1397 if (!lmv_dir_striped(lsm))
1400 /* XXX sigh, this lsm_root initialization should be in
1401 * LMV layer, but it needs ll_iget right now, so we
1402 * put this here right now. */
1403 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1404 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1405 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1407 if (!fid_is_sane(fid))
1410 /* Unfortunately ll_iget will call ll_update_inode,
1411 * where the initialization of slave inode is slightly
1412 * different, so it reset lsm_md to NULL to avoid
1413 * initializing lsm for slave inode. */
1414 lsm->lsm_md_oinfo[i].lmo_root =
1415 ll_iget_anon_dir(inode->i_sb, fid, md);
1416 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1417 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1419 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1421 iput(lsm->lsm_md_oinfo[i].lmo_root);
1422 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1428 lli->lli_lsm_md = lsm;
1433 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1435 struct ll_inode_info *lli = ll_i2info(inode);
1437 if (!md->default_lmv) {
1438 /* clear default lsm */
1439 if (lli->lli_default_lsm_md) {
1440 down_write(&lli->lli_lsm_sem);
1441 if (lli->lli_default_lsm_md) {
1442 lmv_free_memmd(lli->lli_default_lsm_md);
1443 lli->lli_default_lsm_md = NULL;
1445 up_write(&lli->lli_lsm_sem);
1447 } else if (lli->lli_default_lsm_md) {
1448 /* update default lsm if it changes */
1449 down_read(&lli->lli_lsm_sem);
1450 if (lli->lli_default_lsm_md &&
1451 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1452 up_read(&lli->lli_lsm_sem);
1453 down_write(&lli->lli_lsm_sem);
1454 if (lli->lli_default_lsm_md)
1455 lmv_free_memmd(lli->lli_default_lsm_md);
1456 lli->lli_default_lsm_md = md->default_lmv;
1457 lsm_md_dump(D_INODE, md->default_lmv);
1458 md->default_lmv = NULL;
1459 up_write(&lli->lli_lsm_sem);
1461 up_read(&lli->lli_lsm_sem);
1464 /* init default lsm */
1465 down_write(&lli->lli_lsm_sem);
1466 lli->lli_default_lsm_md = md->default_lmv;
1467 lsm_md_dump(D_INODE, md->default_lmv);
1468 md->default_lmv = NULL;
1469 up_write(&lli->lli_lsm_sem);
1473 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1475 struct ll_inode_info *lli = ll_i2info(inode);
1476 struct lmv_stripe_md *lsm = md->lmv;
1477 struct cl_attr *attr;
1482 LASSERT(S_ISDIR(inode->i_mode));
1483 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1484 PFID(ll_inode2fid(inode)));
1486 /* update default LMV */
1487 if (md->default_lmv)
1488 ll_update_default_lsm_md(inode, md);
1491 * no striped information from request, lustre_md from req does not
1492 * include stripeEA, see ll_md_setattr()
1498 * normally dir layout doesn't change, only take read lock to check
1499 * that to avoid blocking other MD operations.
1501 down_read(&lli->lli_lsm_sem);
1503 /* some current lookup initialized lsm, and unchanged */
1504 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1505 GOTO(unlock, rc = 0);
1507 /* if dir layout doesn't match, check whether version is increased,
1508 * which means layout is changed, this happens in dir split/merge and
1511 * foreign LMV should not change.
1513 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1514 lsm->lsm_md_layout_version <=
1515 lli->lli_lsm_md->lsm_md_layout_version) {
1516 CERROR("%s: "DFID" dir layout mismatch:\n",
1517 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1518 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1519 lsm_md_dump(D_ERROR, lsm);
1520 GOTO(unlock, rc = -EINVAL);
1523 up_read(&lli->lli_lsm_sem);
1524 down_write(&lli->lli_lsm_sem);
1525 /* clear existing lsm */
1526 if (lli->lli_lsm_md) {
1527 lmv_free_memmd(lli->lli_lsm_md);
1528 lli->lli_lsm_md = NULL;
1531 rc = ll_init_lsm_md(inode, md);
1532 up_write(&lli->lli_lsm_sem);
1537 /* set md->lmv to NULL, so the following free lustre_md will not free
1542 /* md_merge_attr() may take long, since lsm is already set, switch to
1545 down_read(&lli->lli_lsm_sem);
1547 if (!lmv_dir_striped(lli->lli_lsm_md))
1548 GOTO(unlock, rc = 0);
1550 OBD_ALLOC_PTR(attr);
1552 GOTO(unlock, rc = -ENOMEM);
1554 /* validate the lsm */
1555 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1556 ll_md_blocking_ast);
1558 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1559 md->body->mbo_nlink = attr->cat_nlink;
1560 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1561 md->body->mbo_size = attr->cat_size;
1562 if (md->body->mbo_valid & OBD_MD_FLATIME)
1563 md->body->mbo_atime = attr->cat_atime;
1564 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1565 md->body->mbo_ctime = attr->cat_ctime;
1566 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1567 md->body->mbo_mtime = attr->cat_mtime;
1573 up_read(&lli->lli_lsm_sem);
1578 void ll_clear_inode(struct inode *inode)
1580 struct ll_inode_info *lli = ll_i2info(inode);
1581 struct ll_sb_info *sbi = ll_i2sbi(inode);
1585 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1586 PFID(ll_inode2fid(inode)), inode);
1588 if (S_ISDIR(inode->i_mode)) {
1589 /* these should have been cleared in ll_file_release */
1590 LASSERT(lli->lli_opendir_key == NULL);
1591 LASSERT(lli->lli_sai == NULL);
1592 LASSERT(lli->lli_opendir_pid == 0);
1594 pcc_inode_free(inode);
1597 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1599 LASSERT(!lli->lli_open_fd_write_count);
1600 LASSERT(!lli->lli_open_fd_read_count);
1601 LASSERT(!lli->lli_open_fd_exec_count);
1603 if (lli->lli_mds_write_och)
1604 ll_md_real_close(inode, FMODE_WRITE);
1605 if (lli->lli_mds_exec_och)
1606 ll_md_real_close(inode, FMODE_EXEC);
1607 if (lli->lli_mds_read_och)
1608 ll_md_real_close(inode, FMODE_READ);
1610 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1611 OBD_FREE(lli->lli_symlink_name,
1612 strlen(lli->lli_symlink_name) + 1);
1613 lli->lli_symlink_name = NULL;
1616 ll_xattr_cache_destroy(inode);
1618 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1619 forget_all_cached_acls(inode);
1620 if (lli->lli_posix_acl) {
1621 posix_acl_release(lli->lli_posix_acl);
1622 lli->lli_posix_acl = NULL;
1625 lli->lli_inode_magic = LLI_INODE_DEAD;
1627 if (S_ISDIR(inode->i_mode))
1628 ll_dir_clear_lsm_md(inode);
1629 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1630 LASSERT(list_empty(&lli->lli_agl_list));
1633 * XXX This has to be done before lsm is freed below, because
1634 * cl_object still uses inode lsm.
1636 cl_inode_fini(inode);
1641 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1643 struct lustre_md md;
1644 struct inode *inode = dentry->d_inode;
1645 struct ll_sb_info *sbi = ll_i2sbi(inode);
1646 struct ptlrpc_request *request = NULL;
1650 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1651 LUSTRE_OPC_ANY, NULL);
1652 if (IS_ERR(op_data))
1653 RETURN(PTR_ERR(op_data));
1655 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1657 ptlrpc_req_finished(request);
1658 if (rc == -ENOENT) {
1660 /* Unlinked special device node? Or just a race?
1661 * Pretend we done everything. */
1662 if (!S_ISREG(inode->i_mode) &&
1663 !S_ISDIR(inode->i_mode)) {
1664 ia_valid = op_data->op_attr.ia_valid;
1665 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1666 rc = simple_setattr(dentry, &op_data->op_attr);
1667 op_data->op_attr.ia_valid = ia_valid;
1669 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1670 CERROR("md_setattr fails: rc = %d\n", rc);
1675 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1676 sbi->ll_md_exp, &md);
1678 ptlrpc_req_finished(request);
1682 ia_valid = op_data->op_attr.ia_valid;
1683 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1684 * cache is not cleared yet. */
1685 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1686 if (S_ISREG(inode->i_mode))
1688 rc = simple_setattr(dentry, &op_data->op_attr);
1689 if (S_ISREG(inode->i_mode))
1690 inode_unlock(inode);
1691 op_data->op_attr.ia_valid = ia_valid;
1693 rc = ll_update_inode(inode, &md);
1694 ptlrpc_req_finished(request);
1699 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1700 * object(s) determine the file size and mtime. Otherwise, the MDS will
1701 * keep these values until such a time that objects are allocated for it.
1702 * We do the MDS operations first, as it is checking permissions for us.
1703 * We don't to the MDS RPC if there is nothing that we want to store there,
1704 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1705 * going to do an RPC anyways.
1707 * If we are doing a truncate, we will send the mtime and ctime updates
1708 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1709 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1712 * In case of HSMimport, we only set attr on MDS.
1714 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1715 enum op_xvalid xvalid, bool hsm_import)
1717 struct inode *inode = dentry->d_inode;
1718 struct ll_inode_info *lli = ll_i2info(inode);
1719 struct md_op_data *op_data = NULL;
1720 ktime_t kstart = ktime_get();
1725 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1726 "valid %x, hsm_import %d\n",
1727 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1728 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1731 if (attr->ia_valid & ATTR_SIZE) {
1732 /* Check new size against VFS/VM file size limit and rlimit */
1733 rc = inode_newsize_ok(inode, attr->ia_size);
1737 /* The maximum Lustre file size is variable, based on the
1738 * OST maximum object size and number of stripes. This
1739 * needs another check in addition to the VFS check above. */
1740 if (attr->ia_size > ll_file_maxbytes(inode)) {
1741 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1742 PFID(&lli->lli_fid), attr->ia_size,
1743 ll_file_maxbytes(inode));
1747 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1750 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1751 if (attr->ia_valid & TIMES_SET_FLAGS) {
1752 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1753 !cfs_capable(CFS_CAP_FOWNER))
1757 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1758 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1759 (attr->ia_valid & ATTR_CTIME)) {
1760 attr->ia_ctime = current_time(inode);
1761 xvalid |= OP_XVALID_CTIME_SET;
1763 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1764 (attr->ia_valid & ATTR_ATIME)) {
1765 attr->ia_atime = current_time(inode);
1766 attr->ia_valid |= ATTR_ATIME_SET;
1768 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1769 (attr->ia_valid & ATTR_MTIME)) {
1770 attr->ia_mtime = current_time(inode);
1771 attr->ia_valid |= ATTR_MTIME_SET;
1774 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1775 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1776 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1777 ktime_get_real_seconds());
1779 if (S_ISREG(inode->i_mode))
1780 inode_unlock(inode);
1782 /* We always do an MDS RPC, even if we're only changing the size;
1783 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1785 OBD_ALLOC_PTR(op_data);
1786 if (op_data == NULL)
1787 GOTO(out, rc = -ENOMEM);
1789 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1790 /* If we are changing file size, file content is
1791 * modified, flag it.
1793 xvalid |= OP_XVALID_OWNEROVERRIDE;
1794 op_data->op_bias |= MDS_DATA_MODIFIED;
1795 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1798 if (attr->ia_valid & ATTR_FILE) {
1799 struct ll_file_data *fd = attr->ia_file->private_data;
1801 if (fd->fd_lease_och)
1802 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1805 op_data->op_attr = *attr;
1806 op_data->op_xvalid = xvalid;
1808 rc = ll_md_setattr(dentry, op_data);
1812 if (!S_ISREG(inode->i_mode) || hsm_import)
1815 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1816 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1817 xvalid & OP_XVALID_CTIME_SET) {
1818 bool cached = false;
1820 rc = pcc_inode_setattr(inode, attr, &cached);
1823 CERROR("%s: PCC inode "DFID" setattr failed: "
1825 ll_i2sbi(inode)->ll_fsname,
1826 PFID(&lli->lli_fid), rc);
1830 /* For truncate and utimes sending attributes to OSTs,
1831 * setting mtime/atime to the past will be performed
1832 * under PW [0:EOF] extent lock (new_size:EOF for
1833 * truncate). It may seem excessive to send mtime/atime
1834 * updates to OSTs when not setting times to past, but
1835 * it is necessary due to possible time
1836 * de-synchronization between MDT inode and OST objects
1838 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1842 /* If the file was restored, it needs to set dirty flag.
1844 * We've already sent MDS_DATA_MODIFIED flag in
1845 * ll_md_setattr() for truncate. However, the MDT refuses to
1846 * set the HS_DIRTY flag on released files, so we have to set
1847 * it again if the file has been restored. Please check how
1848 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1850 * Please notice that if the file is not released, the previous
1851 * MDS_DATA_MODIFIED has taken effect and usually
1852 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1853 * This way we can save an RPC for common open + trunc
1855 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1856 struct hsm_state_set hss = {
1857 .hss_valid = HSS_SETMASK,
1858 .hss_setmask = HS_DIRTY,
1862 rc2 = ll_hsm_state_set(inode, &hss);
1863 /* truncate and write can happen at the same time, so that
1864 * the file can be set modified even though the file is not
1865 * restored from released state, and ll_hsm_state_set() is
1866 * not applicable for the file, and rc2 < 0 is normal in this
1869 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1870 PFID(ll_inode2fid(inode)), rc2);
1875 if (op_data != NULL)
1876 ll_finish_md_op_data(op_data);
1878 if (S_ISREG(inode->i_mode)) {
1880 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1881 inode_dio_wait(inode);
1882 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1883 * flag. ll_update_inode (called from ll_md_setattr), clears
1884 * inode flags, so there is a gap where S_NOSEC is not set.
1885 * This can cause a writer to take the i_mutex unnecessarily,
1886 * but this is safe to do and should be rare. */
1887 inode_has_no_xattr(inode);
1891 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
1892 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
1893 ktime_us_delta(ktime_get(), kstart));
1898 int ll_setattr(struct dentry *de, struct iattr *attr)
1900 int mode = de->d_inode->i_mode;
1901 enum op_xvalid xvalid = 0;
1903 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1904 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1905 xvalid |= OP_XVALID_OWNEROVERRIDE;
1907 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1908 (ATTR_SIZE|ATTR_MODE)) &&
1909 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1910 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1911 !(attr->ia_mode & S_ISGID))))
1912 attr->ia_valid |= ATTR_FORCE;
1914 if ((attr->ia_valid & ATTR_MODE) &&
1916 !(attr->ia_mode & S_ISUID) &&
1917 !(attr->ia_valid & ATTR_KILL_SUID))
1918 attr->ia_valid |= ATTR_KILL_SUID;
1920 if ((attr->ia_valid & ATTR_MODE) &&
1921 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1922 !(attr->ia_mode & S_ISGID) &&
1923 !(attr->ia_valid & ATTR_KILL_SGID))
1924 attr->ia_valid |= ATTR_KILL_SGID;
1926 return ll_setattr_raw(de, attr, xvalid, false);
1929 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1932 struct obd_statfs obd_osfs = { 0 };
1937 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1939 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1940 flags |= OBD_STATFS_NODELAY;
1942 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1946 osfs->os_type = LL_SUPER_MAGIC;
1948 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1949 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1951 if (osfs->os_state & OS_STATE_SUM)
1954 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1955 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1958 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1959 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1962 osfs->os_bsize = obd_osfs.os_bsize;
1963 osfs->os_blocks = obd_osfs.os_blocks;
1964 osfs->os_bfree = obd_osfs.os_bfree;
1965 osfs->os_bavail = obd_osfs.os_bavail;
1967 /* If we have _some_ OSTs, but don't have as many free objects on the
1968 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1969 * to compensate, so that the "inodes in use" number is correct.
1970 * This should be kept in sync with lod_statfs() behaviour.
1972 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1973 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1975 osfs->os_ffree = obd_osfs.os_ffree;
1982 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1984 struct super_block *sb = de->d_sb;
1985 struct obd_statfs osfs;
1986 __u64 fsid = huge_encode_dev(sb->s_dev);
1987 ktime_t kstart = ktime_get();
1990 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
1992 /* Some amount of caching on the client is allowed */
1993 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1997 statfs_unpack(sfs, &osfs);
1999 /* We need to downshift for all 32-bit kernels, because we can't
2000 * tell if the kernel is being called via sys_statfs64() or not.
2001 * Stop before overflowing f_bsize - in which case it is better
2002 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2003 if (sizeof(long) < 8) {
2004 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2007 osfs.os_blocks >>= 1;
2008 osfs.os_bfree >>= 1;
2009 osfs.os_bavail >>= 1;
2013 sfs->f_blocks = osfs.os_blocks;
2014 sfs->f_bfree = osfs.os_bfree;
2015 sfs->f_bavail = osfs.os_bavail;
2016 sfs->f_fsid.val[0] = (__u32)fsid;
2017 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2019 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2020 ktime_us_delta(ktime_get(), kstart));
2025 void ll_inode_size_lock(struct inode *inode)
2027 struct ll_inode_info *lli;
2029 LASSERT(!S_ISDIR(inode->i_mode));
2031 lli = ll_i2info(inode);
2032 mutex_lock(&lli->lli_size_mutex);
2035 void ll_inode_size_unlock(struct inode *inode)
2037 struct ll_inode_info *lli;
2039 lli = ll_i2info(inode);
2040 mutex_unlock(&lli->lli_size_mutex);
2043 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2045 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2046 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2047 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2049 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2052 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2054 struct ll_inode_info *lli = ll_i2info(inode);
2055 struct mdt_body *body = md->body;
2056 struct ll_sb_info *sbi = ll_i2sbi(inode);
2059 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2060 rc = cl_file_inode_init(inode, md);
2065 if (S_ISDIR(inode->i_mode)) {
2066 rc = ll_update_lsm_md(inode, md);
2071 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2072 if (body->mbo_valid & OBD_MD_FLACL) {
2073 spin_lock(&lli->lli_lock);
2074 if (lli->lli_posix_acl)
2075 posix_acl_release(lli->lli_posix_acl);
2076 lli->lli_posix_acl = md->posix_acl;
2077 spin_unlock(&lli->lli_lock);
2080 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2081 sbi->ll_flags & LL_SBI_32BIT_API);
2082 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2084 if (body->mbo_valid & OBD_MD_FLATIME) {
2085 if (body->mbo_atime > inode->i_atime.tv_sec)
2086 inode->i_atime.tv_sec = body->mbo_atime;
2087 lli->lli_atime = body->mbo_atime;
2090 if (body->mbo_valid & OBD_MD_FLMTIME) {
2091 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2093 "setting ino %lu mtime from %lld to %llu\n",
2094 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2096 inode->i_mtime.tv_sec = body->mbo_mtime;
2098 lli->lli_mtime = body->mbo_mtime;
2101 if (body->mbo_valid & OBD_MD_FLCTIME) {
2102 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2103 inode->i_ctime.tv_sec = body->mbo_ctime;
2104 lli->lli_ctime = body->mbo_ctime;
2107 if (body->mbo_valid & OBD_MD_FLBTIME)
2108 lli->lli_btime = body->mbo_btime;
2110 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2111 if (body->mbo_valid & OBD_MD_FLFLAGS)
2112 ll_update_inode_flags(inode, body->mbo_flags);
2113 if (body->mbo_valid & OBD_MD_FLMODE)
2114 inode->i_mode = (inode->i_mode & S_IFMT) |
2115 (body->mbo_mode & ~S_IFMT);
2117 if (body->mbo_valid & OBD_MD_FLTYPE)
2118 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2119 (body->mbo_mode & S_IFMT);
2121 LASSERT(inode->i_mode != 0);
2122 if (body->mbo_valid & OBD_MD_FLUID)
2123 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2124 if (body->mbo_valid & OBD_MD_FLGID)
2125 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2126 if (body->mbo_valid & OBD_MD_FLPROJID)
2127 lli->lli_projid = body->mbo_projid;
2128 if (body->mbo_valid & OBD_MD_FLNLINK)
2129 set_nlink(inode, body->mbo_nlink);
2130 if (body->mbo_valid & OBD_MD_FLRDEV)
2131 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2133 if (body->mbo_valid & OBD_MD_FLID) {
2134 /* FID shouldn't be changed! */
2135 if (fid_is_sane(&lli->lli_fid)) {
2136 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2137 "Trying to change FID "DFID
2138 " to the "DFID", inode "DFID"(%p)\n",
2139 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2140 PFID(ll_inode2fid(inode)), inode);
2142 lli->lli_fid = body->mbo_fid1;
2146 LASSERT(fid_seq(&lli->lli_fid) != 0);
2148 lli->lli_attr_valid = body->mbo_valid;
2149 if (body->mbo_valid & OBD_MD_FLSIZE) {
2150 i_size_write(inode, body->mbo_size);
2152 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2153 PFID(ll_inode2fid(inode)),
2154 (unsigned long long)body->mbo_size);
2156 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2157 inode->i_blocks = body->mbo_blocks;
2159 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2160 lli->lli_lazysize = body->mbo_size;
2161 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2162 lli->lli_lazyblocks = body->mbo_blocks;
2165 if (body->mbo_valid & OBD_MD_TSTATE) {
2166 /* Set LLIF_FILE_RESTORING if restore ongoing and
2167 * clear it when done to ensure to start again
2168 * glimpsing updated attrs
2170 if (body->mbo_t_state & MS_RESTORE)
2171 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2173 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2179 int ll_read_inode2(struct inode *inode, void *opaque)
2181 struct lustre_md *md = opaque;
2182 struct ll_inode_info *lli = ll_i2info(inode);
2186 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2187 PFID(&lli->lli_fid), inode);
2189 /* Core attributes from the MDS first. This is a new inode, and
2190 * the VFS doesn't zero times in the core inode so we have to do
2191 * it ourselves. They will be overwritten by either MDS or OST
2192 * attributes - we just need to make sure they aren't newer.
2194 inode->i_mtime.tv_sec = 0;
2195 inode->i_atime.tv_sec = 0;
2196 inode->i_ctime.tv_sec = 0;
2198 rc = ll_update_inode(inode, md);
2202 /* OIDEBUG(inode); */
2204 #ifdef HAVE_BACKING_DEV_INFO
2205 /* initializing backing dev info. */
2206 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2208 if (S_ISREG(inode->i_mode)) {
2209 struct ll_sb_info *sbi = ll_i2sbi(inode);
2210 inode->i_op = &ll_file_inode_operations;
2211 inode->i_fop = sbi->ll_fop;
2212 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2214 } else if (S_ISDIR(inode->i_mode)) {
2215 inode->i_op = &ll_dir_inode_operations;
2216 inode->i_fop = &ll_dir_operations;
2218 } else if (S_ISLNK(inode->i_mode)) {
2219 inode->i_op = &ll_fast_symlink_inode_operations;
2222 inode->i_op = &ll_special_inode_operations;
2224 init_special_inode(inode, inode->i_mode,
2233 void ll_delete_inode(struct inode *inode)
2235 struct ll_inode_info *lli = ll_i2info(inode);
2236 struct address_space *mapping = &inode->i_data;
2237 unsigned long nrpages;
2238 unsigned long flags;
2242 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2243 /* It is last chance to write out dirty pages,
2244 * otherwise we may lose data while umount.
2246 * If i_nlink is 0 then just discard data. This is safe because
2247 * local inode gets i_nlink 0 from server only for the last
2248 * unlink, so that file is not opened somewhere else
2250 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2251 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2253 truncate_inode_pages_final(mapping);
2255 /* Workaround for LU-118: Note nrpages may not be totally updated when
2256 * truncate_inode_pages() returns, as there can be a page in the process
2257 * of deletion (inside __delete_from_page_cache()) in the specified
2258 * range. Thus mapping->nrpages can be non-zero when this function
2259 * returns even after truncation of the whole mapping. Only do this if
2260 * npages isn't already zero.
2262 nrpages = mapping->nrpages;
2264 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2265 nrpages = mapping->nrpages;
2266 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2267 } /* Workaround end */
2269 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2270 "see https://jira.whamcloud.com/browse/LU-118\n",
2271 ll_i2sbi(inode)->ll_fsname,
2272 PFID(ll_inode2fid(inode)), inode, nrpages);
2274 ll_clear_inode(inode);
2280 int ll_iocontrol(struct inode *inode, struct file *file,
2281 unsigned int cmd, unsigned long arg)
2283 struct ll_sb_info *sbi = ll_i2sbi(inode);
2284 struct ptlrpc_request *req = NULL;
2289 case FS_IOC_GETFLAGS: {
2290 struct mdt_body *body;
2291 struct md_op_data *op_data;
2293 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2294 0, 0, LUSTRE_OPC_ANY,
2296 if (IS_ERR(op_data))
2297 RETURN(PTR_ERR(op_data));
2299 op_data->op_valid = OBD_MD_FLFLAGS;
2300 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2301 ll_finish_md_op_data(op_data);
2303 CERROR("%s: failure inode "DFID": rc = %d\n",
2304 sbi->ll_md_exp->exp_obd->obd_name,
2305 PFID(ll_inode2fid(inode)), rc);
2309 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2311 flags = body->mbo_flags;
2313 ptlrpc_req_finished(req);
2315 RETURN(put_user(flags, (int __user *)arg));
2317 case FS_IOC_SETFLAGS: {
2319 struct md_op_data *op_data;
2320 struct cl_object *obj;
2321 struct fsxattr fa = { 0 };
2323 if (get_user(flags, (int __user *)arg))
2326 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2327 if (flags & LUSTRE_PROJINHERIT_FL)
2328 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2330 rc = ll_ioctl_check_project(inode, &fa);
2334 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2335 LUSTRE_OPC_ANY, NULL);
2336 if (IS_ERR(op_data))
2337 RETURN(PTR_ERR(op_data));
2339 op_data->op_attr_flags = flags;
2340 op_data->op_xvalid |= OP_XVALID_FLAGS;
2341 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2342 ll_finish_md_op_data(op_data);
2343 ptlrpc_req_finished(req);
2347 ll_update_inode_flags(inode, flags);
2349 obj = ll_i2info(inode)->lli_clob;
2353 OBD_ALLOC_PTR(attr);
2357 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2369 int ll_flush_ctx(struct inode *inode)
2371 struct ll_sb_info *sbi = ll_i2sbi(inode);
2373 CDEBUG(D_SEC, "flush context for user %d\n",
2374 from_kuid(&init_user_ns, current_uid()));
2376 obd_set_info_async(NULL, sbi->ll_md_exp,
2377 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2379 obd_set_info_async(NULL, sbi->ll_dt_exp,
2380 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2385 /* umount -f client means force down, don't save state */
2386 void ll_umount_begin(struct super_block *sb)
2388 struct ll_sb_info *sbi = ll_s2sbi(sb);
2389 struct obd_device *obd;
2390 struct obd_ioctl_data *ioc_data;
2394 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2395 sb->s_count, atomic_read(&sb->s_active));
2397 obd = class_exp2obd(sbi->ll_md_exp);
2399 CERROR("Invalid MDC connection handle %#llx\n",
2400 sbi->ll_md_exp->exp_handle.h_cookie);
2406 obd = class_exp2obd(sbi->ll_dt_exp);
2408 CERROR("Invalid LOV connection handle %#llx\n",
2409 sbi->ll_dt_exp->exp_handle.h_cookie);
2415 OBD_ALLOC_PTR(ioc_data);
2417 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2418 sizeof *ioc_data, ioc_data, NULL);
2420 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2421 sizeof *ioc_data, ioc_data, NULL);
2423 OBD_FREE_PTR(ioc_data);
2426 /* Really, we'd like to wait until there are no requests outstanding,
2427 * and then continue. For now, we just periodically checking for vfs
2428 * to decrement mnt_cnt and hope to finish it within 10sec.
2432 !may_umount(sbi->ll_mnt.mnt)) {
2440 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2442 struct ll_sb_info *sbi = ll_s2sbi(sb);
2443 char *profilenm = get_profile_name(sb);
2447 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2448 read_only = *flags & MS_RDONLY;
2449 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2450 sizeof(KEY_READ_ONLY),
2451 KEY_READ_ONLY, sizeof(read_only),
2454 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2455 profilenm, read_only ?
2456 "read-only" : "read-write", err);
2461 sb->s_flags |= SB_RDONLY;
2463 sb->s_flags &= ~SB_RDONLY;
2465 if (sbi->ll_flags & LL_SBI_VERBOSE)
2466 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2467 read_only ? "read-only" : "read-write");
2473 * Cleanup the open handle that is cached on MDT-side.
2475 * For open case, the client side open handling thread may hit error
2476 * after the MDT grant the open. Under such case, the client should
2477 * send close RPC to the MDT as cleanup; otherwise, the open handle
2478 * on the MDT will be leaked there until the client umount or evicted.
2480 * In further, if someone unlinked the file, because the open handle
2481 * holds the reference on such file/object, then it will block the
2482 * subsequent threads that want to locate such object via FID.
2484 * \param[in] sb super block for this file-system
2485 * \param[in] open_req pointer to the original open request
2487 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2489 struct mdt_body *body;
2490 struct md_op_data *op_data;
2491 struct ptlrpc_request *close_req = NULL;
2492 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2495 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2496 OBD_ALLOC_PTR(op_data);
2497 if (op_data == NULL) {
2498 CWARN("%s: cannot allocate op_data to release open handle for "
2499 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2504 op_data->op_fid1 = body->mbo_fid1;
2505 op_data->op_open_handle = body->mbo_open_handle;
2506 op_data->op_mod_time = ktime_get_real_seconds();
2507 md_close(exp, op_data, NULL, &close_req);
2508 ptlrpc_req_finished(close_req);
2509 ll_finish_md_op_data(op_data);
2514 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2515 struct super_block *sb, struct lookup_intent *it)
2517 struct ll_sb_info *sbi = NULL;
2518 struct lustre_md md = { NULL };
2519 bool default_lmv_deleted = false;
2524 LASSERT(*inode || sb);
2525 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2526 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2527 sbi->ll_md_exp, &md);
2532 * clear default_lmv only if intent_getattr reply doesn't contain it.
2533 * but it needs to be done after iget, check this early because
2534 * ll_update_lsm_md() may change md.
2536 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2537 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2538 default_lmv_deleted = true;
2541 rc = ll_update_inode(*inode, &md);
2545 LASSERT(sb != NULL);
2548 * At this point server returns to client's same fid as client
2549 * generated for creating. So using ->fid1 is okay here.
2551 if (!fid_is_sane(&md.body->mbo_fid1)) {
2552 CERROR("%s: Fid is insane "DFID"\n",
2554 PFID(&md.body->mbo_fid1));
2555 GOTO(out, rc = -EINVAL);
2558 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2559 sbi->ll_flags & LL_SBI_32BIT_API),
2561 if (IS_ERR(*inode)) {
2562 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2564 posix_acl_release(md.posix_acl);
2565 md.posix_acl = NULL;
2568 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2570 CERROR("new_inode -fatal: rc %d\n", rc);
2575 /* Handling piggyback layout lock.
2576 * Layout lock can be piggybacked by getattr and open request.
2577 * The lsm can be applied to inode only if it comes with a layout lock
2578 * otherwise correct layout may be overwritten, for example:
2579 * 1. proc1: mdt returns a lsm but not granting layout
2580 * 2. layout was changed by another client
2581 * 3. proc2: refresh layout and layout lock granted
2582 * 4. proc1: to apply a stale layout */
2583 if (it != NULL && it->it_lock_mode != 0) {
2584 struct lustre_handle lockh;
2585 struct ldlm_lock *lock;
2587 lockh.cookie = it->it_lock_handle;
2588 lock = ldlm_handle2lock(&lockh);
2589 LASSERT(lock != NULL);
2590 if (ldlm_has_layout(lock)) {
2591 struct cl_object_conf conf;
2593 memset(&conf, 0, sizeof(conf));
2594 conf.coc_opc = OBJECT_CONF_SET;
2595 conf.coc_inode = *inode;
2596 conf.coc_lock = lock;
2597 conf.u.coc_layout = md.layout;
2598 (void)ll_layout_conf(*inode, &conf);
2600 LDLM_LOCK_PUT(lock);
2603 if (default_lmv_deleted)
2604 ll_update_default_lsm_md(*inode, &md);
2609 /* cleanup will be done if necessary */
2610 md_free_lustre_md(sbi->ll_md_exp, &md);
2612 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2613 ll_intent_drop_lock(it);
2614 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2620 int ll_obd_statfs(struct inode *inode, void __user *arg)
2622 struct ll_sb_info *sbi = NULL;
2623 struct obd_export *exp;
2625 struct obd_ioctl_data *data = NULL;
2629 if (!inode || !(sbi = ll_i2sbi(inode)))
2630 GOTO(out_statfs, rc = -EINVAL);
2632 rc = obd_ioctl_getdata(&buf, &len, arg);
2634 GOTO(out_statfs, rc);
2637 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2638 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2639 GOTO(out_statfs, rc = -EINVAL);
2641 if (data->ioc_inllen1 != sizeof(__u32) ||
2642 data->ioc_inllen2 != sizeof(__u32) ||
2643 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2644 data->ioc_plen2 != sizeof(struct obd_uuid))
2645 GOTO(out_statfs, rc = -EINVAL);
2647 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2648 if (type & LL_STATFS_LMV)
2649 exp = sbi->ll_md_exp;
2650 else if (type & LL_STATFS_LOV)
2651 exp = sbi->ll_dt_exp;
2653 GOTO(out_statfs, rc = -ENODEV);
2655 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2657 GOTO(out_statfs, rc);
2659 OBD_FREE_LARGE(buf, len);
2664 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2665 * be called early to avoid deadlock.
2667 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2669 if (op_data->op_mea2_sem) {
2670 up_read(op_data->op_mea2_sem);
2671 op_data->op_mea2_sem = NULL;
2674 if (op_data->op_mea1_sem) {
2675 up_read(op_data->op_mea1_sem);
2676 op_data->op_mea1_sem = NULL;
2680 /* this function prepares md_op_data hint for passing it down to MD stack. */
2681 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2682 struct inode *i1, struct inode *i2,
2683 const char *name, size_t namelen,
2684 __u32 mode, enum md_op_code opc,
2687 LASSERT(i1 != NULL);
2690 /* Do not reuse namelen for something else. */
2692 return ERR_PTR(-EINVAL);
2694 if (namelen > ll_i2sbi(i1)->ll_namelen)
2695 return ERR_PTR(-ENAMETOOLONG);
2697 if (!lu_name_is_valid_2(name, namelen))
2698 return ERR_PTR(-EINVAL);
2701 if (op_data == NULL)
2702 OBD_ALLOC_PTR(op_data);
2704 if (op_data == NULL)
2705 return ERR_PTR(-ENOMEM);
2707 ll_i2gids(op_data->op_suppgids, i1, i2);
2708 op_data->op_fid1 = *ll_inode2fid(i1);
2709 op_data->op_code = opc;
2711 if (S_ISDIR(i1->i_mode)) {
2712 down_read(&ll_i2info(i1)->lli_lsm_sem);
2713 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2714 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2715 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2719 op_data->op_fid2 = *ll_inode2fid(i2);
2720 if (S_ISDIR(i2->i_mode)) {
2722 down_read(&ll_i2info(i2)->lli_lsm_sem);
2723 op_data->op_mea2_sem =
2724 &ll_i2info(i2)->lli_lsm_sem;
2726 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2729 fid_zero(&op_data->op_fid2);
2732 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2733 op_data->op_cli_flags |= CLI_HASH64;
2735 if (ll_need_32bit_api(ll_i2sbi(i1)))
2736 op_data->op_cli_flags |= CLI_API32;
2738 op_data->op_name = name;
2739 op_data->op_namelen = namelen;
2740 op_data->op_mode = mode;
2741 op_data->op_mod_time = ktime_get_real_seconds();
2742 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2743 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2744 op_data->op_cap = cfs_curproc_cap_pack();
2745 op_data->op_mds = 0;
2746 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2747 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2748 op_data->op_bias |= MDS_CREATE_VOLATILE;
2750 op_data->op_data = data;
2755 void ll_finish_md_op_data(struct md_op_data *op_data)
2757 ll_unlock_md_op_lsm(op_data);
2758 security_release_secctx(op_data->op_file_secctx,
2759 op_data->op_file_secctx_size);
2760 OBD_FREE_PTR(op_data);
2763 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2765 struct ll_sb_info *sbi;
2767 LASSERT(seq && dentry);
2768 sbi = ll_s2sbi(dentry->d_sb);
2770 if (sbi->ll_flags & LL_SBI_NOLCK)
2771 seq_puts(seq, ",nolock");
2773 /* "flock" is the default since 2.13, but it wasn't for many years,
2774 * so it is still useful to print this to show it is enabled.
2775 * Start to print "noflock" so it is now clear when flock is disabled.
2777 if (sbi->ll_flags & LL_SBI_FLOCK)
2778 seq_puts(seq, ",flock");
2779 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2780 seq_puts(seq, ",localflock");
2782 seq_puts(seq, ",noflock");
2784 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2785 seq_puts(seq, ",user_xattr");
2787 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2788 seq_puts(seq, ",lazystatfs");
2790 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2791 seq_puts(seq, ",user_fid2path");
2793 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2794 seq_puts(seq, ",always_ping");
2800 * Get obd name by cmd, and copy out to user space
2802 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2804 struct ll_sb_info *sbi = ll_i2sbi(inode);
2805 struct obd_device *obd;
2808 if (cmd == OBD_IOC_GETDTNAME)
2809 obd = class_exp2obd(sbi->ll_dt_exp);
2810 else if (cmd == OBD_IOC_GETMDNAME)
2811 obd = class_exp2obd(sbi->ll_md_exp);
2818 if (copy_to_user((void __user *)arg, obd->obd_name,
2819 strlen(obd->obd_name) + 1))
2825 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2832 p.mnt = current->fs->root.mnt;
2834 path = d_path(&p, buf, bufsize);
2839 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2841 char *buf, *path = NULL;
2842 struct dentry *dentry = NULL;
2843 struct inode *inode = page->mapping->host;
2845 /* this can be called inside spin lock so use GFP_ATOMIC. */
2846 buf = (char *)__get_free_page(GFP_ATOMIC);
2848 dentry = d_find_alias(page->mapping->host);
2850 path = ll_d_path(dentry, buf, PAGE_SIZE);
2853 /* The below message is checked in recovery-small.sh test_24b */
2855 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2856 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2857 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2858 PFID(ll_inode2fid(inode)),
2859 (path && !IS_ERR(path)) ? path : "", ioret);
2865 free_page((unsigned long)buf);
2868 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2869 struct lov_user_md **kbuf)
2871 struct lov_user_md lum;
2875 if (copy_from_user(&lum, md, sizeof(lum)))
2878 lum_size = ll_lov_user_md_size(&lum);
2882 OBD_ALLOC_LARGE(*kbuf, lum_size);
2886 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2887 OBD_FREE_LARGE(*kbuf, lum_size);
2895 * Compute llite root squash state after a change of root squash
2896 * configuration setting or add/remove of a lnet nid
2898 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2900 struct root_squash_info *squash = &sbi->ll_squash;
2903 struct lnet_process_id id;
2905 /* Update norootsquash flag */
2906 spin_lock(&squash->rsi_lock);
2907 if (list_empty(&squash->rsi_nosquash_nids))
2908 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2910 /* Do not apply root squash as soon as one of our NIDs is
2911 * in the nosquash_nids list */
2914 while (LNetGetId(i++, &id) != -ENOENT) {
2915 if (id.nid == LNET_NID_LO_0)
2917 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2923 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2925 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2927 spin_unlock(&squash->rsi_lock);
2931 * Parse linkea content to extract information about a given hardlink
2933 * \param[in] ldata - Initialized linkea data
2934 * \param[in] linkno - Link identifier
2935 * \param[out] parent_fid - The entry's parent FID
2936 * \param[out] ln - Entry name destination buffer
2938 * \retval 0 on success
2939 * \retval Appropriate negative error code on failure
2941 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2942 struct lu_fid *parent_fid, struct lu_name *ln)
2948 rc = linkea_init_with_rec(ldata);
2952 if (linkno >= ldata->ld_leh->leh_reccount)
2953 /* beyond last link */
2956 linkea_first_entry(ldata);
2957 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2958 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2963 linkea_next_entry(ldata);
2973 * Get parent FID and name of an identified link. Operation is performed for
2974 * a given link number, letting the caller iterate over linkno to list one or
2975 * all links of an entry.
2977 * \param[in] file - File descriptor against which to perform the operation
2978 * \param[in,out] arg - User-filled structure containing the linkno to operate
2979 * on and the available size. It is eventually filled with
2980 * the requested information or left untouched on error
2982 * \retval - 0 on success
2983 * \retval - Appropriate negative error code on failure
2985 int ll_getparent(struct file *file, struct getparent __user *arg)
2987 struct inode *inode = file_inode(file);
2988 struct linkea_data *ldata;
2989 struct lu_buf buf = LU_BUF_NULL;
2991 struct lu_fid parent_fid;
2998 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2999 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3002 if (get_user(name_size, &arg->gp_name_size))
3005 if (get_user(linkno, &arg->gp_linkno))
3008 if (name_size > PATH_MAX)
3011 OBD_ALLOC(ldata, sizeof(*ldata));
3015 rc = linkea_data_new(ldata, &buf);
3017 GOTO(ldata_free, rc);
3019 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3020 buf.lb_len, OBD_MD_FLXATTR);
3024 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3028 if (ln.ln_namelen >= name_size)
3029 GOTO(lb_free, rc = -EOVERFLOW);
3031 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3032 GOTO(lb_free, rc = -EFAULT);
3034 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3035 GOTO(lb_free, rc = -EFAULT);
3037 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3038 GOTO(lb_free, rc = -EFAULT);
3043 OBD_FREE(ldata, sizeof(*ldata));