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))
74 * If there is only one number of core visible to Lustre,
75 * async readahead will be disabled, to avoid massive over
76 * subscription, we use 1/2 of active cores as default max
77 * async readahead requests.
79 static inline unsigned int ll_get_ra_async_max_active(void)
81 return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
84 static struct ll_sb_info *ll_init_sbi(void)
86 struct ll_sb_info *sbi = NULL;
88 unsigned long lru_page_max;
97 RETURN(ERR_PTR(-ENOMEM));
99 rc = pcc_super_init(&sbi->ll_pcc_super);
103 spin_lock_init(&sbi->ll_lock);
104 mutex_init(&sbi->ll_lco.lco_lock);
105 spin_lock_init(&sbi->ll_pp_extent_lock);
106 spin_lock_init(&sbi->ll_process_lock);
107 sbi->ll_rw_stats_on = 0;
108 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
111 pages = si.totalram - si.totalhigh;
112 lru_page_max = pages / 2;
114 sbi->ll_ra_info.ra_async_max_active = 0;
115 sbi->ll_ra_info.ll_readahead_wq =
116 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
117 sbi->ll_ra_info.ra_async_max_active);
118 if (!sbi->ll_ra_info.ll_readahead_wq)
119 GOTO(out_pcc, rc = -ENOMEM);
121 /* initialize ll_cache data */
122 sbi->ll_cache = cl_cache_init(lru_page_max);
123 if (sbi->ll_cache == NULL)
124 GOTO(out_destroy_ra, rc = -ENOMEM);
126 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
127 SBI_DEFAULT_READ_AHEAD_MAX);
128 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
129 sbi->ll_ra_info.ra_max_pages_per_file;
130 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
131 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
132 sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
133 atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
135 sbi->ll_flags |= LL_SBI_VERBOSE;
136 #ifdef ENABLE_CHECKSUM
137 sbi->ll_flags |= LL_SBI_CHECKSUM;
140 sbi->ll_flags |= LL_SBI_FLOCK;
143 #ifdef HAVE_LRU_RESIZE_SUPPORT
144 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
146 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
148 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
149 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
151 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
155 /* metadata statahead is enabled by default */
156 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
157 sbi->ll_sa_max = LL_SA_RPC_DEF;
158 atomic_set(&sbi->ll_sa_total, 0);
159 atomic_set(&sbi->ll_sa_wrong, 0);
160 atomic_set(&sbi->ll_sa_running, 0);
161 atomic_set(&sbi->ll_agl_total, 0);
162 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
163 sbi->ll_flags |= LL_SBI_FAST_READ;
164 sbi->ll_flags |= LL_SBI_TINY_WRITE;
167 sbi->ll_squash.rsi_uid = 0;
168 sbi->ll_squash.rsi_gid = 0;
169 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
170 spin_lock_init(&sbi->ll_squash.rsi_lock);
172 /* Per-filesystem file heat */
173 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
174 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
177 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
179 pcc_super_fini(&sbi->ll_pcc_super);
185 static void ll_free_sbi(struct super_block *sb)
187 struct ll_sb_info *sbi = ll_s2sbi(sb);
191 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
192 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
193 if (sbi->ll_ra_info.ll_readahead_wq)
194 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
195 if (sbi->ll_cache != NULL) {
196 cl_cache_decref(sbi->ll_cache);
197 sbi->ll_cache = NULL;
199 pcc_super_fini(&sbi->ll_pcc_super);
200 OBD_FREE(sbi, sizeof(*sbi));
205 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
207 struct inode *root = NULL;
208 struct ll_sb_info *sbi = ll_s2sbi(sb);
209 struct obd_statfs *osfs = NULL;
210 struct ptlrpc_request *request = NULL;
211 struct obd_connect_data *data = NULL;
212 struct obd_uuid *uuid;
213 struct md_op_data *op_data;
214 struct lustre_md lmd;
216 int size, err, checksum;
219 sbi->ll_md_obd = class_name2obd(md);
220 if (!sbi->ll_md_obd) {
221 CERROR("MD %s: not setup or attached\n", md);
235 /* pass client page size via ocd_grant_blkbits, the server should report
236 * back its backend blocksize for grant calculation purpose */
237 data->ocd_grant_blkbits = PAGE_SHIFT;
239 /* indicate MDT features supported by this client */
240 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
241 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
242 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
243 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
244 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
245 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
246 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
247 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
248 OBD_CONNECT_64BITHASH |
249 OBD_CONNECT_EINPROGRESS |
250 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
251 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
252 OBD_CONNECT_MAX_EASIZE |
253 OBD_CONNECT_FLOCK_DEAD |
254 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
255 OBD_CONNECT_OPEN_BY_FID |
256 OBD_CONNECT_DIR_STRIPE |
257 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
258 OBD_CONNECT_SUBTREE |
259 OBD_CONNECT_MULTIMODRPCS |
260 OBD_CONNECT_GRANT_PARAM |
261 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
263 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
264 OBD_CONNECT2_SUM_STATFS |
265 OBD_CONNECT2_OVERSTRIPING |
267 OBD_CONNECT2_LOCK_CONVERT |
268 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
269 OBD_CONNECT2_INC_XID |
271 OBD_CONNECT2_ASYNC_DISCARD |
275 #ifdef HAVE_LRU_RESIZE_SUPPORT
276 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
277 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
279 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
280 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
281 OBD_CONNECT_LARGE_ACL;
284 data->ocd_cksum_types = obd_cksum_types_supported_client();
286 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
287 /* flag mdc connection as lightweight, only used for test
288 * purpose, use with care */
289 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
291 data->ocd_ibits_known = MDS_INODELOCK_FULL;
292 data->ocd_version = LUSTRE_VERSION_CODE;
294 if (sb->s_flags & SB_RDONLY)
295 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
296 if (sbi->ll_flags & LL_SBI_USER_XATTR)
297 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
300 /* Setting this indicates we correctly support S_NOSEC (See kernel
301 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
303 sb->s_flags |= SB_NOSEC;
306 if (sbi->ll_flags & LL_SBI_FLOCK)
307 sbi->ll_fop = &ll_file_operations_flock;
308 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
309 sbi->ll_fop = &ll_file_operations;
311 sbi->ll_fop = &ll_file_operations_noflock;
313 /* always ping even if server suppress_pings */
314 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
315 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
317 obd_connect_set_secctx(data);
319 #if defined(CONFIG_SECURITY)
320 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
323 data->ocd_brw_size = MD_MAX_BRW_SIZE;
325 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
326 &sbi->ll_sb_uuid, data, sbi->ll_cache);
328 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
329 "recovery, of which this client is not a "
330 "part. Please wait for recovery to complete,"
331 " abort, or time out.\n", md);
334 CERROR("cannot connect to %s: rc = %d\n", md, err);
338 sbi->ll_md_exp->exp_connect_data = *data;
340 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
341 LUSTRE_SEQ_METADATA);
343 CERROR("%s: Can't init metadata layer FID infrastructure, "
344 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
348 /* For mount, we only need fs info from MDT0, and also in DNE, it
349 * can make sure the client can be mounted as long as MDT0 is
351 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
352 ktime_get_seconds() - sbi->ll_statfs_max_age,
353 OBD_STATFS_FOR_MDT0);
355 GOTO(out_md_fid, err);
357 /* This needs to be after statfs to ensure connect has finished.
358 * Note that "data" does NOT contain the valid connect reply.
359 * If connecting to a 1.8 server there will be no LMV device, so
360 * we can access the MDC export directly and exp_connect_flags will
361 * be non-zero, but if accessing an upgraded 2.1 server it will
362 * have the correct flags filled in.
363 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
364 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
365 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
366 valid != CLIENT_CONNECT_MDT_REQD) {
369 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
370 obd_connect_flags2str(buf, PAGE_SIZE,
371 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
372 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
373 "feature(s) needed for correct operation "
374 "of this client (%s). Please upgrade "
375 "server or downgrade client.\n",
376 sbi->ll_md_exp->exp_obd->obd_name, buf);
377 OBD_FREE(buf, PAGE_SIZE);
378 GOTO(out_md_fid, err = -EPROTO);
381 size = sizeof(*data);
382 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
383 KEY_CONN_DATA, &size, data);
385 CERROR("%s: Get connect data failed: rc = %d\n",
386 sbi->ll_md_exp->exp_obd->obd_name, err);
387 GOTO(out_md_fid, err);
390 LASSERT(osfs->os_bsize);
391 sb->s_blocksize = osfs->os_bsize;
392 sb->s_blocksize_bits = log2(osfs->os_bsize);
393 sb->s_magic = LL_SUPER_MAGIC;
394 sb->s_maxbytes = MAX_LFS_FILESIZE;
395 sbi->ll_namelen = osfs->os_namelen;
396 sbi->ll_mnt.mnt = current->fs->root.mnt;
398 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
399 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
400 LCONSOLE_INFO("Disabling user_xattr feature because "
401 "it is not supported on the server\n");
402 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
405 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
407 sb->s_flags |= SB_POSIXACL;
409 sbi->ll_flags |= LL_SBI_ACL;
411 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
413 sb->s_flags &= ~SB_POSIXACL;
415 sbi->ll_flags &= ~LL_SBI_ACL;
418 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
419 sbi->ll_flags |= LL_SBI_64BIT_HASH;
421 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
422 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
424 if (obd_connect_has_secctx(data))
425 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
427 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
428 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
429 LCONSOLE_INFO("%s: disabling xattr cache due to "
430 "unknown maximum xattr size.\n", dt);
431 } else if (!sbi->ll_xattr_cache_set) {
432 /* If xattr_cache is already set (no matter 0 or 1)
433 * during processing llog, it won't be enabled here. */
434 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
435 sbi->ll_xattr_cache_enabled = 1;
439 sbi->ll_dt_obd = class_name2obd(dt);
440 if (!sbi->ll_dt_obd) {
441 CERROR("DT %s: not setup or attached\n", dt);
442 GOTO(out_md_fid, err = -ENODEV);
445 /* pass client page size via ocd_grant_blkbits, the server should report
446 * back its backend blocksize for grant calculation purpose */
447 data->ocd_grant_blkbits = PAGE_SHIFT;
449 /* indicate OST features supported by this client */
450 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
451 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
452 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
453 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
454 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
455 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
456 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
457 OBD_CONNECT_EINPROGRESS |
458 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
459 OBD_CONNECT_LAYOUTLOCK |
460 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
461 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
462 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
463 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
464 OBD_CONNECT2_INC_XID;
466 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
467 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
469 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
470 * disabled by default, because it can still be enabled on the
471 * fly via /sys. As a consequence, we still need to come to an
472 * agreement on the supported algorithms at connect time
474 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
476 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
477 data->ocd_cksum_types = OBD_CKSUM_ADLER;
479 data->ocd_cksum_types = obd_cksum_types_supported_client();
481 #ifdef HAVE_LRU_RESIZE_SUPPORT
482 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
484 /* always ping even if server suppress_pings */
485 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
486 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
488 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
489 "ocd_grant: %d\n", data->ocd_connect_flags,
490 data->ocd_version, data->ocd_grant);
492 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
493 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
495 data->ocd_brw_size = DT_MAX_BRW_SIZE;
497 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
498 &sbi->ll_sb_uuid, data, sbi->ll_cache);
500 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
501 "recovery, of which this client is not a "
502 "part. Please wait for recovery to "
503 "complete, abort, or time out.\n", dt);
506 CERROR("%s: Cannot connect to %s: rc = %d\n",
507 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
511 sbi->ll_dt_exp->exp_connect_data = *data;
513 /* Don't change value if it was specified in the config log */
514 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
515 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
516 max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
517 (data->ocd_brw_size >> PAGE_SHIFT));
518 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
519 sbi->ll_ra_info.ra_max_pages_per_file)
520 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
521 sbi->ll_ra_info.ra_max_pages_per_file;
524 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
525 LUSTRE_SEQ_METADATA);
527 CERROR("%s: Can't init data layer FID infrastructure, "
528 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
532 mutex_lock(&sbi->ll_lco.lco_lock);
533 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
534 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
535 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
536 mutex_unlock(&sbi->ll_lco.lco_lock);
538 fid_zero(&sbi->ll_root_fid);
539 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
542 CERROR("cannot mds_connect: rc = %d\n", err);
543 GOTO(out_lock_cn_cb, err);
545 if (!fid_is_sane(&sbi->ll_root_fid)) {
546 CERROR("%s: Invalid root fid "DFID" during mount\n",
547 sbi->ll_md_exp->exp_obd->obd_name,
548 PFID(&sbi->ll_root_fid));
549 GOTO(out_lock_cn_cb, err = -EINVAL);
551 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
553 sb->s_op = &lustre_super_operations;
554 sb->s_xattr = ll_xattr_handlers;
555 #if THREAD_SIZE >= 8192 /*b=17630*/
556 sb->s_export_op = &lustre_export_operations;
560 * XXX: move this to after cbd setup? */
561 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
562 if (sbi->ll_flags & LL_SBI_ACL)
563 valid |= OBD_MD_FLACL;
565 OBD_ALLOC_PTR(op_data);
567 GOTO(out_lock_cn_cb, err = -ENOMEM);
569 op_data->op_fid1 = sbi->ll_root_fid;
570 op_data->op_mode = 0;
571 op_data->op_valid = valid;
573 err = md_getattr(sbi->ll_md_exp, op_data, &request);
575 OBD_FREE_PTR(op_data);
577 CERROR("%s: md_getattr failed for root: rc = %d\n",
578 sbi->ll_md_exp->exp_obd->obd_name, err);
579 GOTO(out_lock_cn_cb, err);
582 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
583 sbi->ll_md_exp, &lmd);
585 CERROR("failed to understand root inode md: rc = %d\n", err);
586 ptlrpc_req_finished(request);
587 GOTO(out_lock_cn_cb, err);
590 LASSERT(fid_is_sane(&sbi->ll_root_fid));
591 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
592 sbi->ll_flags & LL_SBI_32BIT_API),
594 md_free_lustre_md(sbi->ll_md_exp, &lmd);
595 ptlrpc_req_finished(request);
598 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
600 posix_acl_release(lmd.posix_acl);
601 lmd.posix_acl = NULL;
604 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
606 CERROR("%s: bad ll_iget() for root: rc = %d\n",
607 sbi->ll_fsname, err);
611 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
612 if (sbi->ll_checksum_set) {
613 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
614 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
615 sizeof(checksum), &checksum, NULL);
617 CERROR("%s: Set checksum failed: rc = %d\n",
618 sbi->ll_dt_exp->exp_obd->obd_name, err);
624 sb->s_root = d_make_root(root);
625 if (sb->s_root == NULL) {
627 CERROR("%s: can't make root dentry: rc = %d\n",
628 sbi->ll_fsname, err);
632 sbi->ll_sdev_orig = sb->s_dev;
634 /* We set sb->s_dev equal on all lustre clients in order to support
635 * NFS export clustering. NFSD requires that the FSID be the same
637 /* s_dev is also used in lt_compare() to compare two fs, but that is
638 * only a node-local comparison. */
639 uuid = obd_get_uuid(sbi->ll_md_exp);
641 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
648 if (sbi->ll_dt_obd) {
649 err = sysfs_create_link(&sbi->ll_kset.kobj,
650 &sbi->ll_dt_obd->obd_kset.kobj,
651 sbi->ll_dt_obd->obd_type->typ_name);
653 CERROR("%s: could not register %s in llite: rc = %d\n",
654 dt, sbi->ll_fsname, err);
659 if (sbi->ll_md_obd) {
660 err = sysfs_create_link(&sbi->ll_kset.kobj,
661 &sbi->ll_md_obd->obd_kset.kobj,
662 sbi->ll_md_obd->obd_type->typ_name);
664 CERROR("%s: could not register %s in llite: rc = %d\n",
665 md, sbi->ll_fsname, err);
675 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
677 obd_disconnect(sbi->ll_dt_exp);
678 sbi->ll_dt_exp = NULL;
679 sbi->ll_dt_obd = NULL;
681 obd_fid_fini(sbi->ll_md_exp->exp_obd);
683 obd_disconnect(sbi->ll_md_exp);
684 sbi->ll_md_exp = NULL;
685 sbi->ll_md_obd = NULL;
694 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
698 size = sizeof(*lmmsize);
699 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
700 KEY_MAX_EASIZE, &size, lmmsize);
702 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
703 sbi->ll_dt_exp->exp_obd->obd_name, rc);
707 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
710 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
711 KEY_MAX_EASIZE, &size, lmmsize);
713 CERROR("Get max mdsize error rc %d\n", rc);
715 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
721 * Get the value of the default_easize parameter.
723 * \see client_obd::cl_default_mds_easize
725 * \param[in] sbi superblock info for this filesystem
726 * \param[out] lmmsize pointer to storage location for value
728 * \retval 0 on success
729 * \retval negative negated errno on failure
731 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
736 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
737 KEY_DEFAULT_EASIZE, &size, lmmsize);
739 CERROR("Get default mdsize error rc %d\n", rc);
745 * Set the default_easize parameter to the given value.
747 * \see client_obd::cl_default_mds_easize
749 * \param[in] sbi superblock info for this filesystem
750 * \param[in] lmmsize the size to set
752 * \retval 0 on success
753 * \retval negative negated errno on failure
755 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
759 if (lmmsize < sizeof(struct lov_mds_md) ||
760 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
763 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
764 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
765 sizeof(int), &lmmsize, NULL);
770 static void client_common_put_super(struct super_block *sb)
772 struct ll_sb_info *sbi = ll_s2sbi(sb);
777 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
778 obd_disconnect(sbi->ll_dt_exp);
779 sbi->ll_dt_exp = NULL;
781 ll_debugfs_unregister_super(sb);
783 obd_fid_fini(sbi->ll_md_exp->exp_obd);
784 obd_disconnect(sbi->ll_md_exp);
785 sbi->ll_md_exp = NULL;
790 void ll_kill_super(struct super_block *sb)
792 struct ll_sb_info *sbi;
796 if (!(sb->s_flags & SB_ACTIVE))
800 /* we need restore s_dev from changed for clustred NFS before put_super
801 * because new kernels have cached s_dev and change sb->s_dev in
802 * put_super not affected real removing devices */
804 sb->s_dev = sbi->ll_sdev_orig;
806 /* wait running statahead threads to quit */
807 while (atomic_read(&sbi->ll_sa_running) > 0)
808 schedule_timeout_uninterruptible(
809 cfs_time_seconds(1) >> 3);
815 static inline int ll_set_opt(const char *opt, char *data, int fl)
817 if (strncmp(opt, data, strlen(opt)) != 0)
823 /* non-client-specific mount options are parsed in lmd_parse */
824 static int ll_options(char *options, struct ll_sb_info *sbi)
827 char *s1 = options, *s2;
828 int *flags = &sbi->ll_flags;
834 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
837 CDEBUG(D_SUPER, "next opt=%s\n", s1);
838 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
843 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
845 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
848 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
850 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
853 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
858 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
863 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
868 tmp = ll_set_opt("context", s1, 1);
871 tmp = ll_set_opt("fscontext", s1, 1);
874 tmp = ll_set_opt("defcontext", s1, 1);
877 tmp = ll_set_opt("rootcontext", s1, 1);
880 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
885 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
891 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
894 sbi->ll_checksum_set = 1;
897 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
900 sbi->ll_checksum_set = 1;
903 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
908 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
913 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
918 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
923 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
928 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
933 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
938 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
943 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
949 s2 = strchr(s1, ',');
957 void ll_lli_init(struct ll_inode_info *lli)
959 lli->lli_inode_magic = LLI_INODE_MAGIC;
961 spin_lock_init(&lli->lli_lock);
962 lli->lli_posix_acl = NULL;
963 /* Do not set lli_fid, it has been initialized already. */
964 fid_zero(&lli->lli_pfid);
965 lli->lli_mds_read_och = NULL;
966 lli->lli_mds_write_och = NULL;
967 lli->lli_mds_exec_och = NULL;
968 lli->lli_open_fd_read_count = 0;
969 lli->lli_open_fd_write_count = 0;
970 lli->lli_open_fd_exec_count = 0;
971 mutex_init(&lli->lli_och_mutex);
972 spin_lock_init(&lli->lli_agl_lock);
973 spin_lock_init(&lli->lli_layout_lock);
974 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
975 lli->lli_clob = NULL;
977 init_rwsem(&lli->lli_xattrs_list_rwsem);
978 mutex_init(&lli->lli_xattrs_enq_lock);
980 LASSERT(lli->lli_vfs_inode.i_mode != 0);
981 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
982 lli->lli_opendir_key = NULL;
984 spin_lock_init(&lli->lli_sa_lock);
985 lli->lli_opendir_pid = 0;
986 lli->lli_sa_enabled = 0;
987 init_rwsem(&lli->lli_lsm_sem);
989 mutex_init(&lli->lli_size_mutex);
990 lli->lli_symlink_name = NULL;
991 ll_trunc_sem_init(&lli->lli_trunc_sem);
992 range_lock_tree_init(&lli->lli_write_tree);
993 init_rwsem(&lli->lli_glimpse_sem);
994 lli->lli_glimpse_time = ktime_set(0, 0);
995 INIT_LIST_HEAD(&lli->lli_agl_list);
996 lli->lli_agl_index = 0;
997 lli->lli_async_rc = 0;
998 spin_lock_init(&lli->lli_heat_lock);
999 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1000 lli->lli_heat_flags = 0;
1001 mutex_init(&lli->lli_pcc_lock);
1002 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1003 lli->lli_pcc_inode = NULL;
1004 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1005 lli->lli_pcc_generation = 0;
1006 mutex_init(&lli->lli_group_mutex);
1007 lli->lli_group_users = 0;
1008 lli->lli_group_gid = 0;
1010 mutex_init(&lli->lli_layout_mutex);
1011 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1014 #define MAX_STRING_SIZE 128
1016 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1018 #define LSI_BDI_INITIALIZED 0x00400000
1020 #ifndef HAVE_BDI_CAP_MAP_COPY
1021 # define BDI_CAP_MAP_COPY 0
1024 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1026 struct lustre_sb_info *lsi = s2lsi(sb);
1027 char buf[MAX_STRING_SIZE];
1031 err = bdi_init(&lsi->lsi_bdi);
1035 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1036 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1037 lsi->lsi_bdi.name = "lustre";
1038 va_start(args, fmt);
1039 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1041 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1044 sb->s_bdi = &lsi->lsi_bdi;
1048 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1050 int ll_fill_super(struct super_block *sb)
1052 struct lustre_profile *lprof = NULL;
1053 struct lustre_sb_info *lsi = s2lsi(sb);
1054 struct ll_sb_info *sbi = NULL;
1055 char *dt = NULL, *md = NULL;
1056 char *profilenm = get_profile_name(sb);
1057 struct config_llog_instance *cfg;
1058 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1059 const int instlen = LUSTRE_MAXINSTANCE + 2;
1060 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1061 char name[MAX_STRING_SIZE];
1070 /* for ASLR, to map between cfg_instance and hashed ptr */
1071 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1072 profilenm, cfg_instance, sb);
1076 GOTO(out_free_cfg, err = -ENOMEM);
1078 /* client additional sb info */
1079 lsi->lsi_llsbi = sbi = ll_init_sbi();
1081 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1083 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1085 GOTO(out_free_cfg, err);
1087 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1088 sb->s_d_op = &ll_d_ops;
1091 generate_random_uuid(uuid.b);
1092 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1094 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1097 len = strlen(profilenm);
1098 ptr = strrchr(profilenm, '-');
1099 if (ptr && (strcmp(ptr, "-client") == 0))
1102 if (len > LUSTRE_MAXFSNAME) {
1103 if (unlikely(len >= MAX_STRING_SIZE))
1104 len = MAX_STRING_SIZE - 1;
1105 strncpy(name, profilenm, len);
1107 err = -ENAMETOOLONG;
1108 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1109 name, LUSTRE_MAXFSNAME, err);
1110 GOTO(out_free_cfg, err);
1112 strncpy(sbi->ll_fsname, profilenm, len);
1113 sbi->ll_fsname[len] = '\0';
1116 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1117 profilenm, cfg_instance);
1119 err = super_setup_bdi_name(sb, "%s", name);
1121 GOTO(out_free_cfg, err);
1123 /* Call ll_debugfs_register_super() before lustre_process_log()
1124 * so that "llite.*.*" params can be processed correctly.
1126 err = ll_debugfs_register_super(sb, name);
1128 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1129 sbi->ll_fsname, err);
1133 /* The cfg_instance is a value unique to this super, in case some
1134 * joker tries to mount the same fs at two mount points.
1136 cfg->cfg_instance = cfg_instance;
1137 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1138 cfg->cfg_callback = class_config_llog_handler;
1139 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1140 /* set up client obds */
1141 err = lustre_process_log(sb, profilenm, cfg);
1143 GOTO(out_debugfs, err);
1145 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1146 lprof = class_get_profile(profilenm);
1147 if (lprof == NULL) {
1148 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1149 " read from the MGS. Does that filesystem "
1150 "exist?\n", profilenm);
1151 GOTO(out_debugfs, err = -EINVAL);
1153 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1154 lprof->lp_md, lprof->lp_dt);
1156 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1157 OBD_ALLOC(dt, dt_len);
1159 GOTO(out_profile, err = -ENOMEM);
1160 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1162 md_len = strlen(lprof->lp_md) + instlen + 2;
1163 OBD_ALLOC(md, md_len);
1165 GOTO(out_free_dt, err = -ENOMEM);
1166 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1168 /* connections, registrations, sb setup */
1169 err = client_common_fill_super(sb, md, dt);
1171 GOTO(out_free_md, err);
1173 sbi->ll_client_common_fill_super_succeeded = 1;
1177 OBD_FREE(md, md_len);
1180 OBD_FREE(dt, dt_len);
1183 class_put_profile(lprof);
1186 ll_debugfs_unregister_super(sb);
1193 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1194 LCONSOLE_WARN("Mounted %s\n", profilenm);
1196 } /* ll_fill_super */
1198 void ll_put_super(struct super_block *sb)
1200 struct config_llog_instance cfg, params_cfg;
1201 struct obd_device *obd;
1202 struct lustre_sb_info *lsi = s2lsi(sb);
1203 struct ll_sb_info *sbi = ll_s2sbi(sb);
1204 char *profilenm = get_profile_name(sb);
1205 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1207 int next, force = 1, rc = 0;
1211 GOTO(out_no_sbi, 0);
1213 /* Should replace instance_id with something better for ASLR */
1214 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1215 profilenm, cfg_instance, sb);
1217 cfg.cfg_instance = cfg_instance;
1218 lustre_end_log(sb, profilenm, &cfg);
1220 params_cfg.cfg_instance = cfg_instance;
1221 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1223 if (sbi->ll_md_exp) {
1224 obd = class_exp2obd(sbi->ll_md_exp);
1226 force = obd->obd_force;
1229 /* Wait for unstable pages to be committed to stable storage */
1231 rc = l_wait_event_abortable(
1232 sbi->ll_cache->ccc_unstable_waitq,
1233 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1236 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1237 if (force == 0 && rc != -ERESTARTSYS)
1238 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1240 /* We need to set force before the lov_disconnect in
1241 * lustre_common_put_super, since l_d cleans up osc's as well.
1245 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1247 obd->obd_force = force;
1251 if (sbi->ll_client_common_fill_super_succeeded) {
1252 /* Only if client_common_fill_super succeeded */
1253 client_common_put_super(sb);
1257 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1258 class_manual_cleanup(obd);
1260 if (sbi->ll_flags & LL_SBI_VERBOSE)
1261 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1264 class_del_profile(profilenm);
1266 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1267 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1268 bdi_destroy(&lsi->lsi_bdi);
1269 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1274 lsi->lsi_llsbi = NULL;
1276 lustre_common_put_super(sb);
1278 cl_env_cache_purge(~0);
1280 module_put(THIS_MODULE);
1283 } /* client_put_super */
1285 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1287 struct inode *inode = NULL;
1289 /* NOTE: we depend on atomic igrab() -bzzz */
1290 lock_res_and_lock(lock);
1291 if (lock->l_resource->lr_lvb_inode) {
1292 struct ll_inode_info * lli;
1293 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1294 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1295 inode = igrab(lock->l_resource->lr_lvb_inode);
1297 inode = lock->l_resource->lr_lvb_inode;
1298 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1299 D_WARNING, lock, "lr_lvb_inode %p is "
1300 "bogus: magic %08x",
1301 lock->l_resource->lr_lvb_inode,
1302 lli->lli_inode_magic);
1306 unlock_res_and_lock(lock);
1310 void ll_dir_clear_lsm_md(struct inode *inode)
1312 struct ll_inode_info *lli = ll_i2info(inode);
1314 LASSERT(S_ISDIR(inode->i_mode));
1316 if (lli->lli_lsm_md) {
1317 lmv_free_memmd(lli->lli_lsm_md);
1318 lli->lli_lsm_md = NULL;
1321 if (lli->lli_default_lsm_md) {
1322 lmv_free_memmd(lli->lli_default_lsm_md);
1323 lli->lli_default_lsm_md = NULL;
1327 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1328 const struct lu_fid *fid,
1329 struct lustre_md *md)
1331 struct ll_sb_info *sbi = ll_s2sbi(sb);
1332 struct mdt_body *body = md->body;
1333 struct inode *inode;
1337 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1338 inode = iget_locked(sb, ino);
1339 if (inode == NULL) {
1340 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1341 sbi->ll_fsname, PFID(fid));
1342 RETURN(ERR_PTR(-ENOENT));
1345 if (inode->i_state & I_NEW) {
1346 struct ll_inode_info *lli = ll_i2info(inode);
1347 struct lmv_stripe_md *lsm = md->lmv;
1349 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1350 (body->mbo_mode & S_IFMT);
1351 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1354 inode->i_mtime.tv_sec = 0;
1355 inode->i_atime.tv_sec = 0;
1356 inode->i_ctime.tv_sec = 0;
1359 #ifdef HAVE_BACKING_DEV_INFO
1360 /* initializing backing dev info. */
1361 inode->i_mapping->backing_dev_info =
1362 &s2lsi(inode->i_sb)->lsi_bdi;
1364 inode->i_op = &ll_dir_inode_operations;
1365 inode->i_fop = &ll_dir_operations;
1366 lli->lli_fid = *fid;
1369 LASSERT(lsm != NULL);
1370 /* master object FID */
1371 lli->lli_pfid = body->mbo_fid1;
1372 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1373 lli, PFID(fid), PFID(&lli->lli_pfid));
1374 unlock_new_inode(inode);
1380 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1383 struct lmv_stripe_md *lsm = md->lmv;
1384 struct ll_inode_info *lli = ll_i2info(inode);
1387 LASSERT(lsm != NULL);
1389 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1390 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1391 lsm_md_dump(D_INODE, lsm);
1393 if (!lmv_dir_striped(lsm))
1396 /* XXX sigh, this lsm_root initialization should be in
1397 * LMV layer, but it needs ll_iget right now, so we
1398 * put this here right now. */
1399 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1400 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1401 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1403 if (!fid_is_sane(fid))
1406 /* Unfortunately ll_iget will call ll_update_inode,
1407 * where the initialization of slave inode is slightly
1408 * different, so it reset lsm_md to NULL to avoid
1409 * initializing lsm for slave inode. */
1410 lsm->lsm_md_oinfo[i].lmo_root =
1411 ll_iget_anon_dir(inode->i_sb, fid, md);
1412 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1413 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1415 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1417 iput(lsm->lsm_md_oinfo[i].lmo_root);
1418 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1424 lli->lli_lsm_md = lsm;
1429 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1431 struct ll_inode_info *lli = ll_i2info(inode);
1433 if (!md->default_lmv) {
1434 /* clear default lsm */
1435 if (lli->lli_default_lsm_md) {
1436 down_write(&lli->lli_lsm_sem);
1437 if (lli->lli_default_lsm_md) {
1438 lmv_free_memmd(lli->lli_default_lsm_md);
1439 lli->lli_default_lsm_md = NULL;
1441 up_write(&lli->lli_lsm_sem);
1443 } else if (lli->lli_default_lsm_md) {
1444 /* update default lsm if it changes */
1445 down_read(&lli->lli_lsm_sem);
1446 if (lli->lli_default_lsm_md &&
1447 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1448 up_read(&lli->lli_lsm_sem);
1449 down_write(&lli->lli_lsm_sem);
1450 if (lli->lli_default_lsm_md)
1451 lmv_free_memmd(lli->lli_default_lsm_md);
1452 lli->lli_default_lsm_md = md->default_lmv;
1453 lsm_md_dump(D_INODE, md->default_lmv);
1454 md->default_lmv = NULL;
1455 up_write(&lli->lli_lsm_sem);
1457 up_read(&lli->lli_lsm_sem);
1460 /* init default lsm */
1461 down_write(&lli->lli_lsm_sem);
1462 lli->lli_default_lsm_md = md->default_lmv;
1463 lsm_md_dump(D_INODE, md->default_lmv);
1464 md->default_lmv = NULL;
1465 up_write(&lli->lli_lsm_sem);
1469 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1471 struct ll_inode_info *lli = ll_i2info(inode);
1472 struct lmv_stripe_md *lsm = md->lmv;
1473 struct cl_attr *attr;
1478 LASSERT(S_ISDIR(inode->i_mode));
1479 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1480 PFID(ll_inode2fid(inode)));
1482 /* update default LMV */
1483 if (md->default_lmv)
1484 ll_update_default_lsm_md(inode, md);
1487 * no striped information from request, lustre_md from req does not
1488 * include stripeEA, see ll_md_setattr()
1494 * normally dir layout doesn't change, only take read lock to check
1495 * that to avoid blocking other MD operations.
1497 down_read(&lli->lli_lsm_sem);
1499 /* some current lookup initialized lsm, and unchanged */
1500 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1501 GOTO(unlock, rc = 0);
1503 /* if dir layout doesn't match, check whether version is increased,
1504 * which means layout is changed, this happens in dir split/merge and
1507 * foreign LMV should not change.
1509 if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1510 lsm->lsm_md_layout_version <=
1511 lli->lli_lsm_md->lsm_md_layout_version) {
1512 CERROR("%s: "DFID" dir layout mismatch:\n",
1513 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1514 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1515 lsm_md_dump(D_ERROR, lsm);
1516 GOTO(unlock, rc = -EINVAL);
1519 up_read(&lli->lli_lsm_sem);
1520 down_write(&lli->lli_lsm_sem);
1521 /* clear existing lsm */
1522 if (lli->lli_lsm_md) {
1523 lmv_free_memmd(lli->lli_lsm_md);
1524 lli->lli_lsm_md = NULL;
1527 rc = ll_init_lsm_md(inode, md);
1528 up_write(&lli->lli_lsm_sem);
1533 /* set md->lmv to NULL, so the following free lustre_md will not free
1538 /* md_merge_attr() may take long, since lsm is already set, switch to
1541 down_read(&lli->lli_lsm_sem);
1543 if (!lmv_dir_striped(lli->lli_lsm_md))
1544 GOTO(unlock, rc = 0);
1546 OBD_ALLOC_PTR(attr);
1548 GOTO(unlock, rc = -ENOMEM);
1550 /* validate the lsm */
1551 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1552 ll_md_blocking_ast);
1554 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1555 md->body->mbo_nlink = attr->cat_nlink;
1556 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1557 md->body->mbo_size = attr->cat_size;
1558 if (md->body->mbo_valid & OBD_MD_FLATIME)
1559 md->body->mbo_atime = attr->cat_atime;
1560 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1561 md->body->mbo_ctime = attr->cat_ctime;
1562 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1563 md->body->mbo_mtime = attr->cat_mtime;
1569 up_read(&lli->lli_lsm_sem);
1574 void ll_clear_inode(struct inode *inode)
1576 struct ll_inode_info *lli = ll_i2info(inode);
1577 struct ll_sb_info *sbi = ll_i2sbi(inode);
1581 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1582 PFID(ll_inode2fid(inode)), inode);
1584 if (S_ISDIR(inode->i_mode)) {
1585 /* these should have been cleared in ll_file_release */
1586 LASSERT(lli->lli_opendir_key == NULL);
1587 LASSERT(lli->lli_sai == NULL);
1588 LASSERT(lli->lli_opendir_pid == 0);
1590 pcc_inode_free(inode);
1593 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1595 LASSERT(!lli->lli_open_fd_write_count);
1596 LASSERT(!lli->lli_open_fd_read_count);
1597 LASSERT(!lli->lli_open_fd_exec_count);
1599 if (lli->lli_mds_write_och)
1600 ll_md_real_close(inode, FMODE_WRITE);
1601 if (lli->lli_mds_exec_och)
1602 ll_md_real_close(inode, FMODE_EXEC);
1603 if (lli->lli_mds_read_och)
1604 ll_md_real_close(inode, FMODE_READ);
1606 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1607 OBD_FREE(lli->lli_symlink_name,
1608 strlen(lli->lli_symlink_name) + 1);
1609 lli->lli_symlink_name = NULL;
1612 ll_xattr_cache_destroy(inode);
1614 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1615 forget_all_cached_acls(inode);
1616 if (lli->lli_posix_acl) {
1617 posix_acl_release(lli->lli_posix_acl);
1618 lli->lli_posix_acl = NULL;
1621 lli->lli_inode_magic = LLI_INODE_DEAD;
1623 if (S_ISDIR(inode->i_mode))
1624 ll_dir_clear_lsm_md(inode);
1625 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1626 LASSERT(list_empty(&lli->lli_agl_list));
1629 * XXX This has to be done before lsm is freed below, because
1630 * cl_object still uses inode lsm.
1632 cl_inode_fini(inode);
1637 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1639 struct lustre_md md;
1640 struct inode *inode = dentry->d_inode;
1641 struct ll_sb_info *sbi = ll_i2sbi(inode);
1642 struct ptlrpc_request *request = NULL;
1646 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1647 LUSTRE_OPC_ANY, NULL);
1648 if (IS_ERR(op_data))
1649 RETURN(PTR_ERR(op_data));
1651 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1653 ptlrpc_req_finished(request);
1654 if (rc == -ENOENT) {
1656 /* Unlinked special device node? Or just a race?
1657 * Pretend we done everything. */
1658 if (!S_ISREG(inode->i_mode) &&
1659 !S_ISDIR(inode->i_mode)) {
1660 ia_valid = op_data->op_attr.ia_valid;
1661 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1662 rc = simple_setattr(dentry, &op_data->op_attr);
1663 op_data->op_attr.ia_valid = ia_valid;
1665 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1666 CERROR("md_setattr fails: rc = %d\n", rc);
1671 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1672 sbi->ll_md_exp, &md);
1674 ptlrpc_req_finished(request);
1678 ia_valid = op_data->op_attr.ia_valid;
1679 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1680 * cache is not cleared yet. */
1681 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1682 if (S_ISREG(inode->i_mode))
1684 rc = simple_setattr(dentry, &op_data->op_attr);
1685 if (S_ISREG(inode->i_mode))
1686 inode_unlock(inode);
1687 op_data->op_attr.ia_valid = ia_valid;
1689 rc = ll_update_inode(inode, &md);
1690 ptlrpc_req_finished(request);
1695 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1696 * object(s) determine the file size and mtime. Otherwise, the MDS will
1697 * keep these values until such a time that objects are allocated for it.
1698 * We do the MDS operations first, as it is checking permissions for us.
1699 * We don't to the MDS RPC if there is nothing that we want to store there,
1700 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1701 * going to do an RPC anyways.
1703 * If we are doing a truncate, we will send the mtime and ctime updates
1704 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1705 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1708 * In case of HSMimport, we only set attr on MDS.
1710 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1711 enum op_xvalid xvalid, bool hsm_import)
1713 struct inode *inode = dentry->d_inode;
1714 struct ll_inode_info *lli = ll_i2info(inode);
1715 struct md_op_data *op_data = NULL;
1716 ktime_t kstart = ktime_get();
1721 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1722 "valid %x, hsm_import %d\n",
1723 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1724 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1727 if (attr->ia_valid & ATTR_SIZE) {
1728 /* Check new size against VFS/VM file size limit and rlimit */
1729 rc = inode_newsize_ok(inode, attr->ia_size);
1733 /* The maximum Lustre file size is variable, based on the
1734 * OST maximum object size and number of stripes. This
1735 * needs another check in addition to the VFS check above. */
1736 if (attr->ia_size > ll_file_maxbytes(inode)) {
1737 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1738 PFID(&lli->lli_fid), attr->ia_size,
1739 ll_file_maxbytes(inode));
1743 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1746 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1747 if (attr->ia_valid & TIMES_SET_FLAGS) {
1748 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1749 !cfs_capable(CFS_CAP_FOWNER))
1753 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1754 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1755 (attr->ia_valid & ATTR_CTIME)) {
1756 attr->ia_ctime = current_time(inode);
1757 xvalid |= OP_XVALID_CTIME_SET;
1759 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1760 (attr->ia_valid & ATTR_ATIME)) {
1761 attr->ia_atime = current_time(inode);
1762 attr->ia_valid |= ATTR_ATIME_SET;
1764 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1765 (attr->ia_valid & ATTR_MTIME)) {
1766 attr->ia_mtime = current_time(inode);
1767 attr->ia_valid |= ATTR_MTIME_SET;
1770 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1771 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1772 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1773 ktime_get_real_seconds());
1775 if (S_ISREG(inode->i_mode))
1776 inode_unlock(inode);
1778 /* We always do an MDS RPC, even if we're only changing the size;
1779 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1781 OBD_ALLOC_PTR(op_data);
1782 if (op_data == NULL)
1783 GOTO(out, rc = -ENOMEM);
1785 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1786 /* If we are changing file size, file content is
1787 * modified, flag it.
1789 xvalid |= OP_XVALID_OWNEROVERRIDE;
1790 op_data->op_bias |= MDS_DATA_MODIFIED;
1791 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1794 if (attr->ia_valid & ATTR_FILE) {
1795 struct ll_file_data *fd = attr->ia_file->private_data;
1797 if (fd->fd_lease_och)
1798 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1801 op_data->op_attr = *attr;
1802 op_data->op_xvalid = xvalid;
1804 rc = ll_md_setattr(dentry, op_data);
1808 if (!S_ISREG(inode->i_mode) || hsm_import)
1811 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1812 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1813 xvalid & OP_XVALID_CTIME_SET) {
1814 bool cached = false;
1816 rc = pcc_inode_setattr(inode, attr, &cached);
1819 CERROR("%s: PCC inode "DFID" setattr failed: "
1821 ll_i2sbi(inode)->ll_fsname,
1822 PFID(&lli->lli_fid), rc);
1826 /* For truncate and utimes sending attributes to OSTs,
1827 * setting mtime/atime to the past will be performed
1828 * under PW [0:EOF] extent lock (new_size:EOF for
1829 * truncate). It may seem excessive to send mtime/atime
1830 * updates to OSTs when not setting times to past, but
1831 * it is necessary due to possible time
1832 * de-synchronization between MDT inode and OST objects
1834 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1838 /* If the file was restored, it needs to set dirty flag.
1840 * We've already sent MDS_DATA_MODIFIED flag in
1841 * ll_md_setattr() for truncate. However, the MDT refuses to
1842 * set the HS_DIRTY flag on released files, so we have to set
1843 * it again if the file has been restored. Please check how
1844 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1846 * Please notice that if the file is not released, the previous
1847 * MDS_DATA_MODIFIED has taken effect and usually
1848 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1849 * This way we can save an RPC for common open + trunc
1851 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1852 struct hsm_state_set hss = {
1853 .hss_valid = HSS_SETMASK,
1854 .hss_setmask = HS_DIRTY,
1858 rc2 = ll_hsm_state_set(inode, &hss);
1859 /* truncate and write can happen at the same time, so that
1860 * the file can be set modified even though the file is not
1861 * restored from released state, and ll_hsm_state_set() is
1862 * not applicable for the file, and rc2 < 0 is normal in this
1865 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1866 PFID(ll_inode2fid(inode)), rc2);
1871 if (op_data != NULL)
1872 ll_finish_md_op_data(op_data);
1874 if (S_ISREG(inode->i_mode)) {
1876 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1877 inode_dio_wait(inode);
1878 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1879 * flag. ll_update_inode (called from ll_md_setattr), clears
1880 * inode flags, so there is a gap where S_NOSEC is not set.
1881 * This can cause a writer to take the i_mutex unnecessarily,
1882 * but this is safe to do and should be rare. */
1883 inode_has_no_xattr(inode);
1887 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
1888 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
1889 ktime_us_delta(ktime_get(), kstart));
1894 int ll_setattr(struct dentry *de, struct iattr *attr)
1896 int mode = de->d_inode->i_mode;
1897 enum op_xvalid xvalid = 0;
1899 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1900 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1901 xvalid |= OP_XVALID_OWNEROVERRIDE;
1903 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1904 (ATTR_SIZE|ATTR_MODE)) &&
1905 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1906 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1907 !(attr->ia_mode & S_ISGID))))
1908 attr->ia_valid |= ATTR_FORCE;
1910 if ((attr->ia_valid & ATTR_MODE) &&
1912 !(attr->ia_mode & S_ISUID) &&
1913 !(attr->ia_valid & ATTR_KILL_SUID))
1914 attr->ia_valid |= ATTR_KILL_SUID;
1916 if ((attr->ia_valid & ATTR_MODE) &&
1917 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1918 !(attr->ia_mode & S_ISGID) &&
1919 !(attr->ia_valid & ATTR_KILL_SGID))
1920 attr->ia_valid |= ATTR_KILL_SGID;
1922 return ll_setattr_raw(de, attr, xvalid, false);
1925 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1928 struct obd_statfs obd_osfs = { 0 };
1933 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1935 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1936 flags |= OBD_STATFS_NODELAY;
1938 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1942 osfs->os_type = LL_SUPER_MAGIC;
1944 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1945 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1947 if (osfs->os_state & OS_STATE_SUM)
1950 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1951 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1954 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1955 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1958 osfs->os_bsize = obd_osfs.os_bsize;
1959 osfs->os_blocks = obd_osfs.os_blocks;
1960 osfs->os_bfree = obd_osfs.os_bfree;
1961 osfs->os_bavail = obd_osfs.os_bavail;
1963 /* If we have _some_ OSTs, but don't have as many free objects on the
1964 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1965 * to compensate, so that the "inodes in use" number is correct.
1966 * This should be kept in sync with lod_statfs() behaviour.
1968 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1969 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1971 osfs->os_ffree = obd_osfs.os_ffree;
1978 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1980 struct super_block *sb = de->d_sb;
1981 struct obd_statfs osfs;
1982 __u64 fsid = huge_encode_dev(sb->s_dev);
1983 ktime_t kstart = ktime_get();
1986 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
1988 /* Some amount of caching on the client is allowed */
1989 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1993 statfs_unpack(sfs, &osfs);
1995 /* We need to downshift for all 32-bit kernels, because we can't
1996 * tell if the kernel is being called via sys_statfs64() or not.
1997 * Stop before overflowing f_bsize - in which case it is better
1998 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1999 if (sizeof(long) < 8) {
2000 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2003 osfs.os_blocks >>= 1;
2004 osfs.os_bfree >>= 1;
2005 osfs.os_bavail >>= 1;
2009 sfs->f_blocks = osfs.os_blocks;
2010 sfs->f_bfree = osfs.os_bfree;
2011 sfs->f_bavail = osfs.os_bavail;
2012 sfs->f_fsid.val[0] = (__u32)fsid;
2013 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2015 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2016 ktime_us_delta(ktime_get(), kstart));
2021 void ll_inode_size_lock(struct inode *inode)
2023 struct ll_inode_info *lli;
2025 LASSERT(!S_ISDIR(inode->i_mode));
2027 lli = ll_i2info(inode);
2028 mutex_lock(&lli->lli_size_mutex);
2031 void ll_inode_size_unlock(struct inode *inode)
2033 struct ll_inode_info *lli;
2035 lli = ll_i2info(inode);
2036 mutex_unlock(&lli->lli_size_mutex);
2039 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2041 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2042 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2043 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2045 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2048 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2050 struct ll_inode_info *lli = ll_i2info(inode);
2051 struct mdt_body *body = md->body;
2052 struct ll_sb_info *sbi = ll_i2sbi(inode);
2055 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2056 rc = cl_file_inode_init(inode, md);
2061 if (S_ISDIR(inode->i_mode)) {
2062 rc = ll_update_lsm_md(inode, md);
2067 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2068 if (body->mbo_valid & OBD_MD_FLACL) {
2069 spin_lock(&lli->lli_lock);
2070 if (lli->lli_posix_acl)
2071 posix_acl_release(lli->lli_posix_acl);
2072 lli->lli_posix_acl = md->posix_acl;
2073 spin_unlock(&lli->lli_lock);
2076 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2077 sbi->ll_flags & LL_SBI_32BIT_API);
2078 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2080 if (body->mbo_valid & OBD_MD_FLATIME) {
2081 if (body->mbo_atime > inode->i_atime.tv_sec)
2082 inode->i_atime.tv_sec = body->mbo_atime;
2083 lli->lli_atime = body->mbo_atime;
2086 if (body->mbo_valid & OBD_MD_FLMTIME) {
2087 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2089 "setting ino %lu mtime from %lld to %llu\n",
2090 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2092 inode->i_mtime.tv_sec = body->mbo_mtime;
2094 lli->lli_mtime = body->mbo_mtime;
2097 if (body->mbo_valid & OBD_MD_FLCTIME) {
2098 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2099 inode->i_ctime.tv_sec = body->mbo_ctime;
2100 lli->lli_ctime = body->mbo_ctime;
2103 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2104 if (body->mbo_valid & OBD_MD_FLFLAGS)
2105 ll_update_inode_flags(inode, body->mbo_flags);
2106 if (body->mbo_valid & OBD_MD_FLMODE)
2107 inode->i_mode = (inode->i_mode & S_IFMT) |
2108 (body->mbo_mode & ~S_IFMT);
2110 if (body->mbo_valid & OBD_MD_FLTYPE)
2111 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2112 (body->mbo_mode & S_IFMT);
2114 LASSERT(inode->i_mode != 0);
2115 if (S_ISREG(inode->i_mode))
2116 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2117 LL_MAX_BLKSIZE_BITS);
2119 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2121 if (body->mbo_valid & OBD_MD_FLUID)
2122 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2123 if (body->mbo_valid & OBD_MD_FLGID)
2124 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2125 if (body->mbo_valid & OBD_MD_FLPROJID)
2126 lli->lli_projid = body->mbo_projid;
2127 if (body->mbo_valid & OBD_MD_FLNLINK)
2128 set_nlink(inode, body->mbo_nlink);
2129 if (body->mbo_valid & OBD_MD_FLRDEV)
2130 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2132 if (body->mbo_valid & OBD_MD_FLID) {
2133 /* FID shouldn't be changed! */
2134 if (fid_is_sane(&lli->lli_fid)) {
2135 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2136 "Trying to change FID "DFID
2137 " to the "DFID", inode "DFID"(%p)\n",
2138 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2139 PFID(ll_inode2fid(inode)), inode);
2141 lli->lli_fid = body->mbo_fid1;
2145 LASSERT(fid_seq(&lli->lli_fid) != 0);
2147 if (body->mbo_valid & OBD_MD_FLSIZE) {
2148 i_size_write(inode, body->mbo_size);
2150 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2151 PFID(ll_inode2fid(inode)),
2152 (unsigned long long)body->mbo_size);
2154 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2155 inode->i_blocks = body->mbo_blocks;
2158 if (body->mbo_valid & OBD_MD_TSTATE) {
2159 /* Set LLIF_FILE_RESTORING if restore ongoing and
2160 * clear it when done to ensure to start again
2161 * glimpsing updated attrs
2163 if (body->mbo_t_state & MS_RESTORE)
2164 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2166 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2172 int ll_read_inode2(struct inode *inode, void *opaque)
2174 struct lustre_md *md = opaque;
2175 struct ll_inode_info *lli = ll_i2info(inode);
2179 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2180 PFID(&lli->lli_fid), inode);
2182 /* Core attributes from the MDS first. This is a new inode, and
2183 * the VFS doesn't zero times in the core inode so we have to do
2184 * it ourselves. They will be overwritten by either MDS or OST
2185 * attributes - we just need to make sure they aren't newer.
2187 inode->i_mtime.tv_sec = 0;
2188 inode->i_atime.tv_sec = 0;
2189 inode->i_ctime.tv_sec = 0;
2191 rc = ll_update_inode(inode, md);
2195 /* OIDEBUG(inode); */
2197 #ifdef HAVE_BACKING_DEV_INFO
2198 /* initializing backing dev info. */
2199 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2201 if (S_ISREG(inode->i_mode)) {
2202 struct ll_sb_info *sbi = ll_i2sbi(inode);
2203 inode->i_op = &ll_file_inode_operations;
2204 inode->i_fop = sbi->ll_fop;
2205 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2207 } else if (S_ISDIR(inode->i_mode)) {
2208 inode->i_op = &ll_dir_inode_operations;
2209 inode->i_fop = &ll_dir_operations;
2211 } else if (S_ISLNK(inode->i_mode)) {
2212 inode->i_op = &ll_fast_symlink_inode_operations;
2215 inode->i_op = &ll_special_inode_operations;
2217 init_special_inode(inode, inode->i_mode,
2226 void ll_delete_inode(struct inode *inode)
2228 struct ll_inode_info *lli = ll_i2info(inode);
2229 struct address_space *mapping = &inode->i_data;
2230 unsigned long nrpages;
2231 unsigned long flags;
2235 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2236 /* It is last chance to write out dirty pages,
2237 * otherwise we may lose data while umount.
2239 * If i_nlink is 0 then just discard data. This is safe because
2240 * local inode gets i_nlink 0 from server only for the last
2241 * unlink, so that file is not opened somewhere else
2243 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2244 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2246 truncate_inode_pages_final(mapping);
2248 /* Workaround for LU-118: Note nrpages may not be totally updated when
2249 * truncate_inode_pages() returns, as there can be a page in the process
2250 * of deletion (inside __delete_from_page_cache()) in the specified
2251 * range. Thus mapping->nrpages can be non-zero when this function
2252 * returns even after truncation of the whole mapping. Only do this if
2253 * npages isn't already zero.
2255 nrpages = mapping->nrpages;
2257 xa_lock_irqsave(&mapping->i_pages, flags);
2258 nrpages = mapping->nrpages;
2259 xa_unlock_irqrestore(&mapping->i_pages, flags);
2260 } /* Workaround end */
2262 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2263 "see https://jira.whamcloud.com/browse/LU-118\n",
2264 ll_i2sbi(inode)->ll_fsname,
2265 PFID(ll_inode2fid(inode)), inode, nrpages);
2267 ll_clear_inode(inode);
2273 int ll_iocontrol(struct inode *inode, struct file *file,
2274 unsigned int cmd, unsigned long arg)
2276 struct ll_sb_info *sbi = ll_i2sbi(inode);
2277 struct ptlrpc_request *req = NULL;
2282 case FS_IOC_GETFLAGS: {
2283 struct mdt_body *body;
2284 struct md_op_data *op_data;
2286 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2287 0, 0, LUSTRE_OPC_ANY,
2289 if (IS_ERR(op_data))
2290 RETURN(PTR_ERR(op_data));
2292 op_data->op_valid = OBD_MD_FLFLAGS;
2293 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2294 ll_finish_md_op_data(op_data);
2296 CERROR("%s: failure inode "DFID": rc = %d\n",
2297 sbi->ll_md_exp->exp_obd->obd_name,
2298 PFID(ll_inode2fid(inode)), rc);
2302 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2304 flags = body->mbo_flags;
2306 ptlrpc_req_finished(req);
2308 RETURN(put_user(flags, (int __user *)arg));
2310 case FS_IOC_SETFLAGS: {
2312 struct md_op_data *op_data;
2313 struct cl_object *obj;
2314 struct fsxattr fa = { 0 };
2316 if (get_user(flags, (int __user *)arg))
2319 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2320 if (flags & LUSTRE_PROJINHERIT_FL)
2321 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2323 rc = ll_ioctl_check_project(inode, &fa);
2327 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2328 LUSTRE_OPC_ANY, NULL);
2329 if (IS_ERR(op_data))
2330 RETURN(PTR_ERR(op_data));
2332 op_data->op_attr_flags = flags;
2333 op_data->op_xvalid |= OP_XVALID_FLAGS;
2334 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2335 ll_finish_md_op_data(op_data);
2336 ptlrpc_req_finished(req);
2340 ll_update_inode_flags(inode, flags);
2342 obj = ll_i2info(inode)->lli_clob;
2346 OBD_ALLOC_PTR(attr);
2350 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2362 int ll_flush_ctx(struct inode *inode)
2364 struct ll_sb_info *sbi = ll_i2sbi(inode);
2366 CDEBUG(D_SEC, "flush context for user %d\n",
2367 from_kuid(&init_user_ns, current_uid()));
2369 obd_set_info_async(NULL, sbi->ll_md_exp,
2370 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2372 obd_set_info_async(NULL, sbi->ll_dt_exp,
2373 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2378 /* umount -f client means force down, don't save state */
2379 void ll_umount_begin(struct super_block *sb)
2381 struct ll_sb_info *sbi = ll_s2sbi(sb);
2382 struct obd_device *obd;
2383 struct obd_ioctl_data *ioc_data;
2387 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2388 sb->s_count, atomic_read(&sb->s_active));
2390 obd = class_exp2obd(sbi->ll_md_exp);
2392 CERROR("Invalid MDC connection handle %#llx\n",
2393 sbi->ll_md_exp->exp_handle.h_cookie);
2399 obd = class_exp2obd(sbi->ll_dt_exp);
2401 CERROR("Invalid LOV connection handle %#llx\n",
2402 sbi->ll_dt_exp->exp_handle.h_cookie);
2408 OBD_ALLOC_PTR(ioc_data);
2410 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2411 sizeof *ioc_data, ioc_data, NULL);
2413 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2414 sizeof *ioc_data, ioc_data, NULL);
2416 OBD_FREE_PTR(ioc_data);
2419 /* Really, we'd like to wait until there are no requests outstanding,
2420 * and then continue. For now, we just periodically checking for vfs
2421 * to decrement mnt_cnt and hope to finish it within 10sec.
2425 !may_umount(sbi->ll_mnt.mnt)) {
2433 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2435 struct ll_sb_info *sbi = ll_s2sbi(sb);
2436 char *profilenm = get_profile_name(sb);
2440 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2441 read_only = *flags & MS_RDONLY;
2442 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2443 sizeof(KEY_READ_ONLY),
2444 KEY_READ_ONLY, sizeof(read_only),
2447 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2448 profilenm, read_only ?
2449 "read-only" : "read-write", err);
2454 sb->s_flags |= SB_RDONLY;
2456 sb->s_flags &= ~SB_RDONLY;
2458 if (sbi->ll_flags & LL_SBI_VERBOSE)
2459 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2460 read_only ? "read-only" : "read-write");
2466 * Cleanup the open handle that is cached on MDT-side.
2468 * For open case, the client side open handling thread may hit error
2469 * after the MDT grant the open. Under such case, the client should
2470 * send close RPC to the MDT as cleanup; otherwise, the open handle
2471 * on the MDT will be leaked there until the client umount or evicted.
2473 * In further, if someone unlinked the file, because the open handle
2474 * holds the reference on such file/object, then it will block the
2475 * subsequent threads that want to locate such object via FID.
2477 * \param[in] sb super block for this file-system
2478 * \param[in] open_req pointer to the original open request
2480 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2482 struct mdt_body *body;
2483 struct md_op_data *op_data;
2484 struct ptlrpc_request *close_req = NULL;
2485 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2488 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2489 OBD_ALLOC_PTR(op_data);
2490 if (op_data == NULL) {
2491 CWARN("%s: cannot allocate op_data to release open handle for "
2492 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2497 op_data->op_fid1 = body->mbo_fid1;
2498 op_data->op_open_handle = body->mbo_open_handle;
2499 op_data->op_mod_time = ktime_get_real_seconds();
2500 md_close(exp, op_data, NULL, &close_req);
2501 ptlrpc_req_finished(close_req);
2502 ll_finish_md_op_data(op_data);
2507 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2508 struct super_block *sb, struct lookup_intent *it)
2510 struct ll_sb_info *sbi = NULL;
2511 struct lustre_md md = { NULL };
2512 bool default_lmv_deleted = false;
2517 LASSERT(*inode || sb);
2518 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2519 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2520 sbi->ll_md_exp, &md);
2525 * clear default_lmv only if intent_getattr reply doesn't contain it.
2526 * but it needs to be done after iget, check this early because
2527 * ll_update_lsm_md() may change md.
2529 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2530 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2531 default_lmv_deleted = true;
2534 rc = ll_update_inode(*inode, &md);
2538 LASSERT(sb != NULL);
2541 * At this point server returns to client's same fid as client
2542 * generated for creating. So using ->fid1 is okay here.
2544 if (!fid_is_sane(&md.body->mbo_fid1)) {
2545 CERROR("%s: Fid is insane "DFID"\n",
2547 PFID(&md.body->mbo_fid1));
2548 GOTO(out, rc = -EINVAL);
2551 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2552 sbi->ll_flags & LL_SBI_32BIT_API),
2554 if (IS_ERR(*inode)) {
2555 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2557 posix_acl_release(md.posix_acl);
2558 md.posix_acl = NULL;
2561 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2563 CERROR("new_inode -fatal: rc %d\n", rc);
2568 /* Handling piggyback layout lock.
2569 * Layout lock can be piggybacked by getattr and open request.
2570 * The lsm can be applied to inode only if it comes with a layout lock
2571 * otherwise correct layout may be overwritten, for example:
2572 * 1. proc1: mdt returns a lsm but not granting layout
2573 * 2. layout was changed by another client
2574 * 3. proc2: refresh layout and layout lock granted
2575 * 4. proc1: to apply a stale layout */
2576 if (it != NULL && it->it_lock_mode != 0) {
2577 struct lustre_handle lockh;
2578 struct ldlm_lock *lock;
2580 lockh.cookie = it->it_lock_handle;
2581 lock = ldlm_handle2lock(&lockh);
2582 LASSERT(lock != NULL);
2583 if (ldlm_has_layout(lock)) {
2584 struct cl_object_conf conf;
2586 memset(&conf, 0, sizeof(conf));
2587 conf.coc_opc = OBJECT_CONF_SET;
2588 conf.coc_inode = *inode;
2589 conf.coc_lock = lock;
2590 conf.u.coc_layout = md.layout;
2591 (void)ll_layout_conf(*inode, &conf);
2593 LDLM_LOCK_PUT(lock);
2596 if (default_lmv_deleted)
2597 ll_update_default_lsm_md(*inode, &md);
2602 /* cleanup will be done if necessary */
2603 md_free_lustre_md(sbi->ll_md_exp, &md);
2605 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2606 ll_intent_drop_lock(it);
2607 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2613 int ll_obd_statfs(struct inode *inode, void __user *arg)
2615 struct ll_sb_info *sbi = NULL;
2616 struct obd_export *exp;
2618 struct obd_ioctl_data *data = NULL;
2622 if (!inode || !(sbi = ll_i2sbi(inode)))
2623 GOTO(out_statfs, rc = -EINVAL);
2625 rc = obd_ioctl_getdata(&buf, &len, arg);
2627 GOTO(out_statfs, rc);
2630 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2631 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2632 GOTO(out_statfs, rc = -EINVAL);
2634 if (data->ioc_inllen1 != sizeof(__u32) ||
2635 data->ioc_inllen2 != sizeof(__u32) ||
2636 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2637 data->ioc_plen2 != sizeof(struct obd_uuid))
2638 GOTO(out_statfs, rc = -EINVAL);
2640 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2641 if (type & LL_STATFS_LMV)
2642 exp = sbi->ll_md_exp;
2643 else if (type & LL_STATFS_LOV)
2644 exp = sbi->ll_dt_exp;
2646 GOTO(out_statfs, rc = -ENODEV);
2648 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2650 GOTO(out_statfs, rc);
2652 OBD_FREE_LARGE(buf, len);
2657 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2658 * be called early to avoid deadlock.
2660 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2662 if (op_data->op_mea2_sem) {
2663 up_read(op_data->op_mea2_sem);
2664 op_data->op_mea2_sem = NULL;
2667 if (op_data->op_mea1_sem) {
2668 up_read(op_data->op_mea1_sem);
2669 op_data->op_mea1_sem = NULL;
2673 /* this function prepares md_op_data hint for passing it down to MD stack. */
2674 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2675 struct inode *i1, struct inode *i2,
2676 const char *name, size_t namelen,
2677 __u32 mode, enum md_op_code opc,
2680 LASSERT(i1 != NULL);
2683 /* Do not reuse namelen for something else. */
2685 return ERR_PTR(-EINVAL);
2687 if (namelen > ll_i2sbi(i1)->ll_namelen)
2688 return ERR_PTR(-ENAMETOOLONG);
2690 if (!lu_name_is_valid_2(name, namelen))
2691 return ERR_PTR(-EINVAL);
2694 if (op_data == NULL)
2695 OBD_ALLOC_PTR(op_data);
2697 if (op_data == NULL)
2698 return ERR_PTR(-ENOMEM);
2700 ll_i2gids(op_data->op_suppgids, i1, i2);
2701 op_data->op_fid1 = *ll_inode2fid(i1);
2702 op_data->op_code = opc;
2704 if (S_ISDIR(i1->i_mode)) {
2705 down_read(&ll_i2info(i1)->lli_lsm_sem);
2706 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2707 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2708 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2712 op_data->op_fid2 = *ll_inode2fid(i2);
2713 if (S_ISDIR(i2->i_mode)) {
2715 down_read(&ll_i2info(i2)->lli_lsm_sem);
2716 op_data->op_mea2_sem =
2717 &ll_i2info(i2)->lli_lsm_sem;
2719 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2722 fid_zero(&op_data->op_fid2);
2725 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2726 op_data->op_cli_flags |= CLI_HASH64;
2728 if (ll_need_32bit_api(ll_i2sbi(i1)))
2729 op_data->op_cli_flags |= CLI_API32;
2731 op_data->op_name = name;
2732 op_data->op_namelen = namelen;
2733 op_data->op_mode = mode;
2734 op_data->op_mod_time = ktime_get_real_seconds();
2735 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2736 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2737 op_data->op_cap = cfs_curproc_cap_pack();
2738 op_data->op_mds = 0;
2739 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2740 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2741 op_data->op_bias |= MDS_CREATE_VOLATILE;
2743 op_data->op_data = data;
2748 void ll_finish_md_op_data(struct md_op_data *op_data)
2750 ll_unlock_md_op_lsm(op_data);
2751 security_release_secctx(op_data->op_file_secctx,
2752 op_data->op_file_secctx_size);
2753 OBD_FREE_PTR(op_data);
2756 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2758 struct ll_sb_info *sbi;
2760 LASSERT(seq && dentry);
2761 sbi = ll_s2sbi(dentry->d_sb);
2763 if (sbi->ll_flags & LL_SBI_NOLCK)
2764 seq_puts(seq, ",nolock");
2766 /* "flock" is the default since 2.13, but it wasn't for many years,
2767 * so it is still useful to print this to show it is enabled.
2768 * Start to print "noflock" so it is now clear when flock is disabled.
2770 if (sbi->ll_flags & LL_SBI_FLOCK)
2771 seq_puts(seq, ",flock");
2772 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2773 seq_puts(seq, ",localflock");
2775 seq_puts(seq, ",noflock");
2777 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2778 seq_puts(seq, ",user_xattr");
2780 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2781 seq_puts(seq, ",lazystatfs");
2783 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2784 seq_puts(seq, ",user_fid2path");
2786 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2787 seq_puts(seq, ",always_ping");
2793 * Get obd name by cmd, and copy out to user space
2795 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2797 struct ll_sb_info *sbi = ll_i2sbi(inode);
2798 struct obd_device *obd;
2801 if (cmd == OBD_IOC_GETDTNAME)
2802 obd = class_exp2obd(sbi->ll_dt_exp);
2803 else if (cmd == OBD_IOC_GETMDNAME)
2804 obd = class_exp2obd(sbi->ll_md_exp);
2811 if (copy_to_user((void __user *)arg, obd->obd_name,
2812 strlen(obd->obd_name) + 1))
2818 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2825 p.mnt = current->fs->root.mnt;
2827 path = d_path(&p, buf, bufsize);
2832 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2834 char *buf, *path = NULL;
2835 struct dentry *dentry = NULL;
2836 struct inode *inode = page->mapping->host;
2838 /* this can be called inside spin lock so use GFP_ATOMIC. */
2839 buf = (char *)__get_free_page(GFP_ATOMIC);
2841 dentry = d_find_alias(page->mapping->host);
2843 path = ll_d_path(dentry, buf, PAGE_SIZE);
2846 /* The below message is checked in recovery-small.sh test_24b */
2848 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2849 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2850 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2851 PFID(ll_inode2fid(inode)),
2852 (path && !IS_ERR(path)) ? path : "", ioret);
2858 free_page((unsigned long)buf);
2861 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2862 struct lov_user_md **kbuf)
2864 struct lov_user_md lum;
2868 if (copy_from_user(&lum, md, sizeof(lum)))
2871 lum_size = ll_lov_user_md_size(&lum);
2875 OBD_ALLOC_LARGE(*kbuf, lum_size);
2879 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2880 OBD_FREE_LARGE(*kbuf, lum_size);
2888 * Compute llite root squash state after a change of root squash
2889 * configuration setting or add/remove of a lnet nid
2891 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2893 struct root_squash_info *squash = &sbi->ll_squash;
2896 struct lnet_process_id id;
2898 /* Update norootsquash flag */
2899 spin_lock(&squash->rsi_lock);
2900 if (list_empty(&squash->rsi_nosquash_nids))
2901 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2903 /* Do not apply root squash as soon as one of our NIDs is
2904 * in the nosquash_nids list */
2907 while (LNetGetId(i++, &id) != -ENOENT) {
2908 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2910 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2916 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2918 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2920 spin_unlock(&squash->rsi_lock);
2924 * Parse linkea content to extract information about a given hardlink
2926 * \param[in] ldata - Initialized linkea data
2927 * \param[in] linkno - Link identifier
2928 * \param[out] parent_fid - The entry's parent FID
2929 * \param[out] ln - Entry name destination buffer
2931 * \retval 0 on success
2932 * \retval Appropriate negative error code on failure
2934 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2935 struct lu_fid *parent_fid, struct lu_name *ln)
2941 rc = linkea_init_with_rec(ldata);
2945 if (linkno >= ldata->ld_leh->leh_reccount)
2946 /* beyond last link */
2949 linkea_first_entry(ldata);
2950 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2951 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2956 linkea_next_entry(ldata);
2966 * Get parent FID and name of an identified link. Operation is performed for
2967 * a given link number, letting the caller iterate over linkno to list one or
2968 * all links of an entry.
2970 * \param[in] file - File descriptor against which to perform the operation
2971 * \param[in,out] arg - User-filled structure containing the linkno to operate
2972 * on and the available size. It is eventually filled with
2973 * the requested information or left untouched on error
2975 * \retval - 0 on success
2976 * \retval - Appropriate negative error code on failure
2978 int ll_getparent(struct file *file, struct getparent __user *arg)
2980 struct inode *inode = file_inode(file);
2981 struct linkea_data *ldata;
2982 struct lu_buf buf = LU_BUF_NULL;
2984 struct lu_fid parent_fid;
2991 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2992 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2995 if (get_user(name_size, &arg->gp_name_size))
2998 if (get_user(linkno, &arg->gp_linkno))
3001 if (name_size > PATH_MAX)
3004 OBD_ALLOC(ldata, sizeof(*ldata));
3008 rc = linkea_data_new(ldata, &buf);
3010 GOTO(ldata_free, rc);
3012 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3013 buf.lb_len, OBD_MD_FLXATTR);
3017 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3021 if (ln.ln_namelen >= name_size)
3022 GOTO(lb_free, rc = -EOVERFLOW);
3024 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3025 GOTO(lb_free, rc = -EFAULT);
3027 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3028 GOTO(lb_free, rc = -EFAULT);
3030 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3031 GOTO(lb_free, rc = -EFAULT);
3036 OBD_FREE(ldata, sizeof(*ldata));