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/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <linux/security.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 struct kmem_cache *ll_file_data_slab;
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 sbi->ll_cache = cl_cache_init(lru_page_max);
94 if (sbi->ll_cache == NULL) {
95 OBD_FREE(sbi, sizeof(*sbi));
99 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
104 ll_generate_random_uuid(uuid);
105 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
106 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
108 sbi->ll_flags |= LL_SBI_VERBOSE;
109 #ifdef ENABLE_CHECKSUM
110 sbi->ll_flags |= LL_SBI_CHECKSUM;
113 #ifdef HAVE_LRU_RESIZE_SUPPORT
114 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
116 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
118 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
119 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
121 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 /* metadata statahead is enabled by default */
126 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
134 sbi->ll_flags |= LL_SBI_TINY_WRITE;
137 sbi->ll_squash.rsi_uid = 0;
138 sbi->ll_squash.rsi_gid = 0;
139 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
140 init_rwsem(&sbi->ll_squash.rsi_sem);
145 static void ll_free_sbi(struct super_block *sb)
147 struct ll_sb_info *sbi = ll_s2sbi(sb);
151 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
152 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
153 if (sbi->ll_cache != NULL) {
154 cl_cache_decref(sbi->ll_cache);
155 sbi->ll_cache = NULL;
157 OBD_FREE(sbi, sizeof(*sbi));
162 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
164 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
165 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
169 struct vfsmount *mnt)
171 struct inode *root = NULL;
172 struct ll_sb_info *sbi = ll_s2sbi(sb);
173 struct obd_device *obd;
174 struct obd_statfs *osfs = NULL;
175 struct ptlrpc_request *request = NULL;
176 struct obd_connect_data *data = NULL;
177 struct obd_uuid *uuid;
178 struct md_op_data *op_data;
179 struct lustre_md lmd;
181 int size, err, checksum;
184 obd = class_name2obd(md);
186 CERROR("MD %s: not setup or attached\n", md);
200 /* pass client page size via ocd_grant_blkbits, the server should report
201 * back its backend blocksize for grant calculation purpose */
202 data->ocd_grant_blkbits = PAGE_SHIFT;
204 /* indicate MDT features supported by this client */
205 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
206 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
207 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
208 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
209 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
210 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
211 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
212 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
213 OBD_CONNECT_64BITHASH |
214 OBD_CONNECT_EINPROGRESS |
215 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
216 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
217 OBD_CONNECT_MAX_EASIZE |
218 OBD_CONNECT_FLOCK_DEAD |
219 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
220 OBD_CONNECT_OPEN_BY_FID |
221 OBD_CONNECT_DIR_STRIPE |
222 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
223 OBD_CONNECT_SUBTREE |
224 OBD_CONNECT_MULTIMODRPCS |
225 OBD_CONNECT_GRANT_PARAM |
226 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
228 data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
229 OBD_CONNECT2_LOCK_CONVERT |
230 OBD_CONNECT2_DIR_MIGRATE;
232 #ifdef HAVE_LRU_RESIZE_SUPPORT
233 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
234 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
236 #ifdef CONFIG_FS_POSIX_ACL
237 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
238 OBD_CONNECT_LARGE_ACL;
241 data->ocd_cksum_types = obd_cksum_types_supported_client();
243 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
244 /* flag mdc connection as lightweight, only used for test
245 * purpose, use with care */
246 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
248 data->ocd_ibits_known = MDS_INODELOCK_FULL;
249 data->ocd_version = LUSTRE_VERSION_CODE;
251 if (sb->s_flags & MS_RDONLY)
252 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
253 if (sbi->ll_flags & LL_SBI_USER_XATTR)
254 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
257 /* Setting this indicates we correctly support S_NOSEC (See kernel
258 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
260 sb->s_flags |= MS_NOSEC;
263 if (sbi->ll_flags & LL_SBI_FLOCK)
264 sbi->ll_fop = &ll_file_operations_flock;
265 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
266 sbi->ll_fop = &ll_file_operations;
268 sbi->ll_fop = &ll_file_operations_noflock;
270 /* always ping even if server suppress_pings */
271 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
272 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
274 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
275 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
276 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
278 data->ocd_brw_size = MD_MAX_BRW_SIZE;
280 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
282 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
283 "recovery, of which this client is not a "
284 "part. Please wait for recovery to complete,"
285 " abort, or time out.\n", md);
288 CERROR("cannot connect to %s: rc = %d\n", md, err);
292 sbi->ll_md_exp->exp_connect_data = *data;
294 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
295 LUSTRE_SEQ_METADATA);
297 CERROR("%s: Can't init metadata layer FID infrastructure, "
298 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
302 /* For mount, we only need fs info from MDT0, and also in DNE, it
303 * can make sure the client can be mounted as long as MDT0 is
305 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
306 ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
307 OBD_STATFS_FOR_MDT0);
309 GOTO(out_md_fid, err);
311 /* This needs to be after statfs to ensure connect has finished.
312 * Note that "data" does NOT contain the valid connect reply.
313 * If connecting to a 1.8 server there will be no LMV device, so
314 * we can access the MDC export directly and exp_connect_flags will
315 * be non-zero, but if accessing an upgraded 2.1 server it will
316 * have the correct flags filled in.
317 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
318 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
319 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
320 valid != CLIENT_CONNECT_MDT_REQD) {
323 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
324 obd_connect_flags2str(buf, PAGE_SIZE,
325 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
326 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
327 "feature(s) needed for correct operation "
328 "of this client (%s). Please upgrade "
329 "server or downgrade client.\n",
330 sbi->ll_md_exp->exp_obd->obd_name, buf);
331 OBD_FREE(buf, PAGE_SIZE);
332 GOTO(out_md_fid, err = -EPROTO);
335 size = sizeof(*data);
336 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
337 KEY_CONN_DATA, &size, data);
339 CERROR("%s: Get connect data failed: rc = %d\n",
340 sbi->ll_md_exp->exp_obd->obd_name, err);
341 GOTO(out_md_fid, err);
344 LASSERT(osfs->os_bsize);
345 sb->s_blocksize = osfs->os_bsize;
346 sb->s_blocksize_bits = log2(osfs->os_bsize);
347 sb->s_magic = LL_SUPER_MAGIC;
348 sb->s_maxbytes = MAX_LFS_FILESIZE;
349 sbi->ll_namelen = osfs->os_namelen;
350 sbi->ll_mnt.mnt = current->fs->root.mnt;
352 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
353 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
354 LCONSOLE_INFO("Disabling user_xattr feature because "
355 "it is not supported on the server\n");
356 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
359 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
361 sb->s_flags |= MS_POSIXACL;
363 sbi->ll_flags |= LL_SBI_ACL;
365 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
367 sb->s_flags &= ~MS_POSIXACL;
369 sbi->ll_flags &= ~LL_SBI_ACL;
372 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
373 sbi->ll_flags |= LL_SBI_64BIT_HASH;
375 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
376 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
378 if (obd_connect_has_secctx(data))
379 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
381 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
382 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
383 LCONSOLE_INFO("%s: disabling xattr cache due to "
384 "unknown maximum xattr size.\n", dt);
385 } else if (!sbi->ll_xattr_cache_set) {
386 /* If xattr_cache is already set (no matter 0 or 1)
387 * during processing llog, it won't be enabled here. */
388 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
389 sbi->ll_xattr_cache_enabled = 1;
393 obd = class_name2obd(dt);
395 CERROR("DT %s: not setup or attached\n", dt);
396 GOTO(out_md_fid, err = -ENODEV);
399 /* pass client page size via ocd_grant_blkbits, the server should report
400 * back its backend blocksize for grant calculation purpose */
401 data->ocd_grant_blkbits = PAGE_SHIFT;
403 /* indicate OST features supported by this client */
404 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
405 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
406 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
407 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
408 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
409 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
410 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
411 OBD_CONNECT_EINPROGRESS |
412 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
413 OBD_CONNECT_LAYOUTLOCK |
414 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
415 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
416 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
418 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
419 * can interoperate with an older version of lockahead which was released prior
420 * to landing in master. This support will be dropped when 2.13 development
421 * starts. At the point, we should not just drop the connect flag (below), we
422 * should also remove the support in the code.
424 * Removing it means a few things:
425 * 1. Remove this section here
426 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
427 * 3. Remove function exp_connect_lockahead_old
428 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
430 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
431 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
434 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
436 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
437 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
439 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
440 * disabled by default, because it can still be enabled on the
441 * fly via /proc. As a consequence, we still need to come to an
442 * agreement on the supported algorithms at connect time */
443 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
445 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
446 data->ocd_cksum_types = OBD_CKSUM_ADLER;
448 data->ocd_cksum_types = obd_cksum_types_supported_client();
450 #ifdef HAVE_LRU_RESIZE_SUPPORT
451 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
453 /* always ping even if server suppress_pings */
454 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
455 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
457 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
458 "ocd_grant: %d\n", data->ocd_connect_flags,
459 data->ocd_version, data->ocd_grant);
461 obd->obd_upcall.onu_owner = &sbi->ll_lco;
462 obd->obd_upcall.onu_upcall = cl_ocd_update;
464 data->ocd_brw_size = DT_MAX_BRW_SIZE;
466 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
469 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
470 "recovery, of which this client is not a "
471 "part. Please wait for recovery to "
472 "complete, abort, or time out.\n", dt);
475 CERROR("%s: Cannot connect to %s: rc = %d\n",
476 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
480 sbi->ll_dt_exp->exp_connect_data = *data;
482 /* Don't change value if it was specified in the config log */
483 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
484 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
485 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
486 (data->ocd_brw_size >> PAGE_SHIFT));
488 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
489 LUSTRE_SEQ_METADATA);
491 CERROR("%s: Can't init data layer FID infrastructure, "
492 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
496 mutex_lock(&sbi->ll_lco.lco_lock);
497 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
498 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
499 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
500 mutex_unlock(&sbi->ll_lco.lco_lock);
502 fid_zero(&sbi->ll_root_fid);
503 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
506 CERROR("cannot mds_connect: rc = %d\n", err);
507 GOTO(out_lock_cn_cb, err);
509 if (!fid_is_sane(&sbi->ll_root_fid)) {
510 CERROR("%s: Invalid root fid "DFID" during mount\n",
511 sbi->ll_md_exp->exp_obd->obd_name,
512 PFID(&sbi->ll_root_fid));
513 GOTO(out_lock_cn_cb, err = -EINVAL);
515 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
517 sb->s_op = &lustre_super_operations;
518 #ifdef HAVE_XATTR_HANDLER_FLAGS
519 sb->s_xattr = ll_xattr_handlers;
521 #if THREAD_SIZE >= 8192 /*b=17630*/
522 sb->s_export_op = &lustre_export_operations;
526 * XXX: move this to after cbd setup? */
527 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
528 if (sbi->ll_flags & LL_SBI_ACL)
529 valid |= OBD_MD_FLACL;
531 OBD_ALLOC_PTR(op_data);
533 GOTO(out_lock_cn_cb, err = -ENOMEM);
535 op_data->op_fid1 = sbi->ll_root_fid;
536 op_data->op_mode = 0;
537 op_data->op_valid = valid;
539 err = md_getattr(sbi->ll_md_exp, op_data, &request);
541 OBD_FREE_PTR(op_data);
543 CERROR("%s: md_getattr failed for root: rc = %d\n",
544 sbi->ll_md_exp->exp_obd->obd_name, err);
545 GOTO(out_lock_cn_cb, err);
548 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
549 sbi->ll_md_exp, &lmd);
551 CERROR("failed to understand root inode md: rc = %d\n", err);
552 ptlrpc_req_finished(request);
553 GOTO(out_lock_cn_cb, err);
556 LASSERT(fid_is_sane(&sbi->ll_root_fid));
557 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
558 sbi->ll_flags & LL_SBI_32BIT_API),
560 md_free_lustre_md(sbi->ll_md_exp, &lmd);
561 ptlrpc_req_finished(request);
564 #ifdef CONFIG_FS_POSIX_ACL
566 posix_acl_release(lmd.posix_acl);
567 lmd.posix_acl = NULL;
570 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
572 CERROR("lustre_lite: bad iget4 for root\n");
576 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
577 if (sbi->ll_checksum_set) {
578 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
579 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
580 sizeof(checksum), &checksum, NULL);
582 CERROR("%s: Set checksum failed: rc = %d\n",
583 sbi->ll_dt_exp->exp_obd->obd_name, err);
589 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
590 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
591 sbi->ll_cache, NULL);
593 CERROR("%s: Set cache_set failed: rc = %d\n",
594 sbi->ll_dt_exp->exp_obd->obd_name, err);
598 sb->s_root = d_make_root(root);
599 if (sb->s_root == NULL) {
600 CERROR("%s: can't make root dentry\n",
601 ll_get_fsname(sb, NULL, 0));
602 GOTO(out_root, err = -ENOMEM);
604 #ifdef HAVE_DCACHE_LOCK
605 sb->s_root->d_op = &ll_d_ops;
608 sbi->ll_sdev_orig = sb->s_dev;
610 /* We set sb->s_dev equal on all lustre clients in order to support
611 * NFS export clustering. NFSD requires that the FSID be the same
613 /* s_dev is also used in lt_compare() to compare two fs, but that is
614 * only a node-local comparison. */
615 uuid = obd_get_uuid(sbi->ll_md_exp);
617 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
624 if (sbi->ll_proc_root != NULL) {
625 err = lprocfs_ll_register_obd(sb, dt);
627 CERROR("%s: could not register %s in llite: rc = %d\n",
628 dt, ll_get_fsname(sb, NULL, 0), err);
631 err = lprocfs_ll_register_obd(sb, md);
633 CERROR("%s: could not register %s in llite: rc = %d\n",
634 md, ll_get_fsname(sb, NULL, 0), err);
644 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
646 obd_disconnect(sbi->ll_dt_exp);
647 sbi->ll_dt_exp = NULL;
649 obd_fid_fini(sbi->ll_md_exp->exp_obd);
651 obd_disconnect(sbi->ll_md_exp);
652 sbi->ll_md_exp = NULL;
661 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
665 size = sizeof(*lmmsize);
666 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
667 KEY_MAX_EASIZE, &size, lmmsize);
669 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
670 sbi->ll_dt_exp->exp_obd->obd_name, rc);
675 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
676 KEY_MAX_EASIZE, &size, lmmsize);
678 CERROR("Get max mdsize error rc %d\n", rc);
684 * Get the value of the default_easize parameter.
686 * \see client_obd::cl_default_mds_easize
688 * \param[in] sbi superblock info for this filesystem
689 * \param[out] lmmsize pointer to storage location for value
691 * \retval 0 on success
692 * \retval negative negated errno on failure
694 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
699 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
700 KEY_DEFAULT_EASIZE, &size, lmmsize);
702 CERROR("Get default mdsize error rc %d\n", rc);
708 * Set the default_easize parameter to the given value.
710 * \see client_obd::cl_default_mds_easize
712 * \param[in] sbi superblock info for this filesystem
713 * \param[in] lmmsize the size to set
715 * \retval 0 on success
716 * \retval negative negated errno on failure
718 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
722 if (lmmsize < sizeof(struct lov_mds_md) ||
723 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
726 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
727 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
728 sizeof(int), &lmmsize, NULL);
733 static void client_common_put_super(struct super_block *sb)
735 struct ll_sb_info *sbi = ll_s2sbi(sb);
740 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
741 obd_disconnect(sbi->ll_dt_exp);
742 sbi->ll_dt_exp = NULL;
744 ll_debugfs_unregister_super(sb);
746 obd_fid_fini(sbi->ll_md_exp->exp_obd);
747 obd_disconnect(sbi->ll_md_exp);
748 sbi->ll_md_exp = NULL;
753 void ll_kill_super(struct super_block *sb)
755 struct ll_sb_info *sbi;
759 if (!(sb->s_flags & MS_ACTIVE))
763 /* we need restore s_dev from changed for clustred NFS before put_super
764 * because new kernels have cached s_dev and change sb->s_dev in
765 * put_super not affected real removing devices */
767 sb->s_dev = sbi->ll_sdev_orig;
768 sbi->ll_umounting = 1;
770 /* wait running statahead threads to quit */
771 while (atomic_read(&sbi->ll_sa_running) > 0) {
772 set_current_state(TASK_UNINTERRUPTIBLE);
773 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
780 static inline int ll_set_opt(const char *opt, char *data, int fl)
782 if (strncmp(opt, data, strlen(opt)) != 0)
788 /* non-client-specific mount options are parsed in lmd_parse */
789 static int ll_options(char *options, struct ll_sb_info *sbi)
792 char *s1 = options, *s2;
793 int *flags = &sbi->ll_flags;
799 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
802 CDEBUG(D_SUPER, "next opt=%s\n", s1);
803 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
808 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
813 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
818 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
823 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
828 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
833 tmp = ll_set_opt("context", s1, 1);
836 tmp = ll_set_opt("fscontext", s1, 1);
839 tmp = ll_set_opt("defcontext", s1, 1);
842 tmp = ll_set_opt("rootcontext", s1, 1);
845 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
850 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
856 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
859 sbi->ll_checksum_set = 1;
862 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
865 sbi->ll_checksum_set = 1;
868 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
873 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
878 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
883 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
888 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
893 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
898 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
903 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
908 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
914 s2 = strchr(s1, ',');
922 void ll_lli_init(struct ll_inode_info *lli)
924 lli->lli_inode_magic = LLI_INODE_MAGIC;
926 spin_lock_init(&lli->lli_lock);
927 lli->lli_posix_acl = NULL;
928 /* Do not set lli_fid, it has been initialized already. */
929 fid_zero(&lli->lli_pfid);
930 lli->lli_mds_read_och = NULL;
931 lli->lli_mds_write_och = NULL;
932 lli->lli_mds_exec_och = NULL;
933 lli->lli_open_fd_read_count = 0;
934 lli->lli_open_fd_write_count = 0;
935 lli->lli_open_fd_exec_count = 0;
936 mutex_init(&lli->lli_och_mutex);
937 spin_lock_init(&lli->lli_agl_lock);
938 spin_lock_init(&lli->lli_layout_lock);
939 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
940 lli->lli_clob = NULL;
942 init_rwsem(&lli->lli_xattrs_list_rwsem);
943 mutex_init(&lli->lli_xattrs_enq_lock);
945 LASSERT(lli->lli_vfs_inode.i_mode != 0);
946 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
947 mutex_init(&lli->lli_readdir_mutex);
948 lli->lli_opendir_key = NULL;
950 spin_lock_init(&lli->lli_sa_lock);
951 lli->lli_opendir_pid = 0;
952 lli->lli_sa_enabled = 0;
953 lli->lli_def_stripe_offset = -1;
955 mutex_init(&lli->lli_size_mutex);
956 lli->lli_symlink_name = NULL;
957 init_rwsem(&lli->lli_trunc_sem);
958 range_lock_tree_init(&lli->lli_write_tree);
959 init_rwsem(&lli->lli_glimpse_sem);
960 lli->lli_glimpse_time = ktime_set(0, 0);
961 INIT_LIST_HEAD(&lli->lli_agl_list);
962 lli->lli_agl_index = 0;
963 lli->lli_async_rc = 0;
965 mutex_init(&lli->lli_layout_mutex);
966 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
969 #define MAX_STRING_SIZE 128
971 #ifndef HAVE_SUPER_SETUP_BDI_NAME
973 #define LSI_BDI_INITIALIZED 0x00400000
975 #ifndef HAVE_BDI_CAP_MAP_COPY
976 # define BDI_CAP_MAP_COPY 0
979 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
981 struct lustre_sb_info *lsi = s2lsi(sb);
982 char buf[MAX_STRING_SIZE];
986 err = bdi_init(&lsi->lsi_bdi);
990 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
991 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
992 lsi->lsi_bdi.name = "lustre";
994 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
996 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
999 sb->s_bdi = &lsi->lsi_bdi;
1003 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1005 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1007 struct lustre_profile *lprof = NULL;
1008 struct lustre_sb_info *lsi = s2lsi(sb);
1009 struct ll_sb_info *sbi = NULL;
1010 char *dt = NULL, *md = NULL;
1011 char *profilenm = get_profile_name(sb);
1012 struct config_llog_instance *cfg;
1013 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1014 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1015 char name[MAX_STRING_SIZE];
1023 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1025 try_module_get(THIS_MODULE);
1029 GOTO(out_free, err = -ENOMEM);
1031 /* client additional sb info */
1032 lsi->lsi_llsbi = sbi = ll_init_sbi();
1034 GOTO(out_free, err = -ENOMEM);
1036 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1038 GOTO(out_free, err);
1040 err = super_setup_bdi_name(sb, "lustre-%p", sb);
1042 GOTO(out_free, err);
1044 #ifndef HAVE_DCACHE_LOCK
1045 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1046 sb->s_d_op = &ll_d_ops;
1049 len = strlen(lsi->lsi_lmd->lmd_profile);
1050 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1051 if (ptr && (strcmp(ptr, "-client") == 0))
1055 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1056 lsi->lsi_lmd->lmd_profile, sb);
1058 /* Call ll_debugfs_register_super() before lustre_process_log()
1059 * so that "llite.*.*" params can be processed correctly.
1061 err = ll_debugfs_register_super(sb, name);
1063 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1064 ll_get_fsname(sb, NULL, 0), err);
1068 /* Generate a string unique to this super, in case some joker tries
1069 to mount the same fs at two mount points.
1070 Use the address of the super itself.*/
1071 cfg->cfg_instance = sb;
1072 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1073 cfg->cfg_callback = class_config_llog_handler;
1074 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1075 /* set up client obds */
1076 err = lustre_process_log(sb, profilenm, cfg);
1078 GOTO(out_proc, err);
1080 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1081 lprof = class_get_profile(profilenm);
1082 if (lprof == NULL) {
1083 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1084 " read from the MGS. Does that filesystem "
1085 "exist?\n", profilenm);
1086 GOTO(out_proc, err = -EINVAL);
1088 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1089 lprof->lp_md, lprof->lp_dt);
1091 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1092 OBD_ALLOC(dt, dt_len);
1094 GOTO(out_proc, err = -ENOMEM);
1095 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1097 md_len = strlen(lprof->lp_md) + instlen + 2;
1098 OBD_ALLOC(md, md_len);
1100 GOTO(out_proc, err = -ENOMEM);
1101 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1103 /* connections, registrations, sb setup */
1104 err = client_common_fill_super(sb, md, dt, mnt);
1106 GOTO(out_proc, err);
1108 sbi->ll_client_common_fill_super_succeeded = 1;
1112 ll_debugfs_unregister_super(sb);
1115 OBD_FREE(md, md_len);
1117 OBD_FREE(dt, dt_len);
1119 class_put_profile(lprof);
1124 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1125 LCONSOLE_WARN("Mounted %s\n", profilenm);
1127 } /* ll_fill_super */
1129 void ll_put_super(struct super_block *sb)
1131 struct config_llog_instance cfg, params_cfg;
1132 struct obd_device *obd;
1133 struct lustre_sb_info *lsi = s2lsi(sb);
1134 struct ll_sb_info *sbi = ll_s2sbi(sb);
1135 char *profilenm = get_profile_name(sb);
1137 int next, force = 1, rc = 0;
1141 GOTO(out_no_sbi, 0);
1143 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1145 cfg.cfg_instance = sb;
1146 lustre_end_log(sb, profilenm, &cfg);
1148 params_cfg.cfg_instance = sb;
1149 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1151 if (sbi->ll_md_exp) {
1152 obd = class_exp2obd(sbi->ll_md_exp);
1154 force = obd->obd_force;
1157 /* Wait for unstable pages to be committed to stable storage */
1159 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1160 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1161 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1165 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1166 if (force == 0 && rc != -EINTR)
1167 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1170 /* We need to set force before the lov_disconnect in
1171 lustre_common_put_super, since l_d cleans up osc's as well. */
1174 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1176 obd->obd_force = force;
1180 if (sbi->ll_client_common_fill_super_succeeded) {
1181 /* Only if client_common_fill_super succeeded */
1182 client_common_put_super(sb);
1186 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1187 class_manual_cleanup(obd);
1190 if (sbi->ll_flags & LL_SBI_VERBOSE)
1191 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1194 class_del_profile(profilenm);
1196 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1197 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1198 bdi_destroy(&lsi->lsi_bdi);
1199 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1204 lsi->lsi_llsbi = NULL;
1206 lustre_common_put_super(sb);
1208 cl_env_cache_purge(~0);
1210 module_put(THIS_MODULE);
1213 } /* client_put_super */
1215 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1217 struct inode *inode = NULL;
1219 /* NOTE: we depend on atomic igrab() -bzzz */
1220 lock_res_and_lock(lock);
1221 if (lock->l_resource->lr_lvb_inode) {
1222 struct ll_inode_info * lli;
1223 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1224 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1225 inode = igrab(lock->l_resource->lr_lvb_inode);
1227 inode = lock->l_resource->lr_lvb_inode;
1228 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1229 D_WARNING, lock, "lr_lvb_inode %p is "
1230 "bogus: magic %08x",
1231 lock->l_resource->lr_lvb_inode,
1232 lli->lli_inode_magic);
1236 unlock_res_and_lock(lock);
1240 void ll_dir_clear_lsm_md(struct inode *inode)
1242 struct ll_inode_info *lli = ll_i2info(inode);
1244 LASSERT(S_ISDIR(inode->i_mode));
1246 if (lli->lli_lsm_md != NULL) {
1247 lmv_free_memmd(lli->lli_lsm_md);
1248 lli->lli_lsm_md = NULL;
1252 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1253 const struct lu_fid *fid,
1254 struct lustre_md *md)
1256 struct ll_sb_info *sbi = ll_s2sbi(sb);
1257 struct mdt_body *body = md->body;
1258 struct inode *inode;
1262 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1263 inode = iget_locked(sb, ino);
1264 if (inode == NULL) {
1265 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1266 ll_get_fsname(sb, NULL, 0), PFID(fid));
1267 RETURN(ERR_PTR(-ENOENT));
1270 if (inode->i_state & I_NEW) {
1271 struct ll_inode_info *lli = ll_i2info(inode);
1272 struct lmv_stripe_md *lsm = md->lmv;
1274 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1275 (body->mbo_mode & S_IFMT);
1276 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1279 LTIME_S(inode->i_mtime) = 0;
1280 LTIME_S(inode->i_atime) = 0;
1281 LTIME_S(inode->i_ctime) = 0;
1284 #ifdef HAVE_BACKING_DEV_INFO
1285 /* initializing backing dev info. */
1286 inode->i_mapping->backing_dev_info =
1287 &s2lsi(inode->i_sb)->lsi_bdi;
1289 inode->i_op = &ll_dir_inode_operations;
1290 inode->i_fop = &ll_dir_operations;
1291 lli->lli_fid = *fid;
1294 LASSERT(lsm != NULL);
1295 /* master object FID */
1296 lli->lli_pfid = body->mbo_fid1;
1297 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1298 lli, PFID(fid), PFID(&lli->lli_pfid));
1299 unlock_new_inode(inode);
1305 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1308 struct lmv_stripe_md *lsm = md->lmv;
1311 LASSERT(lsm != NULL);
1312 /* XXX sigh, this lsm_root initialization should be in
1313 * LMV layer, but it needs ll_iget right now, so we
1314 * put this here right now. */
1315 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1316 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1317 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1318 /* Unfortunately ll_iget will call ll_update_inode,
1319 * where the initialization of slave inode is slightly
1320 * different, so it reset lsm_md to NULL to avoid
1321 * initializing lsm for slave inode. */
1322 /* For migrating inode, master stripe and master object will
1323 * be same, so we only need assign this inode */
1324 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1325 lsm->lsm_md_oinfo[i].lmo_root = inode;
1327 lsm->lsm_md_oinfo[i].lmo_root =
1328 ll_iget_anon_dir(inode->i_sb, fid, md);
1330 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1331 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1333 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1341 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1342 const struct lmv_stripe_md *lsm_md2)
1344 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1345 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1346 lsm_md1->lsm_md_master_mdt_index ==
1347 lsm_md2->lsm_md_master_mdt_index &&
1348 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1349 lsm_md1->lsm_md_layout_version ==
1350 lsm_md2->lsm_md_layout_version &&
1351 strcmp(lsm_md1->lsm_md_pool_name,
1352 lsm_md2->lsm_md_pool_name) == 0;
1355 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1357 struct ll_inode_info *lli = ll_i2info(inode);
1358 struct lmv_stripe_md *lsm = md->lmv;
1362 LASSERT(S_ISDIR(inode->i_mode));
1363 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1364 PFID(ll_inode2fid(inode)));
1366 /* no striped information from request. */
1368 if (lli->lli_lsm_md == NULL) {
1370 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1371 LMV_HASH_FLAG_MIGRATION) {
1372 /* migration is done, the temporay MIGRATE layout has
1374 CDEBUG(D_INODE, DFID" finish migration.\n",
1375 PFID(ll_inode2fid(inode)));
1376 lmv_free_memmd(lli->lli_lsm_md);
1377 lli->lli_lsm_md = NULL;
1380 /* The lustre_md from req does not include stripeEA,
1381 * see ll_md_setattr */
1386 /* set the directory layout */
1387 if (lli->lli_lsm_md == NULL) {
1388 struct cl_attr *attr;
1390 rc = ll_init_lsm_md(inode, md);
1394 /* set md->lmv to NULL, so the following free lustre_md
1395 * will not free this lsm */
1397 lli->lli_lsm_md = lsm;
1399 OBD_ALLOC_PTR(attr);
1403 /* validate the lsm */
1404 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1405 ll_md_blocking_ast);
1411 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1412 md->body->mbo_nlink = attr->cat_nlink;
1413 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1414 md->body->mbo_size = attr->cat_size;
1415 if (md->body->mbo_valid & OBD_MD_FLATIME)
1416 md->body->mbo_atime = attr->cat_atime;
1417 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1418 md->body->mbo_ctime = attr->cat_ctime;
1419 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1420 md->body->mbo_mtime = attr->cat_mtime;
1424 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1425 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1429 /* Compare the old and new stripe information */
1430 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1431 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1434 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1435 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1436 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1437 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1438 inode, lsm, old_lsm,
1439 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1440 lsm->lsm_md_stripe_count,
1441 old_lsm->lsm_md_stripe_count,
1442 lsm->lsm_md_master_mdt_index,
1443 old_lsm->lsm_md_master_mdt_index,
1444 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1445 lsm->lsm_md_layout_version,
1446 old_lsm->lsm_md_layout_version,
1447 lsm->lsm_md_pool_name,
1448 old_lsm->lsm_md_pool_name);
1450 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1451 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1452 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1453 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1456 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1457 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1458 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1459 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1468 void ll_clear_inode(struct inode *inode)
1470 struct ll_inode_info *lli = ll_i2info(inode);
1471 struct ll_sb_info *sbi = ll_i2sbi(inode);
1474 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1475 PFID(ll_inode2fid(inode)), inode);
1477 if (S_ISDIR(inode->i_mode)) {
1478 /* these should have been cleared in ll_file_release */
1479 LASSERT(lli->lli_opendir_key == NULL);
1480 LASSERT(lli->lli_sai == NULL);
1481 LASSERT(lli->lli_opendir_pid == 0);
1484 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1486 LASSERT(!lli->lli_open_fd_write_count);
1487 LASSERT(!lli->lli_open_fd_read_count);
1488 LASSERT(!lli->lli_open_fd_exec_count);
1490 if (lli->lli_mds_write_och)
1491 ll_md_real_close(inode, FMODE_WRITE);
1492 if (lli->lli_mds_exec_och)
1493 ll_md_real_close(inode, FMODE_EXEC);
1494 if (lli->lli_mds_read_och)
1495 ll_md_real_close(inode, FMODE_READ);
1497 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1498 OBD_FREE(lli->lli_symlink_name,
1499 strlen(lli->lli_symlink_name) + 1);
1500 lli->lli_symlink_name = NULL;
1503 ll_xattr_cache_destroy(inode);
1505 #ifdef CONFIG_FS_POSIX_ACL
1506 forget_all_cached_acls(inode);
1507 if (lli->lli_posix_acl) {
1508 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1509 posix_acl_release(lli->lli_posix_acl);
1510 lli->lli_posix_acl = NULL;
1513 lli->lli_inode_magic = LLI_INODE_DEAD;
1515 if (S_ISDIR(inode->i_mode))
1516 ll_dir_clear_lsm_md(inode);
1517 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1518 LASSERT(list_empty(&lli->lli_agl_list));
1521 * XXX This has to be done before lsm is freed below, because
1522 * cl_object still uses inode lsm.
1524 cl_inode_fini(inode);
1529 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1531 struct lustre_md md;
1532 struct inode *inode = dentry->d_inode;
1533 struct ll_sb_info *sbi = ll_i2sbi(inode);
1534 struct ptlrpc_request *request = NULL;
1538 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1539 LUSTRE_OPC_ANY, NULL);
1540 if (IS_ERR(op_data))
1541 RETURN(PTR_ERR(op_data));
1543 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1545 ptlrpc_req_finished(request);
1546 if (rc == -ENOENT) {
1548 /* Unlinked special device node? Or just a race?
1549 * Pretend we done everything. */
1550 if (!S_ISREG(inode->i_mode) &&
1551 !S_ISDIR(inode->i_mode)) {
1552 ia_valid = op_data->op_attr.ia_valid;
1553 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1554 rc = simple_setattr(dentry, &op_data->op_attr);
1555 op_data->op_attr.ia_valid = ia_valid;
1557 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1558 CERROR("md_setattr fails: rc = %d\n", rc);
1563 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1564 sbi->ll_md_exp, &md);
1566 ptlrpc_req_finished(request);
1570 ia_valid = op_data->op_attr.ia_valid;
1571 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1572 * cache is not cleared yet. */
1573 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1574 if (S_ISREG(inode->i_mode))
1576 rc = simple_setattr(dentry, &op_data->op_attr);
1577 if (S_ISREG(inode->i_mode))
1578 inode_unlock(inode);
1579 op_data->op_attr.ia_valid = ia_valid;
1581 rc = ll_update_inode(inode, &md);
1582 ptlrpc_req_finished(request);
1587 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1588 * object(s) determine the file size and mtime. Otherwise, the MDS will
1589 * keep these values until such a time that objects are allocated for it.
1590 * We do the MDS operations first, as it is checking permissions for us.
1591 * We don't to the MDS RPC if there is nothing that we want to store there,
1592 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1593 * going to do an RPC anyways.
1595 * If we are doing a truncate, we will send the mtime and ctime updates
1596 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1597 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1600 * In case of HSMimport, we only set attr on MDS.
1602 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1604 struct inode *inode = dentry->d_inode;
1605 struct ll_inode_info *lli = ll_i2info(inode);
1606 struct md_op_data *op_data = NULL;
1610 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1611 "valid %x, hsm_import %d\n",
1612 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1613 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1616 if (attr->ia_valid & ATTR_SIZE) {
1617 /* Check new size against VFS/VM file size limit and rlimit */
1618 rc = inode_newsize_ok(inode, attr->ia_size);
1622 /* The maximum Lustre file size is variable, based on the
1623 * OST maximum object size and number of stripes. This
1624 * needs another check in addition to the VFS check above. */
1625 if (attr->ia_size > ll_file_maxbytes(inode)) {
1626 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1627 PFID(&lli->lli_fid), attr->ia_size,
1628 ll_file_maxbytes(inode));
1632 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1635 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1636 if (attr->ia_valid & TIMES_SET_FLAGS) {
1637 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1638 !cfs_capable(CFS_CAP_FOWNER))
1642 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1643 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1644 (attr->ia_valid & ATTR_CTIME)) {
1645 attr->ia_ctime = current_time(inode);
1646 attr->ia_valid |= ATTR_CTIME_SET;
1648 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1649 (attr->ia_valid & ATTR_ATIME)) {
1650 attr->ia_atime = current_time(inode);
1651 attr->ia_valid |= ATTR_ATIME_SET;
1653 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1654 (attr->ia_valid & ATTR_MTIME)) {
1655 attr->ia_mtime = current_time(inode);
1656 attr->ia_valid |= ATTR_MTIME_SET;
1659 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1660 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1661 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1662 (s64)ktime_get_real_seconds());
1664 if (S_ISREG(inode->i_mode)) {
1665 if (attr->ia_valid & ATTR_SIZE)
1666 inode_dio_write_done(inode);
1667 inode_unlock(inode);
1670 /* We always do an MDS RPC, even if we're only changing the size;
1671 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1673 OBD_ALLOC_PTR(op_data);
1674 if (op_data == NULL)
1675 GOTO(out, rc = -ENOMEM);
1677 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1678 /* If we are changing file size, file content is
1679 * modified, flag it. */
1680 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1681 op_data->op_bias |= MDS_DATA_MODIFIED;
1682 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1685 op_data->op_attr = *attr;
1687 rc = ll_md_setattr(dentry, op_data);
1691 if (!S_ISREG(inode->i_mode) || hsm_import)
1694 if (attr->ia_valid & (ATTR_SIZE |
1695 ATTR_ATIME | ATTR_ATIME_SET |
1696 ATTR_MTIME | ATTR_MTIME_SET |
1697 ATTR_CTIME | ATTR_CTIME_SET)) {
1698 /* For truncate and utimes sending attributes to OSTs, setting
1699 * mtime/atime to the past will be performed under PW [0:EOF]
1700 * extent lock (new_size:EOF for truncate). It may seem
1701 * excessive to send mtime/atime updates to OSTs when not
1702 * setting times to past, but it is necessary due to possible
1703 * time de-synchronization between MDT inode and OST objects */
1704 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1707 /* If the file was restored, it needs to set dirty flag.
1709 * We've already sent MDS_DATA_MODIFIED flag in
1710 * ll_md_setattr() for truncate. However, the MDT refuses to
1711 * set the HS_DIRTY flag on released files, so we have to set
1712 * it again if the file has been restored. Please check how
1713 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1715 * Please notice that if the file is not released, the previous
1716 * MDS_DATA_MODIFIED has taken effect and usually
1717 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1718 * This way we can save an RPC for common open + trunc
1720 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1721 struct hsm_state_set hss = {
1722 .hss_valid = HSS_SETMASK,
1723 .hss_setmask = HS_DIRTY,
1727 rc2 = ll_hsm_state_set(inode, &hss);
1728 /* truncate and write can happen at the same time, so that
1729 * the file can be set modified even though the file is not
1730 * restored from released state, and ll_hsm_state_set() is
1731 * not applicable for the file, and rc2 < 0 is normal in this
1734 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1735 PFID(ll_inode2fid(inode)), rc2);
1740 if (op_data != NULL)
1741 ll_finish_md_op_data(op_data);
1743 if (S_ISREG(inode->i_mode)) {
1745 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1746 inode_dio_wait(inode);
1747 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1748 * flag. ll_update_inode (called from ll_md_setattr), clears
1749 * inode flags, so there is a gap where S_NOSEC is not set.
1750 * This can cause a writer to take the i_mutex unnecessarily,
1751 * but this is safe to do and should be rare. */
1752 inode_has_no_xattr(inode);
1755 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1756 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1761 int ll_setattr(struct dentry *de, struct iattr *attr)
1763 int mode = de->d_inode->i_mode;
1765 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1766 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1767 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1769 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1770 (ATTR_SIZE|ATTR_MODE)) &&
1771 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1772 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1773 !(attr->ia_mode & S_ISGID))))
1774 attr->ia_valid |= ATTR_FORCE;
1776 if ((attr->ia_valid & ATTR_MODE) &&
1778 !(attr->ia_mode & S_ISUID) &&
1779 !(attr->ia_valid & ATTR_KILL_SUID))
1780 attr->ia_valid |= ATTR_KILL_SUID;
1782 if ((attr->ia_valid & ATTR_MODE) &&
1783 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1784 !(attr->ia_mode & S_ISGID) &&
1785 !(attr->ia_valid & ATTR_KILL_SGID))
1786 attr->ia_valid |= ATTR_KILL_SGID;
1788 /* avoid polluted from ATTR_TIMES_SET,
1789 * projid is not expected to be set here */
1790 attr->ia_valid &= ~MDS_ATTR_PROJID;
1792 return ll_setattr_raw(de, attr, false);
1795 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1798 struct obd_statfs obd_osfs;
1803 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1805 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1807 CERROR("md_statfs fails: rc = %d\n", rc);
1811 osfs->os_type = LL_SUPER_MAGIC;
1813 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1814 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1816 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1817 flags |= OBD_STATFS_NODELAY;
1819 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1821 CERROR("obd_statfs fails: rc = %d\n", rc);
1825 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1826 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1829 osfs->os_bsize = obd_osfs.os_bsize;
1830 osfs->os_blocks = obd_osfs.os_blocks;
1831 osfs->os_bfree = obd_osfs.os_bfree;
1832 osfs->os_bavail = obd_osfs.os_bavail;
1834 /* If we don't have as many objects free on the OST as inodes
1835 * on the MDS, we reduce the total number of inodes to
1836 * compensate, so that the "inodes in use" number is correct.
1838 if (obd_osfs.os_ffree < osfs->os_ffree) {
1839 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1841 osfs->os_ffree = obd_osfs.os_ffree;
1846 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1848 struct super_block *sb = de->d_sb;
1849 struct obd_statfs osfs;
1850 __u64 fsid = huge_encode_dev(sb->s_dev);
1853 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1854 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1856 /* Some amount of caching on the client is allowed */
1857 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, 0);
1861 statfs_unpack(sfs, &osfs);
1863 /* We need to downshift for all 32-bit kernels, because we can't
1864 * tell if the kernel is being called via sys_statfs64() or not.
1865 * Stop before overflowing f_bsize - in which case it is better
1866 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1867 if (sizeof(long) < 8) {
1868 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1871 osfs.os_blocks >>= 1;
1872 osfs.os_bfree >>= 1;
1873 osfs.os_bavail >>= 1;
1877 sfs->f_blocks = osfs.os_blocks;
1878 sfs->f_bfree = osfs.os_bfree;
1879 sfs->f_bavail = osfs.os_bavail;
1880 sfs->f_fsid.val[0] = (__u32)fsid;
1881 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1885 void ll_inode_size_lock(struct inode *inode)
1887 struct ll_inode_info *lli;
1889 LASSERT(!S_ISDIR(inode->i_mode));
1891 lli = ll_i2info(inode);
1892 mutex_lock(&lli->lli_size_mutex);
1895 void ll_inode_size_unlock(struct inode *inode)
1897 struct ll_inode_info *lli;
1899 lli = ll_i2info(inode);
1900 mutex_unlock(&lli->lli_size_mutex);
1903 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1905 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1906 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1907 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1909 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1912 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1914 struct ll_inode_info *lli = ll_i2info(inode);
1915 struct mdt_body *body = md->body;
1916 struct ll_sb_info *sbi = ll_i2sbi(inode);
1919 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1920 rc = cl_file_inode_init(inode, md);
1925 if (S_ISDIR(inode->i_mode)) {
1926 rc = ll_update_lsm_md(inode, md);
1931 #ifdef CONFIG_FS_POSIX_ACL
1932 if (body->mbo_valid & OBD_MD_FLACL) {
1933 spin_lock(&lli->lli_lock);
1934 if (lli->lli_posix_acl)
1935 posix_acl_release(lli->lli_posix_acl);
1936 lli->lli_posix_acl = md->posix_acl;
1937 spin_unlock(&lli->lli_lock);
1940 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1941 sbi->ll_flags & LL_SBI_32BIT_API);
1942 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1944 if (body->mbo_valid & OBD_MD_FLATIME) {
1945 if (body->mbo_atime > LTIME_S(inode->i_atime))
1946 LTIME_S(inode->i_atime) = body->mbo_atime;
1947 lli->lli_atime = body->mbo_atime;
1950 if (body->mbo_valid & OBD_MD_FLMTIME) {
1951 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1952 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1953 "to %llu\n", inode->i_ino,
1954 LTIME_S(inode->i_mtime), body->mbo_mtime);
1955 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1957 lli->lli_mtime = body->mbo_mtime;
1960 if (body->mbo_valid & OBD_MD_FLCTIME) {
1961 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1962 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1963 lli->lli_ctime = body->mbo_ctime;
1966 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1967 if (body->mbo_valid & OBD_MD_FLFLAGS)
1968 ll_update_inode_flags(inode, body->mbo_flags);
1969 if (body->mbo_valid & OBD_MD_FLMODE)
1970 inode->i_mode = (inode->i_mode & S_IFMT) |
1971 (body->mbo_mode & ~S_IFMT);
1973 if (body->mbo_valid & OBD_MD_FLTYPE)
1974 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1975 (body->mbo_mode & S_IFMT);
1977 LASSERT(inode->i_mode != 0);
1978 if (S_ISREG(inode->i_mode))
1979 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1980 LL_MAX_BLKSIZE_BITS);
1982 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1984 if (body->mbo_valid & OBD_MD_FLUID)
1985 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1986 if (body->mbo_valid & OBD_MD_FLGID)
1987 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1988 if (body->mbo_valid & OBD_MD_FLPROJID)
1989 lli->lli_projid = body->mbo_projid;
1990 if (body->mbo_valid & OBD_MD_FLNLINK)
1991 set_nlink(inode, body->mbo_nlink);
1992 if (body->mbo_valid & OBD_MD_FLRDEV)
1993 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1995 if (body->mbo_valid & OBD_MD_FLID) {
1996 /* FID shouldn't be changed! */
1997 if (fid_is_sane(&lli->lli_fid)) {
1998 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1999 "Trying to change FID "DFID
2000 " to the "DFID", inode "DFID"(%p)\n",
2001 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2002 PFID(ll_inode2fid(inode)), inode);
2004 lli->lli_fid = body->mbo_fid1;
2008 LASSERT(fid_seq(&lli->lli_fid) != 0);
2010 if (body->mbo_valid & OBD_MD_FLSIZE) {
2011 i_size_write(inode, body->mbo_size);
2013 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2014 PFID(ll_inode2fid(inode)),
2015 (unsigned long long)body->mbo_size);
2017 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2018 inode->i_blocks = body->mbo_blocks;
2021 if (body->mbo_valid & OBD_MD_TSTATE) {
2022 /* Set LLIF_FILE_RESTORING if restore ongoing and
2023 * clear it when done to ensure to start again
2024 * glimpsing updated attrs
2026 if (body->mbo_t_state & MS_RESTORE)
2027 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2029 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2035 int ll_read_inode2(struct inode *inode, void *opaque)
2037 struct lustre_md *md = opaque;
2038 struct ll_inode_info *lli = ll_i2info(inode);
2042 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2043 PFID(&lli->lli_fid), inode);
2045 /* Core attributes from the MDS first. This is a new inode, and
2046 * the VFS doesn't zero times in the core inode so we have to do
2047 * it ourselves. They will be overwritten by either MDS or OST
2048 * attributes - we just need to make sure they aren't newer. */
2049 LTIME_S(inode->i_mtime) = 0;
2050 LTIME_S(inode->i_atime) = 0;
2051 LTIME_S(inode->i_ctime) = 0;
2053 rc = ll_update_inode(inode, md);
2057 /* OIDEBUG(inode); */
2059 #ifdef HAVE_BACKING_DEV_INFO
2060 /* initializing backing dev info. */
2061 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2063 if (S_ISREG(inode->i_mode)) {
2064 struct ll_sb_info *sbi = ll_i2sbi(inode);
2065 inode->i_op = &ll_file_inode_operations;
2066 inode->i_fop = sbi->ll_fop;
2067 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2069 } else if (S_ISDIR(inode->i_mode)) {
2070 inode->i_op = &ll_dir_inode_operations;
2071 inode->i_fop = &ll_dir_operations;
2073 } else if (S_ISLNK(inode->i_mode)) {
2074 inode->i_op = &ll_fast_symlink_inode_operations;
2077 inode->i_op = &ll_special_inode_operations;
2079 init_special_inode(inode, inode->i_mode,
2088 void ll_delete_inode(struct inode *inode)
2090 struct ll_inode_info *lli = ll_i2info(inode);
2093 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2094 /* It is last chance to write out dirty pages,
2095 * otherwise we may lose data while umount */
2096 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2098 truncate_inode_pages_final(&inode->i_data);
2100 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2101 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2102 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2104 #ifdef HAVE_SBOPS_EVICT_INODE
2105 ll_clear_inode(inode);
2112 int ll_iocontrol(struct inode *inode, struct file *file,
2113 unsigned int cmd, unsigned long arg)
2115 struct ll_sb_info *sbi = ll_i2sbi(inode);
2116 struct ptlrpc_request *req = NULL;
2121 case FS_IOC_GETFLAGS: {
2122 struct mdt_body *body;
2123 struct md_op_data *op_data;
2125 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2126 0, 0, LUSTRE_OPC_ANY,
2128 if (IS_ERR(op_data))
2129 RETURN(PTR_ERR(op_data));
2131 op_data->op_valid = OBD_MD_FLFLAGS;
2132 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2133 ll_finish_md_op_data(op_data);
2135 CERROR("%s: failure inode "DFID": rc = %d\n",
2136 sbi->ll_md_exp->exp_obd->obd_name,
2137 PFID(ll_inode2fid(inode)), rc);
2141 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2143 flags = body->mbo_flags;
2145 ptlrpc_req_finished(req);
2147 RETURN(put_user(flags, (int __user *)arg));
2149 case FS_IOC_SETFLAGS: {
2151 struct md_op_data *op_data;
2152 struct cl_object *obj;
2154 if (get_user(flags, (int __user *)arg))
2157 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2158 LUSTRE_OPC_ANY, NULL);
2159 if (IS_ERR(op_data))
2160 RETURN(PTR_ERR(op_data));
2162 op_data->op_attr_flags = flags;
2163 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2164 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2165 ll_finish_md_op_data(op_data);
2166 ptlrpc_req_finished(req);
2170 ll_update_inode_flags(inode, flags);
2172 obj = ll_i2info(inode)->lli_clob;
2176 OBD_ALLOC_PTR(attr);
2180 attr->ia_valid = ATTR_ATTR_FLAG;
2181 rc = cl_setattr_ost(obj, attr, flags);
2193 int ll_flush_ctx(struct inode *inode)
2195 struct ll_sb_info *sbi = ll_i2sbi(inode);
2197 CDEBUG(D_SEC, "flush context for user %d\n",
2198 from_kuid(&init_user_ns, current_uid()));
2200 obd_set_info_async(NULL, sbi->ll_md_exp,
2201 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2203 obd_set_info_async(NULL, sbi->ll_dt_exp,
2204 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2209 /* umount -f client means force down, don't save state */
2210 void ll_umount_begin(struct super_block *sb)
2212 struct ll_sb_info *sbi = ll_s2sbi(sb);
2213 struct obd_device *obd;
2214 struct obd_ioctl_data *ioc_data;
2215 struct l_wait_info lwi;
2216 wait_queue_head_t waitq;
2219 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2220 sb->s_count, atomic_read(&sb->s_active));
2222 obd = class_exp2obd(sbi->ll_md_exp);
2224 CERROR("Invalid MDC connection handle %#llx\n",
2225 sbi->ll_md_exp->exp_handle.h_cookie);
2231 obd = class_exp2obd(sbi->ll_dt_exp);
2233 CERROR("Invalid LOV connection handle %#llx\n",
2234 sbi->ll_dt_exp->exp_handle.h_cookie);
2240 OBD_ALLOC_PTR(ioc_data);
2242 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2243 sizeof *ioc_data, ioc_data, NULL);
2245 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2246 sizeof *ioc_data, ioc_data, NULL);
2248 OBD_FREE_PTR(ioc_data);
2251 /* Really, we'd like to wait until there are no requests outstanding,
2252 * and then continue. For now, we just periodically checking for vfs
2253 * to decrement mnt_cnt and hope to finish it within 10sec.
2255 init_waitqueue_head(&waitq);
2256 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2257 cfs_time_seconds(1), NULL, NULL);
2258 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2263 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2265 struct ll_sb_info *sbi = ll_s2sbi(sb);
2266 char *profilenm = get_profile_name(sb);
2270 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2271 read_only = *flags & MS_RDONLY;
2272 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2273 sizeof(KEY_READ_ONLY),
2274 KEY_READ_ONLY, sizeof(read_only),
2277 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2278 profilenm, read_only ?
2279 "read-only" : "read-write", err);
2284 sb->s_flags |= MS_RDONLY;
2286 sb->s_flags &= ~MS_RDONLY;
2288 if (sbi->ll_flags & LL_SBI_VERBOSE)
2289 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2290 read_only ? "read-only" : "read-write");
2296 * Cleanup the open handle that is cached on MDT-side.
2298 * For open case, the client side open handling thread may hit error
2299 * after the MDT grant the open. Under such case, the client should
2300 * send close RPC to the MDT as cleanup; otherwise, the open handle
2301 * on the MDT will be leaked there until the client umount or evicted.
2303 * In further, if someone unlinked the file, because the open handle
2304 * holds the reference on such file/object, then it will block the
2305 * subsequent threads that want to locate such object via FID.
2307 * \param[in] sb super block for this file-system
2308 * \param[in] open_req pointer to the original open request
2310 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2312 struct mdt_body *body;
2313 struct md_op_data *op_data;
2314 struct ptlrpc_request *close_req = NULL;
2315 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2318 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2319 OBD_ALLOC_PTR(op_data);
2320 if (op_data == NULL) {
2321 CWARN("%s: cannot allocate op_data to release open handle for "
2323 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2328 op_data->op_fid1 = body->mbo_fid1;
2329 op_data->op_handle = body->mbo_handle;
2330 op_data->op_mod_time = ktime_get_real_seconds();
2331 md_close(exp, op_data, NULL, &close_req);
2332 ptlrpc_req_finished(close_req);
2333 ll_finish_md_op_data(op_data);
2338 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2339 struct super_block *sb, struct lookup_intent *it)
2341 struct ll_sb_info *sbi = NULL;
2342 struct lustre_md md = { NULL };
2346 LASSERT(*inode || sb);
2347 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2348 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2349 sbi->ll_md_exp, &md);
2354 rc = ll_update_inode(*inode, &md);
2358 LASSERT(sb != NULL);
2361 * At this point server returns to client's same fid as client
2362 * generated for creating. So using ->fid1 is okay here.
2364 if (!fid_is_sane(&md.body->mbo_fid1)) {
2365 CERROR("%s: Fid is insane "DFID"\n",
2366 ll_get_fsname(sb, NULL, 0),
2367 PFID(&md.body->mbo_fid1));
2368 GOTO(out, rc = -EINVAL);
2371 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2372 sbi->ll_flags & LL_SBI_32BIT_API),
2374 if (IS_ERR(*inode)) {
2375 #ifdef CONFIG_FS_POSIX_ACL
2377 posix_acl_release(md.posix_acl);
2378 md.posix_acl = NULL;
2381 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2383 CERROR("new_inode -fatal: rc %d\n", rc);
2388 /* Handling piggyback layout lock.
2389 * Layout lock can be piggybacked by getattr and open request.
2390 * The lsm can be applied to inode only if it comes with a layout lock
2391 * otherwise correct layout may be overwritten, for example:
2392 * 1. proc1: mdt returns a lsm but not granting layout
2393 * 2. layout was changed by another client
2394 * 3. proc2: refresh layout and layout lock granted
2395 * 4. proc1: to apply a stale layout */
2396 if (it != NULL && it->it_lock_mode != 0) {
2397 struct lustre_handle lockh;
2398 struct ldlm_lock *lock;
2400 lockh.cookie = it->it_lock_handle;
2401 lock = ldlm_handle2lock(&lockh);
2402 LASSERT(lock != NULL);
2403 if (ldlm_has_layout(lock)) {
2404 struct cl_object_conf conf;
2406 memset(&conf, 0, sizeof(conf));
2407 conf.coc_opc = OBJECT_CONF_SET;
2408 conf.coc_inode = *inode;
2409 conf.coc_lock = lock;
2410 conf.u.coc_layout = md.layout;
2411 (void)ll_layout_conf(*inode, &conf);
2413 LDLM_LOCK_PUT(lock);
2419 md_free_lustre_md(sbi->ll_md_exp, &md);
2422 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2423 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2428 int ll_obd_statfs(struct inode *inode, void __user *arg)
2430 struct ll_sb_info *sbi = NULL;
2431 struct obd_export *exp;
2433 struct obd_ioctl_data *data = NULL;
2437 if (!inode || !(sbi = ll_i2sbi(inode)))
2438 GOTO(out_statfs, rc = -EINVAL);
2440 rc = obd_ioctl_getdata(&buf, &len, arg);
2442 GOTO(out_statfs, rc);
2445 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2446 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2447 GOTO(out_statfs, rc = -EINVAL);
2449 if (data->ioc_inllen1 != sizeof(__u32) ||
2450 data->ioc_inllen2 != sizeof(__u32) ||
2451 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2452 data->ioc_plen2 != sizeof(struct obd_uuid))
2453 GOTO(out_statfs, rc = -EINVAL);
2455 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2456 if (type & LL_STATFS_LMV)
2457 exp = sbi->ll_md_exp;
2458 else if (type & LL_STATFS_LOV)
2459 exp = sbi->ll_dt_exp;
2461 GOTO(out_statfs, rc = -ENODEV);
2463 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2465 GOTO(out_statfs, rc);
2467 OBD_FREE_LARGE(buf, len);
2471 /* this function prepares md_op_data hint for passing it down to MD stack. */
2472 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2473 struct inode *i1, struct inode *i2,
2474 const char *name, size_t namelen,
2475 __u32 mode, __u32 opc, void *data)
2477 LASSERT(i1 != NULL);
2480 /* Do not reuse namelen for something else. */
2482 return ERR_PTR(-EINVAL);
2484 if (namelen > ll_i2sbi(i1)->ll_namelen)
2485 return ERR_PTR(-ENAMETOOLONG);
2487 if (!lu_name_is_valid_2(name, namelen))
2488 return ERR_PTR(-EINVAL);
2491 if (op_data == NULL)
2492 OBD_ALLOC_PTR(op_data);
2494 if (op_data == NULL)
2495 return ERR_PTR(-ENOMEM);
2497 ll_i2gids(op_data->op_suppgids, i1, i2);
2498 op_data->op_fid1 = *ll_inode2fid(i1);
2499 op_data->op_default_stripe_offset = -1;
2500 if (S_ISDIR(i1->i_mode)) {
2501 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2502 if (opc == LUSTRE_OPC_MKDIR)
2503 op_data->op_default_stripe_offset =
2504 ll_i2info(i1)->lli_def_stripe_offset;
2508 op_data->op_fid2 = *ll_inode2fid(i2);
2509 if (S_ISDIR(i2->i_mode))
2510 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2512 fid_zero(&op_data->op_fid2);
2515 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2516 op_data->op_cli_flags |= CLI_HASH64;
2518 if (ll_need_32bit_api(ll_i2sbi(i1)))
2519 op_data->op_cli_flags |= CLI_API32;
2521 op_data->op_name = name;
2522 op_data->op_namelen = namelen;
2523 op_data->op_mode = mode;
2524 op_data->op_mod_time = ktime_get_real_seconds();
2525 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2526 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2527 op_data->op_cap = cfs_curproc_cap_pack();
2528 op_data->op_mds = 0;
2529 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2530 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2531 op_data->op_bias |= MDS_CREATE_VOLATILE;
2533 op_data->op_data = data;
2538 void ll_finish_md_op_data(struct md_op_data *op_data)
2540 security_release_secctx(op_data->op_file_secctx,
2541 op_data->op_file_secctx_size);
2542 OBD_FREE_PTR(op_data);
2545 #ifdef HAVE_SUPEROPS_USE_DENTRY
2546 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2548 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2551 struct ll_sb_info *sbi;
2553 #ifdef HAVE_SUPEROPS_USE_DENTRY
2554 LASSERT((seq != NULL) && (dentry != NULL));
2555 sbi = ll_s2sbi(dentry->d_sb);
2557 LASSERT((seq != NULL) && (vfs != NULL));
2558 sbi = ll_s2sbi(vfs->mnt_sb);
2561 if (sbi->ll_flags & LL_SBI_NOLCK)
2562 seq_puts(seq, ",nolock");
2564 if (sbi->ll_flags & LL_SBI_FLOCK)
2565 seq_puts(seq, ",flock");
2567 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2568 seq_puts(seq, ",localflock");
2570 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2571 seq_puts(seq, ",user_xattr");
2573 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2574 seq_puts(seq, ",lazystatfs");
2576 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2577 seq_puts(seq, ",user_fid2path");
2579 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2580 seq_puts(seq, ",always_ping");
2586 * Get obd name by cmd, and copy out to user space
2588 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2590 struct ll_sb_info *sbi = ll_i2sbi(inode);
2591 struct obd_device *obd;
2594 if (cmd == OBD_IOC_GETDTNAME)
2595 obd = class_exp2obd(sbi->ll_dt_exp);
2596 else if (cmd == OBD_IOC_GETMDNAME)
2597 obd = class_exp2obd(sbi->ll_md_exp);
2604 if (copy_to_user((void __user *)arg, obd->obd_name,
2605 strlen(obd->obd_name) + 1))
2612 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2613 * fsname will be returned in this buffer; otherwise, a static buffer will be
2614 * used to store the fsname and returned to caller.
2616 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2618 static char fsname_static[MTI_NAME_MAXLEN];
2619 struct lustre_sb_info *lsi = s2lsi(sb);
2624 /* this means the caller wants to use static buffer
2625 * and it doesn't care about race. Usually this is
2626 * in error reporting path */
2627 buf = fsname_static;
2628 buflen = sizeof(fsname_static);
2631 len = strlen(lsi->lsi_lmd->lmd_profile);
2632 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2633 if (ptr && (strcmp(ptr, "-client") == 0))
2636 if (unlikely(len >= buflen))
2638 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2644 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2651 p.mnt = current->fs->root.mnt;
2653 path = d_path(&p, buf, bufsize);
2658 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2660 char *buf, *path = NULL;
2661 struct dentry *dentry = NULL;
2662 struct inode *inode = page->mapping->host;
2664 /* this can be called inside spin lock so use GFP_ATOMIC. */
2665 buf = (char *)__get_free_page(GFP_ATOMIC);
2667 dentry = d_find_alias(page->mapping->host);
2669 path = ll_d_path(dentry, buf, PAGE_SIZE);
2673 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2674 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2675 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2676 PFID(ll_inode2fid(inode)),
2677 (path && !IS_ERR(path)) ? path : "", ioret);
2683 free_page((unsigned long)buf);
2686 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2687 struct lov_user_md **kbuf)
2689 struct lov_user_md lum;
2693 if (copy_from_user(&lum, md, sizeof(lum)))
2696 lum_size = ll_lov_user_md_size(&lum);
2700 OBD_ALLOC(*kbuf, lum_size);
2704 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2705 OBD_FREE(*kbuf, lum_size);
2713 * Compute llite root squash state after a change of root squash
2714 * configuration setting or add/remove of a lnet nid
2716 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2718 struct root_squash_info *squash = &sbi->ll_squash;
2721 struct lnet_process_id id;
2723 /* Update norootsquash flag */
2724 down_write(&squash->rsi_sem);
2725 if (list_empty(&squash->rsi_nosquash_nids))
2726 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2728 /* Do not apply root squash as soon as one of our NIDs is
2729 * in the nosquash_nids list */
2732 while (LNetGetId(i++, &id) != -ENOENT) {
2733 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2735 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2741 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2743 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2745 up_write(&squash->rsi_sem);
2749 * Parse linkea content to extract information about a given hardlink
2751 * \param[in] ldata - Initialized linkea data
2752 * \param[in] linkno - Link identifier
2753 * \param[out] parent_fid - The entry's parent FID
2754 * \param[out] ln - Entry name destination buffer
2756 * \retval 0 on success
2757 * \retval Appropriate negative error code on failure
2759 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2760 struct lu_fid *parent_fid, struct lu_name *ln)
2766 rc = linkea_init_with_rec(ldata);
2770 if (linkno >= ldata->ld_leh->leh_reccount)
2771 /* beyond last link */
2774 linkea_first_entry(ldata);
2775 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2776 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2781 linkea_next_entry(ldata);
2791 * Get parent FID and name of an identified link. Operation is performed for
2792 * a given link number, letting the caller iterate over linkno to list one or
2793 * all links of an entry.
2795 * \param[in] file - File descriptor against which to perform the operation
2796 * \param[in,out] arg - User-filled structure containing the linkno to operate
2797 * on and the available size. It is eventually filled with
2798 * the requested information or left untouched on error
2800 * \retval - 0 on success
2801 * \retval - Appropriate negative error code on failure
2803 int ll_getparent(struct file *file, struct getparent __user *arg)
2805 struct inode *inode = file_inode(file);
2806 struct linkea_data *ldata;
2807 struct lu_buf buf = LU_BUF_NULL;
2809 struct lu_fid parent_fid;
2816 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2817 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2820 if (get_user(name_size, &arg->gp_name_size))
2823 if (get_user(linkno, &arg->gp_linkno))
2826 if (name_size > PATH_MAX)
2829 OBD_ALLOC(ldata, sizeof(*ldata));
2833 rc = linkea_data_new(ldata, &buf);
2835 GOTO(ldata_free, rc);
2837 #ifdef HAVE_XATTR_HANDLER_FLAGS
2838 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2839 buf.lb_len, OBD_MD_FLXATTR);
2841 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2843 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2847 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2851 if (ln.ln_namelen >= name_size)
2852 GOTO(lb_free, rc = -EOVERFLOW);
2854 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2855 GOTO(lb_free, rc = -EFAULT);
2857 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2858 GOTO(lb_free, rc = -EFAULT);
2860 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2861 GOTO(lb_free, rc = -EFAULT);
2866 OBD_FREE(ldata, sizeof(*ldata));