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, 2016, 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_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_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 =
103 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
105 ll_generate_random_uuid(uuid);
106 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
107 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
109 sbi->ll_flags |= LL_SBI_VERBOSE;
110 #ifdef ENABLE_CHECKSUM
111 sbi->ll_flags |= LL_SBI_CHECKSUM;
114 #ifdef HAVE_LRU_RESIZE_SUPPORT
115 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
117 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
119 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
120 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
122 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 /* metadata statahead is enabled by default */
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;
136 sbi->ll_squash.rsi_uid = 0;
137 sbi->ll_squash.rsi_gid = 0;
138 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
139 init_rwsem(&sbi->ll_squash.rsi_sem);
144 static void ll_free_sbi(struct super_block *sb)
146 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
151 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
152 if (sbi->ll_cache != NULL) {
153 cl_cache_decref(sbi->ll_cache);
154 sbi->ll_cache = NULL;
156 OBD_FREE(sbi, sizeof(*sbi));
161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
163 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
164 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168 struct vfsmount *mnt)
170 struct inode *root = NULL;
171 struct ll_sb_info *sbi = ll_s2sbi(sb);
172 struct obd_device *obd;
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 obd = class_name2obd(md);
185 CERROR("MD %s: not setup or attached\n", md);
199 /* indicate the features supported by this client */
200 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
201 OBD_CONNECT_ATTRFID |
202 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
203 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
204 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
205 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
206 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
207 OBD_CONNECT_64BITHASH |
208 OBD_CONNECT_EINPROGRESS |
209 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
210 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
211 OBD_CONNECT_MAX_EASIZE |
212 OBD_CONNECT_FLOCK_DEAD |
213 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
214 OBD_CONNECT_OPEN_BY_FID |
215 OBD_CONNECT_DIR_STRIPE |
216 OBD_CONNECT_BULK_MBITS |
217 OBD_CONNECT_SUBTREE |
218 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
220 data->ocd_connect_flags2 = 0;
222 #ifdef HAVE_LRU_RESIZE_SUPPORT
223 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
224 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
226 #ifdef CONFIG_FS_POSIX_ACL
227 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
228 OBD_CONNECT_LARGE_ACL;
231 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
232 /* flag mdc connection as lightweight, only used for test
233 * purpose, use with care */
234 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
236 data->ocd_ibits_known = MDS_INODELOCK_FULL;
237 data->ocd_version = LUSTRE_VERSION_CODE;
239 if (sb->s_flags & MS_RDONLY)
240 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
241 if (sbi->ll_flags & LL_SBI_USER_XATTR)
242 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
245 /* Setting this indicates we correctly support S_NOSEC (See kernel
246 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
248 sb->s_flags |= MS_NOSEC;
251 if (sbi->ll_flags & LL_SBI_FLOCK)
252 sbi->ll_fop = &ll_file_operations_flock;
253 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
254 sbi->ll_fop = &ll_file_operations;
256 sbi->ll_fop = &ll_file_operations_noflock;
258 /* always ping even if server suppress_pings */
259 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
260 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
262 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
263 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
264 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
266 data->ocd_brw_size = MD_MAX_BRW_SIZE;
268 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
270 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
271 "recovery, of which this client is not a "
272 "part. Please wait for recovery to complete,"
273 " abort, or time out.\n", md);
276 CERROR("cannot connect to %s: rc = %d\n", md, err);
280 sbi->ll_md_exp->exp_connect_data = *data;
282 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
283 LUSTRE_SEQ_METADATA);
285 CERROR("%s: Can't init metadata layer FID infrastructure, "
286 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
290 /* For mount, we only need fs info from MDT0, and also in DNE, it
291 * can make sure the client can be mounted as long as MDT0 is
293 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
294 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
295 OBD_STATFS_FOR_MDT0);
297 GOTO(out_md_fid, err);
299 /* This needs to be after statfs to ensure connect has finished.
300 * Note that "data" does NOT contain the valid connect reply.
301 * If connecting to a 1.8 server there will be no LMV device, so
302 * we can access the MDC export directly and exp_connect_flags will
303 * be non-zero, but if accessing an upgraded 2.1 server it will
304 * have the correct flags filled in.
305 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
306 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
307 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
308 valid != CLIENT_CONNECT_MDT_REQD) {
311 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
312 obd_connect_flags2str(buf, PAGE_SIZE,
313 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
314 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
315 "feature(s) needed for correct operation "
316 "of this client (%s). Please upgrade "
317 "server or downgrade client.\n",
318 sbi->ll_md_exp->exp_obd->obd_name, buf);
319 OBD_FREE(buf, PAGE_SIZE);
320 GOTO(out_md_fid, err = -EPROTO);
323 size = sizeof(*data);
324 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
325 KEY_CONN_DATA, &size, data);
327 CERROR("%s: Get connect data failed: rc = %d\n",
328 sbi->ll_md_exp->exp_obd->obd_name, err);
329 GOTO(out_md_fid, err);
332 LASSERT(osfs->os_bsize);
333 sb->s_blocksize = osfs->os_bsize;
334 sb->s_blocksize_bits = log2(osfs->os_bsize);
335 sb->s_magic = LL_SUPER_MAGIC;
336 sb->s_maxbytes = MAX_LFS_FILESIZE;
337 sbi->ll_namelen = osfs->os_namelen;
338 sbi->ll_mnt.mnt = current->fs->root.mnt;
340 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
341 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
342 LCONSOLE_INFO("Disabling user_xattr feature because "
343 "it is not supported on the server\n");
344 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
347 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
349 sb->s_flags |= MS_POSIXACL;
351 sbi->ll_flags |= LL_SBI_ACL;
353 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
355 sb->s_flags &= ~MS_POSIXACL;
357 sbi->ll_flags &= ~LL_SBI_ACL;
360 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
361 sbi->ll_flags |= LL_SBI_64BIT_HASH;
363 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
364 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
366 if (obd_connect_has_secctx(data))
367 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
369 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
370 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
371 LCONSOLE_INFO("%s: disabling xattr cache due to "
372 "unknown maximum xattr size.\n", dt);
374 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
375 sbi->ll_xattr_cache_enabled = 1;
379 obd = class_name2obd(dt);
381 CERROR("DT %s: not setup or attached\n", dt);
382 GOTO(out_md_fid, err = -ENODEV);
385 /* pass client page size via ocd_grant_blkbits, the server should report
386 * back its backend blocksize for grant calculation purpose */
387 data->ocd_grant_blkbits = PAGE_SHIFT;
389 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
390 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
391 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
392 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
393 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
394 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
395 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
396 OBD_CONNECT_EINPROGRESS |
397 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
398 OBD_CONNECT_LAYOUTLOCK |
399 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
400 OBD_CONNECT_BULK_MBITS;
402 data->ocd_connect_flags2 = 0;
404 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
405 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
407 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
408 * disabled by default, because it can still be enabled on the
409 * fly via /proc. As a consequence, we still need to come to an
410 * agreement on the supported algorithms at connect time */
411 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
413 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
414 data->ocd_cksum_types = OBD_CKSUM_ADLER;
416 data->ocd_cksum_types = cksum_types_supported_client();
418 #ifdef HAVE_LRU_RESIZE_SUPPORT
419 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
421 /* always ping even if server suppress_pings */
422 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
423 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
425 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
426 "ocd_grant: %d\n", data->ocd_connect_flags,
427 data->ocd_version, data->ocd_grant);
429 obd->obd_upcall.onu_owner = &sbi->ll_lco;
430 obd->obd_upcall.onu_upcall = cl_ocd_update;
432 data->ocd_brw_size = DT_MAX_BRW_SIZE;
434 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
437 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
438 "recovery, of which this client is not a "
439 "part. Please wait for recovery to "
440 "complete, abort, or time out.\n", dt);
443 CERROR("%s: Cannot connect to %s: rc = %d\n",
444 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
448 sbi->ll_dt_exp->exp_connect_data = *data;
450 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
451 LUSTRE_SEQ_METADATA);
453 CERROR("%s: Can't init data layer FID infrastructure, "
454 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
458 mutex_lock(&sbi->ll_lco.lco_lock);
459 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
460 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
461 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
462 mutex_unlock(&sbi->ll_lco.lco_lock);
464 fid_zero(&sbi->ll_root_fid);
465 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
468 CERROR("cannot mds_connect: rc = %d\n", err);
469 GOTO(out_lock_cn_cb, err);
471 if (!fid_is_sane(&sbi->ll_root_fid)) {
472 CERROR("%s: Invalid root fid "DFID" during mount\n",
473 sbi->ll_md_exp->exp_obd->obd_name,
474 PFID(&sbi->ll_root_fid));
475 GOTO(out_lock_cn_cb, err = -EINVAL);
477 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
479 sb->s_op = &lustre_super_operations;
480 #if THREAD_SIZE >= 8192 /*b=17630*/
481 sb->s_export_op = &lustre_export_operations;
485 * XXX: move this to after cbd setup? */
486 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
487 if (sbi->ll_flags & LL_SBI_ACL)
488 valid |= OBD_MD_FLACL;
490 OBD_ALLOC_PTR(op_data);
492 GOTO(out_lock_cn_cb, err = -ENOMEM);
494 op_data->op_fid1 = sbi->ll_root_fid;
495 op_data->op_mode = 0;
496 op_data->op_valid = valid;
498 err = md_getattr(sbi->ll_md_exp, op_data, &request);
500 OBD_FREE_PTR(op_data);
502 CERROR("%s: md_getattr failed for root: rc = %d\n",
503 sbi->ll_md_exp->exp_obd->obd_name, err);
504 GOTO(out_lock_cn_cb, err);
507 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
508 sbi->ll_md_exp, &lmd);
510 CERROR("failed to understand root inode md: rc = %d\n", err);
511 ptlrpc_req_finished(request);
512 GOTO(out_lock_cn_cb, err);
515 LASSERT(fid_is_sane(&sbi->ll_root_fid));
516 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
517 sbi->ll_flags & LL_SBI_32BIT_API),
519 md_free_lustre_md(sbi->ll_md_exp, &lmd);
520 ptlrpc_req_finished(request);
523 #ifdef CONFIG_FS_POSIX_ACL
525 posix_acl_release(lmd.posix_acl);
526 lmd.posix_acl = NULL;
529 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
531 CERROR("lustre_lite: bad iget4 for root\n");
535 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
536 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
537 KEY_CHECKSUM, sizeof(checksum), &checksum,
540 CERROR("%s: Set checksum failed: rc = %d\n",
541 sbi->ll_dt_exp->exp_obd->obd_name, err);
546 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
547 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
548 sbi->ll_cache, NULL);
550 CERROR("%s: Set cache_set failed: rc = %d\n",
551 sbi->ll_dt_exp->exp_obd->obd_name, err);
555 sb->s_root = d_make_root(root);
556 if (sb->s_root == NULL) {
557 CERROR("%s: can't make root dentry\n",
558 ll_get_fsname(sb, NULL, 0));
559 GOTO(out_root, err = -ENOMEM);
561 #ifdef HAVE_DCACHE_LOCK
562 sb->s_root->d_op = &ll_d_ops;
565 sbi->ll_sdev_orig = sb->s_dev;
567 /* We set sb->s_dev equal on all lustre clients in order to support
568 * NFS export clustering. NFSD requires that the FSID be the same
570 /* s_dev is also used in lt_compare() to compare two fs, but that is
571 * only a node-local comparison. */
572 uuid = obd_get_uuid(sbi->ll_md_exp);
574 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
580 if (proc_lustre_fs_root != NULL) {
581 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
584 CERROR("%s: could not register mount in lprocfs: "
585 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
595 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
597 obd_disconnect(sbi->ll_dt_exp);
598 sbi->ll_dt_exp = NULL;
600 obd_fid_fini(sbi->ll_md_exp->exp_obd);
602 obd_disconnect(sbi->ll_md_exp);
603 sbi->ll_md_exp = NULL;
612 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
616 size = sizeof(*lmmsize);
617 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
618 KEY_MAX_EASIZE, &size, lmmsize);
620 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
621 sbi->ll_dt_exp->exp_obd->obd_name, rc);
626 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
627 KEY_MAX_EASIZE, &size, lmmsize);
629 CERROR("Get max mdsize error rc %d\n", rc);
635 * Get the value of the default_easize parameter.
637 * \see client_obd::cl_default_mds_easize
639 * \param[in] sbi superblock info for this filesystem
640 * \param[out] lmmsize pointer to storage location for value
642 * \retval 0 on success
643 * \retval negative negated errno on failure
645 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
650 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
651 KEY_DEFAULT_EASIZE, &size, lmmsize);
653 CERROR("Get default mdsize error rc %d\n", rc);
659 * Set the default_easize parameter to the given value.
661 * \see client_obd::cl_default_mds_easize
663 * \param[in] sbi superblock info for this filesystem
664 * \param[in] lmmsize the size to set
666 * \retval 0 on success
667 * \retval negative negated errno on failure
669 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
673 if (lmmsize < sizeof(struct lov_mds_md) ||
674 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
677 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
678 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
679 sizeof(int), &lmmsize, NULL);
684 static void client_common_put_super(struct super_block *sb)
686 struct ll_sb_info *sbi = ll_s2sbi(sb);
691 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
692 obd_disconnect(sbi->ll_dt_exp);
693 sbi->ll_dt_exp = NULL;
695 lprocfs_unregister_mountpoint(sbi);
697 obd_fid_fini(sbi->ll_md_exp->exp_obd);
698 obd_disconnect(sbi->ll_md_exp);
699 sbi->ll_md_exp = NULL;
704 void ll_kill_super(struct super_block *sb)
706 struct ll_sb_info *sbi;
710 if (!(sb->s_flags & MS_ACTIVE))
714 /* we need restore s_dev from changed for clustred NFS before put_super
715 * because new kernels have cached s_dev and change sb->s_dev in
716 * put_super not affected real removing devices */
718 sb->s_dev = sbi->ll_sdev_orig;
719 sbi->ll_umounting = 1;
721 /* wait running statahead threads to quit */
722 while (atomic_read(&sbi->ll_sa_running) > 0) {
723 set_current_state(TASK_UNINTERRUPTIBLE);
724 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
731 static inline int ll_set_opt(const char *opt, char *data, int fl)
733 if (strncmp(opt, data, strlen(opt)) != 0)
739 /* non-client-specific mount options are parsed in lmd_parse */
740 static int ll_options(char *options, int *flags)
743 char *s1 = options, *s2;
749 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
752 CDEBUG(D_SUPER, "next opt=%s\n", s1);
753 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
758 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
763 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
768 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
773 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
778 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
783 tmp = ll_set_opt("context", s1, 1);
786 tmp = ll_set_opt("fscontext", s1, 1);
789 tmp = ll_set_opt("defcontext", s1, 1);
792 tmp = ll_set_opt("rootcontext", s1, 1);
795 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
800 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
806 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
811 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
816 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
821 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
826 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
831 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
836 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
841 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
846 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
851 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
856 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
862 s2 = strchr(s1, ',');
870 void ll_lli_init(struct ll_inode_info *lli)
872 lli->lli_inode_magic = LLI_INODE_MAGIC;
874 spin_lock_init(&lli->lli_lock);
875 lli->lli_posix_acl = NULL;
876 /* Do not set lli_fid, it has been initialized already. */
877 fid_zero(&lli->lli_pfid);
878 lli->lli_mds_read_och = NULL;
879 lli->lli_mds_write_och = NULL;
880 lli->lli_mds_exec_och = NULL;
881 lli->lli_open_fd_read_count = 0;
882 lli->lli_open_fd_write_count = 0;
883 lli->lli_open_fd_exec_count = 0;
884 mutex_init(&lli->lli_och_mutex);
885 spin_lock_init(&lli->lli_agl_lock);
886 spin_lock_init(&lli->lli_layout_lock);
887 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
888 lli->lli_clob = NULL;
890 init_rwsem(&lli->lli_xattrs_list_rwsem);
891 mutex_init(&lli->lli_xattrs_enq_lock);
893 LASSERT(lli->lli_vfs_inode.i_mode != 0);
894 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
895 mutex_init(&lli->lli_readdir_mutex);
896 lli->lli_opendir_key = NULL;
898 spin_lock_init(&lli->lli_sa_lock);
899 lli->lli_opendir_pid = 0;
900 lli->lli_sa_enabled = 0;
901 lli->lli_def_stripe_offset = -1;
903 mutex_init(&lli->lli_size_mutex);
904 lli->lli_symlink_name = NULL;
905 init_rwsem(&lli->lli_trunc_sem);
906 range_lock_tree_init(&lli->lli_write_tree);
907 init_rwsem(&lli->lli_glimpse_sem);
908 lli->lli_glimpse_time = 0;
909 INIT_LIST_HEAD(&lli->lli_agl_list);
910 lli->lli_agl_index = 0;
911 lli->lli_async_rc = 0;
913 mutex_init(&lli->lli_layout_mutex);
914 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
917 static inline int ll_bdi_register(struct backing_dev_info *bdi)
919 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
921 bdi->name = "lustre";
922 return bdi_register(bdi, NULL, "lustre-%d",
923 atomic_inc_return(&ll_bdi_num));
926 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
928 struct lustre_profile *lprof = NULL;
929 struct lustre_sb_info *lsi = s2lsi(sb);
930 struct ll_sb_info *sbi;
931 char *dt = NULL, *md = NULL;
932 char *profilenm = get_profile_name(sb);
933 struct config_llog_instance *cfg;
934 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
935 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
939 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
945 try_module_get(THIS_MODULE);
947 /* client additional sb info */
948 lsi->lsi_llsbi = sbi = ll_init_sbi();
950 module_put(THIS_MODULE);
955 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
959 err = bdi_init(&lsi->lsi_bdi);
962 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
963 #ifdef HAVE_BDI_CAP_MAP_COPY
964 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
966 lsi->lsi_bdi.capabilities = 0;
968 err = ll_bdi_register(&lsi->lsi_bdi);
972 sb->s_bdi = &lsi->lsi_bdi;
973 #ifndef HAVE_DCACHE_LOCK
974 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
975 sb->s_d_op = &ll_d_ops;
978 /* Generate a string unique to this super, in case some joker tries
979 to mount the same fs at two mount points.
980 Use the address of the super itself.*/
981 cfg->cfg_instance = sb;
982 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
983 cfg->cfg_callback = class_config_llog_handler;
984 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
985 /* set up client obds */
986 err = lustre_process_log(sb, profilenm, cfg);
990 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
991 lprof = class_get_profile(profilenm);
993 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
994 " read from the MGS. Does that filesystem "
995 "exist?\n", profilenm);
996 GOTO(out_free, err = -EINVAL);
998 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
999 lprof->lp_md, lprof->lp_dt);
1001 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1003 GOTO(out_free, err = -ENOMEM);
1004 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1006 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1008 GOTO(out_free, err = -ENOMEM);
1009 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1011 /* connections, registrations, sb setup */
1012 err = client_common_fill_super(sb, md, dt, mnt);
1014 GOTO(out_free, err);
1016 sbi->ll_client_common_fill_super_succeeded = 1;
1020 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1022 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1024 class_put_profile(lprof);
1027 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1028 LCONSOLE_WARN("Mounted %s\n", profilenm);
1032 } /* ll_fill_super */
1034 void ll_put_super(struct super_block *sb)
1036 struct config_llog_instance cfg, params_cfg;
1037 struct obd_device *obd;
1038 struct lustre_sb_info *lsi = s2lsi(sb);
1039 struct ll_sb_info *sbi = ll_s2sbi(sb);
1040 char *profilenm = get_profile_name(sb);
1042 int next, force = 1, rc = 0;
1045 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1047 cfg.cfg_instance = sb;
1048 lustre_end_log(sb, profilenm, &cfg);
1050 params_cfg.cfg_instance = sb;
1051 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1053 if (sbi->ll_md_exp) {
1054 obd = class_exp2obd(sbi->ll_md_exp);
1056 force = obd->obd_force;
1059 /* Wait for unstable pages to be committed to stable storage */
1061 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1062 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1063 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1067 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1068 if (force == 0 && rc != -EINTR)
1069 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1072 /* We need to set force before the lov_disconnect in
1073 lustre_common_put_super, since l_d cleans up osc's as well. */
1076 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1078 obd->obd_force = force;
1082 if (sbi->ll_client_common_fill_super_succeeded) {
1083 /* Only if client_common_fill_super succeeded */
1084 client_common_put_super(sb);
1088 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1089 class_manual_cleanup(obd);
1092 if (sbi->ll_flags & LL_SBI_VERBOSE)
1093 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1096 class_del_profile(profilenm);
1098 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1099 bdi_destroy(&lsi->lsi_bdi);
1100 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1104 lsi->lsi_llsbi = NULL;
1106 lustre_common_put_super(sb);
1108 cl_env_cache_purge(~0);
1110 module_put(THIS_MODULE);
1113 } /* client_put_super */
1115 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1117 struct inode *inode = NULL;
1119 /* NOTE: we depend on atomic igrab() -bzzz */
1120 lock_res_and_lock(lock);
1121 if (lock->l_resource->lr_lvb_inode) {
1122 struct ll_inode_info * lli;
1123 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1124 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1125 inode = igrab(lock->l_resource->lr_lvb_inode);
1127 inode = lock->l_resource->lr_lvb_inode;
1128 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1129 D_WARNING, lock, "lr_lvb_inode %p is "
1130 "bogus: magic %08x",
1131 lock->l_resource->lr_lvb_inode,
1132 lli->lli_inode_magic);
1136 unlock_res_and_lock(lock);
1140 void ll_dir_clear_lsm_md(struct inode *inode)
1142 struct ll_inode_info *lli = ll_i2info(inode);
1144 LASSERT(S_ISDIR(inode->i_mode));
1146 if (lli->lli_lsm_md != NULL) {
1147 lmv_free_memmd(lli->lli_lsm_md);
1148 lli->lli_lsm_md = NULL;
1152 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1153 const struct lu_fid *fid,
1154 struct lustre_md *md)
1156 struct ll_sb_info *sbi = ll_s2sbi(sb);
1157 struct mdt_body *body = md->body;
1158 struct inode *inode;
1162 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1163 inode = iget_locked(sb, ino);
1164 if (inode == NULL) {
1165 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1166 ll_get_fsname(sb, NULL, 0), PFID(fid));
1167 RETURN(ERR_PTR(-ENOENT));
1170 if (inode->i_state & I_NEW) {
1171 struct ll_inode_info *lli = ll_i2info(inode);
1172 struct lmv_stripe_md *lsm = md->lmv;
1174 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1175 (body->mbo_mode & S_IFMT);
1176 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1179 LTIME_S(inode->i_mtime) = 0;
1180 LTIME_S(inode->i_atime) = 0;
1181 LTIME_S(inode->i_ctime) = 0;
1184 #ifdef HAVE_BACKING_DEV_INFO
1185 /* initializing backing dev info. */
1186 inode->i_mapping->backing_dev_info =
1187 &s2lsi(inode->i_sb)->lsi_bdi;
1189 inode->i_op = &ll_dir_inode_operations;
1190 inode->i_fop = &ll_dir_operations;
1191 lli->lli_fid = *fid;
1194 LASSERT(lsm != NULL);
1195 /* master object FID */
1196 lli->lli_pfid = body->mbo_fid1;
1197 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1198 lli, PFID(fid), PFID(&lli->lli_pfid));
1199 unlock_new_inode(inode);
1205 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1208 struct lmv_stripe_md *lsm = md->lmv;
1211 LASSERT(lsm != NULL);
1212 /* XXX sigh, this lsm_root initialization should be in
1213 * LMV layer, but it needs ll_iget right now, so we
1214 * put this here right now. */
1215 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1216 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1217 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1218 /* Unfortunately ll_iget will call ll_update_inode,
1219 * where the initialization of slave inode is slightly
1220 * different, so it reset lsm_md to NULL to avoid
1221 * initializing lsm for slave inode. */
1222 /* For migrating inode, master stripe and master object will
1223 * be same, so we only need assign this inode */
1224 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1225 lsm->lsm_md_oinfo[i].lmo_root = inode;
1227 lsm->lsm_md_oinfo[i].lmo_root =
1228 ll_iget_anon_dir(inode->i_sb, fid, md);
1230 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1231 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1233 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1241 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1242 const struct lmv_stripe_md *lsm_md2)
1244 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1245 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1246 lsm_md1->lsm_md_master_mdt_index ==
1247 lsm_md2->lsm_md_master_mdt_index &&
1248 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1249 lsm_md1->lsm_md_layout_version ==
1250 lsm_md2->lsm_md_layout_version &&
1251 strcmp(lsm_md1->lsm_md_pool_name,
1252 lsm_md2->lsm_md_pool_name) == 0;
1255 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1257 struct ll_inode_info *lli = ll_i2info(inode);
1258 struct lmv_stripe_md *lsm = md->lmv;
1262 LASSERT(S_ISDIR(inode->i_mode));
1263 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1264 PFID(ll_inode2fid(inode)));
1266 /* no striped information from request. */
1268 if (lli->lli_lsm_md == NULL) {
1270 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1271 LMV_HASH_FLAG_MIGRATION) {
1272 /* migration is done, the temporay MIGRATE layout has
1274 CDEBUG(D_INODE, DFID" finish migration.\n",
1275 PFID(ll_inode2fid(inode)));
1276 lmv_free_memmd(lli->lli_lsm_md);
1277 lli->lli_lsm_md = NULL;
1280 /* The lustre_md from req does not include stripeEA,
1281 * see ll_md_setattr */
1286 /* set the directory layout */
1287 if (lli->lli_lsm_md == NULL) {
1288 struct cl_attr *attr;
1290 rc = ll_init_lsm_md(inode, md);
1294 /* set md->lmv to NULL, so the following free lustre_md
1295 * will not free this lsm */
1297 lli->lli_lsm_md = lsm;
1299 OBD_ALLOC_PTR(attr);
1303 /* validate the lsm */
1304 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1305 ll_md_blocking_ast);
1311 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1312 md->body->mbo_nlink = attr->cat_nlink;
1313 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1314 md->body->mbo_size = attr->cat_size;
1315 if (md->body->mbo_valid & OBD_MD_FLATIME)
1316 md->body->mbo_atime = attr->cat_atime;
1317 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1318 md->body->mbo_ctime = attr->cat_ctime;
1319 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1320 md->body->mbo_mtime = attr->cat_mtime;
1324 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1325 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1329 /* Compare the old and new stripe information */
1330 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1331 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1334 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1335 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1336 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1337 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1338 inode, lsm, old_lsm,
1339 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1340 lsm->lsm_md_stripe_count,
1341 old_lsm->lsm_md_stripe_count,
1342 lsm->lsm_md_master_mdt_index,
1343 old_lsm->lsm_md_master_mdt_index,
1344 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1345 lsm->lsm_md_layout_version,
1346 old_lsm->lsm_md_layout_version,
1347 lsm->lsm_md_pool_name,
1348 old_lsm->lsm_md_pool_name);
1350 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1351 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1352 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1353 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1356 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1357 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1358 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1359 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1368 void ll_clear_inode(struct inode *inode)
1370 struct ll_inode_info *lli = ll_i2info(inode);
1371 struct ll_sb_info *sbi = ll_i2sbi(inode);
1374 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1375 PFID(ll_inode2fid(inode)), inode);
1377 if (S_ISDIR(inode->i_mode)) {
1378 /* these should have been cleared in ll_file_release */
1379 LASSERT(lli->lli_opendir_key == NULL);
1380 LASSERT(lli->lli_sai == NULL);
1381 LASSERT(lli->lli_opendir_pid == 0);
1384 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1386 LASSERT(!lli->lli_open_fd_write_count);
1387 LASSERT(!lli->lli_open_fd_read_count);
1388 LASSERT(!lli->lli_open_fd_exec_count);
1390 if (lli->lli_mds_write_och)
1391 ll_md_real_close(inode, FMODE_WRITE);
1392 if (lli->lli_mds_exec_och)
1393 ll_md_real_close(inode, FMODE_EXEC);
1394 if (lli->lli_mds_read_och)
1395 ll_md_real_close(inode, FMODE_READ);
1397 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1398 OBD_FREE(lli->lli_symlink_name,
1399 strlen(lli->lli_symlink_name) + 1);
1400 lli->lli_symlink_name = NULL;
1403 ll_xattr_cache_destroy(inode);
1405 #ifdef CONFIG_FS_POSIX_ACL
1406 if (lli->lli_posix_acl) {
1407 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1408 posix_acl_release(lli->lli_posix_acl);
1409 lli->lli_posix_acl = NULL;
1412 lli->lli_inode_magic = LLI_INODE_DEAD;
1414 if (S_ISDIR(inode->i_mode))
1415 ll_dir_clear_lsm_md(inode);
1416 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1417 LASSERT(list_empty(&lli->lli_agl_list));
1420 * XXX This has to be done before lsm is freed below, because
1421 * cl_object still uses inode lsm.
1423 cl_inode_fini(inode);
1428 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1430 struct lustre_md md;
1431 struct inode *inode = dentry->d_inode;
1432 struct ll_sb_info *sbi = ll_i2sbi(inode);
1433 struct ptlrpc_request *request = NULL;
1437 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1438 LUSTRE_OPC_ANY, NULL);
1439 if (IS_ERR(op_data))
1440 RETURN(PTR_ERR(op_data));
1442 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1444 ptlrpc_req_finished(request);
1445 if (rc == -ENOENT) {
1447 /* Unlinked special device node? Or just a race?
1448 * Pretend we done everything. */
1449 if (!S_ISREG(inode->i_mode) &&
1450 !S_ISDIR(inode->i_mode)) {
1451 ia_valid = op_data->op_attr.ia_valid;
1452 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1453 rc = simple_setattr(dentry, &op_data->op_attr);
1454 op_data->op_attr.ia_valid = ia_valid;
1456 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1457 CERROR("md_setattr fails: rc = %d\n", rc);
1462 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1463 sbi->ll_md_exp, &md);
1465 ptlrpc_req_finished(request);
1469 ia_valid = op_data->op_attr.ia_valid;
1470 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1471 * cache is not cleared yet. */
1472 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1473 if (S_ISREG(inode->i_mode))
1475 rc = simple_setattr(dentry, &op_data->op_attr);
1476 if (S_ISREG(inode->i_mode))
1477 inode_unlock(inode);
1478 op_data->op_attr.ia_valid = ia_valid;
1480 rc = ll_update_inode(inode, &md);
1481 ptlrpc_req_finished(request);
1486 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1487 * object(s) determine the file size and mtime. Otherwise, the MDS will
1488 * keep these values until such a time that objects are allocated for it.
1489 * We do the MDS operations first, as it is checking permissions for us.
1490 * We don't to the MDS RPC if there is nothing that we want to store there,
1491 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1492 * going to do an RPC anyways.
1494 * If we are doing a truncate, we will send the mtime and ctime updates
1495 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1496 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1499 * In case of HSMimport, we only set attr on MDS.
1501 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1503 struct inode *inode = dentry->d_inode;
1504 struct ll_inode_info *lli = ll_i2info(inode);
1505 struct md_op_data *op_data = NULL;
1509 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1510 "valid %x, hsm_import %d\n",
1511 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1512 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1515 if (attr->ia_valid & ATTR_SIZE) {
1516 /* Check new size against VFS/VM file size limit and rlimit */
1517 rc = inode_newsize_ok(inode, attr->ia_size);
1521 /* The maximum Lustre file size is variable, based on the
1522 * OST maximum object size and number of stripes. This
1523 * needs another check in addition to the VFS check above. */
1524 if (attr->ia_size > ll_file_maxbytes(inode)) {
1525 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1526 PFID(&lli->lli_fid), attr->ia_size,
1527 ll_file_maxbytes(inode));
1531 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1534 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1535 if (attr->ia_valid & TIMES_SET_FLAGS) {
1536 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1537 !cfs_capable(CFS_CAP_FOWNER))
1541 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1542 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1543 (attr->ia_valid & ATTR_CTIME)) {
1544 attr->ia_ctime = CURRENT_TIME;
1545 attr->ia_valid |= ATTR_CTIME_SET;
1547 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1548 (attr->ia_valid & ATTR_ATIME)) {
1549 attr->ia_atime = CURRENT_TIME;
1550 attr->ia_valid |= ATTR_ATIME_SET;
1552 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1553 (attr->ia_valid & ATTR_MTIME)) {
1554 attr->ia_mtime = CURRENT_TIME;
1555 attr->ia_valid |= ATTR_MTIME_SET;
1558 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1559 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1560 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1561 (s64)ktime_get_real_seconds());
1563 if (S_ISREG(inode->i_mode)) {
1564 if (attr->ia_valid & ATTR_SIZE)
1565 inode_dio_write_done(inode);
1566 inode_unlock(inode);
1569 /* We always do an MDS RPC, even if we're only changing the size;
1570 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1572 OBD_ALLOC_PTR(op_data);
1573 if (op_data == NULL)
1574 GOTO(out, rc = -ENOMEM);
1576 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1577 /* If we are changing file size, file content is
1578 * modified, flag it. */
1579 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1580 op_data->op_bias |= MDS_DATA_MODIFIED;
1581 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1584 op_data->op_attr = *attr;
1586 rc = ll_md_setattr(dentry, op_data);
1590 if (!S_ISREG(inode->i_mode) || hsm_import)
1593 if (attr->ia_valid & (ATTR_SIZE |
1594 ATTR_ATIME | ATTR_ATIME_SET |
1595 ATTR_MTIME | ATTR_MTIME_SET |
1596 ATTR_CTIME | ATTR_CTIME_SET)) {
1597 /* For truncate and utimes sending attributes to OSTs, setting
1598 * mtime/atime to the past will be performed under PW [0:EOF]
1599 * extent lock (new_size:EOF for truncate). It may seem
1600 * excessive to send mtime/atime updates to OSTs when not
1601 * setting times to past, but it is necessary due to possible
1602 * time de-synchronization between MDT inode and OST objects */
1603 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1606 /* If the file was restored, it needs to set dirty flag.
1608 * We've already sent MDS_DATA_MODIFIED flag in
1609 * ll_md_setattr() for truncate. However, the MDT refuses to
1610 * set the HS_DIRTY flag on released files, so we have to set
1611 * it again if the file has been restored. Please check how
1612 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1614 * Please notice that if the file is not released, the previous
1615 * MDS_DATA_MODIFIED has taken effect and usually
1616 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1617 * This way we can save an RPC for common open + trunc
1619 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1620 struct hsm_state_set hss = {
1621 .hss_valid = HSS_SETMASK,
1622 .hss_setmask = HS_DIRTY,
1626 rc2 = ll_hsm_state_set(inode, &hss);
1627 /* truncate and write can happen at the same time, so that
1628 * the file can be set modified even though the file is not
1629 * restored from released state, and ll_hsm_state_set() is
1630 * not applicable for the file, and rc2 < 0 is normal in this
1633 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1634 PFID(ll_inode2fid(inode)), rc2);
1639 if (op_data != NULL)
1640 ll_finish_md_op_data(op_data);
1642 if (S_ISREG(inode->i_mode)) {
1644 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1645 inode_dio_wait(inode);
1646 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1647 * flag. ll_update_inode (called from ll_md_setattr), clears
1648 * inode flags, so there is a gap where S_NOSEC is not set.
1649 * This can cause a writer to take the i_mutex unnecessarily,
1650 * but this is safe to do and should be rare. */
1651 inode_has_no_xattr(inode);
1654 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1655 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1660 int ll_setattr(struct dentry *de, struct iattr *attr)
1662 int mode = de->d_inode->i_mode;
1664 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1665 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1666 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1668 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1669 (ATTR_SIZE|ATTR_MODE)) &&
1670 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1671 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1672 !(attr->ia_mode & S_ISGID))))
1673 attr->ia_valid |= ATTR_FORCE;
1675 if ((attr->ia_valid & ATTR_MODE) &&
1677 !(attr->ia_mode & S_ISUID) &&
1678 !(attr->ia_valid & ATTR_KILL_SUID))
1679 attr->ia_valid |= ATTR_KILL_SUID;
1681 if ((attr->ia_valid & ATTR_MODE) &&
1682 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1683 !(attr->ia_mode & S_ISGID) &&
1684 !(attr->ia_valid & ATTR_KILL_SGID))
1685 attr->ia_valid |= ATTR_KILL_SGID;
1687 /* avoid polluted from ATTR_TIMES_SET,
1688 * projid is not expected to be set here */
1689 attr->ia_valid &= ~MDS_ATTR_PROJID;
1691 return ll_setattr_raw(de, attr, false);
1694 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1695 __u64 max_age, __u32 flags)
1697 struct ll_sb_info *sbi = ll_s2sbi(sb);
1698 struct obd_statfs obd_osfs;
1702 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1704 CERROR("md_statfs fails: rc = %d\n", rc);
1708 osfs->os_type = sb->s_magic;
1710 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1711 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1713 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1714 flags |= OBD_STATFS_NODELAY;
1716 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1718 CERROR("obd_statfs fails: rc = %d\n", rc);
1722 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1723 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1726 osfs->os_bsize = obd_osfs.os_bsize;
1727 osfs->os_blocks = obd_osfs.os_blocks;
1728 osfs->os_bfree = obd_osfs.os_bfree;
1729 osfs->os_bavail = obd_osfs.os_bavail;
1731 /* If we don't have as many objects free on the OST as inodes
1732 * on the MDS, we reduce the total number of inodes to
1733 * compensate, so that the "inodes in use" number is correct.
1735 if (obd_osfs.os_ffree < osfs->os_ffree) {
1736 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1738 osfs->os_ffree = obd_osfs.os_ffree;
1743 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1745 struct super_block *sb = de->d_sb;
1746 struct obd_statfs osfs;
1747 __u64 fsid = huge_encode_dev(sb->s_dev);
1750 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1751 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1753 /* Some amount of caching on the client is allowed */
1754 rc = ll_statfs_internal(sb, &osfs,
1755 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1760 statfs_unpack(sfs, &osfs);
1762 /* We need to downshift for all 32-bit kernels, because we can't
1763 * tell if the kernel is being called via sys_statfs64() or not.
1764 * Stop before overflowing f_bsize - in which case it is better
1765 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1766 if (sizeof(long) < 8) {
1767 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1770 osfs.os_blocks >>= 1;
1771 osfs.os_bfree >>= 1;
1772 osfs.os_bavail >>= 1;
1776 sfs->f_blocks = osfs.os_blocks;
1777 sfs->f_bfree = osfs.os_bfree;
1778 sfs->f_bavail = osfs.os_bavail;
1779 sfs->f_fsid.val[0] = (__u32)fsid;
1780 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1784 void ll_inode_size_lock(struct inode *inode)
1786 struct ll_inode_info *lli;
1788 LASSERT(!S_ISDIR(inode->i_mode));
1790 lli = ll_i2info(inode);
1791 mutex_lock(&lli->lli_size_mutex);
1794 void ll_inode_size_unlock(struct inode *inode)
1796 struct ll_inode_info *lli;
1798 lli = ll_i2info(inode);
1799 mutex_unlock(&lli->lli_size_mutex);
1802 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1804 struct ll_inode_info *lli = ll_i2info(inode);
1805 struct mdt_body *body = md->body;
1806 struct ll_sb_info *sbi = ll_i2sbi(inode);
1808 if (body->mbo_valid & OBD_MD_FLEASIZE)
1809 cl_file_inode_init(inode, md);
1811 if (S_ISDIR(inode->i_mode)) {
1814 rc = ll_update_lsm_md(inode, md);
1819 #ifdef CONFIG_FS_POSIX_ACL
1820 if (body->mbo_valid & OBD_MD_FLACL) {
1821 spin_lock(&lli->lli_lock);
1822 if (lli->lli_posix_acl)
1823 posix_acl_release(lli->lli_posix_acl);
1824 lli->lli_posix_acl = md->posix_acl;
1825 spin_unlock(&lli->lli_lock);
1828 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1829 sbi->ll_flags & LL_SBI_32BIT_API);
1830 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1832 if (body->mbo_valid & OBD_MD_FLATIME) {
1833 if (body->mbo_atime > LTIME_S(inode->i_atime))
1834 LTIME_S(inode->i_atime) = body->mbo_atime;
1835 lli->lli_atime = body->mbo_atime;
1838 if (body->mbo_valid & OBD_MD_FLMTIME) {
1839 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1840 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1841 "to %llu\n", inode->i_ino,
1842 LTIME_S(inode->i_mtime), body->mbo_mtime);
1843 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1845 lli->lli_mtime = body->mbo_mtime;
1848 if (body->mbo_valid & OBD_MD_FLCTIME) {
1849 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1850 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1851 lli->lli_ctime = body->mbo_ctime;
1854 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1855 if (body->mbo_valid & OBD_MD_FLFLAGS)
1856 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1857 if (body->mbo_valid & OBD_MD_FLMODE)
1858 inode->i_mode = (inode->i_mode & S_IFMT) |
1859 (body->mbo_mode & ~S_IFMT);
1861 if (body->mbo_valid & OBD_MD_FLTYPE)
1862 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1863 (body->mbo_mode & S_IFMT);
1865 LASSERT(inode->i_mode != 0);
1866 if (S_ISREG(inode->i_mode))
1867 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1868 LL_MAX_BLKSIZE_BITS);
1870 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1872 if (body->mbo_valid & OBD_MD_FLUID)
1873 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1874 if (body->mbo_valid & OBD_MD_FLGID)
1875 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1876 if (body->mbo_valid & OBD_MD_FLPROJID)
1877 lli->lli_projid = body->mbo_projid;
1878 if (body->mbo_valid & OBD_MD_FLNLINK)
1879 set_nlink(inode, body->mbo_nlink);
1880 if (body->mbo_valid & OBD_MD_FLRDEV)
1881 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1883 if (body->mbo_valid & OBD_MD_FLID) {
1884 /* FID shouldn't be changed! */
1885 if (fid_is_sane(&lli->lli_fid)) {
1886 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1887 "Trying to change FID "DFID
1888 " to the "DFID", inode "DFID"(%p)\n",
1889 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1890 PFID(ll_inode2fid(inode)), inode);
1892 lli->lli_fid = body->mbo_fid1;
1896 LASSERT(fid_seq(&lli->lli_fid) != 0);
1898 if (body->mbo_valid & OBD_MD_FLSIZE) {
1899 i_size_write(inode, body->mbo_size);
1901 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1902 PFID(ll_inode2fid(inode)),
1903 (unsigned long long)body->mbo_size);
1905 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1906 inode->i_blocks = body->mbo_blocks;
1909 if (body->mbo_valid & OBD_MD_TSTATE) {
1910 /* Set LLIF_FILE_RESTORING if restore ongoing and
1911 * clear it when done to ensure to start again
1912 * glimpsing updated attrs
1914 if (body->mbo_t_state & MS_RESTORE)
1915 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1917 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1923 int ll_read_inode2(struct inode *inode, void *opaque)
1925 struct lustre_md *md = opaque;
1926 struct ll_inode_info *lli = ll_i2info(inode);
1930 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1931 PFID(&lli->lli_fid), inode);
1933 /* Core attributes from the MDS first. This is a new inode, and
1934 * the VFS doesn't zero times in the core inode so we have to do
1935 * it ourselves. They will be overwritten by either MDS or OST
1936 * attributes - we just need to make sure they aren't newer. */
1937 LTIME_S(inode->i_mtime) = 0;
1938 LTIME_S(inode->i_atime) = 0;
1939 LTIME_S(inode->i_ctime) = 0;
1941 rc = ll_update_inode(inode, md);
1945 /* OIDEBUG(inode); */
1947 #ifdef HAVE_BACKING_DEV_INFO
1948 /* initializing backing dev info. */
1949 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1951 if (S_ISREG(inode->i_mode)) {
1952 struct ll_sb_info *sbi = ll_i2sbi(inode);
1953 inode->i_op = &ll_file_inode_operations;
1954 inode->i_fop = sbi->ll_fop;
1955 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1957 } else if (S_ISDIR(inode->i_mode)) {
1958 inode->i_op = &ll_dir_inode_operations;
1959 inode->i_fop = &ll_dir_operations;
1961 } else if (S_ISLNK(inode->i_mode)) {
1962 inode->i_op = &ll_fast_symlink_inode_operations;
1965 inode->i_op = &ll_special_inode_operations;
1967 init_special_inode(inode, inode->i_mode,
1976 void ll_delete_inode(struct inode *inode)
1978 struct ll_inode_info *lli = ll_i2info(inode);
1981 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1982 /* It is last chance to write out dirty pages,
1983 * otherwise we may lose data while umount */
1984 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
1986 truncate_inode_pages_final(&inode->i_data);
1988 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
1989 "see https://jira.hpdd.intel.com/browse/LU-118\n",
1990 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
1992 #ifdef HAVE_SBOPS_EVICT_INODE
1993 ll_clear_inode(inode);
2000 int ll_iocontrol(struct inode *inode, struct file *file,
2001 unsigned int cmd, unsigned long arg)
2003 struct ll_sb_info *sbi = ll_i2sbi(inode);
2004 struct ptlrpc_request *req = NULL;
2009 case FSFILT_IOC_GETFLAGS: {
2010 struct mdt_body *body;
2011 struct md_op_data *op_data;
2013 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2014 0, 0, LUSTRE_OPC_ANY,
2016 if (IS_ERR(op_data))
2017 RETURN(PTR_ERR(op_data));
2019 op_data->op_valid = OBD_MD_FLFLAGS;
2020 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2021 ll_finish_md_op_data(op_data);
2023 CERROR("%s: failure inode "DFID": rc = %d\n",
2024 sbi->ll_md_exp->exp_obd->obd_name,
2025 PFID(ll_inode2fid(inode)), rc);
2029 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2031 flags = body->mbo_flags;
2033 ptlrpc_req_finished(req);
2035 RETURN(put_user(flags, (int __user *)arg));
2037 case FSFILT_IOC_SETFLAGS: {
2039 struct md_op_data *op_data;
2040 struct cl_object *obj;
2042 if (get_user(flags, (int __user *)arg))
2045 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2046 LUSTRE_OPC_ANY, NULL);
2047 if (IS_ERR(op_data))
2048 RETURN(PTR_ERR(op_data));
2050 op_data->op_attr_flags = flags;
2051 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2052 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2053 ll_finish_md_op_data(op_data);
2054 ptlrpc_req_finished(req);
2058 inode->i_flags = ll_ext_to_inode_flags(flags);
2060 obj = ll_i2info(inode)->lli_clob;
2064 OBD_ALLOC_PTR(attr);
2068 attr->ia_valid = ATTR_ATTR_FLAG;
2069 rc = cl_setattr_ost(obj, attr, flags);
2081 int ll_flush_ctx(struct inode *inode)
2083 struct ll_sb_info *sbi = ll_i2sbi(inode);
2085 CDEBUG(D_SEC, "flush context for user %d\n",
2086 from_kuid(&init_user_ns, current_uid()));
2088 obd_set_info_async(NULL, sbi->ll_md_exp,
2089 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2091 obd_set_info_async(NULL, sbi->ll_dt_exp,
2092 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2097 /* umount -f client means force down, don't save state */
2098 void ll_umount_begin(struct super_block *sb)
2100 struct ll_sb_info *sbi = ll_s2sbi(sb);
2101 struct obd_device *obd;
2102 struct obd_ioctl_data *ioc_data;
2103 struct l_wait_info lwi;
2104 wait_queue_head_t waitq;
2107 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2108 sb->s_count, atomic_read(&sb->s_active));
2110 obd = class_exp2obd(sbi->ll_md_exp);
2112 CERROR("Invalid MDC connection handle %#llx\n",
2113 sbi->ll_md_exp->exp_handle.h_cookie);
2119 obd = class_exp2obd(sbi->ll_dt_exp);
2121 CERROR("Invalid LOV connection handle %#llx\n",
2122 sbi->ll_dt_exp->exp_handle.h_cookie);
2128 OBD_ALLOC_PTR(ioc_data);
2130 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2131 sizeof *ioc_data, ioc_data, NULL);
2133 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2134 sizeof *ioc_data, ioc_data, NULL);
2136 OBD_FREE_PTR(ioc_data);
2139 /* Really, we'd like to wait until there are no requests outstanding,
2140 * and then continue. For now, we just periodically checking for vfs
2141 * to decrement mnt_cnt and hope to finish it within 10sec.
2143 init_waitqueue_head(&waitq);
2144 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2145 cfs_time_seconds(1), NULL, NULL);
2146 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2151 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2153 struct ll_sb_info *sbi = ll_s2sbi(sb);
2154 char *profilenm = get_profile_name(sb);
2158 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2159 read_only = *flags & MS_RDONLY;
2160 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2161 sizeof(KEY_READ_ONLY),
2162 KEY_READ_ONLY, sizeof(read_only),
2165 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2166 profilenm, read_only ?
2167 "read-only" : "read-write", err);
2172 sb->s_flags |= MS_RDONLY;
2174 sb->s_flags &= ~MS_RDONLY;
2176 if (sbi->ll_flags & LL_SBI_VERBOSE)
2177 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2178 read_only ? "read-only" : "read-write");
2184 * Cleanup the open handle that is cached on MDT-side.
2186 * For open case, the client side open handling thread may hit error
2187 * after the MDT grant the open. Under such case, the client should
2188 * send close RPC to the MDT as cleanup; otherwise, the open handle
2189 * on the MDT will be leaked there until the client umount or evicted.
2191 * In further, if someone unlinked the file, because the open handle
2192 * holds the reference on such file/object, then it will block the
2193 * subsequent threads that want to locate such object via FID.
2195 * \param[in] sb super block for this file-system
2196 * \param[in] open_req pointer to the original open request
2198 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2200 struct mdt_body *body;
2201 struct md_op_data *op_data;
2202 struct ptlrpc_request *close_req = NULL;
2203 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2206 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2207 OBD_ALLOC_PTR(op_data);
2208 if (op_data == NULL) {
2209 CWARN("%s: cannot allocate op_data to release open handle for "
2211 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2216 op_data->op_fid1 = body->mbo_fid1;
2217 op_data->op_handle = body->mbo_handle;
2218 op_data->op_mod_time = ktime_get_real_seconds();
2219 md_close(exp, op_data, NULL, &close_req);
2220 ptlrpc_req_finished(close_req);
2221 ll_finish_md_op_data(op_data);
2226 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2227 struct super_block *sb, struct lookup_intent *it)
2229 struct ll_sb_info *sbi = NULL;
2230 struct lustre_md md = { NULL };
2234 LASSERT(*inode || sb);
2235 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2236 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2237 sbi->ll_md_exp, &md);
2242 rc = ll_update_inode(*inode, &md);
2246 LASSERT(sb != NULL);
2249 * At this point server returns to client's same fid as client
2250 * generated for creating. So using ->fid1 is okay here.
2252 if (!fid_is_sane(&md.body->mbo_fid1)) {
2253 CERROR("%s: Fid is insane "DFID"\n",
2254 ll_get_fsname(sb, NULL, 0),
2255 PFID(&md.body->mbo_fid1));
2256 GOTO(out, rc = -EINVAL);
2259 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2260 sbi->ll_flags & LL_SBI_32BIT_API),
2262 if (IS_ERR(*inode)) {
2263 #ifdef CONFIG_FS_POSIX_ACL
2265 posix_acl_release(md.posix_acl);
2266 md.posix_acl = NULL;
2269 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2271 CERROR("new_inode -fatal: rc %d\n", rc);
2276 /* Handling piggyback layout lock.
2277 * Layout lock can be piggybacked by getattr and open request.
2278 * The lsm can be applied to inode only if it comes with a layout lock
2279 * otherwise correct layout may be overwritten, for example:
2280 * 1. proc1: mdt returns a lsm but not granting layout
2281 * 2. layout was changed by another client
2282 * 3. proc2: refresh layout and layout lock granted
2283 * 4. proc1: to apply a stale layout */
2284 if (it != NULL && it->it_lock_mode != 0) {
2285 struct lustre_handle lockh;
2286 struct ldlm_lock *lock;
2288 lockh.cookie = it->it_lock_handle;
2289 lock = ldlm_handle2lock(&lockh);
2290 LASSERT(lock != NULL);
2291 if (ldlm_has_layout(lock)) {
2292 struct cl_object_conf conf;
2294 memset(&conf, 0, sizeof(conf));
2295 conf.coc_opc = OBJECT_CONF_SET;
2296 conf.coc_inode = *inode;
2297 conf.coc_lock = lock;
2298 conf.u.coc_layout = md.layout;
2299 (void)ll_layout_conf(*inode, &conf);
2301 LDLM_LOCK_PUT(lock);
2307 md_free_lustre_md(sbi->ll_md_exp, &md);
2310 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2311 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2316 int ll_obd_statfs(struct inode *inode, void __user *arg)
2318 struct ll_sb_info *sbi = NULL;
2319 struct obd_export *exp;
2321 struct obd_ioctl_data *data = NULL;
2325 if (!inode || !(sbi = ll_i2sbi(inode)))
2326 GOTO(out_statfs, rc = -EINVAL);
2328 rc = obd_ioctl_getdata(&buf, &len, arg);
2330 GOTO(out_statfs, rc);
2333 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2334 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2335 GOTO(out_statfs, rc = -EINVAL);
2337 if (data->ioc_inllen1 != sizeof(__u32) ||
2338 data->ioc_inllen2 != sizeof(__u32) ||
2339 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2340 data->ioc_plen2 != sizeof(struct obd_uuid))
2341 GOTO(out_statfs, rc = -EINVAL);
2343 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2344 if (type & LL_STATFS_LMV)
2345 exp = sbi->ll_md_exp;
2346 else if (type & LL_STATFS_LOV)
2347 exp = sbi->ll_dt_exp;
2349 GOTO(out_statfs, rc = -ENODEV);
2351 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2353 GOTO(out_statfs, rc);
2355 OBD_FREE_LARGE(buf, len);
2359 int ll_process_config(struct lustre_cfg *lcfg)
2361 struct super_block *sb;
2366 /* The instance name contains the sb: lustre-client-aacfe000 */
2367 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2368 if (!ptr || !*(++ptr))
2370 if (sscanf(ptr, "%lx", &x) != 1)
2372 sb = (struct super_block *)x;
2373 /* This better be a real Lustre superblock! */
2374 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2376 /* Note we have not called client_common_fill_super yet, so
2377 proc fns must be able to handle that! */
2378 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2385 /* this function prepares md_op_data hint for passing it down to MD stack. */
2386 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2387 struct inode *i1, struct inode *i2,
2388 const char *name, size_t namelen,
2389 __u32 mode, __u32 opc, void *data)
2391 LASSERT(i1 != NULL);
2394 /* Do not reuse namelen for something else. */
2396 return ERR_PTR(-EINVAL);
2398 if (namelen > ll_i2sbi(i1)->ll_namelen)
2399 return ERR_PTR(-ENAMETOOLONG);
2401 if (!lu_name_is_valid_2(name, namelen))
2402 return ERR_PTR(-EINVAL);
2405 if (op_data == NULL)
2406 OBD_ALLOC_PTR(op_data);
2408 if (op_data == NULL)
2409 return ERR_PTR(-ENOMEM);
2411 ll_i2gids(op_data->op_suppgids, i1, i2);
2412 op_data->op_fid1 = *ll_inode2fid(i1);
2413 op_data->op_default_stripe_offset = -1;
2414 if (S_ISDIR(i1->i_mode)) {
2415 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2416 if (opc == LUSTRE_OPC_MKDIR)
2417 op_data->op_default_stripe_offset =
2418 ll_i2info(i1)->lli_def_stripe_offset;
2422 op_data->op_fid2 = *ll_inode2fid(i2);
2423 if (S_ISDIR(i2->i_mode))
2424 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2426 fid_zero(&op_data->op_fid2);
2429 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2430 op_data->op_cli_flags |= CLI_HASH64;
2432 if (ll_need_32bit_api(ll_i2sbi(i1)))
2433 op_data->op_cli_flags |= CLI_API32;
2435 op_data->op_name = name;
2436 op_data->op_namelen = namelen;
2437 op_data->op_mode = mode;
2438 op_data->op_mod_time = cfs_time_current_sec();
2439 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2440 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2441 op_data->op_cap = cfs_curproc_cap_pack();
2442 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2443 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2444 op_data->op_bias |= MDS_CREATE_VOLATILE;
2446 op_data->op_mds = 0;
2448 op_data->op_data = data;
2453 void ll_finish_md_op_data(struct md_op_data *op_data)
2455 security_release_secctx(op_data->op_file_secctx,
2456 op_data->op_file_secctx_size);
2457 OBD_FREE_PTR(op_data);
2460 #ifdef HAVE_SUPEROPS_USE_DENTRY
2461 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2463 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2466 struct ll_sb_info *sbi;
2468 #ifdef HAVE_SUPEROPS_USE_DENTRY
2469 LASSERT((seq != NULL) && (dentry != NULL));
2470 sbi = ll_s2sbi(dentry->d_sb);
2472 LASSERT((seq != NULL) && (vfs != NULL));
2473 sbi = ll_s2sbi(vfs->mnt_sb);
2476 if (sbi->ll_flags & LL_SBI_NOLCK)
2477 seq_puts(seq, ",nolock");
2479 if (sbi->ll_flags & LL_SBI_FLOCK)
2480 seq_puts(seq, ",flock");
2482 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2483 seq_puts(seq, ",localflock");
2485 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2486 seq_puts(seq, ",user_xattr");
2488 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2489 seq_puts(seq, ",lazystatfs");
2491 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2492 seq_puts(seq, ",user_fid2path");
2494 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2495 seq_puts(seq, ",always_ping");
2501 * Get obd name by cmd, and copy out to user space
2503 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2505 struct ll_sb_info *sbi = ll_i2sbi(inode);
2506 struct obd_device *obd;
2509 if (cmd == OBD_IOC_GETDTNAME)
2510 obd = class_exp2obd(sbi->ll_dt_exp);
2511 else if (cmd == OBD_IOC_GETMDNAME)
2512 obd = class_exp2obd(sbi->ll_md_exp);
2519 if (copy_to_user((void __user *)arg, obd->obd_name,
2520 strlen(obd->obd_name) + 1))
2527 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2528 * fsname will be returned in this buffer; otherwise, a static buffer will be
2529 * used to store the fsname and returned to caller.
2531 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2533 static char fsname_static[MTI_NAME_MAXLEN];
2534 struct lustre_sb_info *lsi = s2lsi(sb);
2539 /* this means the caller wants to use static buffer
2540 * and it doesn't care about race. Usually this is
2541 * in error reporting path */
2542 buf = fsname_static;
2543 buflen = sizeof(fsname_static);
2546 len = strlen(lsi->lsi_lmd->lmd_profile);
2547 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2548 if (ptr && (strcmp(ptr, "-client") == 0))
2551 if (unlikely(len >= buflen))
2553 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2559 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2566 p.mnt = current->fs->root.mnt;
2568 path = d_path(&p, buf, bufsize);
2573 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2575 char *buf, *path = NULL;
2576 struct dentry *dentry = NULL;
2577 struct inode *inode = page->mapping->host;
2579 /* this can be called inside spin lock so use GFP_ATOMIC. */
2580 buf = (char *)__get_free_page(GFP_ATOMIC);
2582 dentry = d_find_alias(page->mapping->host);
2584 path = ll_d_path(dentry, buf, PAGE_SIZE);
2588 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2589 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2590 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2591 PFID(ll_inode2fid(inode)),
2592 (path && !IS_ERR(path)) ? path : "", ioret);
2598 free_page((unsigned long)buf);
2601 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2602 struct lov_user_md **kbuf)
2604 struct lov_user_md lum;
2608 if (copy_from_user(&lum, md, sizeof(lum)))
2611 lum_size = ll_lov_user_md_size(&lum);
2615 OBD_ALLOC(*kbuf, lum_size);
2619 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2620 OBD_FREE(*kbuf, lum_size);
2628 * Compute llite root squash state after a change of root squash
2629 * configuration setting or add/remove of a lnet nid
2631 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2633 struct root_squash_info *squash = &sbi->ll_squash;
2636 struct lnet_process_id id;
2638 /* Update norootsquash flag */
2639 down_write(&squash->rsi_sem);
2640 if (list_empty(&squash->rsi_nosquash_nids))
2641 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2643 /* Do not apply root squash as soon as one of our NIDs is
2644 * in the nosquash_nids list */
2647 while (LNetGetId(i++, &id) != -ENOENT) {
2648 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2650 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2656 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2658 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2660 up_write(&squash->rsi_sem);
2664 * Parse linkea content to extract information about a given hardlink
2666 * \param[in] ldata - Initialized linkea data
2667 * \param[in] linkno - Link identifier
2668 * \param[out] parent_fid - The entry's parent FID
2669 * \param[out] ln - Entry name destination buffer
2671 * \retval 0 on success
2672 * \retval Appropriate negative error code on failure
2674 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2675 struct lu_fid *parent_fid, struct lu_name *ln)
2681 rc = linkea_init_with_rec(ldata);
2685 if (linkno >= ldata->ld_leh->leh_reccount)
2686 /* beyond last link */
2689 linkea_first_entry(ldata);
2690 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2691 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2696 linkea_next_entry(ldata);
2706 * Get parent FID and name of an identified link. Operation is performed for
2707 * a given link number, letting the caller iterate over linkno to list one or
2708 * all links of an entry.
2710 * \param[in] file - File descriptor against which to perform the operation
2711 * \param[in,out] arg - User-filled structure containing the linkno to operate
2712 * on and the available size. It is eventually filled with
2713 * the requested information or left untouched on error
2715 * \retval - 0 on success
2716 * \retval - Appropriate negative error code on failure
2718 int ll_getparent(struct file *file, struct getparent __user *arg)
2720 struct dentry *dentry = file_dentry(file);
2721 struct inode *inode = file_inode(file);
2722 struct linkea_data *ldata;
2723 struct lu_buf buf = LU_BUF_NULL;
2725 struct lu_fid parent_fid;
2732 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2733 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2736 if (get_user(name_size, &arg->gp_name_size))
2739 if (get_user(linkno, &arg->gp_linkno))
2742 if (name_size > PATH_MAX)
2745 OBD_ALLOC(ldata, sizeof(*ldata));
2749 rc = linkea_data_new(ldata, &buf);
2751 GOTO(ldata_free, rc);
2753 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2757 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2761 if (ln.ln_namelen >= name_size)
2762 GOTO(lb_free, rc = -EOVERFLOW);
2764 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2765 GOTO(lb_free, rc = -EFAULT);
2767 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2768 GOTO(lb_free, rc = -EFAULT);
2770 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2771 GOTO(lb_free, rc = -EFAULT);
2776 OBD_FREE(ldata, sizeof(*ldata));