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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/statfs.h>
45 #include <linux/time.h>
46 #include <linux/types.h>
47 #include <linux/version.h>
49 #include <linux/user_namespace.h>
50 #ifdef HAVE_UIDGID_HEADER
51 # include <linux/uidgid.h>
54 #include <lustre_ioctl.h>
55 #include <lustre_ha.h>
56 #include <lustre_dlm.h>
57 #include <lprocfs_status.h>
58 #include <lustre_disk.h>
59 #include <lustre_param.h>
60 #include <lustre_log.h>
61 #include <cl_object.h>
62 #include <obd_cksum.h>
63 #include "llite_internal.h"
65 struct kmem_cache *ll_file_data_slab;
68 #define log2(n) ffz(~(n))
71 static struct ll_sb_info *ll_init_sbi(void)
73 struct ll_sb_info *sbi = NULL;
75 unsigned long lru_page_max;
85 spin_lock_init(&sbi->ll_lock);
86 mutex_init(&sbi->ll_lco.lco_lock);
87 spin_lock_init(&sbi->ll_pp_extent_lock);
88 spin_lock_init(&sbi->ll_process_lock);
89 sbi->ll_rw_stats_on = 0;
92 pages = si.totalram - si.totalhigh;
93 lru_page_max = pages / 2;
95 /* initialize ll_cache data */
96 sbi->ll_cache = cl_cache_init(lru_page_max);
97 if (sbi->ll_cache == NULL) {
98 OBD_FREE(sbi, sizeof(*sbi));
102 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
103 SBI_DEFAULT_READAHEAD_MAX);
104 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
105 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
106 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
108 ll_generate_random_uuid(uuid);
109 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
110 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
112 sbi->ll_flags |= LL_SBI_VERBOSE;
113 #ifdef ENABLE_CHECKSUM
114 sbi->ll_flags |= LL_SBI_CHECKSUM;
117 #ifdef HAVE_LRU_RESIZE_SUPPORT
118 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
120 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
122 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
123 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
129 /* metadata statahead is enabled by default */
130 sbi->ll_sa_max = LL_SA_RPC_DEF;
131 atomic_set(&sbi->ll_sa_total, 0);
132 atomic_set(&sbi->ll_sa_wrong, 0);
133 atomic_set(&sbi->ll_sa_running, 0);
134 atomic_set(&sbi->ll_agl_total, 0);
135 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
136 sbi->ll_flags |= LL_SBI_FAST_READ;
139 sbi->ll_squash.rsi_uid = 0;
140 sbi->ll_squash.rsi_gid = 0;
141 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
142 init_rwsem(&sbi->ll_squash.rsi_sem);
147 static void ll_free_sbi(struct super_block *sb)
149 struct ll_sb_info *sbi = ll_s2sbi(sb);
153 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
154 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
155 if (sbi->ll_cache != NULL) {
156 cl_cache_decref(sbi->ll_cache);
157 sbi->ll_cache = NULL;
159 OBD_FREE(sbi, sizeof(*sbi));
164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
165 struct vfsmount *mnt)
167 struct inode *root = NULL;
168 struct ll_sb_info *sbi = ll_s2sbi(sb);
169 struct obd_device *obd;
170 struct obd_statfs *osfs = NULL;
171 struct ptlrpc_request *request = NULL;
172 struct obd_connect_data *data = NULL;
173 struct obd_uuid *uuid;
174 struct md_op_data *op_data;
175 struct lustre_md lmd;
177 int size, err, checksum;
180 obd = class_name2obd(md);
182 CERROR("MD %s: not setup or attached\n", md);
196 /* indicate the features supported by this client */
197 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
198 OBD_CONNECT_ATTRFID |
199 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
200 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
201 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
202 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
203 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
204 OBD_CONNECT_64BITHASH |
205 OBD_CONNECT_EINPROGRESS |
206 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
207 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
208 OBD_CONNECT_MAX_EASIZE |
209 OBD_CONNECT_FLOCK_DEAD |
210 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
211 OBD_CONNECT_OPEN_BY_FID |
212 OBD_CONNECT_DIR_STRIPE |
213 OBD_CONNECT_BULK_MBITS |
214 OBD_CONNECT_SUBTREE |
217 data->ocd_connect_flags2 = 0;
219 #ifdef HAVE_LRU_RESIZE_SUPPORT
220 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
221 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
223 #ifdef CONFIG_FS_POSIX_ACL
224 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
227 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
228 /* flag mdc connection as lightweight, only used for test
229 * purpose, use with care */
230 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
232 data->ocd_ibits_known = MDS_INODELOCK_FULL;
233 data->ocd_version = LUSTRE_VERSION_CODE;
235 if (sb->s_flags & MS_RDONLY)
236 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
237 if (sbi->ll_flags & LL_SBI_USER_XATTR)
238 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
240 if (sbi->ll_flags & LL_SBI_FLOCK)
241 sbi->ll_fop = &ll_file_operations_flock;
242 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
243 sbi->ll_fop = &ll_file_operations;
245 sbi->ll_fop = &ll_file_operations_noflock;
248 data->ocd_connect_flags |= OBD_CONNECT_REAL;
250 /* always ping even if server suppress_pings */
251 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
252 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
254 data->ocd_brw_size = MD_MAX_BRW_SIZE;
256 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
258 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
259 "recovery, of which this client is not a "
260 "part. Please wait for recovery to complete,"
261 " abort, or time out.\n", md);
264 CERROR("cannot connect to %s: rc = %d\n", md, err);
268 sbi->ll_md_exp->exp_connect_data = *data;
270 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
271 LUSTRE_SEQ_METADATA);
273 CERROR("%s: Can't init metadata layer FID infrastructure, "
274 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
278 /* For mount, we only need fs info from MDT0, and also in DNE, it
279 * can make sure the client can be mounted as long as MDT0 is
281 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
282 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
283 OBD_STATFS_FOR_MDT0);
285 GOTO(out_md_fid, err);
287 /* This needs to be after statfs to ensure connect has finished.
288 * Note that "data" does NOT contain the valid connect reply.
289 * If connecting to a 1.8 server there will be no LMV device, so
290 * we can access the MDC export directly and exp_connect_flags will
291 * be non-zero, but if accessing an upgraded 2.1 server it will
292 * have the correct flags filled in.
293 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
294 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
295 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
296 valid != CLIENT_CONNECT_MDT_REQD) {
299 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
300 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
301 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
302 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
303 "feature(s) needed for correct operation "
304 "of this client (%s). Please upgrade "
305 "server or downgrade client.\n",
306 sbi->ll_md_exp->exp_obd->obd_name, buf);
307 OBD_FREE(buf, PAGE_CACHE_SIZE);
308 GOTO(out_md_fid, err = -EPROTO);
311 size = sizeof(*data);
312 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
313 KEY_CONN_DATA, &size, data);
315 CERROR("%s: Get connect data failed: rc = %d\n",
316 sbi->ll_md_exp->exp_obd->obd_name, err);
317 GOTO(out_md_fid, err);
320 LASSERT(osfs->os_bsize);
321 sb->s_blocksize = osfs->os_bsize;
322 sb->s_blocksize_bits = log2(osfs->os_bsize);
323 sb->s_magic = LL_SUPER_MAGIC;
324 sb->s_maxbytes = MAX_LFS_FILESIZE;
325 sbi->ll_namelen = osfs->os_namelen;
327 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
328 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
329 LCONSOLE_INFO("Disabling user_xattr feature because "
330 "it is not supported on the server\n");
331 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
334 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
336 sb->s_flags |= MS_POSIXACL;
338 sbi->ll_flags |= LL_SBI_ACL;
340 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
342 sb->s_flags &= ~MS_POSIXACL;
344 sbi->ll_flags &= ~LL_SBI_ACL;
347 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
348 sbi->ll_flags |= LL_SBI_64BIT_HASH;
350 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
351 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
353 sbi->ll_md_brw_pages = 1;
355 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
356 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
358 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
359 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
360 LCONSOLE_INFO("%s: disabling xattr cache due to "
361 "unknown maximum xattr size.\n", dt);
363 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
364 sbi->ll_xattr_cache_enabled = 1;
368 obd = class_name2obd(dt);
370 CERROR("DT %s: not setup or attached\n", dt);
371 GOTO(out_md_fid, err = -ENODEV);
374 /* pass client page size via ocd_grant_blkbits, the server should report
375 * back its backend blocksize for grant calculation purpose */
376 data->ocd_grant_blkbits = PAGE_SHIFT;
378 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
379 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
380 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
381 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
382 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
383 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
384 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
385 OBD_CONNECT_EINPROGRESS |
386 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
387 OBD_CONNECT_LAYOUTLOCK |
388 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
389 OBD_CONNECT_BULK_MBITS;
391 data->ocd_connect_flags2 = 0;
393 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
394 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
396 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
397 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
398 * disabled by default, because it can still be enabled on the
399 * fly via /proc. As a consequence, we still need to come to an
400 * agreement on the supported algorithms at connect time */
401 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
403 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
404 data->ocd_cksum_types = OBD_CKSUM_ADLER;
406 data->ocd_cksum_types = cksum_types_supported_client();
409 #ifdef HAVE_LRU_RESIZE_SUPPORT
410 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
412 /* always ping even if server suppress_pings */
413 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
414 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
416 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
417 "ocd_grant: %d\n", data->ocd_connect_flags,
418 data->ocd_version, data->ocd_grant);
420 obd->obd_upcall.onu_owner = &sbi->ll_lco;
421 obd->obd_upcall.onu_upcall = cl_ocd_update;
423 data->ocd_brw_size = DT_MAX_BRW_SIZE;
425 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
428 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
429 "recovery, of which this client is not a "
430 "part. Please wait for recovery to "
431 "complete, abort, or time out.\n", dt);
434 CERROR("%s: Cannot connect to %s: rc = %d\n",
435 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
439 sbi->ll_dt_exp->exp_connect_data = *data;
441 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
442 LUSTRE_SEQ_METADATA);
444 CERROR("%s: Can't init data layer FID infrastructure, "
445 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
449 mutex_lock(&sbi->ll_lco.lco_lock);
450 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
451 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
452 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
453 mutex_unlock(&sbi->ll_lco.lco_lock);
455 fid_zero(&sbi->ll_root_fid);
456 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
459 CERROR("cannot mds_connect: rc = %d\n", err);
460 GOTO(out_lock_cn_cb, err);
462 if (!fid_is_sane(&sbi->ll_root_fid)) {
463 CERROR("%s: Invalid root fid "DFID" during mount\n",
464 sbi->ll_md_exp->exp_obd->obd_name,
465 PFID(&sbi->ll_root_fid));
466 GOTO(out_lock_cn_cb, err = -EINVAL);
468 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
470 sb->s_op = &lustre_super_operations;
471 #if THREAD_SIZE >= 8192 /*b=17630*/
472 sb->s_export_op = &lustre_export_operations;
476 * XXX: move this to after cbd setup? */
477 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
478 if (sbi->ll_flags & LL_SBI_ACL)
479 valid |= OBD_MD_FLACL;
481 OBD_ALLOC_PTR(op_data);
483 GOTO(out_lock_cn_cb, err = -ENOMEM);
485 op_data->op_fid1 = sbi->ll_root_fid;
486 op_data->op_mode = 0;
487 op_data->op_valid = valid;
489 err = md_getattr(sbi->ll_md_exp, op_data, &request);
491 OBD_FREE_PTR(op_data);
493 CERROR("%s: md_getattr failed for root: rc = %d\n",
494 sbi->ll_md_exp->exp_obd->obd_name, err);
495 GOTO(out_lock_cn_cb, err);
498 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
499 sbi->ll_md_exp, &lmd);
501 CERROR("failed to understand root inode md: rc = %d\n", err);
502 ptlrpc_req_finished(request);
503 GOTO(out_lock_cn_cb, err);
506 LASSERT(fid_is_sane(&sbi->ll_root_fid));
507 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
508 sbi->ll_flags & LL_SBI_32BIT_API),
510 md_free_lustre_md(sbi->ll_md_exp, &lmd);
511 ptlrpc_req_finished(request);
514 #ifdef CONFIG_FS_POSIX_ACL
516 posix_acl_release(lmd.posix_acl);
517 lmd.posix_acl = NULL;
520 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
522 CERROR("lustre_lite: bad iget4 for root\n");
526 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
527 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
528 KEY_CHECKSUM, sizeof(checksum), &checksum,
532 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
533 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
534 sbi->ll_cache, NULL);
536 sb->s_root = d_make_root(root);
537 if (sb->s_root == NULL) {
538 CERROR("%s: can't make root dentry\n",
539 ll_get_fsname(sb, NULL, 0));
540 GOTO(out_root, err = -ENOMEM);
542 #ifdef HAVE_DCACHE_LOCK
543 sb->s_root->d_op = &ll_d_ops;
546 sbi->ll_sdev_orig = sb->s_dev;
548 /* We set sb->s_dev equal on all lustre clients in order to support
549 * NFS export clustering. NFSD requires that the FSID be the same
551 /* s_dev is also used in lt_compare() to compare two fs, but that is
552 * only a node-local comparison. */
553 uuid = obd_get_uuid(sbi->ll_md_exp);
555 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
561 if (proc_lustre_fs_root != NULL) {
562 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
565 CERROR("%s: could not register mount in lprocfs: "
566 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
576 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
578 obd_disconnect(sbi->ll_dt_exp);
579 sbi->ll_dt_exp = NULL;
581 obd_fid_fini(sbi->ll_md_exp->exp_obd);
583 obd_disconnect(sbi->ll_md_exp);
584 sbi->ll_md_exp = NULL;
593 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
597 size = sizeof(*lmmsize);
598 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
599 KEY_MAX_EASIZE, &size, lmmsize);
601 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
602 sbi->ll_dt_exp->exp_obd->obd_name, rc);
607 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
608 KEY_MAX_EASIZE, &size, lmmsize);
610 CERROR("Get max mdsize error rc %d\n", rc);
616 * Get the value of the default_easize parameter.
618 * \see client_obd::cl_default_mds_easize
620 * \param[in] sbi superblock info for this filesystem
621 * \param[out] lmmsize pointer to storage location for value
623 * \retval 0 on success
624 * \retval negative negated errno on failure
626 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
631 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
632 KEY_DEFAULT_EASIZE, &size, lmmsize);
634 CERROR("Get default mdsize error rc %d\n", rc);
640 * Set the default_easize parameter to the given value.
642 * \see client_obd::cl_default_mds_easize
644 * \param[in] sbi superblock info for this filesystem
645 * \param[in] lmmsize the size to set
647 * \retval 0 on success
648 * \retval negative negated errno on failure
650 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
654 if (lmmsize < sizeof(struct lov_mds_md) ||
655 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
658 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
659 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
660 sizeof(int), &lmmsize, NULL);
665 static void ll_dump_inode(struct inode *inode)
667 struct ll_d_hlist_node *tmp;
668 int dentry_count = 0;
670 LASSERT(inode != NULL);
672 ll_d_hlist_for_each(tmp, &inode->i_dentry)
675 CERROR("%s: inode %p dump: dev=%s fid="DFID
676 " mode=%o count=%u, %d dentries\n",
677 ll_get_fsname(inode->i_sb, NULL, 0), inode,
678 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
679 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
682 void lustre_dump_dentry(struct dentry *dentry, int recur)
684 struct list_head *tmp;
687 LASSERT(dentry != NULL);
689 list_for_each(tmp, &dentry->d_subdirs)
692 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
693 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
694 dentry->d_name.len, dentry->d_name.name,
695 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
696 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
697 dentry->d_flags, dentry->d_fsdata, subdirs);
698 if (dentry->d_inode != NULL)
699 ll_dump_inode(dentry->d_inode);
704 list_for_each(tmp, &dentry->d_subdirs) {
705 struct dentry *d = list_entry(tmp, struct dentry, d_child);
706 lustre_dump_dentry(d, recur - 1);
710 static void client_common_put_super(struct super_block *sb)
712 struct ll_sb_info *sbi = ll_s2sbi(sb);
717 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
718 obd_disconnect(sbi->ll_dt_exp);
719 sbi->ll_dt_exp = NULL;
721 lprocfs_unregister_mountpoint(sbi);
723 obd_fid_fini(sbi->ll_md_exp->exp_obd);
724 obd_disconnect(sbi->ll_md_exp);
725 sbi->ll_md_exp = NULL;
730 void ll_kill_super(struct super_block *sb)
732 struct ll_sb_info *sbi;
736 if (!(sb->s_flags & MS_ACTIVE))
740 /* we need restore s_dev from changed for clustred NFS before put_super
741 * because new kernels have cached s_dev and change sb->s_dev in
742 * put_super not affected real removing devices */
744 sb->s_dev = sbi->ll_sdev_orig;
745 sbi->ll_umounting = 1;
747 /* wait running statahead threads to quit */
748 while (atomic_read(&sbi->ll_sa_running) > 0) {
749 set_current_state(TASK_UNINTERRUPTIBLE);
750 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
757 static inline int ll_set_opt(const char *opt, char *data, int fl)
759 if (strncmp(opt, data, strlen(opt)) != 0)
765 /* non-client-specific mount options are parsed in lmd_parse */
766 static int ll_options(char *options, int *flags)
769 char *s1 = options, *s2;
775 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
778 CDEBUG(D_SUPER, "next opt=%s\n", s1);
779 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
784 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
789 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
794 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
799 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
804 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
809 tmp = ll_set_opt("context", s1, 1);
812 tmp = ll_set_opt("fscontext", s1, 1);
815 tmp = ll_set_opt("defcontext", s1, 1);
818 tmp = ll_set_opt("rootcontext", s1, 1);
821 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
826 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
832 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
837 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
842 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
847 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
852 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
857 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
862 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
867 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
872 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
877 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
882 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
888 s2 = strchr(s1, ',');
896 void ll_lli_init(struct ll_inode_info *lli)
898 lli->lli_inode_magic = LLI_INODE_MAGIC;
900 spin_lock_init(&lli->lli_lock);
901 lli->lli_posix_acl = NULL;
902 /* Do not set lli_fid, it has been initialized already. */
903 fid_zero(&lli->lli_pfid);
904 atomic_set(&lli->lli_open_count, 0);
905 lli->lli_mds_read_och = NULL;
906 lli->lli_mds_write_och = NULL;
907 lli->lli_mds_exec_och = NULL;
908 lli->lli_open_fd_read_count = 0;
909 lli->lli_open_fd_write_count = 0;
910 lli->lli_open_fd_exec_count = 0;
911 mutex_init(&lli->lli_och_mutex);
912 spin_lock_init(&lli->lli_agl_lock);
913 spin_lock_init(&lli->lli_layout_lock);
914 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
915 lli->lli_clob = NULL;
917 init_rwsem(&lli->lli_xattrs_list_rwsem);
918 mutex_init(&lli->lli_xattrs_enq_lock);
920 LASSERT(lli->lli_vfs_inode.i_mode != 0);
921 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
922 mutex_init(&lli->lli_readdir_mutex);
923 lli->lli_opendir_key = NULL;
925 spin_lock_init(&lli->lli_sa_lock);
926 lli->lli_opendir_pid = 0;
927 lli->lli_sa_enabled = 0;
928 lli->lli_def_stripe_offset = -1;
930 mutex_init(&lli->lli_size_mutex);
931 lli->lli_symlink_name = NULL;
932 init_rwsem(&lli->lli_trunc_sem);
933 range_lock_tree_init(&lli->lli_write_tree);
934 init_rwsem(&lli->lli_glimpse_sem);
935 lli->lli_glimpse_time = 0;
936 INIT_LIST_HEAD(&lli->lli_agl_list);
937 lli->lli_agl_index = 0;
938 lli->lli_async_rc = 0;
940 mutex_init(&lli->lli_layout_mutex);
943 static inline int ll_bdi_register(struct backing_dev_info *bdi)
945 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
947 bdi->name = "lustre";
948 return bdi_register(bdi, NULL, "lustre-%d",
949 atomic_inc_return(&ll_bdi_num));
952 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
954 struct lustre_profile *lprof = NULL;
955 struct lustre_sb_info *lsi = s2lsi(sb);
956 struct ll_sb_info *sbi;
957 char *dt = NULL, *md = NULL;
958 char *profilenm = get_profile_name(sb);
959 struct config_llog_instance *cfg;
960 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
961 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
965 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
971 try_module_get(THIS_MODULE);
973 /* client additional sb info */
974 lsi->lsi_llsbi = sbi = ll_init_sbi();
976 module_put(THIS_MODULE);
981 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
985 err = bdi_init(&lsi->lsi_bdi);
988 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
989 #ifdef HAVE_BDI_CAP_MAP_COPY
990 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
992 lsi->lsi_bdi.capabilities = 0;
994 err = ll_bdi_register(&lsi->lsi_bdi);
998 sb->s_bdi = &lsi->lsi_bdi;
999 #ifndef HAVE_DCACHE_LOCK
1000 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1001 sb->s_d_op = &ll_d_ops;
1004 /* Generate a string unique to this super, in case some joker tries
1005 to mount the same fs at two mount points.
1006 Use the address of the super itself.*/
1007 cfg->cfg_instance = sb;
1008 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1009 cfg->cfg_callback = class_config_llog_handler;
1010 /* set up client obds */
1011 err = lustre_process_log(sb, profilenm, cfg);
1013 GOTO(out_free, err);
1015 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1016 lprof = class_get_profile(profilenm);
1017 if (lprof == NULL) {
1018 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1019 " read from the MGS. Does that filesystem "
1020 "exist?\n", profilenm);
1021 GOTO(out_free, err = -EINVAL);
1023 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1024 lprof->lp_md, lprof->lp_dt);
1026 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1028 GOTO(out_free, err = -ENOMEM);
1029 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1031 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1033 GOTO(out_free, err = -ENOMEM);
1034 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1036 /* connections, registrations, sb setup */
1037 err = client_common_fill_super(sb, md, dt, mnt);
1039 GOTO(out_free, err);
1041 sbi->ll_client_common_fill_super_succeeded = 1;
1045 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1047 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1049 class_put_profile(lprof);
1052 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1053 LCONSOLE_WARN("Mounted %s\n", profilenm);
1057 } /* ll_fill_super */
1059 void ll_put_super(struct super_block *sb)
1061 struct config_llog_instance cfg, params_cfg;
1062 struct obd_device *obd;
1063 struct lustre_sb_info *lsi = s2lsi(sb);
1064 struct ll_sb_info *sbi = ll_s2sbi(sb);
1065 char *profilenm = get_profile_name(sb);
1067 int next, force = 1, rc = 0;
1070 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1072 cfg.cfg_instance = sb;
1073 lustre_end_log(sb, profilenm, &cfg);
1075 params_cfg.cfg_instance = sb;
1076 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1078 if (sbi->ll_md_exp) {
1079 obd = class_exp2obd(sbi->ll_md_exp);
1081 force = obd->obd_force;
1084 /* Wait for unstable pages to be committed to stable storage */
1086 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1087 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1088 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1092 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1093 if (force == 0 && rc != -EINTR)
1094 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1097 /* We need to set force before the lov_disconnect in
1098 lustre_common_put_super, since l_d cleans up osc's as well. */
1101 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1103 obd->obd_force = force;
1107 if (sbi->ll_client_common_fill_super_succeeded) {
1108 /* Only if client_common_fill_super succeeded */
1109 client_common_put_super(sb);
1113 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1114 class_manual_cleanup(obd);
1117 if (sbi->ll_flags & LL_SBI_VERBOSE)
1118 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1121 class_del_profile(profilenm);
1123 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1124 bdi_destroy(&lsi->lsi_bdi);
1125 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1129 lsi->lsi_llsbi = NULL;
1131 lustre_common_put_super(sb);
1133 cl_env_cache_purge(~0);
1135 module_put(THIS_MODULE);
1138 } /* client_put_super */
1140 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1142 struct inode *inode = NULL;
1144 /* NOTE: we depend on atomic igrab() -bzzz */
1145 lock_res_and_lock(lock);
1146 if (lock->l_resource->lr_lvb_inode) {
1147 struct ll_inode_info * lli;
1148 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1149 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1150 inode = igrab(lock->l_resource->lr_lvb_inode);
1152 inode = lock->l_resource->lr_lvb_inode;
1153 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1154 D_WARNING, lock, "lr_lvb_inode %p is "
1155 "bogus: magic %08x",
1156 lock->l_resource->lr_lvb_inode,
1157 lli->lli_inode_magic);
1161 unlock_res_and_lock(lock);
1165 void ll_dir_clear_lsm_md(struct inode *inode)
1167 struct ll_inode_info *lli = ll_i2info(inode);
1169 LASSERT(S_ISDIR(inode->i_mode));
1171 if (lli->lli_lsm_md != NULL) {
1172 lmv_free_memmd(lli->lli_lsm_md);
1173 lli->lli_lsm_md = NULL;
1177 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1178 const struct lu_fid *fid,
1179 struct lustre_md *md)
1181 struct ll_sb_info *sbi = ll_s2sbi(sb);
1182 struct mdt_body *body = md->body;
1183 struct inode *inode;
1187 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1188 inode = iget_locked(sb, ino);
1189 if (inode == NULL) {
1190 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1191 ll_get_fsname(sb, NULL, 0), PFID(fid));
1192 RETURN(ERR_PTR(-ENOENT));
1195 if (inode->i_state & I_NEW) {
1196 struct ll_inode_info *lli = ll_i2info(inode);
1197 struct lmv_stripe_md *lsm = md->lmv;
1199 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1200 (body->mbo_mode & S_IFMT);
1201 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1204 LTIME_S(inode->i_mtime) = 0;
1205 LTIME_S(inode->i_atime) = 0;
1206 LTIME_S(inode->i_ctime) = 0;
1209 #ifdef HAVE_BACKING_DEV_INFO
1210 /* initializing backing dev info. */
1211 inode->i_mapping->backing_dev_info =
1212 &s2lsi(inode->i_sb)->lsi_bdi;
1214 inode->i_op = &ll_dir_inode_operations;
1215 inode->i_fop = &ll_dir_operations;
1216 lli->lli_fid = *fid;
1219 LASSERT(lsm != NULL);
1220 /* master object FID */
1221 lli->lli_pfid = body->mbo_fid1;
1222 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1223 lli, PFID(fid), PFID(&lli->lli_pfid));
1224 unlock_new_inode(inode);
1230 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1233 struct lmv_stripe_md *lsm = md->lmv;
1236 LASSERT(lsm != NULL);
1237 /* XXX sigh, this lsm_root initialization should be in
1238 * LMV layer, but it needs ll_iget right now, so we
1239 * put this here right now. */
1240 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1241 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1242 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1243 /* Unfortunately ll_iget will call ll_update_inode,
1244 * where the initialization of slave inode is slightly
1245 * different, so it reset lsm_md to NULL to avoid
1246 * initializing lsm for slave inode. */
1247 /* For migrating inode, master stripe and master object will
1248 * be same, so we only need assign this inode */
1249 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1250 lsm->lsm_md_oinfo[i].lmo_root = inode;
1252 lsm->lsm_md_oinfo[i].lmo_root =
1253 ll_iget_anon_dir(inode->i_sb, fid, md);
1255 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1256 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1258 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1266 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1267 const struct lmv_stripe_md *lsm_md2)
1269 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1270 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1271 lsm_md1->lsm_md_master_mdt_index ==
1272 lsm_md2->lsm_md_master_mdt_index &&
1273 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1274 lsm_md1->lsm_md_layout_version ==
1275 lsm_md2->lsm_md_layout_version &&
1276 strcmp(lsm_md1->lsm_md_pool_name,
1277 lsm_md2->lsm_md_pool_name) == 0;
1280 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1282 struct ll_inode_info *lli = ll_i2info(inode);
1283 struct lmv_stripe_md *lsm = md->lmv;
1287 LASSERT(S_ISDIR(inode->i_mode));
1288 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1289 PFID(ll_inode2fid(inode)));
1291 /* no striped information from request. */
1293 if (lli->lli_lsm_md == NULL) {
1295 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1296 LMV_HASH_FLAG_MIGRATION) {
1297 /* migration is done, the temporay MIGRATE layout has
1299 CDEBUG(D_INODE, DFID" finish migration.\n",
1300 PFID(ll_inode2fid(inode)));
1301 lmv_free_memmd(lli->lli_lsm_md);
1302 lli->lli_lsm_md = NULL;
1305 /* The lustre_md from req does not include stripeEA,
1306 * see ll_md_setattr */
1311 /* set the directory layout */
1312 if (lli->lli_lsm_md == NULL) {
1313 struct cl_attr *attr;
1315 rc = ll_init_lsm_md(inode, md);
1319 /* set md->lmv to NULL, so the following free lustre_md
1320 * will not free this lsm */
1322 lli->lli_lsm_md = lsm;
1324 OBD_ALLOC_PTR(attr);
1328 /* validate the lsm */
1329 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1330 ll_md_blocking_ast);
1336 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1337 md->body->mbo_nlink = attr->cat_nlink;
1338 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1339 md->body->mbo_size = attr->cat_size;
1340 if (md->body->mbo_valid & OBD_MD_FLATIME)
1341 md->body->mbo_atime = attr->cat_atime;
1342 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1343 md->body->mbo_ctime = attr->cat_ctime;
1344 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1345 md->body->mbo_mtime = attr->cat_mtime;
1349 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1350 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1354 /* Compare the old and new stripe information */
1355 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1356 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1359 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1360 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1361 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1362 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1363 inode, lsm, old_lsm,
1364 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1365 lsm->lsm_md_stripe_count,
1366 old_lsm->lsm_md_stripe_count,
1367 lsm->lsm_md_master_mdt_index,
1368 old_lsm->lsm_md_master_mdt_index,
1369 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1370 lsm->lsm_md_layout_version,
1371 old_lsm->lsm_md_layout_version,
1372 lsm->lsm_md_pool_name,
1373 old_lsm->lsm_md_pool_name);
1375 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1376 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1377 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1378 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1381 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1382 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1383 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1384 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1393 void ll_clear_inode(struct inode *inode)
1395 struct ll_inode_info *lli = ll_i2info(inode);
1396 struct ll_sb_info *sbi = ll_i2sbi(inode);
1399 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1400 PFID(ll_inode2fid(inode)), inode);
1402 if (S_ISDIR(inode->i_mode)) {
1403 /* these should have been cleared in ll_file_release */
1404 LASSERT(lli->lli_opendir_key == NULL);
1405 LASSERT(lli->lli_sai == NULL);
1406 LASSERT(lli->lli_opendir_pid == 0);
1409 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1411 LASSERT(!lli->lli_open_fd_write_count);
1412 LASSERT(!lli->lli_open_fd_read_count);
1413 LASSERT(!lli->lli_open_fd_exec_count);
1415 if (lli->lli_mds_write_och)
1416 ll_md_real_close(inode, FMODE_WRITE);
1417 if (lli->lli_mds_exec_och)
1418 ll_md_real_close(inode, FMODE_EXEC);
1419 if (lli->lli_mds_read_och)
1420 ll_md_real_close(inode, FMODE_READ);
1422 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1423 OBD_FREE(lli->lli_symlink_name,
1424 strlen(lli->lli_symlink_name) + 1);
1425 lli->lli_symlink_name = NULL;
1428 ll_xattr_cache_destroy(inode);
1430 #ifdef CONFIG_FS_POSIX_ACL
1431 if (lli->lli_posix_acl) {
1432 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1433 posix_acl_release(lli->lli_posix_acl);
1434 lli->lli_posix_acl = NULL;
1437 lli->lli_inode_magic = LLI_INODE_DEAD;
1439 if (S_ISDIR(inode->i_mode))
1440 ll_dir_clear_lsm_md(inode);
1441 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1442 LASSERT(list_empty(&lli->lli_agl_list));
1445 * XXX This has to be done before lsm is freed below, because
1446 * cl_object still uses inode lsm.
1448 cl_inode_fini(inode);
1453 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1455 struct lustre_md md;
1456 struct inode *inode = dentry->d_inode;
1457 struct ll_sb_info *sbi = ll_i2sbi(inode);
1458 struct ptlrpc_request *request = NULL;
1462 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1463 LUSTRE_OPC_ANY, NULL);
1464 if (IS_ERR(op_data))
1465 RETURN(PTR_ERR(op_data));
1467 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1469 ptlrpc_req_finished(request);
1470 if (rc == -ENOENT) {
1472 /* Unlinked special device node? Or just a race?
1473 * Pretend we done everything. */
1474 if (!S_ISREG(inode->i_mode) &&
1475 !S_ISDIR(inode->i_mode)) {
1476 ia_valid = op_data->op_attr.ia_valid;
1477 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1478 rc = simple_setattr(dentry, &op_data->op_attr);
1479 op_data->op_attr.ia_valid = ia_valid;
1481 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1482 CERROR("md_setattr fails: rc = %d\n", rc);
1487 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1488 sbi->ll_md_exp, &md);
1490 ptlrpc_req_finished(request);
1494 ia_valid = op_data->op_attr.ia_valid;
1495 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1496 * cache is not cleared yet. */
1497 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1498 if (S_ISREG(inode->i_mode))
1499 mutex_lock(&inode->i_mutex);
1500 rc = simple_setattr(dentry, &op_data->op_attr);
1501 if (S_ISREG(inode->i_mode))
1502 mutex_unlock(&inode->i_mutex);
1503 op_data->op_attr.ia_valid = ia_valid;
1505 rc = ll_update_inode(inode, &md);
1506 ptlrpc_req_finished(request);
1511 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1512 * object(s) determine the file size and mtime. Otherwise, the MDS will
1513 * keep these values until such a time that objects are allocated for it.
1514 * We do the MDS operations first, as it is checking permissions for us.
1515 * We don't to the MDS RPC if there is nothing that we want to store there,
1516 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1517 * going to do an RPC anyways.
1519 * If we are doing a truncate, we will send the mtime and ctime updates
1520 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1521 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1524 * In case of HSMimport, we only set attr on MDS.
1526 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1528 struct inode *inode = dentry->d_inode;
1529 struct ll_inode_info *lli = ll_i2info(inode);
1530 struct md_op_data *op_data = NULL;
1534 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1535 "valid %x, hsm_import %d\n",
1536 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1537 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1540 if (attr->ia_valid & ATTR_SIZE) {
1541 /* Check new size against VFS/VM file size limit and rlimit */
1542 rc = inode_newsize_ok(inode, attr->ia_size);
1546 /* The maximum Lustre file size is variable, based on the
1547 * OST maximum object size and number of stripes. This
1548 * needs another check in addition to the VFS check above. */
1549 if (attr->ia_size > ll_file_maxbytes(inode)) {
1550 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1551 PFID(&lli->lli_fid), attr->ia_size,
1552 ll_file_maxbytes(inode));
1556 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1559 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1560 if (attr->ia_valid & TIMES_SET_FLAGS) {
1561 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1562 !cfs_capable(CFS_CAP_FOWNER))
1566 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1567 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1568 (attr->ia_valid & ATTR_CTIME)) {
1569 attr->ia_ctime = CURRENT_TIME;
1570 attr->ia_valid |= ATTR_CTIME_SET;
1572 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1573 (attr->ia_valid & ATTR_ATIME)) {
1574 attr->ia_atime = CURRENT_TIME;
1575 attr->ia_valid |= ATTR_ATIME_SET;
1577 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1578 (attr->ia_valid & ATTR_MTIME)) {
1579 attr->ia_mtime = CURRENT_TIME;
1580 attr->ia_valid |= ATTR_MTIME_SET;
1583 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1584 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1585 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1586 cfs_time_current_sec());
1588 if (S_ISREG(inode->i_mode)) {
1589 if (attr->ia_valid & ATTR_SIZE)
1590 inode_dio_write_done(inode);
1591 mutex_unlock(&inode->i_mutex);
1594 /* We always do an MDS RPC, even if we're only changing the size;
1595 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1597 OBD_ALLOC_PTR(op_data);
1598 if (op_data == NULL)
1599 GOTO(out, rc = -ENOMEM);
1601 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1602 /* If we are changing file size, file content is
1603 * modified, flag it. */
1604 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1605 op_data->op_bias |= MDS_DATA_MODIFIED;
1606 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1609 op_data->op_attr = *attr;
1611 rc = ll_md_setattr(dentry, op_data);
1615 if (!S_ISREG(inode->i_mode) || hsm_import)
1618 if (attr->ia_valid & (ATTR_SIZE |
1619 ATTR_ATIME | ATTR_ATIME_SET |
1620 ATTR_MTIME | ATTR_MTIME_SET |
1621 ATTR_CTIME | ATTR_CTIME_SET)) {
1622 /* For truncate and utimes sending attributes to OSTs, setting
1623 * mtime/atime to the past will be performed under PW [0:EOF]
1624 * extent lock (new_size:EOF for truncate). It may seem
1625 * excessive to send mtime/atime updates to OSTs when not
1626 * setting times to past, but it is necessary due to possible
1627 * time de-synchronization between MDT inode and OST objects */
1628 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1631 /* If the file was restored, it needs to set dirty flag.
1633 * We've already sent MDS_DATA_MODIFIED flag in
1634 * ll_md_setattr() for truncate. However, the MDT refuses to
1635 * set the HS_DIRTY flag on released files, so we have to set
1636 * it again if the file has been restored. Please check how
1637 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1639 * Please notice that if the file is not released, the previous
1640 * MDS_DATA_MODIFIED has taken effect and usually
1641 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1642 * This way we can save an RPC for common open + trunc
1644 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1645 struct hsm_state_set hss = {
1646 .hss_valid = HSS_SETMASK,
1647 .hss_setmask = HS_DIRTY,
1651 rc2 = ll_hsm_state_set(inode, &hss);
1652 /* truncate and write can happen at the same time, so that
1653 * the file can be set modified even though the file is not
1654 * restored from released state, and ll_hsm_state_set() is
1655 * not applicable for the file, and rc2 < 0 is normal in this
1658 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1659 PFID(ll_inode2fid(inode)), rc2);
1664 if (op_data != NULL)
1665 ll_finish_md_op_data(op_data);
1667 if (S_ISREG(inode->i_mode)) {
1668 mutex_lock(&inode->i_mutex);
1669 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1670 inode_dio_wait(inode);
1673 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1674 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1679 int ll_setattr(struct dentry *de, struct iattr *attr)
1681 int mode = de->d_inode->i_mode;
1683 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1684 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1685 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1687 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1688 (ATTR_SIZE|ATTR_MODE)) &&
1689 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1690 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1691 !(attr->ia_mode & S_ISGID))))
1692 attr->ia_valid |= ATTR_FORCE;
1694 if ((attr->ia_valid & ATTR_MODE) &&
1696 !(attr->ia_mode & S_ISUID) &&
1697 !(attr->ia_valid & ATTR_KILL_SUID))
1698 attr->ia_valid |= ATTR_KILL_SUID;
1700 if ((attr->ia_valid & ATTR_MODE) &&
1701 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1702 !(attr->ia_mode & S_ISGID) &&
1703 !(attr->ia_valid & ATTR_KILL_SGID))
1704 attr->ia_valid |= ATTR_KILL_SGID;
1706 return ll_setattr_raw(de, attr, false);
1709 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1710 __u64 max_age, __u32 flags)
1712 struct ll_sb_info *sbi = ll_s2sbi(sb);
1713 struct obd_statfs obd_osfs;
1717 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1719 CERROR("md_statfs fails: rc = %d\n", rc);
1723 osfs->os_type = sb->s_magic;
1725 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1726 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1728 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1729 flags |= OBD_STATFS_NODELAY;
1731 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1733 CERROR("obd_statfs fails: rc = %d\n", rc);
1737 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1738 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1741 osfs->os_bsize = obd_osfs.os_bsize;
1742 osfs->os_blocks = obd_osfs.os_blocks;
1743 osfs->os_bfree = obd_osfs.os_bfree;
1744 osfs->os_bavail = obd_osfs.os_bavail;
1746 /* If we don't have as many objects free on the OST as inodes
1747 * on the MDS, we reduce the total number of inodes to
1748 * compensate, so that the "inodes in use" number is correct.
1750 if (obd_osfs.os_ffree < osfs->os_ffree) {
1751 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1753 osfs->os_ffree = obd_osfs.os_ffree;
1758 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1760 struct super_block *sb = de->d_sb;
1761 struct obd_statfs osfs;
1762 __u64 fsid = huge_encode_dev(sb->s_dev);
1765 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1766 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1768 /* Some amount of caching on the client is allowed */
1769 rc = ll_statfs_internal(sb, &osfs,
1770 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1775 statfs_unpack(sfs, &osfs);
1777 /* We need to downshift for all 32-bit kernels, because we can't
1778 * tell if the kernel is being called via sys_statfs64() or not.
1779 * Stop before overflowing f_bsize - in which case it is better
1780 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1781 if (sizeof(long) < 8) {
1782 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1785 osfs.os_blocks >>= 1;
1786 osfs.os_bfree >>= 1;
1787 osfs.os_bavail >>= 1;
1791 sfs->f_blocks = osfs.os_blocks;
1792 sfs->f_bfree = osfs.os_bfree;
1793 sfs->f_bavail = osfs.os_bavail;
1794 sfs->f_fsid.val[0] = (__u32)fsid;
1795 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1799 void ll_inode_size_lock(struct inode *inode)
1801 struct ll_inode_info *lli;
1803 LASSERT(!S_ISDIR(inode->i_mode));
1805 lli = ll_i2info(inode);
1806 mutex_lock(&lli->lli_size_mutex);
1809 void ll_inode_size_unlock(struct inode *inode)
1811 struct ll_inode_info *lli;
1813 lli = ll_i2info(inode);
1814 mutex_unlock(&lli->lli_size_mutex);
1817 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1819 struct ll_inode_info *lli = ll_i2info(inode);
1820 struct mdt_body *body = md->body;
1821 struct ll_sb_info *sbi = ll_i2sbi(inode);
1823 if (body->mbo_valid & OBD_MD_FLEASIZE)
1824 cl_file_inode_init(inode, md);
1826 if (S_ISDIR(inode->i_mode)) {
1829 rc = ll_update_lsm_md(inode, md);
1834 #ifdef CONFIG_FS_POSIX_ACL
1835 if (body->mbo_valid & OBD_MD_FLACL) {
1836 spin_lock(&lli->lli_lock);
1837 if (lli->lli_posix_acl)
1838 posix_acl_release(lli->lli_posix_acl);
1839 lli->lli_posix_acl = md->posix_acl;
1840 spin_unlock(&lli->lli_lock);
1843 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1844 sbi->ll_flags & LL_SBI_32BIT_API);
1845 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1847 if (body->mbo_valid & OBD_MD_FLATIME) {
1848 if (body->mbo_atime > LTIME_S(inode->i_atime))
1849 LTIME_S(inode->i_atime) = body->mbo_atime;
1850 lli->lli_atime = body->mbo_atime;
1853 if (body->mbo_valid & OBD_MD_FLMTIME) {
1854 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1855 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1856 "to "LPU64"\n", inode->i_ino,
1857 LTIME_S(inode->i_mtime), body->mbo_mtime);
1858 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1860 lli->lli_mtime = body->mbo_mtime;
1863 if (body->mbo_valid & OBD_MD_FLCTIME) {
1864 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1865 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1866 lli->lli_ctime = body->mbo_ctime;
1869 if (body->mbo_valid & OBD_MD_FLMODE)
1870 inode->i_mode = (inode->i_mode & S_IFMT) |
1871 (body->mbo_mode & ~S_IFMT);
1873 if (body->mbo_valid & OBD_MD_FLTYPE)
1874 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1875 (body->mbo_mode & S_IFMT);
1877 LASSERT(inode->i_mode != 0);
1878 if (S_ISREG(inode->i_mode))
1879 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1880 LL_MAX_BLKSIZE_BITS);
1882 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1884 if (body->mbo_valid & OBD_MD_FLUID)
1885 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1886 if (body->mbo_valid & OBD_MD_FLGID)
1887 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1888 if (body->mbo_valid & OBD_MD_FLFLAGS)
1889 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1890 if (body->mbo_valid & OBD_MD_FLNLINK)
1891 set_nlink(inode, body->mbo_nlink);
1892 if (body->mbo_valid & OBD_MD_FLRDEV)
1893 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1895 if (body->mbo_valid & OBD_MD_FLID) {
1896 /* FID shouldn't be changed! */
1897 if (fid_is_sane(&lli->lli_fid)) {
1898 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1899 "Trying to change FID "DFID
1900 " to the "DFID", inode "DFID"(%p)\n",
1901 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1902 PFID(ll_inode2fid(inode)), inode);
1904 lli->lli_fid = body->mbo_fid1;
1908 LASSERT(fid_seq(&lli->lli_fid) != 0);
1910 if (body->mbo_valid & OBD_MD_FLSIZE) {
1911 i_size_write(inode, body->mbo_size);
1913 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1914 PFID(ll_inode2fid(inode)),
1915 (unsigned long long)body->mbo_size);
1917 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1918 inode->i_blocks = body->mbo_blocks;
1921 if (body->mbo_valid & OBD_MD_TSTATE) {
1922 /* Set LLIF_FILE_RESTORING if restore ongoing and
1923 * clear it when done to ensure to start again
1924 * glimpsing updated attrs
1926 if (body->mbo_t_state & MS_RESTORE)
1927 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1929 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1935 int ll_read_inode2(struct inode *inode, void *opaque)
1937 struct lustre_md *md = opaque;
1938 struct ll_inode_info *lli = ll_i2info(inode);
1942 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1943 PFID(&lli->lli_fid), inode);
1945 /* Core attributes from the MDS first. This is a new inode, and
1946 * the VFS doesn't zero times in the core inode so we have to do
1947 * it ourselves. They will be overwritten by either MDS or OST
1948 * attributes - we just need to make sure they aren't newer. */
1949 LTIME_S(inode->i_mtime) = 0;
1950 LTIME_S(inode->i_atime) = 0;
1951 LTIME_S(inode->i_ctime) = 0;
1953 rc = ll_update_inode(inode, md);
1957 /* OIDEBUG(inode); */
1959 #ifdef HAVE_BACKING_DEV_INFO
1960 /* initializing backing dev info. */
1961 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1963 if (S_ISREG(inode->i_mode)) {
1964 struct ll_sb_info *sbi = ll_i2sbi(inode);
1965 inode->i_op = &ll_file_inode_operations;
1966 inode->i_fop = sbi->ll_fop;
1967 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1969 } else if (S_ISDIR(inode->i_mode)) {
1970 inode->i_op = &ll_dir_inode_operations;
1971 inode->i_fop = &ll_dir_operations;
1973 } else if (S_ISLNK(inode->i_mode)) {
1974 inode->i_op = &ll_fast_symlink_inode_operations;
1977 inode->i_op = &ll_special_inode_operations;
1979 init_special_inode(inode, inode->i_mode,
1988 void ll_delete_inode(struct inode *inode)
1990 struct ll_inode_info *lli = ll_i2info(inode);
1993 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1994 /* It is last chance to write out dirty pages,
1995 * otherwise we may lose data while umount */
1996 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
1998 truncate_inode_pages_final(&inode->i_data);
2000 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2001 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2002 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2004 #ifdef HAVE_SBOPS_EVICT_INODE
2005 ll_clear_inode(inode);
2012 int ll_iocontrol(struct inode *inode, struct file *file,
2013 unsigned int cmd, unsigned long arg)
2015 struct ll_sb_info *sbi = ll_i2sbi(inode);
2016 struct ptlrpc_request *req = NULL;
2021 case FSFILT_IOC_GETFLAGS: {
2022 struct mdt_body *body;
2023 struct md_op_data *op_data;
2025 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2026 0, 0, LUSTRE_OPC_ANY,
2028 if (IS_ERR(op_data))
2029 RETURN(PTR_ERR(op_data));
2031 op_data->op_valid = OBD_MD_FLFLAGS;
2032 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2033 ll_finish_md_op_data(op_data);
2035 CERROR("%s: failure inode "DFID": rc = %d\n",
2036 sbi->ll_md_exp->exp_obd->obd_name,
2037 PFID(ll_inode2fid(inode)), rc);
2041 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2043 flags = body->mbo_flags;
2045 ptlrpc_req_finished(req);
2047 RETURN(put_user(flags, (int __user *)arg));
2049 case FSFILT_IOC_SETFLAGS: {
2051 struct md_op_data *op_data;
2052 struct cl_object *obj;
2054 if (get_user(flags, (int __user *)arg))
2057 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2058 LUSTRE_OPC_ANY, NULL);
2059 if (IS_ERR(op_data))
2060 RETURN(PTR_ERR(op_data));
2062 op_data->op_attr_flags = flags;
2063 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2064 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2065 ll_finish_md_op_data(op_data);
2066 ptlrpc_req_finished(req);
2070 inode->i_flags = ll_ext_to_inode_flags(flags);
2072 obj = ll_i2info(inode)->lli_clob;
2076 OBD_ALLOC_PTR(attr);
2080 attr->ia_valid = ATTR_ATTR_FLAG;
2081 rc = cl_setattr_ost(obj, attr, flags);
2093 int ll_flush_ctx(struct inode *inode)
2095 struct ll_sb_info *sbi = ll_i2sbi(inode);
2097 CDEBUG(D_SEC, "flush context for user %d\n",
2098 from_kuid(&init_user_ns, current_uid()));
2100 obd_set_info_async(NULL, sbi->ll_md_exp,
2101 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2103 obd_set_info_async(NULL, sbi->ll_dt_exp,
2104 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2109 /* umount -f client means force down, don't save state */
2110 void ll_umount_begin(struct super_block *sb)
2112 struct ll_sb_info *sbi = ll_s2sbi(sb);
2113 struct obd_device *obd;
2114 struct obd_ioctl_data *ioc_data;
2117 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2118 sb->s_count, atomic_read(&sb->s_active));
2120 obd = class_exp2obd(sbi->ll_md_exp);
2122 CERROR("Invalid MDC connection handle "LPX64"\n",
2123 sbi->ll_md_exp->exp_handle.h_cookie);
2129 obd = class_exp2obd(sbi->ll_dt_exp);
2131 CERROR("Invalid LOV connection handle "LPX64"\n",
2132 sbi->ll_dt_exp->exp_handle.h_cookie);
2138 OBD_ALLOC_PTR(ioc_data);
2140 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2141 sizeof *ioc_data, ioc_data, NULL);
2143 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2144 sizeof *ioc_data, ioc_data, NULL);
2146 OBD_FREE_PTR(ioc_data);
2149 /* Really, we'd like to wait until there are no requests outstanding,
2150 * and then continue. For now, we just invalidate the requests,
2151 * schedule() and sleep one second if needed, and hope.
2157 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2159 struct ll_sb_info *sbi = ll_s2sbi(sb);
2160 char *profilenm = get_profile_name(sb);
2164 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2165 read_only = *flags & MS_RDONLY;
2166 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2167 sizeof(KEY_READ_ONLY),
2168 KEY_READ_ONLY, sizeof(read_only),
2171 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2172 profilenm, read_only ?
2173 "read-only" : "read-write", err);
2178 sb->s_flags |= MS_RDONLY;
2180 sb->s_flags &= ~MS_RDONLY;
2182 if (sbi->ll_flags & LL_SBI_VERBOSE)
2183 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2184 read_only ? "read-only" : "read-write");
2190 * Cleanup the open handle that is cached on MDT-side.
2192 * For open case, the client side open handling thread may hit error
2193 * after the MDT grant the open. Under such case, the client should
2194 * send close RPC to the MDT as cleanup; otherwise, the open handle
2195 * on the MDT will be leaked there until the client umount or evicted.
2197 * In further, if someone unlinked the file, because the open handle
2198 * holds the reference on such file/object, then it will block the
2199 * subsequent threads that want to locate such object via FID.
2201 * \param[in] sb super block for this file-system
2202 * \param[in] open_req pointer to the original open request
2204 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2206 struct mdt_body *body;
2207 struct md_op_data *op_data;
2208 struct ptlrpc_request *close_req = NULL;
2209 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2212 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2213 OBD_ALLOC_PTR(op_data);
2214 if (op_data == NULL) {
2215 CWARN("%s: cannot allocate op_data to release open handle for "
2217 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2222 op_data->op_fid1 = body->mbo_fid1;
2223 op_data->op_handle = body->mbo_handle;
2224 op_data->op_mod_time = cfs_time_current_sec();
2225 md_close(exp, op_data, NULL, &close_req);
2226 ptlrpc_req_finished(close_req);
2227 ll_finish_md_op_data(op_data);
2232 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2233 struct super_block *sb, struct lookup_intent *it)
2235 struct ll_sb_info *sbi = NULL;
2236 struct lustre_md md = { NULL };
2240 LASSERT(*inode || sb);
2241 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2242 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2243 sbi->ll_md_exp, &md);
2248 rc = ll_update_inode(*inode, &md);
2252 LASSERT(sb != NULL);
2255 * At this point server returns to client's same fid as client
2256 * generated for creating. So using ->fid1 is okay here.
2258 if (!fid_is_sane(&md.body->mbo_fid1)) {
2259 CERROR("%s: Fid is insane "DFID"\n",
2260 ll_get_fsname(sb, NULL, 0),
2261 PFID(&md.body->mbo_fid1));
2262 GOTO(out, rc = -EINVAL);
2265 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2266 sbi->ll_flags & LL_SBI_32BIT_API),
2268 if (IS_ERR(*inode)) {
2269 #ifdef CONFIG_FS_POSIX_ACL
2271 posix_acl_release(md.posix_acl);
2272 md.posix_acl = NULL;
2275 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2277 CERROR("new_inode -fatal: rc %d\n", rc);
2282 /* Handling piggyback layout lock.
2283 * Layout lock can be piggybacked by getattr and open request.
2284 * The lsm can be applied to inode only if it comes with a layout lock
2285 * otherwise correct layout may be overwritten, for example:
2286 * 1. proc1: mdt returns a lsm but not granting layout
2287 * 2. layout was changed by another client
2288 * 3. proc2: refresh layout and layout lock granted
2289 * 4. proc1: to apply a stale layout */
2290 if (it != NULL && it->it_lock_mode != 0) {
2291 struct lustre_handle lockh;
2292 struct ldlm_lock *lock;
2294 lockh.cookie = it->it_lock_handle;
2295 lock = ldlm_handle2lock(&lockh);
2296 LASSERT(lock != NULL);
2297 if (ldlm_has_layout(lock)) {
2298 struct cl_object_conf conf;
2300 memset(&conf, 0, sizeof(conf));
2301 conf.coc_opc = OBJECT_CONF_SET;
2302 conf.coc_inode = *inode;
2303 conf.coc_lock = lock;
2304 conf.u.coc_layout = md.layout;
2305 (void)ll_layout_conf(*inode, &conf);
2307 LDLM_LOCK_PUT(lock);
2313 md_free_lustre_md(sbi->ll_md_exp, &md);
2316 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2317 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2322 int ll_obd_statfs(struct inode *inode, void __user *arg)
2324 struct ll_sb_info *sbi = NULL;
2325 struct obd_export *exp;
2327 struct obd_ioctl_data *data = NULL;
2331 if (!inode || !(sbi = ll_i2sbi(inode)))
2332 GOTO(out_statfs, rc = -EINVAL);
2334 rc = obd_ioctl_getdata(&buf, &len, arg);
2336 GOTO(out_statfs, rc);
2339 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2340 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2341 GOTO(out_statfs, rc = -EINVAL);
2343 if (data->ioc_inllen1 != sizeof(__u32) ||
2344 data->ioc_inllen2 != sizeof(__u32) ||
2345 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2346 data->ioc_plen2 != sizeof(struct obd_uuid))
2347 GOTO(out_statfs, rc = -EINVAL);
2349 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2350 if (type & LL_STATFS_LMV)
2351 exp = sbi->ll_md_exp;
2352 else if (type & LL_STATFS_LOV)
2353 exp = sbi->ll_dt_exp;
2355 GOTO(out_statfs, rc = -ENODEV);
2357 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2359 GOTO(out_statfs, rc);
2362 obd_ioctl_freedata(buf, len);
2366 int ll_process_config(struct lustre_cfg *lcfg)
2368 struct super_block *sb;
2373 /* The instance name contains the sb: lustre-client-aacfe000 */
2374 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2375 if (!ptr || !*(++ptr))
2377 if (sscanf(ptr, "%lx", &x) != 1)
2379 sb = (struct super_block *)x;
2380 /* This better be a real Lustre superblock! */
2381 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2383 /* Note we have not called client_common_fill_super yet, so
2384 proc fns must be able to handle that! */
2385 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2392 /* this function prepares md_op_data hint for passing it down to MD stack. */
2393 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2394 struct inode *i1, struct inode *i2,
2395 const char *name, size_t namelen,
2396 __u32 mode, __u32 opc, void *data)
2398 LASSERT(i1 != NULL);
2401 /* Do not reuse namelen for something else. */
2403 return ERR_PTR(-EINVAL);
2405 if (namelen > ll_i2sbi(i1)->ll_namelen)
2406 return ERR_PTR(-ENAMETOOLONG);
2408 if (!lu_name_is_valid_2(name, namelen))
2409 return ERR_PTR(-EINVAL);
2412 if (op_data == NULL)
2413 OBD_ALLOC_PTR(op_data);
2415 if (op_data == NULL)
2416 return ERR_PTR(-ENOMEM);
2418 ll_i2gids(op_data->op_suppgids, i1, i2);
2419 op_data->op_fid1 = *ll_inode2fid(i1);
2420 op_data->op_default_stripe_offset = -1;
2421 if (S_ISDIR(i1->i_mode)) {
2422 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2423 if (opc == LUSTRE_OPC_MKDIR)
2424 op_data->op_default_stripe_offset =
2425 ll_i2info(i1)->lli_def_stripe_offset;
2429 op_data->op_fid2 = *ll_inode2fid(i2);
2430 if (S_ISDIR(i2->i_mode))
2431 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2433 fid_zero(&op_data->op_fid2);
2436 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2437 op_data->op_cli_flags |= CLI_HASH64;
2439 if (ll_need_32bit_api(ll_i2sbi(i1)))
2440 op_data->op_cli_flags |= CLI_API32;
2442 op_data->op_name = name;
2443 op_data->op_namelen = namelen;
2444 op_data->op_mode = mode;
2445 op_data->op_mod_time = cfs_time_current_sec();
2446 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2447 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2448 op_data->op_cap = cfs_curproc_cap_pack();
2449 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2450 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2451 op_data->op_bias |= MDS_CREATE_VOLATILE;
2453 op_data->op_mds = 0;
2455 op_data->op_data = data;
2460 void ll_finish_md_op_data(struct md_op_data *op_data)
2462 OBD_FREE_PTR(op_data);
2465 #ifdef HAVE_SUPEROPS_USE_DENTRY
2466 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2468 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2471 struct ll_sb_info *sbi;
2473 #ifdef HAVE_SUPEROPS_USE_DENTRY
2474 LASSERT((seq != NULL) && (dentry != NULL));
2475 sbi = ll_s2sbi(dentry->d_sb);
2477 LASSERT((seq != NULL) && (vfs != NULL));
2478 sbi = ll_s2sbi(vfs->mnt_sb);
2481 if (sbi->ll_flags & LL_SBI_NOLCK)
2482 seq_puts(seq, ",nolock");
2484 if (sbi->ll_flags & LL_SBI_FLOCK)
2485 seq_puts(seq, ",flock");
2487 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2488 seq_puts(seq, ",localflock");
2490 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2491 seq_puts(seq, ",user_xattr");
2493 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2494 seq_puts(seq, ",lazystatfs");
2496 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2497 seq_puts(seq, ",user_fid2path");
2499 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2500 seq_puts(seq, ",always_ping");
2506 * Get obd name by cmd, and copy out to user space
2508 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2510 struct ll_sb_info *sbi = ll_i2sbi(inode);
2511 struct obd_device *obd;
2514 if (cmd == OBD_IOC_GETDTNAME)
2515 obd = class_exp2obd(sbi->ll_dt_exp);
2516 else if (cmd == OBD_IOC_GETMDNAME)
2517 obd = class_exp2obd(sbi->ll_md_exp);
2524 if (copy_to_user((void __user *)arg, obd->obd_name,
2525 strlen(obd->obd_name) + 1))
2532 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2533 * fsname will be returned in this buffer; otherwise, a static buffer will be
2534 * used to store the fsname and returned to caller.
2536 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2538 static char fsname_static[MTI_NAME_MAXLEN];
2539 struct lustre_sb_info *lsi = s2lsi(sb);
2544 /* this means the caller wants to use static buffer
2545 * and it doesn't care about race. Usually this is
2546 * in error reporting path */
2547 buf = fsname_static;
2548 buflen = sizeof(fsname_static);
2551 len = strlen(lsi->lsi_lmd->lmd_profile);
2552 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2553 if (ptr && (strcmp(ptr, "-client") == 0))
2556 if (unlikely(len >= buflen))
2558 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2564 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2571 p.mnt = current->fs->root.mnt;
2573 path = d_path(&p, buf, bufsize);
2578 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2580 char *buf, *path = NULL;
2581 struct dentry *dentry = NULL;
2582 struct inode *inode = page->mapping->host;
2584 /* this can be called inside spin lock so use GFP_ATOMIC. */
2585 buf = (char *)__get_free_page(GFP_ATOMIC);
2587 dentry = d_find_alias(page->mapping->host);
2589 path = ll_d_path(dentry, buf, PAGE_SIZE);
2593 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2594 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2595 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2596 PFID(ll_inode2fid(inode)),
2597 (path && !IS_ERR(path)) ? path : "", ioret);
2603 free_page((unsigned long)buf);
2606 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2607 struct lov_user_md **kbuf)
2609 struct lov_user_md lum;
2613 if (copy_from_user(&lum, md, sizeof(lum)))
2616 lum_size = ll_lov_user_md_size(&lum);
2620 OBD_ALLOC(*kbuf, lum_size);
2624 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2625 OBD_FREE(*kbuf, lum_size);
2633 * Compute llite root squash state after a change of root squash
2634 * configuration setting or add/remove of a lnet nid
2636 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2638 struct root_squash_info *squash = &sbi->ll_squash;
2641 lnet_process_id_t id;
2643 /* Update norootsquash flag */
2644 down_write(&squash->rsi_sem);
2645 if (list_empty(&squash->rsi_nosquash_nids))
2646 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2648 /* Do not apply root squash as soon as one of our NIDs is
2649 * in the nosquash_nids list */
2652 while (LNetGetId(i++, &id) != -ENOENT) {
2653 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2655 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2661 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2663 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2665 up_write(&squash->rsi_sem);
2669 * Parse linkea content to extract information about a given hardlink
2671 * \param[in] ldata - Initialized linkea data
2672 * \param[in] linkno - Link identifier
2673 * \param[out] parent_fid - The entry's parent FID
2674 * \param[out] ln - Entry name destination buffer
2676 * \retval 0 on success
2677 * \retval Appropriate negative error code on failure
2679 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2680 struct lu_fid *parent_fid, struct lu_name *ln)
2686 rc = linkea_init(ldata);
2690 if (linkno >= ldata->ld_leh->leh_reccount)
2691 /* beyond last link */
2694 linkea_first_entry(ldata);
2695 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2696 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2701 linkea_next_entry(ldata);
2711 * Get parent FID and name of an identified link. Operation is performed for
2712 * a given link number, letting the caller iterate over linkno to list one or
2713 * all links of an entry.
2715 * \param[in] file - File descriptor against which to perform the operation
2716 * \param[in,out] arg - User-filled structure containing the linkno to operate
2717 * on and the available size. It is eventually filled with
2718 * the requested information or left untouched on error
2720 * \retval - 0 on success
2721 * \retval - Appropriate negative error code on failure
2723 int ll_getparent(struct file *file, struct getparent __user *arg)
2725 struct dentry *dentry = file_dentry(file);
2726 struct inode *inode = file_inode(file);
2727 struct linkea_data *ldata;
2728 struct lu_buf buf = LU_BUF_NULL;
2730 struct lu_fid parent_fid;
2737 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2738 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2741 if (get_user(name_size, &arg->gp_name_size))
2744 if (get_user(linkno, &arg->gp_linkno))
2747 if (name_size > PATH_MAX)
2750 OBD_ALLOC(ldata, sizeof(*ldata));
2754 rc = linkea_data_new(ldata, &buf);
2756 GOTO(ldata_free, rc);
2758 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2762 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2766 if (ln.ln_namelen >= name_size)
2767 GOTO(lb_free, rc = -EOVERFLOW);
2769 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2770 GOTO(lb_free, rc = -EFAULT);
2772 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2773 GOTO(lb_free, rc = -EFAULT);
2775 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2776 GOTO(lb_free, rc = -EFAULT);
2781 OBD_FREE(ldata, sizeof(*ldata));