1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
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/types.h>
45 #include <linux/random.h>
46 #include <linux/version.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
57 #include <obd_cksum.h>
58 #include "llite_internal.h"
60 cfs_mem_cache_t *ll_file_data_slab;
62 CFS_LIST_HEAD(ll_super_blocks);
63 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) cfs_ffz(~(n))
72 static struct ll_sb_info *ll_init_sbi(void)
74 struct ll_sb_info *sbi = NULL;
81 OBD_ALLOC(sbi, sizeof(*sbi));
85 cfs_spin_lock_init(&sbi->ll_lock);
86 cfs_init_mutex(&sbi->ll_lco.lco_lock);
87 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
88 cfs_spin_lock_init(&sbi->ll_process_lock);
89 sbi->ll_rw_stats_on = 0;
92 pages = si.totalram - si.totalhigh;
93 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
94 #ifdef HAVE_BGL_SUPPORT
95 sbi->ll_async_page_max = pages / 4;
97 sbi->ll_async_page_max = pages / 2;
100 sbi->ll_async_page_max = (pages / 4) * 3;
103 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
106 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
107 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
108 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
109 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
111 ll_generate_random_uuid(uuid);
112 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
113 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
115 cfs_spin_lock(&ll_sb_lock);
116 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
117 cfs_spin_unlock(&ll_sb_lock);
119 #ifdef ENABLE_LLITE_CHECKSUM
120 sbi->ll_flags |= LL_SBI_CHECKSUM;
123 #ifdef HAVE_LRU_RESIZE_SUPPORT
124 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
127 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
128 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
130 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
134 /* metadata statahead is enabled by default */
135 sbi->ll_sa_max = LL_SA_RPC_DEF;
136 atomic_set(&sbi->ll_sa_total, 0);
137 atomic_set(&sbi->ll_sa_wrong, 0);
142 void ll_free_sbi(struct super_block *sb)
144 struct ll_sb_info *sbi = ll_s2sbi(sb);
148 cfs_spin_lock(&ll_sb_lock);
149 cfs_list_del(&sbi->ll_list);
150 cfs_spin_unlock(&ll_sb_lock);
151 OBD_FREE(sbi, sizeof(*sbi));
156 static struct dentry_operations ll_d_root_ops = {
157 .d_compare = ll_dcompare,
158 .d_revalidate = ll_revalidate_nd,
161 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
163 struct inode *root = 0;
164 struct ll_sb_info *sbi = ll_s2sbi(sb);
165 struct obd_device *obd;
166 struct obd_capa *oc = NULL;
167 struct obd_statfs osfs;
168 struct ptlrpc_request *request = NULL;
169 struct obd_connect_data *data = NULL;
170 struct obd_uuid *uuid;
171 struct md_op_data *op_data;
172 struct lustre_md lmd;
174 int size, err, checksum;
177 obd = class_name2obd(md);
179 CERROR("MD %s: not setup or attached\n", md);
187 if (proc_lustre_fs_root) {
188 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
191 CERROR("could not register mount in /proc/fs/lustre\n");
194 /* indicate the features supported by this client */
195 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
196 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
197 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
198 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
199 OBD_CONNECT_FID | OBD_CONNECT_AT |
200 OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
201 OBD_CONNECT_VBR | OBD_CONNECT_FULL20;
203 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
204 data->ocd_connect_flags |= OBD_CONNECT_SOM;
206 #ifdef HAVE_LRU_RESIZE_SUPPORT
207 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
208 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
210 #ifdef CONFIG_FS_POSIX_ACL
211 data->ocd_connect_flags |= OBD_CONNECT_ACL;
213 data->ocd_ibits_known = MDS_INODELOCK_FULL;
214 data->ocd_version = LUSTRE_VERSION_CODE;
216 if (sb->s_flags & MS_RDONLY)
217 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
218 if (sbi->ll_flags & LL_SBI_USER_XATTR)
219 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
221 #ifdef HAVE_MS_FLOCK_LOCK
222 /* force vfs to use lustre handler for flock() calls - bug 10743 */
223 sb->s_flags |= MS_FLOCK_LOCK;
226 if (sbi->ll_flags & LL_SBI_FLOCK)
227 sbi->ll_fop = &ll_file_operations_flock;
228 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
229 sbi->ll_fop = &ll_file_operations;
231 sbi->ll_fop = &ll_file_operations_noflock;
234 data->ocd_connect_flags |= OBD_CONNECT_REAL;
235 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
236 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
238 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
240 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
241 "recovery, of which this client is not a "
242 "part. Please wait for recovery to complete,"
243 " abort, or time out.\n", md);
246 CERROR("cannot connect to %s: rc = %d\n", md, err);
250 err = obd_fid_init(sbi->ll_md_exp);
252 CERROR("Can't init metadata layer FID infrastructure, "
257 err = obd_statfs(obd, &osfs,
258 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
260 GOTO(out_md_fid, err);
262 size = sizeof(*data);
263 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
264 KEY_CONN_DATA, &size, data, NULL);
266 CERROR("Get connect data failed: %d \n", err);
270 LASSERT(osfs.os_bsize);
271 sb->s_blocksize = osfs.os_bsize;
272 sb->s_blocksize_bits = log2(osfs.os_bsize);
273 sb->s_magic = LL_SUPER_MAGIC;
275 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
276 * retval = in_file->f_op->sendfile(...);
278 * retval = -EOVERFLOW;
280 * it will check if *ppos is greater than max. However, max equals to
281 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
282 * has been defined as a signed long long integer in linux kernel. */
283 #if BITS_PER_LONG == 64
284 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
286 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
288 sbi->ll_namelen = osfs.os_namelen;
289 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
291 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
292 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
293 LCONSOLE_INFO("Disabling user_xattr feature because "
294 "it is not supported on the server\n");
295 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
298 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
300 sb->s_flags |= MS_POSIXACL;
302 sbi->ll_flags |= LL_SBI_ACL;
304 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
306 sb->s_flags &= ~MS_POSIXACL;
308 sbi->ll_flags &= ~LL_SBI_ACL;
311 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
312 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
313 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
314 LCONSOLE_INFO("client is set as remote by default.\n");
317 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
318 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
319 LCONSOLE_INFO("client claims to be remote, but server "
320 "rejected, forced to be local.\n");
324 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
325 LCONSOLE_INFO("client enabled MDS capability!\n");
326 sbi->ll_flags |= LL_SBI_MDS_CAPA;
329 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
330 LCONSOLE_INFO("client enabled OSS capability!\n");
331 sbi->ll_flags |= LL_SBI_OSS_CAPA;
334 obd = class_name2obd(dt);
336 CERROR("DT %s: not setup or attached\n", dt);
337 GOTO(out_md_fid, err = -ENODEV);
340 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
341 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
342 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
343 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
344 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
345 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
348 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
349 data->ocd_connect_flags |= OBD_CONNECT_SOM;
351 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
352 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
353 * disabled by default, because it can still be enabled on the
354 * fly via /proc. As a consequence, we still need to come to an
355 * agreement on the supported algorithms at connect time */
356 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
358 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
359 data->ocd_cksum_types = OBD_CKSUM_ADLER;
361 /* send the list of supported checksum types */
362 data->ocd_cksum_types = OBD_CKSUM_ALL;
365 #ifdef HAVE_LRU_RESIZE_SUPPORT
366 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
368 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
369 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
371 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
372 "ocd_grant: %d\n", data->ocd_connect_flags,
373 data->ocd_version, data->ocd_grant);
375 obd->obd_upcall.onu_owner = &sbi->ll_lco;
376 obd->obd_upcall.onu_upcall = cl_ocd_update;
377 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
379 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
381 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
382 "recovery, of which this client is not a "
383 "part. Please wait for recovery to "
384 "complete, abort, or time out.\n", dt);
385 GOTO(out_md_fid, err);
387 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
388 GOTO(out_md_fid, err);
391 err = obd_fid_init(sbi->ll_dt_exp);
393 CERROR("Can't init data layer FID infrastructure, "
398 cfs_mutex_down(&sbi->ll_lco.lco_lock);
399 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
400 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
401 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
402 cfs_mutex_up(&sbi->ll_lco.lco_lock);
404 fid_zero(&sbi->ll_root_fid);
405 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
407 CERROR("cannot mds_connect: rc = %d\n", err);
408 GOTO(out_lock_cn_cb, err);
410 if (!fid_is_sane(&sbi->ll_root_fid)) {
411 CERROR("Invalid root fid during mount\n");
412 GOTO(out_lock_cn_cb, err = -EINVAL);
414 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
416 sb->s_op = &lustre_super_operations;
417 sb->s_export_op = &lustre_export_operations;
420 * XXX: move this to after cbd setup? */
421 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
422 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
423 valid |= OBD_MD_FLRMTPERM;
424 else if (sbi->ll_flags & LL_SBI_ACL)
425 valid |= OBD_MD_FLACL;
427 OBD_ALLOC_PTR(op_data);
429 GOTO(out_lock_cn_cb, err = -ENOMEM);
431 op_data->op_fid1 = sbi->ll_root_fid;
432 op_data->op_mode = 0;
433 op_data->op_capa1 = oc;
434 op_data->op_valid = valid;
436 err = md_getattr(sbi->ll_md_exp, op_data, &request);
439 OBD_FREE_PTR(op_data);
441 CERROR("md_getattr failed for root: rc = %d\n", err);
442 GOTO(out_lock_cn_cb, err);
444 memset(&lmd, 0, sizeof(lmd));
445 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
446 sbi->ll_md_exp, &lmd);
448 CERROR("failed to understand root inode md: rc = %d\n", err);
449 ptlrpc_req_finished (request);
450 GOTO(out_lock_cn_cb, err);
453 LASSERT(fid_is_sane(&sbi->ll_root_fid));
454 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
455 md_free_lustre_md(sbi->ll_md_exp, &lmd);
456 ptlrpc_req_finished(request);
458 if (root == NULL || IS_ERR(root)) {
460 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
461 #ifdef CONFIG_FS_POSIX_ACL
463 posix_acl_release(lmd.posix_acl);
464 lmd.posix_acl = NULL;
467 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
469 CERROR("lustre_lite: bad iget4 for root\n");
473 err = ll_close_thread_start(&sbi->ll_lcq);
475 CERROR("cannot start close thread: rc %d\n", err);
479 #ifdef CONFIG_FS_POSIX_ACL
480 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
481 rct_init(&sbi->ll_rct);
482 et_init(&sbi->ll_et);
486 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
487 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
488 KEY_CHECKSUM, sizeof(checksum), &checksum,
492 sb->s_root = d_alloc_root(root);
494 OBD_FREE(data, sizeof(*data));
496 sb->s_root->d_op = &ll_d_root_ops;
498 sbi->ll_sdev_orig = sb->s_dev;
500 /* We set sb->s_dev equal on all lustre clients in order to support
501 * NFS export clustering. NFSD requires that the FSID be the same
503 /* s_dev is also used in lt_compare() to compare two fs, but that is
504 * only a node-local comparison. */
505 uuid = obd_get_uuid(sbi->ll_md_exp);
507 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
514 obd_fid_fini(sbi->ll_dt_exp);
516 obd_disconnect(sbi->ll_dt_exp);
517 sbi->ll_dt_exp = NULL;
519 obd_fid_fini(sbi->ll_md_exp);
521 obd_disconnect(sbi->ll_md_exp);
522 sbi->ll_md_exp = NULL;
526 lprocfs_unregister_mountpoint(sbi);
530 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
534 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
536 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
537 KEY_MAX_EASIZE, &size, lmmsize, NULL);
539 CERROR("Get max mdsize error rc %d \n", rc);
544 void ll_dump_inode(struct inode *inode)
546 struct list_head *tmp;
547 int dentry_count = 0;
549 LASSERT(inode != NULL);
551 list_for_each(tmp, &inode->i_dentry)
554 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
555 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
556 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
559 void lustre_dump_dentry(struct dentry *dentry, int recur)
561 struct list_head *tmp;
564 LASSERT(dentry != NULL);
566 list_for_each(tmp, &dentry->d_subdirs)
569 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
570 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
571 dentry->d_name.len, dentry->d_name.name,
572 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
573 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
574 dentry->d_flags, dentry->d_fsdata, subdirs);
575 if (dentry->d_inode != NULL)
576 ll_dump_inode(dentry->d_inode);
581 list_for_each(tmp, &dentry->d_subdirs) {
582 struct dentry *d = list_entry(tmp, struct dentry, d_child);
583 lustre_dump_dentry(d, recur - 1);
587 void client_common_put_super(struct super_block *sb)
589 struct ll_sb_info *sbi = ll_s2sbi(sb);
592 #ifdef CONFIG_FS_POSIX_ACL
593 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
594 et_fini(&sbi->ll_et);
595 rct_fini(&sbi->ll_rct);
599 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
601 ll_close_thread_shutdown(sbi->ll_lcq);
605 cfs_list_del(&sbi->ll_conn_chain);
607 obd_fid_fini(sbi->ll_dt_exp);
608 obd_disconnect(sbi->ll_dt_exp);
609 sbi->ll_dt_exp = NULL;
611 lprocfs_unregister_mountpoint(sbi);
613 obd_fid_fini(sbi->ll_md_exp);
614 obd_disconnect(sbi->ll_md_exp);
615 sbi->ll_md_exp = NULL;
620 void ll_kill_super(struct super_block *sb)
622 struct ll_sb_info *sbi;
627 if (!(sb->s_flags & MS_ACTIVE))
631 /* we need restore s_dev from changed for clustred NFS before put_super
632 * because new kernels have cached s_dev and change sb->s_dev in
633 * put_super not affected real removing devices */
635 sb->s_dev = sbi->ll_sdev_orig;
639 char *ll_read_opt(const char *opt, char *data)
645 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
646 if (strncmp(opt, data, strlen(opt)))
648 if ((value = strchr(data, '=')) == NULL)
652 OBD_ALLOC(retval, strlen(value) + 1);
654 CERROR("out of memory!\n");
658 memcpy(retval, value, strlen(value)+1);
659 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
663 static inline int ll_set_opt(const char *opt, char *data, int fl)
665 if (strncmp(opt, data, strlen(opt)) != 0)
671 /* non-client-specific mount options are parsed in lmd_parse */
672 static int ll_options(char *options, int *flags)
675 char *s1 = options, *s2;
681 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
684 CDEBUG(D_SUPER, "next opt=%s\n", s1);
685 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
690 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
695 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
700 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
705 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
710 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
715 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
716 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
718 /* Ignore deprecated mount option. The client will
719 * always try to mount with ACL support, whether this
720 * is used depends on whether server supports it. */
721 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
722 "mount option 'acl'.\n");
725 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
727 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
728 "mount option 'noacl'.\n");
732 #warning "{no}acl options have been deprecated since 1.8, please remove them"
734 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
740 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
745 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
750 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
755 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
760 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
765 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
770 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
776 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
782 s2 = strchr(s1, ',');
790 void ll_lli_init(struct ll_inode_info *lli)
792 lli->lli_inode_magic = LLI_INODE_MAGIC;
793 cfs_sema_init(&lli->lli_size_sem, 1);
794 cfs_sema_init(&lli->lli_write_sem, 1);
795 cfs_sema_init(&lli->lli_trunc_sem, 1);
797 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
798 cfs_spin_lock_init(&lli->lli_lock);
799 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
800 lli->lli_inode_magic = LLI_INODE_MAGIC;
801 cfs_sema_init(&lli->lli_och_sem, 1);
802 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
803 lli->lli_mds_exec_och = NULL;
804 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
805 lli->lli_open_fd_exec_count = 0;
806 CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
807 lli->lli_remote_perms = NULL;
808 lli->lli_rmtperm_utime = 0;
809 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
810 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
811 cfs_spin_lock_init(&lli->lli_sa_lock);
812 CFS_INIT_LIST_HEAD(&lli->lli_sa_dentry);
815 int ll_fill_super(struct super_block *sb)
817 struct lustre_profile *lprof;
818 struct lustre_sb_info *lsi = s2lsi(sb);
819 struct ll_sb_info *sbi;
820 char *dt = NULL, *md = NULL;
821 char *profilenm = get_profile_name(sb);
822 struct config_llog_instance cfg = {0, };
823 char ll_instance[sizeof(sb) * 2 + 1];
827 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
831 /* client additional sb info */
832 lsi->lsi_llsbi = sbi = ll_init_sbi();
834 cfs_module_put(THIS_MODULE);
838 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
842 /* Generate a string unique to this super, in case some joker tries
843 to mount the same fs at two mount points.
844 Use the address of the super itself.*/
845 sprintf(ll_instance, "%p", sb);
846 cfg.cfg_instance = ll_instance;
847 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
849 /* set up client obds */
850 err = lustre_process_log(sb, profilenm, &cfg);
852 CERROR("Unable to process log: %d\n", err);
856 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
857 lprof = class_get_profile(profilenm);
859 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
860 " read from the MGS. Does that filesystem "
861 "exist?\n", profilenm);
862 GOTO(out_free, err = -EINVAL);
864 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
865 lprof->lp_md, lprof->lp_dt);
867 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
868 strlen(ll_instance) + 2);
870 GOTO(out_free, err = -ENOMEM);
871 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
873 OBD_ALLOC(md, strlen(lprof->lp_md) +
874 strlen(ll_instance) + 2);
876 GOTO(out_free, err = -ENOMEM);
877 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
879 /* connections, registrations, sb setup */
880 err = client_common_fill_super(sb, md, dt);
884 OBD_FREE(md, strlen(md) + 1);
886 OBD_FREE(dt, strlen(dt) + 1);
890 LCONSOLE_WARN("Client %s has started\n", profilenm);
893 } /* ll_fill_super */
896 void lu_context_keys_dump(void);
898 void ll_put_super(struct super_block *sb)
900 struct config_llog_instance cfg;
901 char ll_instance[sizeof(sb) * 2 + 1];
902 struct obd_device *obd;
903 struct lustre_sb_info *lsi = s2lsi(sb);
904 struct ll_sb_info *sbi = ll_s2sbi(sb);
905 char *profilenm = get_profile_name(sb);
909 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
911 ll_print_capa_stat(sbi);
913 sprintf(ll_instance, "%p", sb);
914 cfg.cfg_instance = ll_instance;
915 lustre_end_log(sb, NULL, &cfg);
917 if (sbi->ll_md_exp) {
918 obd = class_exp2obd(sbi->ll_md_exp);
920 force = obd->obd_force;
923 /* We need to set force before the lov_disconnect in
924 lustre_common_put_super, since l_d cleans up osc's as well. */
927 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
929 obd->obd_force = force;
934 /* Only if client_common_fill_super succeeded */
935 client_common_put_super(sb);
939 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
940 class_manual_cleanup(obd);
944 class_del_profile(profilenm);
947 lsi->lsi_llsbi = NULL;
949 lustre_common_put_super(sb);
951 cl_env_cache_purge(~0);
953 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
955 cfs_module_put(THIS_MODULE);
958 } /* client_put_super */
960 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
962 struct inode *inode = NULL;
963 /* NOTE: we depend on atomic igrab() -bzzz */
964 lock_res_and_lock(lock);
965 if (lock->l_ast_data) {
966 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
967 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
968 inode = igrab(lock->l_ast_data);
970 inode = lock->l_ast_data;
971 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
973 lock, __FILE__, __func__, __LINE__,
974 "l_ast_data %p is bogus: magic %08x",
975 lock->l_ast_data, lli->lli_inode_magic);
979 unlock_res_and_lock(lock);
983 static int null_if_equal(struct ldlm_lock *lock, void *data)
985 if (data == lock->l_ast_data) {
986 lock->l_ast_data = NULL;
988 if (lock->l_req_mode != lock->l_granted_mode)
989 LDLM_ERROR(lock,"clearing inode with ungranted lock");
992 return LDLM_ITER_CONTINUE;
995 void ll_clear_inode(struct inode *inode)
997 struct ll_inode_info *lli = ll_i2info(inode);
998 struct ll_sb_info *sbi = ll_i2sbi(inode);
1001 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1002 inode->i_generation, inode);
1004 if (S_ISDIR(inode->i_mode)) {
1005 /* these should have been cleared in ll_file_release */
1006 LASSERT(lli->lli_sai == NULL);
1007 LASSERT(lli->lli_opendir_key == NULL);
1008 LASSERT(lli->lli_opendir_pid == 0);
1011 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1012 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1013 null_if_equal, inode);
1015 LASSERT(!lli->lli_open_fd_write_count);
1016 LASSERT(!lli->lli_open_fd_read_count);
1017 LASSERT(!lli->lli_open_fd_exec_count);
1019 if (lli->lli_mds_write_och)
1020 ll_md_real_close(inode, FMODE_WRITE);
1021 if (lli->lli_mds_exec_och)
1022 ll_md_real_close(inode, FMODE_EXEC);
1023 if (lli->lli_mds_read_och)
1024 ll_md_real_close(inode, FMODE_READ);
1026 if (lli->lli_symlink_name) {
1027 OBD_FREE(lli->lli_symlink_name,
1028 strlen(lli->lli_symlink_name) + 1);
1029 lli->lli_symlink_name = NULL;
1032 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1033 LASSERT(lli->lli_posix_acl == NULL);
1034 if (lli->lli_remote_perms) {
1035 free_rmtperm_hash(lli->lli_remote_perms);
1036 lli->lli_remote_perms = NULL;
1039 #ifdef CONFIG_FS_POSIX_ACL
1040 else if (lli->lli_posix_acl) {
1041 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1042 LASSERT(lli->lli_remote_perms == NULL);
1043 posix_acl_release(lli->lli_posix_acl);
1044 lli->lli_posix_acl = NULL;
1047 lli->lli_inode_magic = LLI_INODE_DEAD;
1049 ll_clear_inode_capas(inode);
1051 * XXX This has to be done before lsm is freed below, because
1052 * cl_object still uses inode lsm.
1054 cl_inode_fini(inode);
1057 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1058 lli->lli_smd = NULL;
1065 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1066 struct md_open_data **mod)
1068 struct lustre_md md;
1069 struct ll_sb_info *sbi = ll_i2sbi(inode);
1070 struct ptlrpc_request *request = NULL;
1074 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1075 LUSTRE_OPC_ANY, NULL);
1076 if (IS_ERR(op_data))
1077 RETURN(PTR_ERR(op_data));
1079 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1082 ptlrpc_req_finished(request);
1083 if (rc == -ENOENT) {
1085 /* Unlinked special device node? Or just a race?
1086 * Pretend we done everything. */
1087 if (!S_ISREG(inode->i_mode) &&
1088 !S_ISDIR(inode->i_mode))
1089 rc = inode_setattr(inode, &op_data->op_attr);
1090 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1091 CERROR("md_setattr fails: rc = %d\n", rc);
1096 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1097 sbi->ll_md_exp, &md);
1099 ptlrpc_req_finished(request);
1103 /* We call inode_setattr to adjust timestamps.
1104 * If there is at least some data in file, we cleared ATTR_SIZE
1105 * above to avoid invoking vmtruncate, otherwise it is important
1106 * to call vmtruncate in inode_setattr to update inode->i_size
1108 rc = inode_setattr(inode, &op_data->op_attr);
1110 /* Extract epoch data if obtained. */
1111 op_data->op_handle = md.body->handle;
1112 op_data->op_ioepoch = md.body->ioepoch;
1114 ll_update_inode(inode, &md);
1115 ptlrpc_req_finished(request);
1120 /* Close IO epoch and send Size-on-MDS attribute update. */
1121 static int ll_setattr_done_writing(struct inode *inode,
1122 struct md_op_data *op_data,
1123 struct md_open_data *mod)
1125 struct ll_inode_info *lli = ll_i2info(inode);
1129 LASSERT(op_data != NULL);
1130 if (!S_ISREG(inode->i_mode))
1133 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1134 op_data->op_ioepoch, PFID(&lli->lli_fid));
1136 op_data->op_flags = MF_EPOCH_CLOSE;
1137 ll_done_writing_attr(inode, op_data);
1138 ll_pack_inode2opdata(inode, op_data, NULL);
1140 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1141 if (rc == -EAGAIN) {
1142 /* MDS has instructed us to obtain Size-on-MDS attribute
1143 * from OSTs and send setattr to back to MDS. */
1144 rc = ll_som_update(inode, op_data);
1146 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1152 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1154 struct obd_capa *capa;
1157 if (attr->ia_valid & ATTR_SIZE)
1158 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1160 capa = ll_mdscapa_get(inode);
1162 rc = cl_setattr_ost(inode, attr, capa);
1164 if (attr->ia_valid & ATTR_SIZE)
1165 ll_truncate_free_capa(capa);
1172 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1173 * object(s) determine the file size and mtime. Otherwise, the MDS will
1174 * keep these values until such a time that objects are allocated for it.
1175 * We do the MDS operations first, as it is checking permissions for us.
1176 * We don't to the MDS RPC if there is nothing that we want to store there,
1177 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1178 * going to do an RPC anyways.
1180 * If we are doing a truncate, we will send the mtime and ctime updates
1181 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1182 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1185 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1187 struct ll_inode_info *lli = ll_i2info(inode);
1188 struct lov_stripe_md *lsm = lli->lli_smd;
1189 struct md_op_data *op_data = NULL;
1190 struct md_open_data *mod = NULL;
1191 int ia_valid = attr->ia_valid;
1192 int rc = 0, rc1 = 0;
1195 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1197 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1199 if (ia_valid & ATTR_SIZE) {
1200 if (attr->ia_size > ll_file_maxbytes(inode)) {
1201 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1202 attr->ia_size, ll_file_maxbytes(inode));
1206 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1209 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1210 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1211 if (cfs_curproc_fsuid() != inode->i_uid &&
1212 !cfs_capable(CFS_CAP_FOWNER))
1216 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1217 if (attr->ia_valid & ATTR_CTIME) {
1218 attr->ia_ctime = CFS_CURRENT_TIME;
1219 attr->ia_valid |= ATTR_CTIME_SET;
1221 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1222 attr->ia_atime = CFS_CURRENT_TIME;
1223 attr->ia_valid |= ATTR_ATIME_SET;
1225 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1226 attr->ia_mtime = CFS_CURRENT_TIME;
1227 attr->ia_valid |= ATTR_MTIME_SET;
1230 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1231 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1232 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1233 cfs_time_current_sec());
1235 /* NB: ATTR_SIZE will only be set after this point if the size
1236 * resides on the MDS, ie, this file has no objects. */
1238 attr->ia_valid &= ~ATTR_SIZE;
1240 /* We always do an MDS RPC, even if we're only changing the size;
1241 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1243 OBD_ALLOC_PTR(op_data);
1244 if (op_data == NULL)
1247 UNLOCK_INODE_MUTEX(inode);
1248 if (ia_valid & ATTR_SIZE)
1249 UP_WRITE_I_ALLOC_SEM(inode);
1250 cfs_down(&lli->lli_trunc_sem);
1251 LOCK_INODE_MUTEX(inode);
1252 if (ia_valid & ATTR_SIZE)
1253 DOWN_WRITE_I_ALLOC_SEM(inode);
1255 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1257 /* Open epoch for truncate. */
1258 if (exp_connect_som(ll_i2mdexp(inode)) &&
1259 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1260 op_data->op_flags = MF_EPOCH_OPEN;
1262 rc = ll_md_setattr(inode, op_data, &mod);
1266 ll_ioepoch_open(lli, op_data->op_ioepoch);
1267 if (!lsm || !S_ISREG(inode->i_mode)) {
1268 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1272 if (ia_valid & ATTR_SIZE)
1273 attr->ia_valid |= ATTR_SIZE;
1274 if ((ia_valid & ATTR_SIZE) |
1275 ((ia_valid | ATTR_ATIME | ATTR_ATIME_SET) &&
1276 LTIME_S(attr->ia_atime) < LTIME_S(attr->ia_ctime)) ||
1277 ((ia_valid | ATTR_MTIME | ATTR_MTIME_SET) &&
1278 LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)))
1279 /* perform truncate and setting mtime/atime to past under PW
1280 * 0:EOF extent lock (new_size:EOF for truncate) */
1281 rc = ll_setattr_ost(inode, attr);
1285 if (op_data->op_ioepoch)
1286 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1287 ll_finish_md_op_data(op_data);
1289 cfs_up(&lli->lli_trunc_sem);
1290 return rc ? rc : rc1;
1293 int ll_setattr(struct dentry *de, struct iattr *attr)
1295 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1296 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1297 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1299 if ((de->d_inode->i_mode & S_ISUID) &&
1300 !(attr->ia_mode & S_ISUID) &&
1301 !(attr->ia_valid & ATTR_KILL_SUID))
1302 attr->ia_valid |= ATTR_KILL_SUID;
1304 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1305 !(attr->ia_mode & S_ISGID) &&
1306 !(attr->ia_valid & ATTR_KILL_SGID))
1307 attr->ia_valid |= ATTR_KILL_SGID;
1309 return ll_setattr_raw(de->d_inode, attr);
1312 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1313 __u64 max_age, __u32 flags)
1315 struct ll_sb_info *sbi = ll_s2sbi(sb);
1316 struct obd_statfs obd_osfs;
1320 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1322 CERROR("md_statfs fails: rc = %d\n", rc);
1326 osfs->os_type = sb->s_magic;
1328 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1329 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1331 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1332 flags |= OBD_STATFS_NODELAY;
1334 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1335 &obd_osfs, max_age, flags);
1337 CERROR("obd_statfs fails: rc = %d\n", rc);
1341 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1342 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1345 osfs->os_bsize = obd_osfs.os_bsize;
1346 osfs->os_blocks = obd_osfs.os_blocks;
1347 osfs->os_bfree = obd_osfs.os_bfree;
1348 osfs->os_bavail = obd_osfs.os_bavail;
1350 /* If we don't have as many objects free on the OST as inodes
1351 * on the MDS, we reduce the total number of inodes to
1352 * compensate, so that the "inodes in use" number is correct.
1354 if (obd_osfs.os_ffree < osfs->os_ffree) {
1355 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1357 osfs->os_ffree = obd_osfs.os_ffree;
1362 #ifndef HAVE_STATFS_DENTRY_PARAM
1363 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1366 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1368 struct super_block *sb = de->d_sb;
1370 struct obd_statfs osfs;
1373 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1374 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1376 /* Some amount of caching on the client is allowed */
1377 rc = ll_statfs_internal(sb, &osfs,
1378 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1383 statfs_unpack(sfs, &osfs);
1385 /* We need to downshift for all 32-bit kernels, because we can't
1386 * tell if the kernel is being called via sys_statfs64() or not.
1387 * Stop before overflowing f_bsize - in which case it is better
1388 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1389 if (sizeof(long) < 8) {
1390 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1393 osfs.os_blocks >>= 1;
1394 osfs.os_bfree >>= 1;
1395 osfs.os_bavail >>= 1;
1399 sfs->f_blocks = osfs.os_blocks;
1400 sfs->f_bfree = osfs.os_bfree;
1401 sfs->f_bavail = osfs.os_bavail;
1406 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1408 struct ll_inode_info *lli;
1409 struct lov_stripe_md *lsm;
1411 lli = ll_i2info(inode);
1412 LASSERT(lli->lli_size_sem_owner != current);
1413 cfs_down(&lli->lli_size_sem);
1414 LASSERT(lli->lli_size_sem_owner == NULL);
1415 lli->lli_size_sem_owner = current;
1417 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1420 lov_stripe_lock(lsm);
1423 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1425 struct ll_inode_info *lli;
1426 struct lov_stripe_md *lsm;
1428 lli = ll_i2info(inode);
1430 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1433 lov_stripe_unlock(lsm);
1434 LASSERT(lli->lli_size_sem_owner == current);
1435 lli->lli_size_sem_owner = NULL;
1436 cfs_up(&lli->lli_size_sem);
1439 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1441 struct ll_inode_info *lli = ll_i2info(inode);
1442 struct mdt_body *body = md->body;
1443 struct lov_stripe_md *lsm = md->lsm;
1444 struct ll_sb_info *sbi = ll_i2sbi(inode);
1446 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1448 cfs_down(&lli->lli_och_sem);
1449 if (lli->lli_smd == NULL) {
1450 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1451 lsm->lsm_magic != LOV_MAGIC_V3) {
1452 dump_lsm(D_ERROR, lsm);
1455 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1456 lsm, inode->i_ino, inode->i_generation, inode);
1457 /* cl_inode_init must go before lli_smd or a race is
1458 * possible where client thinks the file has stripes,
1459 * but lov raid0 is not setup yet and parallel e.g.
1460 * glimpse would try to use uninitialized lov */
1461 cl_inode_init(inode, md);
1463 cfs_up(&lli->lli_och_sem);
1464 lli->lli_maxbytes = lsm->lsm_maxbytes;
1465 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1466 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1468 cfs_up(&lli->lli_och_sem);
1469 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1470 lli->lli_smd->lsm_stripe_count ==
1471 lsm->lsm_stripe_count);
1472 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1473 CERROR("lsm mismatch for inode %ld\n",
1475 CERROR("lli_smd:\n");
1476 dump_lsm(D_ERROR, lli->lli_smd);
1478 dump_lsm(D_ERROR, lsm);
1482 if (lli->lli_smd != lsm)
1483 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1486 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1487 if (body->valid & OBD_MD_FLRMTPERM)
1488 ll_update_remote_perm(inode, md->remote_perm);
1490 #ifdef CONFIG_FS_POSIX_ACL
1491 else if (body->valid & OBD_MD_FLACL) {
1492 cfs_spin_lock(&lli->lli_lock);
1493 if (lli->lli_posix_acl)
1494 posix_acl_release(lli->lli_posix_acl);
1495 lli->lli_posix_acl = md->posix_acl;
1496 cfs_spin_unlock(&lli->lli_lock);
1499 inode->i_ino = cl_fid_build_ino(&body->fid1);
1500 inode->i_generation = cl_fid_build_gen(&body->fid1);
1502 if (body->valid & OBD_MD_FLATIME) {
1503 if (body->atime > LTIME_S(inode->i_atime))
1504 LTIME_S(inode->i_atime) = body->atime;
1505 lli->lli_lvb.lvb_atime = body->atime;
1507 if (body->valid & OBD_MD_FLMTIME) {
1508 if (body->mtime > LTIME_S(inode->i_mtime)) {
1509 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1510 "to "LPU64"\n", inode->i_ino,
1511 LTIME_S(inode->i_mtime), body->mtime);
1512 LTIME_S(inode->i_mtime) = body->mtime;
1514 lli->lli_lvb.lvb_mtime = body->mtime;
1516 if (body->valid & OBD_MD_FLCTIME) {
1517 if (body->ctime > LTIME_S(inode->i_ctime))
1518 LTIME_S(inode->i_ctime) = body->ctime;
1519 lli->lli_lvb.lvb_ctime = body->ctime;
1521 if (body->valid & OBD_MD_FLMODE)
1522 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1523 if (body->valid & OBD_MD_FLTYPE)
1524 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1525 LASSERT(inode->i_mode != 0);
1526 if (S_ISREG(inode->i_mode)) {
1527 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1529 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1531 #ifdef HAVE_INODE_BLKSIZE
1532 inode->i_blksize = 1<<inode->i_blkbits;
1534 if (body->valid & OBD_MD_FLUID)
1535 inode->i_uid = body->uid;
1536 if (body->valid & OBD_MD_FLGID)
1537 inode->i_gid = body->gid;
1538 if (body->valid & OBD_MD_FLFLAGS)
1539 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1540 if (body->valid & OBD_MD_FLNLINK)
1541 inode->i_nlink = body->nlink;
1542 if (body->valid & OBD_MD_FLRDEV)
1543 inode->i_rdev = old_decode_dev(body->rdev);
1545 if (body->valid & OBD_MD_FLID) {
1546 /* FID shouldn't be changed! */
1547 if (fid_is_sane(&lli->lli_fid)) {
1548 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1549 "Trying to change FID "DFID
1550 " to the "DFID", inode %lu/%u(%p)\n",
1551 PFID(&lli->lli_fid), PFID(&body->fid1),
1552 inode->i_ino, inode->i_generation, inode);
1554 lli->lli_fid = body->fid1;
1557 LASSERT(fid_seq(&lli->lli_fid) != 0);
1559 if (body->valid & OBD_MD_FLSIZE) {
1560 if (exp_connect_som(ll_i2mdexp(inode)) &&
1561 S_ISREG(inode->i_mode) && lli->lli_smd) {
1562 struct lustre_handle lockh;
1565 /* As it is possible a blocking ast has been processed
1566 * by this time, we need to check there is an UPDATE
1567 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1569 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1572 if (lli->lli_flags & (LLIF_DONE_WRITING |
1573 LLIF_EPOCH_PENDING |
1575 CERROR("ino %lu flags %lu still has "
1576 "size authority! do not trust "
1577 "the size got from MDS\n",
1578 inode->i_ino, lli->lli_flags);
1580 /* Use old size assignment to avoid
1581 * deadlock bz14138 & bz14326 */
1582 inode->i_size = body->size;
1583 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1585 ldlm_lock_decref(&lockh, mode);
1588 /* Use old size assignment to avoid
1589 * deadlock bz14138 & bz14326 */
1590 inode->i_size = body->size;
1592 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1593 inode->i_ino, body->size);
1596 if (body->valid & OBD_MD_FLBLOCKS)
1597 inode->i_blocks = body->blocks;
1600 if (body->valid & OBD_MD_FLMDSCAPA) {
1601 LASSERT(md->mds_capa);
1602 ll_add_capa(inode, md->mds_capa);
1604 if (body->valid & OBD_MD_FLOSSCAPA) {
1605 LASSERT(md->oss_capa);
1606 ll_add_capa(inode, md->oss_capa);
1610 static struct backing_dev_info ll_backing_dev_info = {
1611 .ra_pages = 0, /* No readahead */
1612 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1613 .capabilities = 0, /* Does contribute to dirty memory */
1615 .memory_backed = 0, /* Does contribute to dirty memory */
1619 void ll_read_inode2(struct inode *inode, void *opaque)
1621 struct lustre_md *md = opaque;
1622 struct ll_inode_info *lli = ll_i2info(inode);
1625 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1626 inode->i_ino, inode->i_generation, inode);
1630 LASSERT(!lli->lli_smd);
1632 /* Core attributes from the MDS first. This is a new inode, and
1633 * the VFS doesn't zero times in the core inode so we have to do
1634 * it ourselves. They will be overwritten by either MDS or OST
1635 * attributes - we just need to make sure they aren't newer. */
1636 LTIME_S(inode->i_mtime) = 0;
1637 LTIME_S(inode->i_atime) = 0;
1638 LTIME_S(inode->i_ctime) = 0;
1640 ll_update_inode(inode, md);
1642 /* OIDEBUG(inode); */
1644 if (S_ISREG(inode->i_mode)) {
1645 struct ll_sb_info *sbi = ll_i2sbi(inode);
1646 inode->i_op = &ll_file_inode_operations;
1647 inode->i_fop = sbi->ll_fop;
1648 inode->i_mapping->a_ops = &ll_aops;
1650 } else if (S_ISDIR(inode->i_mode)) {
1651 inode->i_op = &ll_dir_inode_operations;
1652 inode->i_fop = &ll_dir_operations;
1653 inode->i_mapping->a_ops = &ll_dir_aops;
1655 } else if (S_ISLNK(inode->i_mode)) {
1656 inode->i_op = &ll_fast_symlink_inode_operations;
1659 inode->i_op = &ll_special_inode_operations;
1661 init_special_inode(inode, inode->i_mode,
1662 kdev_t_to_nr(inode->i_rdev));
1664 /* initializing backing dev info. */
1665 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1671 void ll_delete_inode(struct inode *inode)
1673 struct ll_sb_info *sbi = ll_i2sbi(inode);
1677 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1679 CERROR("fid_delete() failed, rc %d\n", rc);
1681 truncate_inode_pages(&inode->i_data, 0);
1687 int ll_iocontrol(struct inode *inode, struct file *file,
1688 unsigned int cmd, unsigned long arg)
1690 struct ll_sb_info *sbi = ll_i2sbi(inode);
1691 struct ptlrpc_request *req = NULL;
1696 case FSFILT_IOC_GETFLAGS: {
1697 struct mdt_body *body;
1698 struct md_op_data *op_data;
1700 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1701 0, 0, LUSTRE_OPC_ANY,
1703 if (op_data == NULL)
1706 op_data->op_valid = OBD_MD_FLFLAGS;
1707 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1708 ll_finish_md_op_data(op_data);
1710 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1714 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1716 flags = body->flags;
1718 ptlrpc_req_finished(req);
1720 RETURN(put_user(flags, (int *)arg));
1722 case FSFILT_IOC_SETFLAGS: {
1723 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1724 struct obd_info oinfo = { { { 0 } } };
1725 struct md_op_data *op_data;
1727 if (get_user(flags, (int *)arg))
1731 OBDO_ALLOC(oinfo.oi_oa);
1735 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1736 LUSTRE_OPC_ANY, NULL);
1737 if (IS_ERR(op_data))
1738 RETURN(PTR_ERR(op_data));
1740 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1741 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1742 rc = md_setattr(sbi->ll_md_exp, op_data,
1743 NULL, 0, NULL, 0, &req, NULL);
1744 ll_finish_md_op_data(op_data);
1745 ptlrpc_req_finished(req);
1747 OBDO_FREE(oinfo.oi_oa);
1752 OBDO_FREE(oinfo.oi_oa);
1753 GOTO(update_cache, rc);
1756 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1757 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1758 oinfo.oi_oa->o_flags = flags;
1759 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1761 oinfo.oi_capa = ll_mdscapa_get(inode);
1762 obdo_from_inode(oinfo.oi_oa, inode,
1763 &ll_i2info(inode)->lli_fid, 0);
1764 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1765 capa_put(oinfo.oi_capa);
1766 OBDO_FREE(oinfo.oi_oa);
1768 if (rc != -EPERM && rc != -EACCES)
1769 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1775 inode->i_flags = ll_ext_to_inode_flags(flags);
1785 int ll_flush_ctx(struct inode *inode)
1787 struct ll_sb_info *sbi = ll_i2sbi(inode);
1789 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1791 obd_set_info_async(sbi->ll_md_exp,
1792 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1794 obd_set_info_async(sbi->ll_dt_exp,
1795 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1800 /* umount -f client means force down, don't save state */
1801 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1802 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1804 struct super_block *sb = vfsmnt->mnt_sb;
1806 void ll_umount_begin(struct super_block *sb)
1809 struct lustre_sb_info *lsi = s2lsi(sb);
1810 struct ll_sb_info *sbi = ll_s2sbi(sb);
1811 struct obd_device *obd;
1812 struct obd_ioctl_data *ioc_data;
1815 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1816 if (!(flags & MNT_FORCE)) {
1822 /* Tell the MGC we got umount -f */
1823 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1825 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1826 sb->s_count, atomic_read(&sb->s_active));
1828 obd = class_exp2obd(sbi->ll_md_exp);
1830 CERROR("Invalid MDC connection handle "LPX64"\n",
1831 sbi->ll_md_exp->exp_handle.h_cookie);
1837 obd = class_exp2obd(sbi->ll_dt_exp);
1839 CERROR("Invalid LOV connection handle "LPX64"\n",
1840 sbi->ll_dt_exp->exp_handle.h_cookie);
1846 OBD_ALLOC_PTR(ioc_data);
1848 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1849 sizeof ioc_data, ioc_data, NULL);
1851 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1852 sizeof ioc_data, ioc_data, NULL);
1854 OBD_FREE_PTR(ioc_data);
1858 /* Really, we'd like to wait until there are no requests outstanding,
1859 * and then continue. For now, we just invalidate the requests,
1860 * schedule() and sleep one second if needed, and hope.
1863 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1864 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1865 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1866 cfs_time_seconds(1));
1867 if (atomic_read(&vfsmnt->mnt_count) > 2)
1868 LCONSOLE_WARN("Mount still busy with %d refs! You "
1869 "may try to umount it a bit later\n",
1870 atomic_read(&vfsmnt->mnt_count));
1877 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1879 struct ll_sb_info *sbi = ll_s2sbi(sb);
1883 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1884 read_only = *flags & MS_RDONLY;
1885 err = obd_set_info_async(sbi->ll_md_exp,
1886 sizeof(KEY_READ_ONLY),
1887 KEY_READ_ONLY, sizeof(read_only),
1890 CERROR("Failed to change the read-only flag during "
1891 "remount: %d\n", err);
1896 sb->s_flags |= MS_RDONLY;
1898 sb->s_flags &= ~MS_RDONLY;
1903 int ll_prep_inode(struct inode **inode,
1904 struct ptlrpc_request *req,
1905 struct super_block *sb)
1907 struct ll_sb_info *sbi = NULL;
1908 struct lustre_md md;
1912 LASSERT(*inode || sb);
1913 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1914 memset(&md, 0, sizeof(struct lustre_md));
1916 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1917 sbi->ll_md_exp, &md);
1922 ll_update_inode(*inode, &md);
1924 LASSERT(sb != NULL);
1927 * At this point server returns to client's same fid as client
1928 * generated for creating. So using ->fid1 is okay here.
1930 LASSERT(fid_is_sane(&md.body->fid1));
1932 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
1933 if (*inode == NULL || IS_ERR(*inode)) {
1935 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1936 #ifdef CONFIG_FS_POSIX_ACL
1938 posix_acl_release(md.posix_acl);
1939 md.posix_acl = NULL;
1942 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
1944 CERROR("new_inode -fatal: rc %d\n", rc);
1950 md_free_lustre_md(sbi->ll_md_exp, &md);
1954 int ll_obd_statfs(struct inode *inode, void *arg)
1956 struct ll_sb_info *sbi = NULL;
1957 struct obd_export *exp;
1959 struct obd_ioctl_data *data = NULL;
1963 if (!inode || !(sbi = ll_i2sbi(inode)))
1964 GOTO(out_statfs, rc = -EINVAL);
1966 rc = obd_ioctl_getdata(&buf, &len, arg);
1968 GOTO(out_statfs, rc);
1971 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1972 !data->ioc_pbuf1 || !data->ioc_pbuf2)
1973 GOTO(out_statfs, rc = -EINVAL);
1975 if (data->ioc_inllen1 != sizeof(__u32) ||
1976 data->ioc_inllen2 != sizeof(__u32) ||
1977 data->ioc_plen1 != sizeof(struct obd_statfs) ||
1978 data->ioc_plen2 != sizeof(struct obd_uuid))
1979 GOTO(out_statfs, rc = -EINVAL);
1981 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
1982 if (type == LL_STATFS_MDC)
1983 exp = sbi->ll_md_exp;
1984 else if (type == LL_STATFS_LOV)
1985 exp = sbi->ll_dt_exp;
1987 GOTO(out_statfs, rc = -ENODEV);
1989 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
1991 GOTO(out_statfs, rc);
1994 obd_ioctl_freedata(buf, len);
1998 int ll_process_config(struct lustre_cfg *lcfg)
2002 struct lprocfs_static_vars lvars;
2006 lprocfs_llite_init_vars(&lvars);
2008 /* The instance name contains the sb: lustre-client-aacfe000 */
2009 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2010 if (!ptr || !*(++ptr))
2012 if (sscanf(ptr, "%lx", &x) != 1)
2015 /* This better be a real Lustre superblock! */
2016 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2018 /* Note we have not called client_common_fill_super yet, so
2019 proc fns must be able to handle that! */
2020 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2027 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2028 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2029 struct inode *i1, struct inode *i2,
2030 const char *name, int namelen,
2031 int mode, __u32 opc, void *data)
2033 LASSERT(i1 != NULL);
2035 if (namelen > ll_i2sbi(i1)->ll_namelen)
2036 return ERR_PTR(-ENAMETOOLONG);
2038 if (op_data == NULL)
2039 OBD_ALLOC_PTR(op_data);
2041 if (op_data == NULL)
2042 return ERR_PTR(-ENOMEM);
2044 ll_i2gids(op_data->op_suppgids, i1, i2);
2045 op_data->op_fid1 = *ll_inode2fid(i1);
2046 op_data->op_capa1 = ll_mdscapa_get(i1);
2049 op_data->op_fid2 = *ll_inode2fid(i2);
2050 op_data->op_capa2 = ll_mdscapa_get(i2);
2052 fid_zero(&op_data->op_fid2);
2053 op_data->op_capa2 = NULL;
2056 op_data->op_name = name;
2057 op_data->op_namelen = namelen;
2058 op_data->op_mode = mode;
2059 op_data->op_mod_time = cfs_time_current_sec();
2060 op_data->op_fsuid = cfs_curproc_fsuid();
2061 op_data->op_fsgid = cfs_curproc_fsgid();
2062 op_data->op_cap = cfs_curproc_cap_pack();
2063 op_data->op_bias = MDS_CHECK_SPLIT;
2064 op_data->op_opc = opc;
2065 op_data->op_mds = 0;
2066 op_data->op_data = data;
2071 void ll_finish_md_op_data(struct md_op_data *op_data)
2073 capa_put(op_data->op_capa1);
2074 capa_put(op_data->op_capa2);
2075 OBD_FREE_PTR(op_data);
2078 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2080 struct ll_sb_info *sbi;
2082 LASSERT((seq != NULL) && (vfs != NULL));
2083 sbi = ll_s2sbi(vfs->mnt_sb);
2085 if (sbi->ll_flags & LL_SBI_NOLCK)
2086 seq_puts(seq, ",nolock");
2088 if (sbi->ll_flags & LL_SBI_FLOCK)
2089 seq_puts(seq, ",flock");
2091 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2092 seq_puts(seq, ",localflock");
2094 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2095 seq_puts(seq, ",user_xattr");
2097 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2098 seq_puts(seq, ",lazystatfs");