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 2008 Sun Microsystems, Inc. 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 <obd_cksum.h>
57 #include <lustre_cache.h>
58 #include "llite_internal.h"
60 cfs_mem_cache_t *ll_file_data_slab;
62 LIST_HEAD(ll_super_blocks);
63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) ffz(~(n))
72 static inline void ll_pglist_fini(struct ll_sb_info *sbi)
77 if (sbi->ll_pglist == NULL)
80 for_each_possible_cpu(i) {
81 page = sbi->ll_pglist[i]->llpd_page;
83 sbi->ll_pglist[i] = NULL;
88 OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
89 sbi->ll_pglist = NULL;
92 static inline int ll_pglist_init(struct ll_sb_info *sbi)
94 struct ll_pglist_data *pd;
99 OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
100 if (sbi->ll_pglist == NULL)
103 budget = sbi->ll_async_page_max / num_online_cpus();
104 for_each_possible_cpu(i) {
105 struct page *page = alloc_pages_node(cpu_to_node(i),
112 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
115 pd = (struct ll_pglist_data *)(page_address(page) + color);
116 memset(pd, 0, sizeof(*pd));
117 spin_lock_init(&pd->llpd_lock);
118 INIT_LIST_HEAD(&pd->llpd_list);
120 pd->llpd_budget = budget;
122 pd->llpd_page = page;
123 atomic_set(&pd->llpd_sample_count, 0);
124 sbi->ll_pglist[i] = pd;
125 color += L1_CACHE_ALIGN(sizeof(*pd));
131 static struct ll_sb_info *ll_init_sbi(void)
133 struct ll_sb_info *sbi = NULL;
140 OBD_ALLOC(sbi, sizeof(*sbi));
144 OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
145 if (sbi->ll_async_page_sample == NULL)
148 spin_lock_init(&sbi->ll_lock);
149 spin_lock_init(&sbi->ll_lco.lco_lock);
150 spin_lock_init(&sbi->ll_pp_extent_lock);
151 spin_lock_init(&sbi->ll_process_lock);
152 sbi->ll_rw_stats_on = 0;
155 pages = si.totalram - si.totalhigh;
156 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
157 #ifdef HAVE_BGL_SUPPORT
158 sbi->ll_async_page_max = pages / 4;
160 sbi->ll_async_page_max = pages / 2;
163 sbi->ll_async_page_max = (pages / 4) * 3;
165 lcounter_init(&sbi->ll_async_page_count);
166 spin_lock_init(&sbi->ll_async_page_reblnc_lock);
167 sbi->ll_async_page_sample_max = 64 * num_online_cpus();
168 sbi->ll_async_page_reblnc_count = 0;
169 sbi->ll_async_page_clock_hand = 0;
170 if (ll_pglist_init(sbi))
173 sbi->ll_ra_info.ra_max_pages = min(pages / 32,
174 SBI_DEFAULT_READAHEAD_MAX);
175 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
176 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
177 sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
178 sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
179 INIT_LIST_HEAD(&sbi->ll_conn_chain);
180 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
182 ll_generate_random_uuid(uuid);
183 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
184 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
186 spin_lock(&ll_sb_lock);
187 list_add_tail(&sbi->ll_list, &ll_super_blocks);
188 spin_unlock(&ll_sb_lock);
190 #ifdef ENABLE_LLITE_CHECKSUM
191 sbi->ll_flags |= LL_SBI_CHECKSUM;
194 #ifdef HAVE_LRU_RESIZE_SUPPORT
195 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
198 #ifdef HAVE_EXPORT___IGET
199 INIT_LIST_HEAD(&sbi->ll_deathrow);
200 spin_lock_init(&sbi->ll_deathrow_lock);
202 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
203 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
204 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
207 /* metadata statahead is enabled by default */
208 sbi->ll_sa_max = LL_SA_RPC_DEF;
213 if (sbi->ll_async_page_sample)
214 OBD_FREE(sbi->ll_async_page_sample,
215 sizeof(long) * num_possible_cpus());
217 OBD_FREE(sbi, sizeof(*sbi));
221 void ll_free_sbi(struct super_block *sb)
223 struct ll_sb_info *sbi = ll_s2sbi(sb);
228 spin_lock(&ll_sb_lock);
229 list_del(&sbi->ll_list);
230 spin_unlock(&ll_sb_lock);
231 lcounter_destroy(&sbi->ll_async_page_count);
232 OBD_FREE(sbi->ll_async_page_sample,
233 sizeof(long) * num_possible_cpus());
234 OBD_FREE(sbi, sizeof(*sbi));
239 static struct dentry_operations ll_d_root_ops = {
240 #ifdef DCACHE_LUSTRE_INVALID
241 .d_compare = ll_dcompare,
245 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
246 * us to make MDS RPCs with large enough reply buffers to hold the
247 * maximum-sized (= maximum striped) EA and cookie without having to
248 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
249 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
251 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
252 __u32 valsize = sizeof(struct lov_desc);
253 int rc, easize, def_easize, cookiesize;
254 struct lov_desc desc;
258 rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
259 &valsize, &desc, NULL);
263 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
264 lsm.lsm_stripe_count = stripes;
265 easize = obd_size_diskmd(dt_exp, &lsm);
267 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
268 def_easize = obd_size_diskmd(dt_exp, &lsm);
270 cookiesize = stripes * sizeof(struct llog_cookie);
272 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
275 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
279 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
281 struct inode *root = 0;
282 struct ll_sb_info *sbi = ll_s2sbi(sb);
283 struct obd_device *obd;
284 struct lu_fid rootfid;
285 struct obd_capa *oc = NULL;
286 struct obd_statfs osfs;
287 struct ptlrpc_request *request = NULL;
288 struct lustre_handle dt_conn = {0, };
289 struct lustre_handle md_conn = {0, };
290 struct obd_connect_data *data = NULL;
291 struct lustre_md lmd;
293 int size, err, checksum;
296 obd = class_name2obd(md);
298 CERROR("MD %s: not setup or attached\n", md);
306 if (proc_lustre_fs_root) {
307 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
310 CERROR("could not register mount in /proc/fs/lustre\n");
313 /* indicate the features supported by this client */
314 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
315 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
316 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
317 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
318 OBD_CONNECT_FID | OBD_CONNECT_AT;
320 #ifdef HAVE_LRU_RESIZE_SUPPORT
321 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
322 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
324 #ifdef CONFIG_FS_POSIX_ACL
325 data->ocd_connect_flags |= OBD_CONNECT_ACL;
327 data->ocd_ibits_known = MDS_INODELOCK_FULL;
328 data->ocd_version = LUSTRE_VERSION_CODE;
330 if (sb->s_flags & MS_RDONLY)
331 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
332 if (sbi->ll_flags & LL_SBI_USER_XATTR)
333 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
335 #ifdef HAVE_MS_FLOCK_LOCK
336 /* force vfs to use lustre handler for flock() calls - bug 10743 */
337 sb->s_flags |= MS_FLOCK_LOCK;
340 if (sbi->ll_flags & LL_SBI_FLOCK)
341 sbi->ll_fop = &ll_file_operations_flock;
342 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
343 sbi->ll_fop = &ll_file_operations;
345 sbi->ll_fop = &ll_file_operations_noflock;
348 data->ocd_connect_flags |= OBD_CONNECT_REAL;
349 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
350 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
351 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
353 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
354 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
357 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
359 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
360 "recovery, of which this client is not a "
361 "part. Please wait for recovery to complete,"
362 " abort, or time out.\n", md);
365 CERROR("cannot connect to %s: rc = %d\n", md, err);
368 sbi->ll_md_exp = class_conn2export(&md_conn);
370 err = obd_fid_init(sbi->ll_md_exp);
372 CERROR("Can't init metadata layer FID infrastructure, "
377 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
379 GOTO(out_md_fid, err);
381 size = sizeof(*data);
382 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
383 KEY_CONN_DATA, &size, data, NULL);
385 CERROR("Get connect data failed: %d \n", err);
389 LASSERT(osfs.os_bsize);
390 sb->s_blocksize = osfs.os_bsize;
391 sb->s_blocksize_bits = log2(osfs.os_bsize);
392 sb->s_magic = LL_SUPER_MAGIC;
394 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
395 * retval = in_file->f_op->sendfile(...);
397 * retval = -EOVERFLOW;
399 * it will check if *ppos is greater than max. However, max equals to
400 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
401 * has been defined as a signed long long ineger in linux kernel. */
402 #if BITS_PER_LONG == 64
403 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
405 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
407 sbi->ll_namelen = osfs.os_namelen;
408 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
410 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
411 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
412 LCONSOLE_INFO("Disabling user_xattr feature because "
413 "it is not supported on the server\n");
414 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
417 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
419 sb->s_flags |= MS_POSIXACL;
421 sbi->ll_flags |= LL_SBI_ACL;
423 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
425 sb->s_flags &= ~MS_POSIXACL;
427 sbi->ll_flags &= ~LL_SBI_ACL;
430 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
431 sbi->ll_flags |= LL_SBI_JOIN;
433 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
434 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
435 /* sometimes local client claims to be remote, but mdt
436 * will disagree when client gss not applied. */
437 LCONSOLE_INFO("client claims to be remote, but server "
438 "rejected, forced to be local.\n");
439 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
442 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
443 /* with gss applied, remote client can not claim to be
444 * local, so mdt maybe force client to be remote. */
445 LCONSOLE_INFO("client claims to be local, but server "
446 "rejected, forced to be remote.\n");
447 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
451 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
452 LCONSOLE_INFO("client enabled MDS capability!\n");
453 sbi->ll_flags |= LL_SBI_MDS_CAPA;
456 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
457 LCONSOLE_INFO("client enabled OSS capability!\n");
458 sbi->ll_flags |= LL_SBI_OSS_CAPA;
461 obd = class_name2obd(dt);
463 CERROR("DT %s: not setup or attached\n", dt);
464 GOTO(out_md_fid, err = -ENODEV);
467 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
468 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
469 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
470 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
472 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
473 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
475 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
476 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
477 * disabled by default, because it can still be enabled on the
478 * fly via /proc. As a consequence, we still need to come to an
479 * agreement on the supported algorithms at connect time */
480 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
482 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
483 data->ocd_cksum_types = OBD_CKSUM_ADLER;
485 /* send the list of supported checksum types */
486 data->ocd_cksum_types = OBD_CKSUM_ALL;
489 #ifdef HAVE_LRU_RESIZE_SUPPORT
490 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
492 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
493 "ocd_grant: %d\n", data->ocd_connect_flags,
494 data->ocd_version, data->ocd_grant);
496 obd->obd_upcall.onu_owner = &sbi->ll_lco;
497 obd->obd_upcall.onu_upcall = ll_ocd_update;
498 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
500 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
502 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
503 "recovery, of which this client is not a "
504 "part. Please wait for recovery to "
505 "complete, abort, or time out.\n", dt);
506 GOTO(out_md_fid, err);
508 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
509 GOTO(out_md_fid, err);
512 sbi->ll_dt_exp = class_conn2export(&dt_conn);
514 err = obd_fid_init(sbi->ll_dt_exp);
516 CERROR("Can't init data layer FID infrastructure, "
521 spin_lock(&sbi->ll_lco.lco_lock);
522 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
523 spin_unlock(&sbi->ll_lco.lco_lock);
525 err = obd_register_page_removal_cb(sbi->ll_dt_exp,
529 CERROR("cannot register page removal callback: rc = %d\n",err);
532 err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
533 ll_extent_lock_cancel_cb);
535 CERROR("cannot register lock cancel callback: rc = %d\n", err);
536 GOTO(out_page_rm_cb, err);
539 err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
541 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
542 GOTO(out_lock_cn_cb, err);
545 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
546 0, NULL, NULL, NULL, 0, NULL);
548 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
549 "filesystem. There must be at least one "
550 "active OST for a client to start.\n");
551 GOTO(out_lock_cn_cb, err);
554 if (!ll_async_page_slab) {
555 ll_async_page_slab_size =
556 size_round(sizeof(struct ll_async_page)) + err;
557 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
558 ll_async_page_slab_size,
560 if (!ll_async_page_slab)
561 GOTO(out_lock_cn_cb, err = -ENOMEM);
564 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
566 CERROR("cannot mds_connect: rc = %d\n", err);
567 GOTO(out_lock_cn_cb, err);
569 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
570 sbi->ll_root_fid = rootfid;
572 sb->s_op = &lustre_super_operations;
573 sb->s_export_op = &lustre_export_operations;
576 * XXX: move this to after cbd setup? */
577 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
578 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
579 valid |= OBD_MD_FLRMTPERM;
580 else if (sbi->ll_flags & LL_SBI_ACL)
581 valid |= OBD_MD_FLACL;
583 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
587 CERROR("md_getattr failed for root: rc = %d\n", err);
588 GOTO(out_lock_cn_cb, err);
590 memset(&lmd, 0, sizeof(lmd));
591 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
592 sbi->ll_md_exp, &lmd);
594 CERROR("failed to understand root inode md: rc = %d\n", err);
595 ptlrpc_req_finished (request);
596 GOTO(out_lock_cn_cb, err);
599 LASSERT(fid_is_sane(&sbi->ll_root_fid));
600 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
601 md_free_lustre_md(sbi->ll_md_exp, &lmd);
602 ptlrpc_req_finished(request);
604 if (root == NULL || is_bad_inode(root)) {
606 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
607 #ifdef CONFIG_FS_POSIX_ACL
609 posix_acl_release(lmd.posix_acl);
610 lmd.posix_acl = NULL;
613 CERROR("lustre_lite: bad iget4 for root\n");
614 GOTO(out_root, err = -EBADF);
617 err = ll_close_thread_start(&sbi->ll_lcq);
619 CERROR("cannot start close thread: rc %d\n", err);
623 #ifdef CONFIG_FS_POSIX_ACL
624 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
625 rct_init(&sbi->ll_rct);
626 et_init(&sbi->ll_et);
630 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
631 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
632 KEY_CHECKSUM, sizeof(checksum), &checksum,
635 sb->s_root = d_alloc_root(root);
637 OBD_FREE(data, sizeof(*data));
639 sb->s_root->d_op = &ll_d_root_ops;
641 sbi->ll_sdev_orig = sb->s_dev;
643 /* We set sb->s_dev equal on all lustre clients in order to support
644 * NFS export clustering. NFSD requires that the FSID be the same
646 /* s_dev is also used in lt_compare() to compare two fs, but that is
647 * only a node-local comparison. */
649 /* XXX: this will not work with LMV */
650 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
651 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
659 obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
660 ll_extent_lock_cancel_cb);
662 obd_unregister_page_removal_cb(sbi->ll_dt_exp,
664 obd_fid_fini(sbi->ll_dt_exp);
666 obd_disconnect(sbi->ll_dt_exp);
667 sbi->ll_dt_exp = NULL;
669 obd_fid_fini(sbi->ll_md_exp);
671 obd_disconnect(sbi->ll_md_exp);
672 sbi->ll_md_exp = NULL;
676 lprocfs_unregister_mountpoint(sbi);
680 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
684 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
686 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
687 KEY_MAX_EASIZE, &size, lmmsize, NULL);
689 CERROR("Get max mdsize error rc %d \n", rc);
694 void ll_dump_inode(struct inode *inode)
696 struct list_head *tmp;
697 int dentry_count = 0;
699 LASSERT(inode != NULL);
701 list_for_each(tmp, &inode->i_dentry)
704 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
705 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
706 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
709 void lustre_dump_dentry(struct dentry *dentry, int recur)
711 struct list_head *tmp;
714 LASSERT(dentry != NULL);
716 list_for_each(tmp, &dentry->d_subdirs)
719 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
720 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
721 dentry->d_name.len, dentry->d_name.name,
722 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
723 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
724 dentry->d_flags, dentry->d_fsdata, subdirs);
725 if (dentry->d_inode != NULL)
726 ll_dump_inode(dentry->d_inode);
731 list_for_each(tmp, &dentry->d_subdirs) {
732 struct dentry *d = list_entry(tmp, struct dentry, d_child);
733 lustre_dump_dentry(d, recur - 1);
737 #ifdef HAVE_EXPORT___IGET
738 static void prune_dir_dentries(struct inode *inode)
740 struct dentry *dentry, *prev = NULL;
742 /* due to lustre specific logic, a directory
743 * can have few dentries - a bug from VFS POV */
745 spin_lock(&dcache_lock);
746 if (!list_empty(&inode->i_dentry)) {
747 dentry = list_entry(inode->i_dentry.prev,
748 struct dentry, d_alias);
749 /* in order to prevent infinite loops we
750 * break if previous dentry is busy */
751 if (dentry != prev) {
754 spin_unlock(&dcache_lock);
756 /* try to kill all child dentries */
758 shrink_dcache_parent(dentry);
759 unlock_dentry(dentry);
762 /* now try to get rid of current dentry */
763 d_prune_aliases(inode);
767 spin_unlock(&dcache_lock);
770 static void prune_deathrow_one(struct ll_inode_info *lli)
772 struct inode *inode = ll_info2i(lli);
774 /* first, try to drop any dentries - they hold a ref on the inode */
775 if (S_ISDIR(inode->i_mode))
776 prune_dir_dentries(inode);
778 d_prune_aliases(inode);
781 /* if somebody still uses it, leave it */
782 LASSERT(atomic_read(&inode->i_count) > 0);
783 if (atomic_read(&inode->i_count) > 1)
786 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
787 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
789 /* seems nobody uses it anymore */
797 static void prune_deathrow(struct ll_sb_info *sbi, int try)
799 struct ll_inode_info *lli;
803 if (need_resched() && try)
807 if (!spin_trylock(&sbi->ll_deathrow_lock))
810 spin_lock(&sbi->ll_deathrow_lock);
815 if (!list_empty(&sbi->ll_deathrow)) {
816 lli = list_entry(sbi->ll_deathrow.next,
817 struct ll_inode_info,
819 list_del_init(&lli->lli_dead_list);
820 if (!list_empty(&sbi->ll_deathrow))
823 spin_unlock(&sbi->ll_deathrow_lock);
826 prune_deathrow_one(lli);
828 } while (empty == 0);
830 #else /* !HAVE_EXPORT___IGET */
831 #define prune_deathrow(sbi, try) do {} while (0)
832 #endif /* HAVE_EXPORT___IGET */
834 void client_common_put_super(struct super_block *sb)
836 struct ll_sb_info *sbi = ll_s2sbi(sb);
839 #ifdef CONFIG_FS_POSIX_ACL
840 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
841 et_fini(&sbi->ll_et);
842 rct_fini(&sbi->ll_rct);
846 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
848 ll_close_thread_shutdown(sbi->ll_lcq);
850 /* destroy inodes in deathrow */
851 prune_deathrow(sbi, 0);
853 list_del(&sbi->ll_conn_chain);
855 obd_fid_fini(sbi->ll_dt_exp);
856 obd_disconnect(sbi->ll_dt_exp);
857 sbi->ll_dt_exp = NULL;
859 lprocfs_unregister_mountpoint(sbi);
861 obd_fid_fini(sbi->ll_md_exp);
862 obd_disconnect(sbi->ll_md_exp);
863 sbi->ll_md_exp = NULL;
868 void ll_kill_super(struct super_block *sb)
870 struct ll_sb_info *sbi;
875 if (!(sb->s_flags & MS_ACTIVE))
879 /* we need restore s_dev from changed for clustred NFS before put_super
880 * because new kernels have cached s_dev and change sb->s_dev in
881 * put_super not affected real removing devices */
883 sb->s_dev = sbi->ll_sdev_orig;
887 char *ll_read_opt(const char *opt, char *data)
893 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
894 if (strncmp(opt, data, strlen(opt)))
896 if ((value = strchr(data, '=')) == NULL)
900 OBD_ALLOC(retval, strlen(value) + 1);
902 CERROR("out of memory!\n");
906 memcpy(retval, value, strlen(value)+1);
907 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
911 static inline int ll_set_opt(const char *opt, char *data, int fl)
913 if (strncmp(opt, data, strlen(opt)) != 0)
919 /* non-client-specific mount options are parsed in lmd_parse */
920 static int ll_options(char *options, int *flags)
923 char *s1 = options, *s2;
929 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
932 CDEBUG(D_SUPER, "next opt=%s\n", s1);
933 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
938 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
943 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
948 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
953 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
958 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
963 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
965 /* Ignore deprecated mount option. The client will
966 * always try to mount with ACL support, whether this
967 * is used depends on whether server supports it. */
970 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
974 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
980 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
985 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
990 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
995 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
1001 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1007 s2 = strchr(s1, ',');
1015 void ll_lli_init(struct ll_inode_info *lli)
1017 lli->lli_inode_magic = LLI_INODE_MAGIC;
1018 sema_init(&lli->lli_size_sem, 1);
1019 sema_init(&lli->lli_write_sem, 1);
1021 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1022 spin_lock_init(&lli->lli_lock);
1023 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
1024 INIT_LIST_HEAD(&lli->lli_close_list);
1025 lli->lli_inode_magic = LLI_INODE_MAGIC;
1026 sema_init(&lli->lli_och_sem, 1);
1027 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
1028 lli->lli_mds_exec_och = NULL;
1029 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
1030 lli->lli_open_fd_exec_count = 0;
1031 INIT_LIST_HEAD(&lli->lli_dead_list);
1032 lli->lli_remote_perms = NULL;
1033 lli->lli_rmtperm_utime = 0;
1034 sema_init(&lli->lli_rmtperm_sem, 1);
1035 INIT_LIST_HEAD(&lli->lli_oss_capas);
1038 int ll_fill_super(struct super_block *sb)
1040 struct lustre_profile *lprof;
1041 struct lustre_sb_info *lsi = s2lsi(sb);
1042 struct ll_sb_info *sbi;
1043 char *dt = NULL, *md = NULL;
1044 char *profilenm = get_profile_name(sb);
1045 struct config_llog_instance cfg = {0, };
1046 char ll_instance[sizeof(sb) * 2 + 1];
1050 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1054 /* client additional sb info */
1055 lsi->lsi_llsbi = sbi = ll_init_sbi();
1061 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1063 GOTO(out_free, err);
1065 /* Generate a string unique to this super, in case some joker tries
1066 to mount the same fs at two mount points.
1067 Use the address of the super itself.*/
1068 sprintf(ll_instance, "%p", sb);
1069 cfg.cfg_instance = ll_instance;
1070 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1072 /* set up client obds */
1073 err = lustre_process_log(sb, profilenm, &cfg);
1075 CERROR("Unable to process log: %d\n", err);
1076 GOTO(out_free, err);
1079 lprof = class_get_profile(profilenm);
1080 if (lprof == NULL) {
1081 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1082 " read from the MGS. Does that filesystem "
1083 "exist?\n", profilenm);
1084 GOTO(out_free, err = -EINVAL);
1086 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1087 lprof->lp_md, lprof->lp_dt);
1089 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1090 strlen(ll_instance) + 2);
1092 GOTO(out_free, err = -ENOMEM);
1093 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1095 OBD_ALLOC(md, strlen(lprof->lp_md) +
1096 strlen(ll_instance) + 2);
1098 GOTO(out_free, err = -ENOMEM);
1099 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1101 /* connections, registrations, sb setup */
1102 err = client_common_fill_super(sb, md, dt);
1106 OBD_FREE(md, strlen(md) + 1);
1108 OBD_FREE(dt, strlen(dt) + 1);
1112 LCONSOLE_WARN("Client %s has started\n", profilenm);
1115 } /* ll_fill_super */
1118 void ll_put_super(struct super_block *sb)
1120 struct config_llog_instance cfg;
1121 char ll_instance[sizeof(sb) * 2 + 1];
1122 struct obd_device *obd;
1123 struct lustre_sb_info *lsi = s2lsi(sb);
1124 struct ll_sb_info *sbi = ll_s2sbi(sb);
1125 char *profilenm = get_profile_name(sb);
1126 int force = 1, next;
1129 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1131 ll_print_capa_stat(sbi);
1133 sprintf(ll_instance, "%p", sb);
1134 cfg.cfg_instance = ll_instance;
1135 lustre_end_log(sb, NULL, &cfg);
1137 if (sbi->ll_md_exp) {
1138 obd = class_exp2obd(sbi->ll_md_exp);
1140 force = obd->obd_force;
1143 /* We need to set force before the lov_disconnect in
1144 lustre_common_put_super, since l_d cleans up osc's as well. */
1147 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1149 obd->obd_force = force;
1154 /* Only if client_common_fill_super succeeded */
1155 client_common_put_super(sb);
1158 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1159 class_manual_cleanup(obd);
1163 class_del_profile(profilenm);
1166 lsi->lsi_llsbi = NULL;
1168 lustre_common_put_super(sb);
1170 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1175 } /* client_put_super */
1177 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1179 #if defined(HAVE_CACHE_RETURN_INT)
1184 ll_shrink_cache(int priority, unsigned int gfp_mask)
1186 struct ll_sb_info *sbi;
1189 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1190 count += llap_shrink_cache(sbi, priority);
1192 #if defined(HAVE_CACHE_RETURN_INT)
1197 struct cache_definition ll_cache_definition = {
1198 .name = "llap_cache",
1199 .shrink = ll_shrink_cache
1201 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1203 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1205 struct inode *inode = NULL;
1206 /* NOTE: we depend on atomic igrab() -bzzz */
1207 lock_res_and_lock(lock);
1208 if (lock->l_ast_data) {
1209 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1210 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1211 inode = igrab(lock->l_ast_data);
1213 inode = lock->l_ast_data;
1214 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1216 lock, __FILE__, __func__, __LINE__,
1217 "l_ast_data %p is bogus: magic %08x",
1218 lock->l_ast_data, lli->lli_inode_magic);
1222 unlock_res_and_lock(lock);
1226 static int null_if_equal(struct ldlm_lock *lock, void *data)
1228 if (data == lock->l_ast_data) {
1229 lock->l_ast_data = NULL;
1231 if (lock->l_req_mode != lock->l_granted_mode)
1232 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1235 return LDLM_ITER_CONTINUE;
1238 void ll_clear_inode(struct inode *inode)
1240 struct ll_inode_info *lli = ll_i2info(inode);
1241 struct ll_sb_info *sbi = ll_i2sbi(inode);
1244 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1245 inode->i_generation, inode);
1247 if (S_ISDIR(inode->i_mode)) {
1248 /* these should have been cleared in ll_file_release */
1249 LASSERT(lli->lli_sai == NULL);
1250 LASSERT(lli->lli_opendir_key == NULL);
1251 LASSERT(lli->lli_opendir_pid == 0);
1254 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1255 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1256 null_if_equal, inode);
1258 LASSERT(!lli->lli_open_fd_write_count);
1259 LASSERT(!lli->lli_open_fd_read_count);
1260 LASSERT(!lli->lli_open_fd_exec_count);
1262 if (lli->lli_mds_write_och)
1263 ll_md_real_close(inode, FMODE_WRITE);
1264 if (lli->lli_mds_exec_och)
1265 ll_md_real_close(inode, FMODE_EXEC);
1266 if (lli->lli_mds_read_och)
1267 ll_md_real_close(inode, FMODE_READ);
1270 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1271 null_if_equal, inode);
1273 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1274 lli->lli_smd = NULL;
1277 if (lli->lli_symlink_name) {
1278 OBD_FREE(lli->lli_symlink_name,
1279 strlen(lli->lli_symlink_name) + 1);
1280 lli->lli_symlink_name = NULL;
1283 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1284 LASSERT(lli->lli_posix_acl == NULL);
1285 if (lli->lli_remote_perms) {
1286 free_rmtperm_hash(lli->lli_remote_perms);
1287 lli->lli_remote_perms = NULL;
1290 #ifdef CONFIG_FS_POSIX_ACL
1291 else if (lli->lli_posix_acl) {
1292 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1293 LASSERT(lli->lli_remote_perms == NULL);
1294 posix_acl_release(lli->lli_posix_acl);
1295 lli->lli_posix_acl = NULL;
1298 lli->lli_inode_magic = LLI_INODE_DEAD;
1300 #ifdef HAVE_EXPORT___IGET
1301 spin_lock(&sbi->ll_deathrow_lock);
1302 list_del_init(&lli->lli_dead_list);
1303 spin_unlock(&sbi->ll_deathrow_lock);
1305 ll_clear_inode_capas(inode);
1310 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1311 struct md_open_data **mod)
1313 struct lustre_md md;
1314 struct ll_sb_info *sbi = ll_i2sbi(inode);
1315 struct ptlrpc_request *request = NULL;
1319 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1320 LUSTRE_OPC_ANY, NULL);
1321 if (IS_ERR(op_data))
1322 RETURN(PTR_ERR(op_data));
1324 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1327 ptlrpc_req_finished(request);
1328 if (rc == -ENOENT) {
1330 /* Unlinked special device node? Or just a race?
1331 * Pretend we done everything. */
1332 if (!S_ISREG(inode->i_mode) &&
1333 !S_ISDIR(inode->i_mode))
1334 rc = inode_setattr(inode, &op_data->op_attr);
1335 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1336 CERROR("md_setattr fails: rc = %d\n", rc);
1341 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1342 sbi->ll_md_exp, &md);
1344 ptlrpc_req_finished(request);
1348 /* We call inode_setattr to adjust timestamps.
1349 * If there is at least some data in file, we cleared ATTR_SIZE
1350 * above to avoid invoking vmtruncate, otherwise it is important
1351 * to call vmtruncate in inode_setattr to update inode->i_size
1353 rc = inode_setattr(inode, &op_data->op_attr);
1355 /* Extract epoch data if obtained. */
1356 op_data->op_handle = md.body->handle;
1357 op_data->op_ioepoch = md.body->ioepoch;
1359 ll_update_inode(inode, &md);
1360 ptlrpc_req_finished(request);
1365 /* Close IO epoch and send Size-on-MDS attribute update. */
1366 static int ll_setattr_done_writing(struct inode *inode,
1367 struct md_op_data *op_data,
1368 struct md_open_data *mod)
1370 struct ll_inode_info *lli = ll_i2info(inode);
1374 LASSERT(op_data != NULL);
1375 if (!S_ISREG(inode->i_mode))
1378 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1379 op_data->op_ioepoch, PFID(&lli->lli_fid));
1381 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1382 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1383 if (rc == -EAGAIN) {
1384 /* MDS has instructed us to obtain Size-on-MDS attribute
1385 * from OSTs and send setattr to back to MDS. */
1386 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1387 op_data->op_ioepoch);
1389 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1395 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1397 struct ll_sb_info *sbi = ll_i2sbi(inode);
1398 struct ll_inode_info *lli = ll_i2info(inode);
1399 struct lov_stripe_md *lsm = lli->lli_smd;
1401 ldlm_policy_data_t policy = { .l_extent = {new_size,
1403 struct lustre_handle lockh = { 0 };
1404 int local_lock = 0; /* 0 - no local lock;
1405 * 1 - lock taken by lock_extent;
1406 * 2 - by obd_match*/
1411 UNLOCK_INODE_MUTEX(inode);
1412 UP_WRITE_I_ALLOC_SEM(inode);
1414 if (sbi->ll_lockless_truncate_enable &&
1415 (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1416 ast_flags = LDLM_FL_BLOCK_GRANTED;
1417 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1418 &policy, LCK_PW, &ast_flags, inode, &lockh);
1422 } else if (rc == 0) {
1423 rc = ll_file_punch(inode, new_size, 1);
1426 /* XXX when we fix the AST intents to pass the discard-range
1427 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1429 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1430 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1432 if (likely(rc == 0))
1436 LOCK_INODE_MUTEX(inode);
1437 DOWN_WRITE_I_ALLOC_SEM(inode);
1439 if (likely(rc == 0)) {
1440 /* Only ll_inode_size_lock is taken at this level.
1441 * lov_stripe_lock() is grabbed by ll_truncate() only over
1442 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1443 * ll_truncate dropped ll_inode_size_lock() */
1444 ll_inode_size_lock(inode, 0);
1446 spin_lock(&lli->lli_lock);
1447 lli->lli_flags |= LLIF_SRVLOCK;
1448 spin_unlock(&lli->lli_lock);
1450 rc = vmtruncate(inode, new_size);
1452 spin_lock(&lli->lli_lock);
1453 lli->lli_flags &= ~LLIF_SRVLOCK;
1454 spin_unlock(&lli->lli_lock);
1457 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1458 ll_inode_size_unlock(inode, 0);
1463 if (local_lock == 2)
1464 err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1466 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1467 if (unlikely(err != 0)){
1468 CERROR("extent unlock failed: err=%d,"
1469 " unlock method =%d\n", err, local_lock);
1477 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1478 * object(s) determine the file size and mtime. Otherwise, the MDS will
1479 * keep these values until such a time that objects are allocated for it.
1480 * We do the MDS operations first, as it is checking permissions for us.
1481 * We don't to the MDS RPC if there is nothing that we want to store there,
1482 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1483 * going to do an RPC anyways.
1485 * If we are doing a truncate, we will send the mtime and ctime updates
1486 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1487 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1490 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1492 struct ll_inode_info *lli = ll_i2info(inode);
1493 struct lov_stripe_md *lsm = lli->lli_smd;
1494 struct ll_sb_info *sbi = ll_i2sbi(inode);
1495 struct md_op_data *op_data = NULL;
1496 struct md_open_data *mod = NULL;
1497 int ia_valid = attr->ia_valid;
1498 int rc = 0, rc1 = 0;
1501 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1503 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1505 if (ia_valid & ATTR_SIZE) {
1506 if (attr->ia_size > ll_file_maxbytes(inode)) {
1507 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1508 attr->ia_size, ll_file_maxbytes(inode));
1512 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1515 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1516 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1517 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1521 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1522 if (attr->ia_valid & ATTR_CTIME) {
1523 attr->ia_ctime = CURRENT_TIME;
1524 attr->ia_valid |= ATTR_CTIME_SET;
1526 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1527 attr->ia_atime = CURRENT_TIME;
1528 attr->ia_valid |= ATTR_ATIME_SET;
1530 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1531 attr->ia_mtime = CURRENT_TIME;
1532 attr->ia_valid |= ATTR_MTIME_SET;
1534 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1535 /* To avoid stale mtime on mds, obtain it from ost and send
1537 rc = ll_glimpse_size(inode, 0);
1541 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1542 attr->ia_mtime = inode->i_mtime;
1545 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1546 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1547 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1548 cfs_time_current_sec());
1550 /* NB: ATTR_SIZE will only be set after this point if the size
1551 * resides on the MDS, ie, this file has no objects. */
1553 attr->ia_valid &= ~ATTR_SIZE;
1555 /* We always do an MDS RPC, even if we're only changing the size;
1556 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1558 OBD_ALLOC_PTR(op_data);
1559 if (op_data == NULL)
1562 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1564 /* Open epoch for truncate. */
1565 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1566 (ia_valid & ATTR_SIZE))
1567 op_data->op_flags = MF_EPOCH_OPEN;
1569 rc = ll_md_setattr(inode, op_data, &mod);
1573 if (op_data->op_ioepoch)
1574 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1575 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1577 if (!lsm || !S_ISREG(inode->i_mode)) {
1578 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1582 /* We really need to get our PW lock before we change inode->i_size.
1583 * If we don't we can race with other i_size updaters on our node, like
1584 * ll_file_read. We can also race with i_size propogation to other
1585 * nodes through dirtying and writeback of final cached pages. This
1586 * last one is especially bad for racing o_append users on other
1588 if (ia_valid & ATTR_SIZE) {
1589 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1590 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1592 struct obd_info oinfo = { { { 0 } } };
1595 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1596 inode->i_ino, LTIME_S(attr->ia_mtime));
1600 oa->o_id = lsm->lsm_object_id;
1601 oa->o_gr = lsm->lsm_object_gr;
1602 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1604 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1605 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1606 OBD_MD_FLFID | OBD_MD_FLGENER |
1609 obdo_from_inode(oa, inode, flags);
1613 oinfo.oi_capa = ll_mdscapa_get(inode);
1615 /* XXX: this looks unnecessary now. */
1616 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1617 capa_put(oinfo.oi_capa);
1619 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1628 if (op_data->op_ioepoch)
1629 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1630 ll_finish_md_op_data(op_data);
1632 return rc ? rc : rc1;
1635 int ll_setattr(struct dentry *de, struct iattr *attr)
1637 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1638 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1639 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1641 return ll_setattr_raw(de->d_inode, attr);
1644 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1645 __u64 max_age, __u32 flags)
1647 struct ll_sb_info *sbi = ll_s2sbi(sb);
1648 struct obd_statfs obd_osfs;
1652 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1654 CERROR("md_statfs fails: rc = %d\n", rc);
1658 osfs->os_type = sb->s_magic;
1660 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1661 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1663 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1664 &obd_osfs, max_age, flags);
1666 CERROR("obd_statfs fails: rc = %d\n", rc);
1670 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1671 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1674 osfs->os_bsize = obd_osfs.os_bsize;
1675 osfs->os_blocks = obd_osfs.os_blocks;
1676 osfs->os_bfree = obd_osfs.os_bfree;
1677 osfs->os_bavail = obd_osfs.os_bavail;
1679 /* If we don't have as many objects free on the OST as inodes
1680 * on the MDS, we reduce the total number of inodes to
1681 * compensate, so that the "inodes in use" number is correct.
1683 if (obd_osfs.os_ffree < osfs->os_ffree) {
1684 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1686 osfs->os_ffree = obd_osfs.os_ffree;
1691 #ifndef HAVE_STATFS_DENTRY_PARAM
1692 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1695 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1697 struct super_block *sb = de->d_sb;
1699 struct obd_statfs osfs;
1702 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1703 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1705 /* For now we will always get up-to-date statfs values, but in the
1706 * future we may allow some amount of caching on the client (e.g.
1707 * from QOS or lprocfs updates). */
1708 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1712 statfs_unpack(sfs, &osfs);
1714 /* We need to downshift for all 32-bit kernels, because we can't
1715 * tell if the kernel is being called via sys_statfs64() or not.
1716 * Stop before overflowing f_bsize - in which case it is better
1717 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1718 if (sizeof(long) < 8) {
1719 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1722 osfs.os_blocks >>= 1;
1723 osfs.os_bfree >>= 1;
1724 osfs.os_bavail >>= 1;
1728 sfs->f_blocks = osfs.os_blocks;
1729 sfs->f_bfree = osfs.os_bfree;
1730 sfs->f_bavail = osfs.os_bavail;
1735 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1737 struct ll_inode_info *lli;
1738 struct lov_stripe_md *lsm;
1740 lli = ll_i2info(inode);
1741 LASSERT(lli->lli_size_sem_owner != current);
1742 down(&lli->lli_size_sem);
1743 LASSERT(lli->lli_size_sem_owner == NULL);
1744 lli->lli_size_sem_owner = current;
1746 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1749 lov_stripe_lock(lsm);
1752 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1754 struct ll_inode_info *lli;
1755 struct lov_stripe_md *lsm;
1757 lli = ll_i2info(inode);
1759 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1762 lov_stripe_unlock(lsm);
1763 LASSERT(lli->lli_size_sem_owner == current);
1764 lli->lli_size_sem_owner = NULL;
1765 up(&lli->lli_size_sem);
1768 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1770 struct ll_inode_info *lli = ll_i2info(inode);
1772 dump_lsm(D_INODE, lsm);
1773 dump_lsm(D_INODE, lli->lli_smd);
1774 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1775 "lsm must be joined lsm %p\n", lsm);
1776 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1777 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1778 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1780 lli->lli_maxbytes = lsm->lsm_maxbytes;
1781 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1782 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1785 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1787 struct ll_inode_info *lli = ll_i2info(inode);
1788 struct mdt_body *body = md->body;
1789 struct lov_stripe_md *lsm = md->lsm;
1790 struct ll_sb_info *sbi = ll_i2sbi(inode);
1792 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1794 if (lli->lli_smd == NULL) {
1795 if (lsm->lsm_magic != LOV_MAGIC &&
1796 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1797 dump_lsm(D_ERROR, lsm);
1800 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1801 lsm, inode->i_ino, inode->i_generation, inode);
1802 /* ll_inode_size_lock() requires it is only called
1803 * with lli_smd != NULL or lock_lsm == 0 or we can
1804 * race between lock/unlock. bug 9547 */
1806 lli->lli_maxbytes = lsm->lsm_maxbytes;
1807 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1808 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1810 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1811 lli->lli_smd->lsm_stripe_count ==
1812 lsm->lsm_stripe_count) {
1813 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1814 CERROR("lsm mismatch for inode %ld\n",
1816 CERROR("lli_smd:\n");
1817 dump_lsm(D_ERROR, lli->lli_smd);
1819 dump_lsm(D_ERROR, lsm);
1823 ll_replace_lsm(inode, lsm);
1825 if (lli->lli_smd != lsm)
1826 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1829 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1830 if (body->valid & OBD_MD_FLRMTPERM)
1831 ll_update_remote_perm(inode, md->remote_perm);
1833 #ifdef CONFIG_FS_POSIX_ACL
1834 else if (body->valid & OBD_MD_FLACL) {
1835 spin_lock(&lli->lli_lock);
1836 if (lli->lli_posix_acl)
1837 posix_acl_release(lli->lli_posix_acl);
1838 lli->lli_posix_acl = md->posix_acl;
1839 spin_unlock(&lli->lli_lock);
1842 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1844 if (body->valid & OBD_MD_FLATIME &&
1845 body->atime > LTIME_S(inode->i_atime))
1846 LTIME_S(inode->i_atime) = body->atime;
1848 /* mtime is always updated with ctime, but can be set in past.
1849 As write and utime(2) may happen within 1 second, and utime's
1850 mtime has a priority over write's one, so take mtime from mds
1851 for the same ctimes. */
1852 if (body->valid & OBD_MD_FLCTIME &&
1853 body->ctime >= LTIME_S(inode->i_ctime)) {
1854 LTIME_S(inode->i_ctime) = body->ctime;
1855 if (body->valid & OBD_MD_FLMTIME) {
1856 CDEBUG(D_INODE, "setting ino %lu mtime "
1857 "from %lu to "LPU64"\n", inode->i_ino,
1858 LTIME_S(inode->i_mtime), body->mtime);
1859 LTIME_S(inode->i_mtime) = body->mtime;
1862 if (body->valid & OBD_MD_FLMODE)
1863 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1864 if (body->valid & OBD_MD_FLTYPE)
1865 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1866 if (S_ISREG(inode->i_mode)) {
1867 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1869 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1871 #ifdef HAVE_INODE_BLKSIZE
1872 inode->i_blksize = 1<<inode->i_blkbits;
1874 if (body->valid & OBD_MD_FLUID)
1875 inode->i_uid = body->uid;
1876 if (body->valid & OBD_MD_FLGID)
1877 inode->i_gid = body->gid;
1878 if (body->valid & OBD_MD_FLFLAGS)
1879 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1880 if (body->valid & OBD_MD_FLNLINK)
1881 inode->i_nlink = body->nlink;
1882 if (body->valid & OBD_MD_FLRDEV)
1883 inode->i_rdev = old_decode_dev(body->rdev);
1885 if (body->valid & OBD_MD_FLID) {
1886 /* FID shouldn't be changed! */
1887 if (fid_is_sane(&lli->lli_fid)) {
1888 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1889 "Trying to change FID "DFID
1890 " to the "DFID", inode %lu/%u(%p)\n",
1891 PFID(&lli->lli_fid), PFID(&body->fid1),
1892 inode->i_ino, inode->i_generation, inode);
1894 lli->lli_fid = body->fid1;
1897 LASSERT(fid_seq(&lli->lli_fid) != 0);
1899 if (body->valid & OBD_MD_FLSIZE) {
1900 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1901 S_ISREG(inode->i_mode) && lli->lli_smd) {
1902 struct lustre_handle lockh;
1905 /* As it is possible a blocking ast has been processed
1906 * by this time, we need to check there is an UPDATE
1907 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1909 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1912 if (lli->lli_flags & (LLIF_DONE_WRITING |
1913 LLIF_EPOCH_PENDING |
1915 CERROR("ino %lu flags %lu still has "
1916 "size authority! do not trust "
1917 "the size got from MDS\n",
1918 inode->i_ino, lli->lli_flags);
1920 /* Use old size assignment to avoid
1921 * deadlock bz14138 & bz14326 */
1922 inode->i_size = body->size;
1923 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1925 ldlm_lock_decref(&lockh, mode);
1928 /* Use old size assignment to avoid
1929 * deadlock bz14138 & bz14326 */
1930 inode->i_size = body->size;
1933 if (body->valid & OBD_MD_FLBLOCKS)
1934 inode->i_blocks = body->blocks;
1937 if (body->valid & OBD_MD_FLMDSCAPA) {
1938 LASSERT(md->mds_capa);
1939 ll_add_capa(inode, md->mds_capa);
1941 if (body->valid & OBD_MD_FLOSSCAPA) {
1942 LASSERT(md->oss_capa);
1943 ll_add_capa(inode, md->oss_capa);
1947 static struct backing_dev_info ll_backing_dev_info = {
1948 .ra_pages = 0, /* No readahead */
1949 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1950 .capabilities = 0, /* Does contribute to dirty memory */
1952 .memory_backed = 0, /* Does contribute to dirty memory */
1956 void ll_read_inode2(struct inode *inode, void *opaque)
1958 struct lustre_md *md = opaque;
1959 struct ll_inode_info *lli = ll_i2info(inode);
1962 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1963 inode->i_ino, inode->i_generation, inode);
1967 LASSERT(!lli->lli_smd);
1969 /* Core attributes from the MDS first. This is a new inode, and
1970 * the VFS doesn't zero times in the core inode so we have to do
1971 * it ourselves. They will be overwritten by either MDS or OST
1972 * attributes - we just need to make sure they aren't newer. */
1973 LTIME_S(inode->i_mtime) = 0;
1974 LTIME_S(inode->i_atime) = 0;
1975 LTIME_S(inode->i_ctime) = 0;
1977 ll_update_inode(inode, md);
1979 /* OIDEBUG(inode); */
1981 if (S_ISREG(inode->i_mode)) {
1982 struct ll_sb_info *sbi = ll_i2sbi(inode);
1983 inode->i_op = &ll_file_inode_operations;
1984 inode->i_fop = sbi->ll_fop;
1985 inode->i_mapping->a_ops = &ll_aops;
1987 } else if (S_ISDIR(inode->i_mode)) {
1988 inode->i_op = &ll_dir_inode_operations;
1989 inode->i_fop = &ll_dir_operations;
1990 inode->i_mapping->a_ops = &ll_dir_aops;
1992 } else if (S_ISLNK(inode->i_mode)) {
1993 inode->i_op = &ll_fast_symlink_inode_operations;
1996 inode->i_op = &ll_special_inode_operations;
1998 init_special_inode(inode, inode->i_mode,
1999 kdev_t_to_nr(inode->i_rdev));
2001 /* initializing backing dev info. */
2002 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2008 void ll_delete_inode(struct inode *inode)
2010 struct ll_sb_info *sbi = ll_i2sbi(inode);
2014 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
2016 CERROR("fid_delete() failed, rc %d\n", rc);
2018 truncate_inode_pages(&inode->i_data, 0);
2024 int ll_iocontrol(struct inode *inode, struct file *file,
2025 unsigned int cmd, unsigned long arg)
2027 struct ll_sb_info *sbi = ll_i2sbi(inode);
2028 struct ptlrpc_request *req = NULL;
2033 case EXT3_IOC_GETFLAGS: {
2034 struct mdt_body *body;
2035 struct obd_capa *oc;
2037 oc = ll_mdscapa_get(inode);
2038 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
2039 OBD_MD_FLFLAGS, 0, &req);
2042 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2046 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2048 flags = body->flags;
2050 ptlrpc_req_finished(req);
2052 RETURN(put_user(flags, (int *)arg));
2054 case EXT3_IOC_SETFLAGS: {
2055 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2056 struct obd_info oinfo = { { { 0 } } };
2057 struct md_op_data *op_data;
2059 if (get_user(flags, (int *)arg))
2063 OBDO_ALLOC(oinfo.oi_oa);
2067 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2068 LUSTRE_OPC_ANY, NULL);
2069 if (IS_ERR(op_data))
2070 RETURN(PTR_ERR(op_data));
2072 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2073 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2074 rc = md_setattr(sbi->ll_md_exp, op_data,
2075 NULL, 0, NULL, 0, &req, NULL);
2076 ll_finish_md_op_data(op_data);
2077 ptlrpc_req_finished(req);
2078 if (rc || lsm == NULL) {
2079 OBDO_FREE(oinfo.oi_oa);
2083 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2084 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2085 oinfo.oi_oa->o_flags = flags;
2086 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2088 oinfo.oi_capa = ll_mdscapa_get(inode);
2090 obdo_from_inode(oinfo.oi_oa, inode,
2091 OBD_MD_FLFID | OBD_MD_FLGENER);
2092 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2093 capa_put(oinfo.oi_capa);
2094 OBDO_FREE(oinfo.oi_oa);
2096 if (rc != -EPERM && rc != -EACCES)
2097 CERROR("md_setattr_async fails: rc = %d\n", rc);
2101 inode->i_flags = ll_ext_to_inode_flags(flags |
2102 MDS_BFLAG_EXT_FLAGS);
2112 int ll_flush_ctx(struct inode *inode)
2114 struct ll_sb_info *sbi = ll_i2sbi(inode);
2116 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2118 obd_set_info_async(sbi->ll_md_exp,
2119 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2121 obd_set_info_async(sbi->ll_dt_exp,
2122 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2127 /* umount -f client means force down, don't save state */
2128 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2129 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2131 struct super_block *sb = vfsmnt->mnt_sb;
2133 void ll_umount_begin(struct super_block *sb)
2136 struct lustre_sb_info *lsi = s2lsi(sb);
2137 struct ll_sb_info *sbi = ll_s2sbi(sb);
2138 struct obd_device *obd;
2139 struct obd_ioctl_data ioc_data = { 0 };
2142 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2143 if (!(flags & MNT_FORCE)) {
2149 /* Tell the MGC we got umount -f */
2150 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2152 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2153 sb->s_count, atomic_read(&sb->s_active));
2155 obd = class_exp2obd(sbi->ll_md_exp);
2157 CERROR("Invalid MDC connection handle "LPX64"\n",
2158 sbi->ll_md_exp->exp_handle.h_cookie);
2163 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2166 obd = class_exp2obd(sbi->ll_dt_exp);
2168 CERROR("Invalid LOV connection handle "LPX64"\n",
2169 sbi->ll_dt_exp->exp_handle.h_cookie);
2175 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2178 /* Really, we'd like to wait until there are no requests outstanding,
2179 * and then continue. For now, we just invalidate the requests,
2180 * schedule, and hope.
2187 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2189 struct ll_sb_info *sbi = ll_s2sbi(sb);
2193 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2194 read_only = *flags & MS_RDONLY;
2195 err = obd_set_info_async(sbi->ll_md_exp,
2196 sizeof(KEY_READ_ONLY),
2197 KEY_READ_ONLY, sizeof(read_only),
2200 CERROR("Failed to change the read-only flag during "
2201 "remount: %d\n", err);
2206 sb->s_flags |= MS_RDONLY;
2208 sb->s_flags &= ~MS_RDONLY;
2213 int ll_prep_inode(struct inode **inode,
2214 struct ptlrpc_request *req,
2215 struct super_block *sb)
2217 struct ll_sb_info *sbi = NULL;
2218 struct lustre_md md;
2222 LASSERT(*inode || sb);
2223 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2224 prune_deathrow(sbi, 1);
2225 memset(&md, 0, sizeof(struct lustre_md));
2227 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2228 sbi->ll_md_exp, &md);
2233 ll_update_inode(*inode, &md);
2235 LASSERT(sb != NULL);
2238 * At this point server returns to client's same fid as client
2239 * generated for creating. So using ->fid1 is okay here.
2241 LASSERT(fid_is_sane(&md.body->fid1));
2243 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2244 if (*inode == NULL || is_bad_inode(*inode)) {
2246 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2247 #ifdef CONFIG_FS_POSIX_ACL
2249 posix_acl_release(md.posix_acl);
2250 md.posix_acl = NULL;
2254 CERROR("new_inode -fatal: rc %d\n", rc);
2259 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2260 ll_i2info(*inode)->lli_smd);
2262 md_free_lustre_md(sbi->ll_md_exp, &md);
2266 char *llap_origins[] = {
2267 [LLAP_ORIGIN_UNKNOWN] = "--",
2268 [LLAP_ORIGIN_READPAGE] = "rp",
2269 [LLAP_ORIGIN_READAHEAD] = "ra",
2270 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2271 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2272 [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2275 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2276 struct list_head *list)
2278 struct ll_async_page *llap;
2279 struct list_head *pos;
2281 list_for_each(pos, list) {
2284 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2285 if (llap->llap_page == NULL)
2293 int ll_obd_statfs(struct inode *inode, void *arg)
2295 struct ll_sb_info *sbi = NULL;
2296 struct obd_export *exp;
2298 struct obd_ioctl_data *data = NULL;
2302 if (!inode || !(sbi = ll_i2sbi(inode)))
2303 GOTO(out_statfs, rc = -EINVAL);
2305 rc = obd_ioctl_getdata(&buf, &len, arg);
2307 GOTO(out_statfs, rc);
2310 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2311 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2312 GOTO(out_statfs, rc = -EINVAL);
2314 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2315 if (type == LL_STATFS_MDC)
2316 exp = sbi->ll_md_exp;
2317 else if (type == LL_STATFS_LOV)
2318 exp = sbi->ll_dt_exp;
2320 GOTO(out_statfs, rc = -ENODEV);
2322 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2324 GOTO(out_statfs, rc);
2327 obd_ioctl_freedata(buf, len);
2331 int ll_process_config(struct lustre_cfg *lcfg)
2335 struct lprocfs_static_vars lvars;
2339 lprocfs_llite_init_vars(&lvars);
2341 /* The instance name contains the sb: lustre-client-aacfe000 */
2342 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2343 if (!ptr || !*(++ptr))
2345 if (sscanf(ptr, "%lx", &x) != 1)
2348 /* This better be a real Lustre superblock! */
2349 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2351 /* Note we have not called client_common_fill_super yet, so
2352 proc fns must be able to handle that! */
2353 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2358 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2359 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2360 struct inode *i1, struct inode *i2,
2361 const char *name, int namelen,
2362 int mode, __u32 opc, void *data)
2364 LASSERT(i1 != NULL);
2366 if (namelen > ll_i2sbi(i1)->ll_namelen)
2367 return ERR_PTR(-ENAMETOOLONG);
2369 if (op_data == NULL)
2370 OBD_ALLOC_PTR(op_data);
2372 if (op_data == NULL)
2373 return ERR_PTR(-ENOMEM);
2375 ll_i2gids(op_data->op_suppgids, i1, i2);
2376 op_data->op_fid1 = *ll_inode2fid(i1);
2377 op_data->op_capa1 = ll_mdscapa_get(i1);
2380 op_data->op_fid2 = *ll_inode2fid(i2);
2381 op_data->op_capa2 = ll_mdscapa_get(i2);
2383 fid_zero(&op_data->op_fid2);
2384 op_data->op_capa2 = NULL;
2387 op_data->op_name = name;
2388 op_data->op_namelen = namelen;
2389 op_data->op_mode = mode;
2390 op_data->op_mod_time = cfs_time_current_sec();
2391 op_data->op_fsuid = current->fsuid;
2392 op_data->op_fsgid = current->fsgid;
2393 op_data->op_cap = current->cap_effective;
2394 op_data->op_bias = MDS_CHECK_SPLIT;
2395 op_data->op_opc = opc;
2396 op_data->op_mds = 0;
2397 op_data->op_data = data;
2402 void ll_finish_md_op_data(struct md_op_data *op_data)
2404 capa_put(op_data->op_capa1);
2405 capa_put(op_data->op_capa2);
2406 OBD_FREE_PTR(op_data);
2409 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2411 struct ll_sb_info *sbi;
2413 LASSERT((seq != NULL) && (vfs != NULL));
2414 sbi = ll_s2sbi(vfs->mnt_sb);
2416 if (sbi->ll_flags & LL_SBI_NOLCK)
2417 seq_puts(seq, ",nolock");
2419 if (sbi->ll_flags & LL_SBI_FLOCK)
2420 seq_puts(seq, ",flock");
2422 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2423 seq_puts(seq, ",localflock");
2425 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2426 seq_puts(seq, ",user_xattr");
2428 if (sbi->ll_flags & LL_SBI_ACL)
2429 seq_puts(seq, ",acl");