4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Directory code for lustre client.
38 #include <linux/pagemap.h>
40 #include <linux/version.h>
41 #include <linux/security.h>
42 #include <linux/user_namespace.h>
43 #ifdef HAVE_UIDGID_HEADER
44 # include <linux/uidgid.h>
46 #include <linux/uaccess.h>
47 #include <linux/buffer_head.h> // for wait_on_buffer
48 #include <linux/pagevec.h>
50 #define DEBUG_SUBSYSTEM S_LLITE
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <uapi/linux/lustre/lustre_ioctl.h>
55 #include <lustre_lib.h>
56 #include <lustre_dlm.h>
57 #include <lustre_fid.h>
58 #include <lustre_kernelcomm.h>
59 #include <lustre_swab.h>
61 #include "llite_internal.h"
64 * (new) readdir implementation overview.
66 * Original lustre readdir implementation cached exact copy of raw directory
67 * pages on the client. These pages were indexed in client page cache by
68 * logical offset in the directory file. This design, while very simple and
69 * intuitive had some inherent problems:
71 * . it implies that byte offset to the directory entry serves as a
72 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
73 * ext3/htree directory entries may move due to splits, and more
76 * . it is incompatible with the design of split directories for cmd3,
77 * that assumes that names are distributed across nodes based on their
78 * hash, and so readdir should be done in hash order.
80 * New readdir implementation does readdir in hash order, and uses hash of a
81 * file name as a telldir/seekdir cookie. This led to number of complications:
83 * . hash is not unique, so it cannot be used to index cached directory
84 * pages on the client (note, that it requires a whole pageful of hash
85 * collided entries to cause two pages to have identical hashes);
87 * . hash is not unique, so it cannot, strictly speaking, be used as an
88 * entry cookie. ext3/htree has the same problem and lustre implementation
89 * mimics their solution: seekdir(hash) positions directory at the first
90 * entry with the given hash.
96 * Client caches directory pages using hash of the first entry as an index. As
97 * noted above hash is not unique, so this solution doesn't work as is:
98 * special processing is needed for "page hash chains" (i.e., sequences of
99 * pages filled with entries all having the same hash value).
101 * First, such chains have to be detected. To this end, server returns to the
102 * client the hash of the first entry on the page next to one returned. When
103 * client detects that this hash is the same as hash of the first entry on the
104 * returned page, page hash collision has to be handled. Pages in the
105 * hash chain, except first one, are termed "overflow pages".
107 * Solution to index uniqueness problem is to not cache overflow
108 * pages. Instead, when page hash collision is detected, all overflow pages
109 * from emerging chain are immediately requested from the server and placed in
110 * a special data structure (struct ll_dir_chain). This data structure is used
111 * by ll_readdir() to process entries from overflow pages. When readdir
112 * invocation finishes, overflow pages are discarded. If page hash collision
113 * chain weren't completely processed, next call to readdir will again detect
114 * page hash collision, again read overflow pages in, process next portion of
115 * entries and again discard the pages. This is not as wasteful as it looks,
116 * because, given reasonable hash, page hash collisions are extremely rare.
118 * 1. directory positioning
120 * When seekdir(hash) is called, original
131 * identification of and access to overflow pages
135 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
136 * a header lu_dirpage which describes the start/end hash, and whether this
137 * page is empty (contains no dir entry) or hash collide with next page.
138 * After client receives reply, several pages will be integrated into dir page
139 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
140 * lu_dirpage for this integrated page will be adjusted. See
141 * mdc_adjust_dirpages().
144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
145 __u64 offset, struct ll_dir_chain *chain)
147 struct md_callback cb_op;
151 cb_op.md_blocking_ast = ll_md_blocking_ast;
152 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
159 void ll_release_page(struct inode *inode, struct page *page,
164 /* Always remove the page for striped dir, because the page is
165 * built from temporarily in LMV layer */
166 if (inode && ll_dir_striped(inode)) {
173 if (likely(page->mapping != NULL))
174 truncate_complete_page(page->mapping, page);
181 * return IF_* type for given lu_dirent entry.
182 * IF_* flag shld be converted to particular OS file type in
183 * platform llite module.
185 static u16 ll_dirent_type_get(struct lu_dirent *ent)
188 struct luda_type *lt;
191 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
192 const unsigned align = sizeof(struct luda_type) - 1;
194 len = le16_to_cpu(ent->lde_namelen);
195 len = (len + align) & ~align;
196 lt = (void *)ent->lde_name + len;
197 type = IFTODT(le16_to_cpu(lt->lt_type));
203 #ifdef HAVE_DIR_CONTEXT
204 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
205 struct dir_context *ctx)
208 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
209 void *cookie, filldir_t filldir)
212 struct ll_sb_info *sbi = ll_i2sbi(inode);
214 bool is_api32 = ll_need_32bit_api(sbi);
215 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
217 struct ll_dir_chain chain;
222 ll_dir_chain_init(&chain);
224 page = ll_get_dir_page(inode, op_data, pos, &chain);
226 while (rc == 0 && !done) {
227 struct lu_dirpage *dp;
228 struct lu_dirent *ent;
237 hash = MDS_DIR_END_OFF;
238 dp = page_address(page);
239 for (ent = lu_dirent_start(dp); ent != NULL && !done;
240 ent = lu_dirent_next(ent)) {
247 hash = le64_to_cpu(ent->lde_hash);
248 if (hash < pos) /* Skip until we find target hash */
251 namelen = le16_to_cpu(ent->lde_namelen);
252 if (namelen == 0) /* Skip dummy record */
255 if (is_api32 && is_hash64)
259 fid_le_to_cpu(&fid, &ent->lde_fid);
260 ino = cl_fid_build_ino(&fid, is_api32);
261 type = ll_dirent_type_get(ent);
262 /* For ll_nfs_get_name_filldir(), it will try to access
263 * 'ent' through 'lde_name', so the parameter 'name'
264 * for 'filldir()' must be part of the 'ent'. */
265 #ifdef HAVE_DIR_CONTEXT
267 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
270 done = filldir(cookie, ent->lde_name, namelen, lhash,
277 ll_release_page(inode, page, false);
281 next = le64_to_cpu(dp->ldp_hash_end);
283 if (pos == MDS_DIR_END_OFF) {
285 * End of directory reached.
288 ll_release_page(inode, page, false);
291 * Normal case: continue to the next
294 ll_release_page(inode, page,
295 le32_to_cpu(dp->ldp_flags) &
298 page = ll_get_dir_page(inode, op_data, pos,
302 #ifdef HAVE_DIR_CONTEXT
307 ll_dir_chain_fini(&chain);
311 #ifdef HAVE_DIR_CONTEXT
312 static int ll_iterate(struct file *filp, struct dir_context *ctx)
314 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
317 struct inode *inode = file_inode(filp);
318 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
319 struct ll_sb_info *sbi = ll_i2sbi(inode);
320 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
321 int api32 = ll_need_32bit_api(sbi);
322 struct md_op_data *op_data;
323 struct lu_fid pfid = { 0 };
333 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
334 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
335 inode, (unsigned long)pos, i_size_read(inode), api32);
337 if (pos == MDS_DIR_END_OFF)
343 if (unlikely(ll_dir_striped(inode))) {
345 * This is only needed for striped dir to fill ..,
346 * see lmv_read_page()
348 if (file_dentry(filp)->d_parent != NULL &&
349 file_dentry(filp)->d_parent->d_inode != NULL) {
350 __u64 ibits = MDS_INODELOCK_LOOKUP;
351 struct inode *parent =
352 file_dentry(filp)->d_parent->d_inode;
354 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
355 pfid = *ll_inode2fid(parent);
358 /* If it can not find in cache, do lookup .. on the master
360 if (fid_is_zero(&pfid)) {
361 rc = ll_dir_get_parent_fid(inode, &pfid);
367 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
368 LUSTRE_OPC_ANY, inode);
370 GOTO(out, rc = PTR_ERR(op_data));
372 /* foreign dirs are browsed out of Lustre */
373 if (unlikely(op_data->op_mea1 != NULL &&
374 op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
375 ll_finish_md_op_data(op_data);
379 op_data->op_fid3 = pfid;
381 #ifdef HAVE_DIR_CONTEXT
383 rc = ll_dir_read(inode, &pos, op_data, ctx);
386 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
391 if (pos == MDS_DIR_END_OFF) {
393 pos = LL_DIR_END_OFF_32BIT;
395 pos = LL_DIR_END_OFF;
400 #ifdef HAVE_DIR_CONTEXT
405 ll_finish_md_op_data(op_data);
409 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
414 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
415 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
417 struct mgs_send_param *msp;
424 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
425 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
426 sizeof(struct mgs_send_param), msp, NULL);
428 CERROR("Failed to set parameter: %d\n", rc);
436 * Create striped directory with specified stripe(@lump)
438 * \param[in] dparent the parent of the directory.
439 * \param[in] lump the specified stripes.
440 * \param[in] dirname the name of the directory.
441 * \param[in] mode the specified mode of the directory.
443 * \retval =0 if striped directory is being created successfully.
444 * <0 if the creation is failed.
446 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
447 size_t len, const char *dirname, umode_t mode)
449 struct inode *parent = dparent->d_inode;
450 struct ptlrpc_request *request = NULL;
451 struct md_op_data *op_data;
452 struct ll_sb_info *sbi = ll_i2sbi(parent);
453 struct inode *inode = NULL;
454 struct dentry dentry = {
458 .len = strlen(dirname),
459 .hash = ll_full_name_hash(dparent, dirname,
466 if (unlikely(!lmv_magic_supported(lump->lum_magic)))
469 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
471 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
472 PFID(ll_inode2fid(parent)), parent, dirname,
473 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
475 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
478 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
479 PFID(ll_inode2fid(parent)), parent, dirname,
480 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
483 if (lump->lum_stripe_count > 1 &&
484 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
487 if (IS_DEADDIR(parent) &&
488 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
491 if (unlikely(!lmv_magic_supported(cpu_to_le32(lump->lum_magic))))
492 lustre_swab_lmv_user_md(lump);
494 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
495 mode &= ~current_umask();
496 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
497 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
498 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
501 RETURN(PTR_ERR(op_data));
503 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
504 /* selinux_dentry_init_security() uses dentry->d_parent and name
505 * to determine the security context for the file. So our fake
506 * dentry should be real enough for this purpose. */
507 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
508 &op_data->op_file_secctx_name,
509 &op_data->op_file_secctx,
510 &op_data->op_file_secctx_size);
512 GOTO(out_op_data, err);
515 op_data->op_cli_flags |= CLI_SET_MEA;
516 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
517 from_kuid(&init_user_ns, current_fsuid()),
518 from_kgid(&init_user_ns, current_fsgid()),
519 cfs_curproc_cap_pack(), 0, &request);
521 GOTO(out_request, err);
523 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
525 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
527 GOTO(out_inode, err);
529 dentry.d_inode = inode;
531 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
533 err = security_inode_notifysecctx(inode,
534 op_data->op_file_secctx,
535 op_data->op_file_secctx_size);
538 err = ll_inode_init_security(&dentry, inode, parent);
541 GOTO(out_inode, err);
547 ptlrpc_req_finished(request);
549 ll_finish_md_op_data(op_data);
554 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
557 struct ll_sb_info *sbi = ll_i2sbi(inode);
558 struct md_op_data *op_data;
559 struct ptlrpc_request *req = NULL;
561 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
562 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
563 struct obd_device *mgc = lsi->lsi_mgc;
570 * This is coming from userspace, so should be in
571 * local endian. But the MDS would like it in little
572 * endian, so we swab it before we send it.
574 switch (lump->lmm_magic) {
575 case LOV_USER_MAGIC_V1: {
576 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
577 lustre_swab_lov_user_md_v1(lump);
578 lum_size = sizeof(struct lov_user_md_v1);
581 case LOV_USER_MAGIC_V3: {
582 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
583 lustre_swab_lov_user_md_v3(
584 (struct lov_user_md_v3 *)lump);
585 lum_size = sizeof(struct lov_user_md_v3);
588 case LOV_USER_MAGIC_COMP_V1: {
589 if (lump->lmm_magic !=
590 cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
591 lustre_swab_lov_comp_md_v1(
592 (struct lov_comp_md_v1 *)lump);
593 lum_size = le32_to_cpu(
594 ((struct lov_comp_md_v1 *)lump)->lcm_size);
597 case LMV_USER_MAGIC: {
598 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
599 lustre_swab_lmv_user_md(
600 (struct lmv_user_md *)lump);
601 lum_size = sizeof(struct lmv_user_md);
604 case LOV_USER_MAGIC_SPECIFIC: {
605 struct lov_user_md_v3 *v3 =
606 (struct lov_user_md_v3 *)lump;
607 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
609 if (lump->lmm_magic !=
610 cpu_to_le32(LOV_USER_MAGIC_SPECIFIC)) {
611 lustre_swab_lov_user_md_v3(v3);
612 lustre_swab_lov_user_md_objects(v3->lmm_objects,
613 v3->lmm_stripe_count);
615 lum_size = lov_user_md_size(v3->lmm_stripe_count,
616 LOV_USER_MAGIC_SPECIFIC);
620 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
621 " %#08x != %#08x nor %#08x\n",
622 lump->lmm_magic, LOV_USER_MAGIC_V1,
628 lum_size = sizeof(struct lov_user_md_v1);
631 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
632 LUSTRE_OPC_ANY, NULL);
634 RETURN(PTR_ERR(op_data));
636 /* swabbing is done in lov_setstripe() on server side */
637 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
638 ll_finish_md_op_data(op_data);
639 ptlrpc_req_finished(req);
643 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
645 * 2.9 server has stored filesystem default stripe in ROOT xattr,
646 * and it's stored into system config for backward compatibility.
648 * In the following we use the fact that LOV_USER_MAGIC_V1 and
649 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
650 * need the make the distiction between the 2 versions
652 if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
654 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
655 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
659 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
661 GOTO(end, rc = -ENOMEM);
664 /* Get fsname and assume devname to be -MDT0000. */
665 snprintf(buf, MGS_PARAM_MAXLEN, "%s-MDT0000.lov",
669 /* Set root stripesize */
670 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
671 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
672 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
676 /* Set root stripecount */
677 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
678 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
679 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
683 /* Set root stripeoffset */
684 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
685 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
686 (typeof(lump->lmm_stripe_offset))(-1));
687 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
691 OBD_FREE(param, MGS_PARAM_MAXLEN);
698 * This function will be used to get default LOV/LMV/Default LMV
699 * @valid will be used to indicate which stripe it will retrieve
700 * OBD_MD_MEA LMV stripe EA
701 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
702 * otherwise Default LOV EA.
703 * Each time, it can only retrieve 1 stripe EA
705 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
706 struct ptlrpc_request **request, u64 valid)
708 struct ll_sb_info *sbi = ll_i2sbi(inode);
709 struct mdt_body *body;
710 struct lov_mds_md *lmm = NULL;
711 struct ptlrpc_request *req = NULL;
713 struct md_op_data *op_data;
716 rc = ll_get_default_mdsize(sbi, &lmm_size);
720 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
721 0, lmm_size, LUSTRE_OPC_ANY,
724 RETURN(PTR_ERR(op_data));
726 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
727 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
728 ll_finish_md_op_data(op_data);
730 CDEBUG(D_INFO, "md_getattr failed on inode "
731 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
735 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
736 LASSERT(body != NULL);
738 lmm_size = body->mbo_eadatasize;
740 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
742 GOTO(out, rc = -ENODATA);
745 lmm = req_capsule_server_sized_get(&req->rq_pill,
746 &RMF_MDT_MD, lmm_size);
747 LASSERT(lmm != NULL);
750 * This is coming from the MDS, so is probably in
751 * little endian. We convert it to host endian before
752 * passing it to userspace.
754 /* We don't swab objects for directories */
755 switch (le32_to_cpu(lmm->lmm_magic)) {
757 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
758 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
761 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
762 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
764 case LOV_MAGIC_COMP_V1:
765 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
766 lustre_swab_lov_comp_md_v1(
767 (struct lov_comp_md_v1 *)lmm);
770 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
771 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
774 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
775 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
777 case LOV_USER_MAGIC_SPECIFIC: {
778 struct lov_user_md_v3 *v3 = (struct lov_user_md_v3 *)lmm;
780 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
781 lustre_swab_lov_user_md_v3(v3);
782 lustre_swab_lov_user_md_objects(v3->lmm_objects,
783 v3->lmm_stripe_count);
787 case LMV_MAGIC_FOREIGN: {
788 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
790 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
791 __swab32s(&lfm->lfm_magic);
792 __swab32s(&lfm->lfm_length);
793 __swab32s(&lfm->lfm_type);
794 __swab32s(&lfm->lfm_flags);
799 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
804 *plmm_size = lmm_size;
809 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
811 struct md_op_data *op_data;
816 OBD_ALLOC_PTR(op_data);
820 op_data->op_flags |= MF_GET_MDT_IDX;
821 op_data->op_fid1 = *fid;
822 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
823 mdt_index = op_data->op_mds;
824 OBD_FREE_PTR(op_data);
832 * Get MDT index for the inode.
834 int ll_get_mdt_idx(struct inode *inode)
836 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
840 * Generic handler to do any pre-copy work.
842 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
843 * first information for it that real work has started.
845 * Moreover, for a ARCHIVE request, it will sample the file data version and
846 * store it in \a copy.
848 * \return 0 on success.
850 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
852 struct ll_sb_info *sbi = ll_s2sbi(sb);
853 struct hsm_progress_kernel hpk;
858 /* Forge a hsm_progress based on data from copy. */
859 hpk.hpk_fid = copy->hc_hai.hai_fid;
860 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
861 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
862 hpk.hpk_extent.length = 0;
865 hpk.hpk_data_version = 0;
868 /* For archive request, we need to read the current file version. */
869 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
871 __u64 data_version = 0;
873 /* Get inode for this fid */
874 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
876 hpk.hpk_flags |= HP_FLAG_RETRY;
877 /* hpk_errval is >= 0 */
878 hpk.hpk_errval = -PTR_ERR(inode);
879 GOTO(progress, rc = PTR_ERR(inode));
882 /* Read current file data version */
883 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
886 CDEBUG(D_HSM, "Could not read file data version of "
887 DFID" (rc = %d). Archive request ("
888 "%#llx) could not be done.\n",
889 PFID(©->hc_hai.hai_fid), rc,
890 copy->hc_hai.hai_cookie);
891 hpk.hpk_flags |= HP_FLAG_RETRY;
892 /* hpk_errval must be >= 0 */
893 hpk.hpk_errval = -rc;
897 /* Store in the hsm_copy for later copytool use.
898 * Always modified even if no lsm. */
899 copy->hc_data_version = data_version;
903 /* On error, the request should be considered as completed */
904 if (hpk.hpk_errval > 0)
905 hpk.hpk_flags |= HP_FLAG_COMPLETED;
907 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
910 /* Return first error */
911 RETURN(rc != 0 ? rc : rc2);
915 * Generic handler to do any post-copy work.
917 * It will send the last hsm_progress update to coordinator to inform it
918 * that copy is finished and whether it was successful or not.
921 * - for ARCHIVE request, it will sample the file data version and compare it
922 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
923 * will be considered as failed.
924 * - for RESTORE request, it will sample the file data version and send it to
925 * coordinator which is useful if the file was imported as 'released'.
927 * \return 0 on success.
929 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
931 struct ll_sb_info *sbi = ll_s2sbi(sb);
932 struct hsm_progress_kernel hpk;
937 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
938 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
939 * initialized if copy_end was called with copy == NULL.
942 /* Forge a hsm_progress based on data from copy. */
943 hpk.hpk_fid = copy->hc_hai.hai_fid;
944 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
945 hpk.hpk_extent = copy->hc_hai.hai_extent;
946 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
947 hpk.hpk_errval = copy->hc_errval;
948 hpk.hpk_data_version = 0;
950 /* For archive request, we need to check the file data was not changed.
952 * For restore request, we need to send the file data version, this is
953 * useful when the file was created using hsm_import.
955 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
956 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
957 (copy->hc_errval == 0)) {
959 __u64 data_version = 0;
961 /* Get lsm for this fid */
962 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
964 hpk.hpk_flags |= HP_FLAG_RETRY;
965 /* hpk_errval must be >= 0 */
966 hpk.hpk_errval = -PTR_ERR(inode);
967 GOTO(progress, rc = PTR_ERR(inode));
970 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
973 CDEBUG(D_HSM, "Could not read file data version. "
974 "Request could not be confirmed.\n");
975 if (hpk.hpk_errval == 0)
976 hpk.hpk_errval = -rc;
980 /* Store in the hsm_copy for later copytool use.
981 * Always modified even if no lsm. */
982 hpk.hpk_data_version = data_version;
984 /* File could have been stripped during archiving, so we need
985 * to check anyway. */
986 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
987 (copy->hc_data_version != data_version)) {
988 CDEBUG(D_HSM, "File data version mismatched. "
989 "File content was changed during archiving. "
990 DFID", start:%#llx current:%#llx\n",
991 PFID(©->hc_hai.hai_fid),
992 copy->hc_data_version, data_version);
993 /* File was changed, send error to cdt. Do not ask for
994 * retry because if a file is modified frequently,
995 * the cdt will loop on retried archive requests.
996 * The policy engine will ask for a new archive later
997 * when the file will not be modified for some tunable
999 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1001 /* hpk_errval must be >= 0 */
1002 hpk.hpk_errval = -rc;
1009 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1012 /* Return first error */
1013 RETURN(rc != 0 ? rc : rc2);
1017 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1018 const struct lustre_kernelcomm __user *data)
1020 struct lustre_kernelcomm *lk;
1021 struct lustre_kernelcomm *tmp;
1022 size_t size = sizeof(*lk);
1027 /* copy data from userspace to get numbers of archive_id */
1028 OBD_ALLOC(lk, size);
1032 if (copy_from_user(lk, data, size))
1033 GOTO(out_lk, rc = -EFAULT);
1035 if (lk->lk_flags & LK_FLG_STOP)
1038 if (!(lk->lk_flags & LK_FLG_DATANR)) {
1039 __u32 archive_mask = lk->lk_data_count;
1042 /* old hsm agent to old MDS */
1043 if (!exp_connect_archive_id_array(exp))
1046 /* old hsm agent to new MDS */
1047 lk->lk_flags |= LK_FLG_DATANR;
1049 if (archive_mask == 0)
1052 count = hweight32(archive_mask);
1053 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1054 OBD_ALLOC(tmp, new_size);
1056 GOTO(out_lk, rc = -ENOMEM);
1058 memcpy(tmp, lk, size);
1059 tmp->lk_data_count = count;
1065 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1066 if ((1 << i) & archive_mask) {
1067 lk->lk_data[count] = i + 1;
1074 /* new hsm agent to new mds */
1075 if (lk->lk_data_count > 0) {
1076 new_size = offsetof(struct lustre_kernelcomm,
1077 lk_data[lk->lk_data_count]);
1078 OBD_ALLOC(tmp, new_size);
1080 GOTO(out_lk, rc = -ENOMEM);
1086 if (copy_from_user(lk, data, size))
1087 GOTO(out_lk, rc = -EFAULT);
1090 /* new hsm agent to old MDS */
1091 if (!exp_connect_archive_id_array(exp)) {
1094 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1095 GOTO(out_lk, rc = -EINVAL);
1097 for (i = 0; i < lk->lk_data_count; i++) {
1098 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1100 CERROR("%s: archive id %d requested but only "
1101 "[0 - %zu] supported: rc = %d\n",
1102 exp->exp_obd->obd_name, lk->lk_data[i],
1103 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1107 if (lk->lk_data[i] == 0) {
1112 archives |= (1 << (lk->lk_data[i] - 1));
1114 lk->lk_flags &= ~LK_FLG_DATANR;
1115 lk->lk_data_count = archives;
1118 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1124 static int check_owner(int type, int id)
1128 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1132 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1141 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1143 int cmd = qctl->qc_cmd;
1144 int type = qctl->qc_type;
1145 int id = qctl->qc_id;
1146 int valid = qctl->qc_valid;
1153 case LUSTRE_Q_SETDEFAULT:
1154 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1158 case LUSTRE_Q_GETDEFAULT:
1159 if (check_owner(type, id) &&
1160 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1166 CERROR("unsupported quotactl op: %#x\n", cmd);
1170 if (valid != QC_GENERAL) {
1171 if (cmd == Q_GETINFO)
1172 qctl->qc_cmd = Q_GETOINFO;
1173 else if (cmd == Q_GETQUOTA)
1174 qctl->qc_cmd = Q_GETOQUOTA;
1180 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1181 sizeof(*qctl), qctl, NULL);
1184 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1185 sizeof(*qctl), qctl, NULL);
1188 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1189 sizeof(*qctl), qctl, NULL);
1191 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1193 sizeof(*qctl), qctl, NULL);
1205 struct obd_quotactl *oqctl;
1207 OBD_ALLOC_PTR(oqctl);
1211 QCTL_COPY(oqctl, qctl);
1212 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1214 OBD_FREE_PTR(oqctl);
1217 /* If QIF_SPACE is not set, client should collect the
1218 * space usage from OSSs by itself */
1219 if (cmd == Q_GETQUOTA &&
1220 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1221 !oqctl->qc_dqblk.dqb_curspace) {
1222 struct obd_quotactl *oqctl_tmp;
1224 OBD_ALLOC_PTR(oqctl_tmp);
1225 if (oqctl_tmp == NULL)
1226 GOTO(out, rc = -ENOMEM);
1228 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1229 oqctl_tmp->qc_id = oqctl->qc_id;
1230 oqctl_tmp->qc_type = oqctl->qc_type;
1232 /* collect space usage from OSTs */
1233 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1234 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1235 if (!rc || rc == -EREMOTEIO) {
1236 oqctl->qc_dqblk.dqb_curspace =
1237 oqctl_tmp->qc_dqblk.dqb_curspace;
1238 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1241 /* collect space & inode usage from MDTs */
1242 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1243 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1244 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1245 if (!rc || rc == -EREMOTEIO) {
1246 oqctl->qc_dqblk.dqb_curspace +=
1247 oqctl_tmp->qc_dqblk.dqb_curspace;
1248 oqctl->qc_dqblk.dqb_curinodes =
1249 oqctl_tmp->qc_dqblk.dqb_curinodes;
1250 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1252 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1255 OBD_FREE_PTR(oqctl_tmp);
1258 QCTL_COPY(qctl, oqctl);
1259 OBD_FREE_PTR(oqctl);
1265 /* This function tries to get a single name component,
1266 * to send to the server. No actual path traversal involved,
1267 * so we limit to NAME_MAX */
1268 static char *ll_getname(const char __user *filename)
1273 OBD_ALLOC(tmp, NAME_MAX + 1);
1276 return ERR_PTR(-ENOMEM);
1278 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1281 else if (len > NAME_MAX)
1282 ret = -ENAMETOOLONG;
1285 OBD_FREE(tmp, NAME_MAX + 1);
1291 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1293 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1295 struct dentry *dentry = file_dentry(file);
1296 struct inode *inode = file_inode(file);
1297 struct ll_sb_info *sbi = ll_i2sbi(inode);
1298 struct obd_ioctl_data *data;
1302 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1303 PFID(ll_inode2fid(inode)), inode, cmd);
1305 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1306 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1309 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1311 case FS_IOC_GETFLAGS:
1312 case FS_IOC_SETFLAGS:
1313 RETURN(ll_iocontrol(inode, file, cmd, arg));
1314 case FSFILT_IOC_GETVERSION:
1315 case FS_IOC_GETVERSION:
1316 RETURN(put_user(inode->i_generation, (int __user *)arg));
1317 /* We need to special case any other ioctls we want to handle,
1318 * to send them to the MDS/OST as appropriate and to properly
1319 * network encode the arg field. */
1320 case FS_IOC_SETVERSION:
1323 case LL_IOC_GET_MDTIDX: {
1326 mdtidx = ll_get_mdt_idx(inode);
1330 if (put_user((int)mdtidx, (int __user *)arg))
1335 case IOC_MDC_LOOKUP: {
1336 int namelen, len = 0;
1340 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1345 filename = data->ioc_inlbuf1;
1346 namelen = strlen(filename);
1348 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1349 GOTO(out_free, rc = -EINVAL);
1352 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1354 CERROR("%s: lookup %.*s failed: rc = %d\n",
1355 sbi->ll_fsname, namelen, filename, rc);
1359 OBD_FREE_LARGE(buf, len);
1362 case LL_IOC_LMV_SETSTRIPE: {
1363 struct lmv_user_md *lum;
1372 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1377 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1378 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1379 GOTO(lmv_out_free, rc = -EINVAL);
1381 filename = data->ioc_inlbuf1;
1382 namelen = data->ioc_inllen1;
1385 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1386 GOTO(lmv_out_free, rc = -EINVAL);
1388 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1389 lumlen = data->ioc_inllen2;
1391 if (!lmv_magic_supported(lum->lum_magic)) {
1392 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1393 lum->lum_magic, -EINVAL);
1394 GOTO(lmv_out_free, rc = -EINVAL);
1397 if ((lum->lum_magic == LMV_USER_MAGIC ||
1398 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1399 lumlen < sizeof(*lum)) {
1400 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1401 filename, lumlen, lum->lum_magic, -EINVAL);
1402 GOTO(lmv_out_free, rc = -EINVAL);
1405 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1406 lumlen < sizeof(struct lmv_foreign_md)) {
1407 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1408 filename, lum->lum_magic, lumlen, -EFAULT);
1409 GOTO(lmv_out_free, rc = -EINVAL);
1412 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1413 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1415 mode = data->ioc_type;
1417 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1419 OBD_FREE_LARGE(buf, len);
1423 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1424 struct lmv_user_md lum;
1425 struct lmv_user_md __user *ulump =
1426 (struct lmv_user_md __user *)arg;
1429 if (copy_from_user(&lum, ulump, sizeof(lum)))
1432 if (lum.lum_magic != LMV_USER_MAGIC)
1435 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1439 case LL_IOC_LOV_SETSTRIPE_NEW:
1440 case LL_IOC_LOV_SETSTRIPE: {
1441 struct lov_user_md_v3 *lumv3 = NULL;
1442 struct lov_user_md_v1 lumv1;
1443 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1444 struct lov_user_md_v1 __user *lumv1p =
1445 (struct lov_user_md_v1 __user *)arg;
1446 struct lov_user_md_v3 __user *lumv3p =
1447 (struct lov_user_md_v3 __user *)arg;
1450 int set_default = 0;
1452 CLASSERT(sizeof(struct lov_user_md_v3) >
1453 sizeof(struct lov_comp_md_v1));
1454 CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
1455 /* first try with v1 which is smaller than v3 */
1456 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1459 if (inode->i_sb->s_root == file_dentry(file))
1462 switch (lumv1.lmm_magic) {
1463 case LOV_USER_MAGIC_V3:
1464 case LOV_USER_MAGIC_SPECIFIC:
1465 lum_size = ll_lov_user_md_size(&lumv1);
1468 OBD_ALLOC(lumv3, lum_size);
1471 if (copy_from_user(lumv3, lumv3p, lum_size))
1472 GOTO(out, rc = -EFAULT);
1473 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1475 case LOV_USER_MAGIC_V1:
1478 GOTO(out, rc = -ENOTSUPP);
1481 /* in v1 and v3 cases lumv1 points to data */
1482 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1485 OBD_FREE(lumv3, lum_size);
1488 case LL_IOC_LMV_GETSTRIPE: {
1489 struct lmv_user_md __user *ulmv =
1490 (struct lmv_user_md __user *)arg;
1491 struct lmv_user_md lum;
1492 struct ptlrpc_request *request = NULL;
1493 union lmv_mds_md *lmm = NULL;
1496 struct lmv_user_md *tmp = NULL;
1500 int max_stripe_count;
1504 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1507 max_stripe_count = lum.lum_stripe_count;
1508 /* lum_magic will indicate which stripe the ioctl will like
1509 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1510 * is for default LMV stripe */
1511 if (lum.lum_magic == LMV_MAGIC_V1)
1512 valid |= OBD_MD_MEA;
1513 else if (lum.lum_magic == LMV_USER_MAGIC)
1514 valid |= OBD_MD_DEFAULT_MEA;
1518 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1521 GOTO(finish_req, rc);
1523 /* Get default LMV EA */
1524 if (lum.lum_magic == LMV_USER_MAGIC) {
1525 if (lmmsize > sizeof(*ulmv))
1526 GOTO(finish_req, rc = -EINVAL);
1528 if (copy_to_user(ulmv, lmm, lmmsize))
1529 GOTO(finish_req, rc = -EFAULT);
1531 GOTO(finish_req, rc);
1534 /* if foreign LMV case, fake stripes number */
1535 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1536 struct lmv_foreign_md *lfm;
1538 lfm = (struct lmv_foreign_md *)lmm;
1539 if (lfm->lfm_length < XATTR_SIZE_MAX -
1540 offsetof(typeof(*lfm), lfm_value)) {
1541 __u32 size = lfm->lfm_length +
1542 offsetof(typeof(*lfm), lfm_value);
1544 stripe_count = lmv_foreign_to_md_stripes(size);
1546 CERROR("invalid %d foreign size returned\n",
1551 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1553 if (max_stripe_count < stripe_count) {
1554 lum.lum_stripe_count = stripe_count;
1555 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1556 GOTO(finish_req, rc = -EFAULT);
1557 GOTO(finish_req, rc = -E2BIG);
1560 /* enough room on user side and foreign case */
1561 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1562 struct lmv_foreign_md *lfm;
1565 lfm = (struct lmv_foreign_md *)lmm;
1566 size = lfm->lfm_length +
1567 offsetof(struct lmv_foreign_md, lfm_value);
1568 if (copy_to_user(ulmv, lfm, size))
1569 GOTO(finish_req, rc = -EFAULT);
1570 GOTO(finish_req, rc);
1573 lum_size = lmv_user_md_size(stripe_count,
1574 LMV_USER_MAGIC_SPECIFIC);
1575 OBD_ALLOC(tmp, lum_size);
1577 GOTO(finish_req, rc = -ENOMEM);
1579 mdt_index = ll_get_mdt_idx(inode);
1581 GOTO(out_tmp, rc = -ENOMEM);
1583 tmp->lum_magic = LMV_MAGIC_V1;
1584 tmp->lum_stripe_count = 0;
1585 tmp->lum_stripe_offset = mdt_index;
1586 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1587 for (i = 0; i < stripe_count; i++) {
1590 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1591 if (fid_is_sane(&fid)) {
1592 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1594 GOTO(out_tmp, rc = mdt_index);
1596 tmp->lum_objects[i].lum_mds = mdt_index;
1597 tmp->lum_objects[i].lum_fid = fid;
1600 tmp->lum_stripe_count++;
1603 if (copy_to_user(ulmv, tmp, lum_size))
1604 GOTO(out_tmp, rc = -EFAULT);
1606 OBD_FREE(tmp, lum_size);
1608 ptlrpc_req_finished(request);
1612 case LL_IOC_REMOVE_ENTRY: {
1613 char *filename = NULL;
1617 /* Here is a little hack to avoid sending REINT_RMENTRY to
1618 * unsupported server, which might crash the server(LU-2730),
1619 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1620 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1621 * server will support REINT_RMENTRY XXX*/
1622 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1625 filename = ll_getname((const char __user *)arg);
1626 if (IS_ERR(filename))
1627 RETURN(PTR_ERR(filename));
1629 namelen = strlen(filename);
1631 GOTO(out_rmdir, rc = -EINVAL);
1633 rc = ll_rmdir_entry(inode, filename, namelen);
1636 ll_putname(filename);
1639 case LL_IOC_LOV_SWAP_LAYOUTS:
1641 case IOC_OBD_STATFS:
1642 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1643 case LL_IOC_LOV_GETSTRIPE:
1644 case LL_IOC_LOV_GETSTRIPE_NEW:
1645 case LL_IOC_MDC_GETINFO:
1646 case IOC_MDC_GETFILEINFO:
1647 case IOC_MDC_GETFILESTRIPE: {
1648 struct ptlrpc_request *request = NULL;
1649 struct lov_user_md __user *lump;
1650 struct lov_mds_md *lmm = NULL;
1651 struct mdt_body *body;
1652 char *filename = NULL;
1655 if (cmd == IOC_MDC_GETFILEINFO ||
1656 cmd == IOC_MDC_GETFILESTRIPE) {
1657 filename = ll_getname((const char __user *)arg);
1658 if (IS_ERR(filename))
1659 RETURN(PTR_ERR(filename));
1661 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1662 &lmmsize, &request);
1664 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1669 body = req_capsule_server_get(&request->rq_pill,
1671 LASSERT(body != NULL);
1676 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1677 cmd == LL_IOC_MDC_GETINFO)) {
1685 if (cmd == IOC_MDC_GETFILESTRIPE ||
1686 cmd == LL_IOC_LOV_GETSTRIPE ||
1687 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1688 lump = (struct lov_user_md __user *)arg;
1690 struct lov_user_mds_data __user *lmdp;
1691 lmdp = (struct lov_user_mds_data __user *)arg;
1692 lump = &lmdp->lmd_lmm;
1696 /* If the file has no striping then zero out *lump so
1697 * that the caller isn't confused by garbage. */
1698 if (clear_user(lump, sizeof(*lump)))
1699 GOTO(out_req, rc = -EFAULT);
1700 } else if (copy_to_user(lump, lmm, lmmsize)) {
1701 if (copy_to_user(lump, lmm, sizeof(*lump)))
1702 GOTO(out_req, rc = -EFAULT);
1706 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1707 struct lov_user_mds_data __user *lmdp;
1710 st.st_dev = inode->i_sb->s_dev;
1711 st.st_mode = body->mbo_mode;
1712 st.st_nlink = body->mbo_nlink;
1713 st.st_uid = body->mbo_uid;
1714 st.st_gid = body->mbo_gid;
1715 st.st_rdev = body->mbo_rdev;
1716 st.st_size = body->mbo_size;
1717 st.st_blksize = PAGE_SIZE;
1718 st.st_blocks = body->mbo_blocks;
1719 st.st_atime = body->mbo_atime;
1720 st.st_mtime = body->mbo_mtime;
1721 st.st_ctime = body->mbo_ctime;
1722 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1726 lmdp = (struct lov_user_mds_data __user *)arg;
1727 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1728 GOTO(out_req, rc = -EFAULT);
1733 ptlrpc_req_finished(request);
1735 ll_putname(filename);
1738 case OBD_IOC_QUOTACTL: {
1739 struct if_quotactl *qctl;
1741 OBD_ALLOC_PTR(qctl);
1745 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1746 GOTO(out_quotactl, rc = -EFAULT);
1748 rc = quotactl_ioctl(sbi, qctl);
1751 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1758 case OBD_IOC_GETDTNAME:
1759 case OBD_IOC_GETMDNAME:
1760 RETURN(ll_get_obd_name(inode, cmd, arg));
1761 case LL_IOC_FLUSHCTX:
1762 RETURN(ll_flush_ctx(inode));
1763 case LL_IOC_GETOBDCOUNT: {
1765 struct obd_export *exp;
1767 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1770 /* get ost count when count is zero, get mdt count otherwise */
1771 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1772 vallen = sizeof(count);
1773 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1774 KEY_TGT_COUNT, &vallen, &count);
1776 CERROR("get target count failed: %d\n", rc);
1780 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1785 case LL_IOC_PATH2FID:
1786 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1787 sizeof(struct lu_fid)))
1790 case LL_IOC_GET_CONNECT_FLAGS: {
1791 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1792 (void __user *)arg));
1794 case OBD_IOC_FID2PATH:
1795 RETURN(ll_fid2path(inode, (void __user *)arg));
1796 case LL_IOC_GETPARENT:
1797 RETURN(ll_getparent(file, (void __user *)arg));
1798 case LL_IOC_FID2MDTIDX: {
1799 struct obd_export *exp = ll_i2mdexp(inode);
1803 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1807 /* Call mdc_iocontrol */
1808 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1809 (__u32 __user *)&index);
1815 case LL_IOC_HSM_REQUEST: {
1816 struct hsm_user_request *hur;
1823 /* We don't know the true size yet; copy the fixed-size part */
1824 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1829 /* Compute the whole struct size */
1830 totalsize = hur_len(hur);
1835 /* Final size will be more than double totalsize */
1836 if (totalsize >= MDS_MAXREQSIZE / 3)
1839 OBD_ALLOC_LARGE(hur, totalsize);
1843 /* Copy the whole struct */
1844 if (copy_from_user(hur, (void __user *)arg, totalsize))
1845 GOTO(out_hur, rc = -EFAULT);
1847 if (hur->hur_request.hr_action == HUA_RELEASE) {
1848 const struct lu_fid *fid;
1852 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1853 fid = &hur->hur_user_item[i].hui_fid;
1854 f = search_inode_for_lustre(inode->i_sb, fid);
1860 rc = ll_hsm_release(f);
1866 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1871 OBD_FREE_LARGE(hur, totalsize);
1875 case LL_IOC_HSM_PROGRESS: {
1876 struct hsm_progress_kernel hpk;
1877 struct hsm_progress hp;
1879 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1882 hpk.hpk_fid = hp.hp_fid;
1883 hpk.hpk_cookie = hp.hp_cookie;
1884 hpk.hpk_extent = hp.hp_extent;
1885 hpk.hpk_flags = hp.hp_flags;
1886 hpk.hpk_errval = hp.hp_errval;
1887 hpk.hpk_data_version = 0;
1889 /* File may not exist in Lustre; all progress
1890 * reported to Lustre root */
1891 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1895 case LL_IOC_HSM_CT_START:
1896 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1899 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
1900 (struct lustre_kernelcomm __user *)arg);
1903 case LL_IOC_HSM_COPY_START: {
1904 struct hsm_copy *copy;
1907 OBD_ALLOC_PTR(copy);
1910 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1915 rc = ll_ioc_copy_start(inode->i_sb, copy);
1916 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1922 case LL_IOC_HSM_COPY_END: {
1923 struct hsm_copy *copy;
1926 OBD_ALLOC_PTR(copy);
1929 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1934 rc = ll_ioc_copy_end(inode->i_sb, copy);
1935 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1941 case LL_IOC_MIGRATE: {
1942 struct lmv_user_md *lum;
1949 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1953 data = (struct obd_ioctl_data *)buf;
1954 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1955 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1956 GOTO(migrate_free, rc = -EINVAL);
1958 filename = data->ioc_inlbuf1;
1959 namelen = data->ioc_inllen1;
1961 if (namelen < 1 || namelen != strlen(filename) + 1) {
1962 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1963 GOTO(migrate_free, rc = -EINVAL);
1966 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1967 if (lum->lum_magic != LMV_USER_MAGIC &&
1968 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
1970 CERROR("%s: wrong lum magic %x: rc = %d\n",
1971 filename, lum->lum_magic, rc);
1972 GOTO(migrate_free, rc);
1975 rc = ll_migrate(inode, file, lum, filename);
1977 OBD_FREE_LARGE(buf, len);
1981 case LL_IOC_FSGETXATTR:
1982 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1983 case LL_IOC_FSSETXATTR:
1984 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1985 case LL_IOC_PCC_DETACH_BY_FID: {
1986 struct lu_pcc_detach_fid *detach;
1988 struct inode *inode2;
1991 OBD_ALLOC_PTR(detach);
1995 if (copy_from_user(detach,
1996 (const struct lu_pcc_detach_fid __user *)arg,
1998 GOTO(out_detach, rc = -EFAULT);
2000 fid = &detach->pccd_fid;
2001 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2002 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2004 /* Target inode is not in inode cache, and PCC file
2005 * has aleady released, return immdiately.
2007 GOTO(out_detach, rc = 0);
2009 if (!S_ISREG(inode2->i_mode))
2010 GOTO(out_iput, rc = -EINVAL);
2012 if (!inode_owner_or_capable(inode2))
2013 GOTO(out_iput, rc = -EPERM);
2015 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2019 OBD_FREE_PTR(detach);
2023 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2024 (void __user *)arg));
2028 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2030 struct inode *inode = file->f_mapping->host;
2031 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2032 struct ll_sb_info *sbi = ll_i2sbi(inode);
2033 int api32 = ll_need_32bit_api(sbi);
2034 loff_t ret = -EINVAL;
2042 offset += file->f_pos;
2048 offset += LL_DIR_END_OFF_32BIT;
2050 offset += LL_DIR_END_OFF;
2057 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2058 (!api32 && offset <= LL_DIR_END_OFF))) {
2059 if (offset != file->f_pos) {
2060 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2061 (!api32 && offset == LL_DIR_END_OFF))
2062 fd->lfd_pos = MDS_DIR_END_OFF;
2063 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2064 fd->lfd_pos = offset << 32;
2066 fd->lfd_pos = offset;
2067 file->f_pos = offset;
2068 file->f_version = 0;
2075 inode_unlock(inode);
2079 static int ll_dir_open(struct inode *inode, struct file *file)
2082 RETURN(ll_file_open(inode, file));
2085 static int ll_dir_release(struct inode *inode, struct file *file)
2088 RETURN(ll_file_release(inode, file));
2091 const struct file_operations ll_dir_operations = {
2092 .llseek = ll_dir_seek,
2093 .open = ll_dir_open,
2094 .release = ll_dir_release,
2095 .read = generic_read_dir,
2096 #ifdef HAVE_DIR_CONTEXT
2097 .iterate = ll_iterate,
2099 .readdir = ll_readdir,
2101 .unlocked_ioctl = ll_dir_ioctl,