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 #include <linux/uidgid.h>
44 #include <linux/uaccess.h>
45 #include <linux/buffer_head.h> // for wait_on_buffer
46 #include <linux/pagevec.h>
48 #define DEBUG_SUBSYSTEM S_LLITE
50 #include <obd_support.h>
51 #include <obd_class.h>
52 #include <uapi/linux/lustre/lustre_ioctl.h>
53 #include <uapi/linux/llcrypt.h>
54 #include <lustre_lib.h>
55 #include <lustre_dlm.h>
56 #include <lustre_fid.h>
57 #include <lustre_kernelcomm.h>
58 #include <lustre_swab.h>
60 #include "llite_internal.h"
63 * (new) readdir implementation overview.
65 * Original lustre readdir implementation cached exact copy of raw directory
66 * pages on the client. These pages were indexed in client page cache by
67 * logical offset in the directory file. This design, while very simple and
68 * intuitive had some inherent problems:
70 * . it implies that byte offset to the directory entry serves as a
71 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
72 * ext3/htree directory entries may move due to splits, and more
75 * . it is incompatible with the design of split directories for cmd3,
76 * that assumes that names are distributed across nodes based on their
77 * hash, and so readdir should be done in hash order.
79 * New readdir implementation does readdir in hash order, and uses hash of a
80 * file name as a telldir/seekdir cookie. This led to number of complications:
82 * . hash is not unique, so it cannot be used to index cached directory
83 * pages on the client (note, that it requires a whole pageful of hash
84 * collided entries to cause two pages to have identical hashes);
86 * . hash is not unique, so it cannot, strictly speaking, be used as an
87 * entry cookie. ext3/htree has the same problem and lustre implementation
88 * mimics their solution: seekdir(hash) positions directory at the first
89 * entry with the given hash.
95 * Client caches directory pages using hash of the first entry as an index. As
96 * noted above hash is not unique, so this solution doesn't work as is:
97 * special processing is needed for "page hash chains" (i.e., sequences of
98 * pages filled with entries all having the same hash value).
100 * First, such chains have to be detected. To this end, server returns to the
101 * client the hash of the first entry on the page next to one returned. When
102 * client detects that this hash is the same as hash of the first entry on the
103 * returned page, page hash collision has to be handled. Pages in the
104 * hash chain, except first one, are termed "overflow pages".
106 * Solution to index uniqueness problem is to not cache overflow
107 * pages. Instead, when page hash collision is detected, all overflow pages
108 * from emerging chain are immediately requested from the server and placed in
109 * a special data structure (struct ll_dir_chain). This data structure is used
110 * by ll_readdir() to process entries from overflow pages. When readdir
111 * invocation finishes, overflow pages are discarded. If page hash collision
112 * chain weren't completely processed, next call to readdir will again detect
113 * page hash collision, again read overflow pages in, process next portion of
114 * entries and again discard the pages. This is not as wasteful as it looks,
115 * because, given reasonable hash, page hash collisions are extremely rare.
117 * 1. directory positioning
119 * When seekdir(hash) is called, original
130 * identification of and access to overflow pages
134 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
135 * a header lu_dirpage which describes the start/end hash, and whether this
136 * page is empty (contains no dir entry) or hash collide with next page.
137 * After client receives reply, several pages will be integrated into dir page
138 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
139 * lu_dirpage for this integrated page will be adjusted. See
140 * mdc_adjust_dirpages().
143 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
144 __u64 offset, struct ll_dir_chain *chain)
146 struct md_callback cb_op;
150 cb_op.md_blocking_ast = ll_md_blocking_ast;
151 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
158 void ll_release_page(struct inode *inode, struct page *page,
163 /* Always remove the page for striped dir, because the page is
164 * built from temporarily in LMV layer */
165 if (inode && ll_dir_striped(inode)) {
172 if (likely(page->mapping != NULL))
173 delete_from_page_cache(page);
179 #ifdef HAVE_DIR_CONTEXT
180 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
181 struct dir_context *ctx)
184 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
185 void *cookie, filldir_t filldir)
188 struct ll_sb_info *sbi = ll_i2sbi(inode);
190 bool is_api32 = ll_need_32bit_api(sbi);
191 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
193 struct ll_dir_chain chain;
198 ll_dir_chain_init(&chain);
200 page = ll_get_dir_page(inode, op_data, pos, &chain);
202 while (rc == 0 && !done) {
203 struct lu_dirpage *dp;
204 struct lu_dirent *ent;
213 hash = MDS_DIR_END_OFF;
214 dp = page_address(page);
215 for (ent = lu_dirent_start(dp); ent != NULL && !done;
216 ent = lu_dirent_next(ent)) {
223 hash = le64_to_cpu(ent->lde_hash);
224 if (hash < pos) /* Skip until we find target hash */
227 namelen = le16_to_cpu(ent->lde_namelen);
228 if (namelen == 0) /* Skip dummy record */
231 if (is_api32 && is_hash64)
235 fid_le_to_cpu(&fid, &ent->lde_fid);
236 ino = cl_fid_build_ino(&fid, is_api32);
237 type = IFTODT(lu_dirent_type_get(ent));
238 /* For ll_nfs_get_name_filldir(), it will try to access
239 * 'ent' through 'lde_name', so the parameter 'name'
240 * for 'filldir()' must be part of the 'ent'. */
241 #ifdef HAVE_DIR_CONTEXT
243 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
246 done = filldir(cookie, ent->lde_name, namelen, lhash,
253 ll_release_page(inode, page, false);
257 next = le64_to_cpu(dp->ldp_hash_end);
259 if (pos == MDS_DIR_END_OFF) {
261 * End of directory reached.
264 ll_release_page(inode, page, false);
267 * Normal case: continue to the next
270 ll_release_page(inode, page,
271 le32_to_cpu(dp->ldp_flags) &
274 page = ll_get_dir_page(inode, op_data, pos,
278 #ifdef HAVE_DIR_CONTEXT
283 ll_dir_chain_fini(&chain);
287 #ifdef HAVE_DIR_CONTEXT
288 static int ll_iterate(struct file *filp, struct dir_context *ctx)
290 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
293 struct inode *inode = file_inode(filp);
294 struct ll_file_data *lfd = filp->private_data;
295 struct ll_sb_info *sbi = ll_i2sbi(inode);
296 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
297 int api32 = ll_need_32bit_api(sbi);
298 struct md_op_data *op_data;
299 struct lu_fid pfid = { 0 };
300 ktime_t kstart = ktime_get();
311 "VFS Op:inode="DFID"(%p) pos/size%lu/%llu 32bit_api %d\n",
312 PFID(ll_inode2fid(inode)),
313 inode, (unsigned long)pos, i_size_read(inode), api32);
315 if (pos == MDS_DIR_END_OFF)
321 if (unlikely(ll_dir_striped(inode))) {
323 * This is only needed for striped dir to fill ..,
324 * see lmv_read_page()
326 if (file_dentry(filp)->d_parent != NULL &&
327 file_dentry(filp)->d_parent->d_inode != NULL) {
328 __u64 ibits = MDS_INODELOCK_LOOKUP;
329 struct inode *parent =
330 file_dentry(filp)->d_parent->d_inode;
332 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
333 pfid = *ll_inode2fid(parent);
336 /* If it can not find in cache, do lookup .. on the master
338 if (fid_is_zero(&pfid)) {
339 rc = ll_dir_get_parent_fid(inode, &pfid);
345 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
346 LUSTRE_OPC_ANY, inode);
348 GOTO(out, rc = PTR_ERR(op_data));
350 /* foreign dirs are browsed out of Lustre */
351 if (unlikely(op_data->op_mea1 != NULL &&
352 op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
353 ll_finish_md_op_data(op_data);
357 op_data->op_fid3 = pfid;
359 #ifdef HAVE_DIR_CONTEXT
361 rc = ll_dir_read(inode, &pos, op_data, ctx);
364 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
369 if (pos == MDS_DIR_END_OFF) {
371 pos = LL_DIR_END_OFF_32BIT;
373 pos = LL_DIR_END_OFF;
378 #ifdef HAVE_DIR_CONTEXT
383 ll_finish_md_op_data(op_data);
387 ll_stats_ops_tally(sbi, LPROC_LL_READDIR,
388 ktime_us_delta(ktime_get(), kstart));
394 * Create striped directory with specified stripe(@lump)
396 * \param[in] dparent the parent of the directory.
397 * \param[in] lump the specified stripes.
398 * \param[in] dirname the name of the directory.
399 * \param[in] mode the specified mode of the directory.
401 * \retval =0 if striped directory is being created successfully.
402 * <0 if the creation is failed.
404 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
405 size_t len, const char *dirname, umode_t mode)
407 struct inode *parent = dparent->d_inode;
408 struct ptlrpc_request *request = NULL;
409 struct md_op_data *op_data;
410 struct ll_sb_info *sbi = ll_i2sbi(parent);
411 struct inode *inode = NULL;
412 struct dentry dentry = {
416 .len = strlen(dirname),
417 .hash = ll_full_name_hash(dparent, dirname,
421 bool encrypt = false;
425 if (unlikely(!lmv_user_magic_supported(lump->lum_magic)))
428 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
430 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
431 PFID(ll_inode2fid(parent)), parent, dirname,
432 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
434 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
437 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
438 PFID(ll_inode2fid(parent)), parent, dirname,
439 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
442 if (lump->lum_stripe_count > 1 &&
443 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
446 if (IS_DEADDIR(parent) &&
447 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
450 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_CRUSH)) {
451 if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
452 LMV_HASH_TYPE_CRUSH) {
453 /* if server doesn't support 'crush' hash type,
454 * switch to fnv_1a_64.
456 lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
457 lump->lum_hash_type |= LMV_HASH_TYPE_FNV_1A_64;
458 } else if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
459 LMV_HASH_TYPE_UNKNOWN) {
460 /* from 2.14 MDT will choose default hash type if client
461 * doesn't set a valid one, while old server doesn't
464 lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
465 lump->lum_hash_type |= LMV_HASH_TYPE_DEFAULT;
469 if (unlikely(!lmv_user_magic_supported(cpu_to_le32(lump->lum_magic))))
470 lustre_swab_lmv_user_md(lump);
472 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
473 mode &= ~current_umask();
474 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
475 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
476 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
479 RETURN(PTR_ERR(op_data));
481 if (IS_ENCRYPTED(parent) ||
482 unlikely(llcrypt_dummy_context_enabled(parent))) {
483 err = llcrypt_get_encryption_info(parent);
485 GOTO(out_op_data, err);
486 if (!llcrypt_has_encryption_key(parent))
487 GOTO(out_op_data, err = -ENOKEY);
491 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
492 /* selinux_dentry_init_security() uses dentry->d_parent and name
493 * to determine the security context for the file. So our fake
494 * dentry should be real enough for this purpose. */
495 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
496 &op_data->op_file_secctx_name,
497 &op_data->op_file_secctx,
498 &op_data->op_file_secctx_size);
500 GOTO(out_op_data, err);
503 op_data->op_cli_flags |= CLI_SET_MEA;
504 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
505 from_kuid(&init_user_ns, current_fsuid()),
506 from_kgid(&init_user_ns, current_fsgid()),
507 cfs_curproc_cap_pack(), 0, &request);
509 GOTO(out_request, err);
511 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
513 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
515 GOTO(out_inode, err);
517 dentry.d_inode = inode;
519 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
521 err = security_inode_notifysecctx(inode,
522 op_data->op_file_secctx,
523 op_data->op_file_secctx_size);
526 err = ll_inode_init_security(&dentry, inode, parent);
529 GOTO(out_inode, err);
532 err = llcrypt_inherit_context(parent, inode, NULL, false);
534 GOTO(out_inode, err);
541 ptlrpc_req_finished(request);
543 ll_finish_md_op_data(op_data);
548 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
551 struct ll_sb_info *sbi = ll_i2sbi(inode);
552 struct md_op_data *op_data;
553 struct ptlrpc_request *req = NULL;
559 switch (lump->lmm_magic) {
560 case LOV_USER_MAGIC_V1:
561 lum_size = sizeof(struct lov_user_md_v1);
563 case LOV_USER_MAGIC_V3:
564 lum_size = sizeof(struct lov_user_md_v3);
566 case LOV_USER_MAGIC_COMP_V1:
567 lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
570 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
571 lustre_swab_lmv_user_md(
572 (struct lmv_user_md *)lump);
573 lum_size = sizeof(struct lmv_user_md);
575 case LOV_USER_MAGIC_SPECIFIC: {
576 struct lov_user_md_v3 *v3 =
577 (struct lov_user_md_v3 *)lump;
578 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
580 lum_size = lov_user_md_size(v3->lmm_stripe_count,
581 LOV_USER_MAGIC_SPECIFIC);
585 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
586 " %#08x != %#08x nor %#08x\n",
587 lump->lmm_magic, LOV_USER_MAGIC_V1,
593 * This is coming from userspace, so should be in
594 * local endian. But the MDS would like it in little
595 * endian, so we swab it before we send it.
597 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
598 le32_to_cpu(LOV_MAGIC_MAGIC))
599 lustre_swab_lov_user_md(lump, 0);
601 lum_size = sizeof(struct lov_user_md_v1);
604 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
605 LUSTRE_OPC_ANY, NULL);
607 RETURN(PTR_ERR(op_data));
609 /* swabbing is done in lov_setstripe() on server side */
610 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
611 ll_finish_md_op_data(op_data);
612 ptlrpc_req_finished(req);
619 static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
621 struct ptlrpc_request **request, u64 valid,
622 enum get_default_layout_type type)
624 struct ll_sb_info *sbi = ll_i2sbi(inode);
625 struct mdt_body *body;
626 struct lov_mds_md *lmm = NULL;
627 struct ptlrpc_request *req = NULL;
629 struct md_op_data *op_data;
633 rc = ll_get_default_mdsize(sbi, &lmm_size);
637 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
638 0, lmm_size, LUSTRE_OPC_ANY,
641 RETURN(PTR_ERR(op_data));
643 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
645 if (type == GET_DEFAULT_LAYOUT_ROOT) {
646 lu_root_fid(&op_data->op_fid1);
647 fid = op_data->op_fid1;
649 fid = *ll_inode2fid(inode);
652 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
653 ll_finish_md_op_data(op_data);
655 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
660 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
661 LASSERT(body != NULL);
663 lmm_size = body->mbo_eadatasize;
665 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
667 GOTO(out, rc = -ENODATA);
670 lmm = req_capsule_server_sized_get(&req->rq_pill,
671 &RMF_MDT_MD, lmm_size);
672 LASSERT(lmm != NULL);
675 * This is coming from the MDS, so is probably in
676 * little endian. We convert it to host endian before
677 * passing it to userspace.
679 /* We don't swab objects for directories */
680 switch (le32_to_cpu(lmm->lmm_magic)) {
683 case LOV_MAGIC_COMP_V1:
684 case LOV_USER_MAGIC_SPECIFIC:
685 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
686 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
689 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
690 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
693 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
694 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
696 case LMV_MAGIC_FOREIGN: {
697 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
699 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
700 __swab32s(&lfm->lfm_magic);
701 __swab32s(&lfm->lfm_length);
702 __swab32s(&lfm->lfm_type);
703 __swab32s(&lfm->lfm_flags);
708 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
713 *plmm_size = lmm_size;
719 * This function will be used to get default LOV/LMV/Default LMV
720 * @valid will be used to indicate which stripe it will retrieve.
721 * If the directory does not have its own default layout, then the
722 * function will request the default layout from root FID.
723 * OBD_MD_MEA LMV stripe EA
724 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
725 * otherwise Default LOV EA.
726 * Each time, it can only retrieve 1 stripe EA
728 int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
729 struct ptlrpc_request **request,
730 struct ptlrpc_request **root_request,
733 struct ptlrpc_request *req = NULL;
734 struct ptlrpc_request *root_req = NULL;
735 struct lov_mds_md *lmm = NULL;
740 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
742 if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
743 !(valid & (OBD_MD_MEA|OBD_MD_DEFAULT_MEA)) && root_request != NULL)
744 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
746 GET_DEFAULT_LAYOUT_ROOT);
749 *plmm_size = lmm_size;
751 if (root_request != NULL)
752 *root_request = root_req;
758 * This function will be used to get default LOV/LMV/Default LMV
759 * @valid will be used to indicate which stripe it will retrieve
760 * OBD_MD_MEA LMV stripe EA
761 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
762 * otherwise Default LOV EA.
763 * Each time, it can only retrieve 1 stripe EA
765 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
766 struct ptlrpc_request **request, u64 valid)
768 struct ptlrpc_request *req = NULL;
769 struct lov_mds_md *lmm = NULL;
774 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
778 *plmm_size = lmm_size;
784 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
786 struct md_op_data *op_data;
791 OBD_ALLOC_PTR(op_data);
795 op_data->op_flags |= MF_GET_MDT_IDX;
796 op_data->op_fid1 = *fid;
797 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
798 mdt_index = op_data->op_mds;
799 OBD_FREE_PTR(op_data);
807 * Get MDT index for the inode.
809 int ll_get_mdt_idx(struct inode *inode)
811 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
815 * Generic handler to do any pre-copy work.
817 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
818 * first information for it that real work has started.
820 * Moreover, for a ARCHIVE request, it will sample the file data version and
821 * store it in \a copy.
823 * \return 0 on success.
825 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
827 struct ll_sb_info *sbi = ll_s2sbi(sb);
828 struct hsm_progress_kernel hpk;
833 /* Forge a hsm_progress based on data from copy. */
834 hpk.hpk_fid = copy->hc_hai.hai_fid;
835 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
836 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
837 hpk.hpk_extent.length = 0;
840 hpk.hpk_data_version = 0;
843 /* For archive request, we need to read the current file version. */
844 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
846 __u64 data_version = 0;
848 /* Get inode for this fid */
849 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
851 hpk.hpk_flags |= HP_FLAG_RETRY;
852 /* hpk_errval is >= 0 */
853 hpk.hpk_errval = -PTR_ERR(inode);
854 GOTO(progress, rc = PTR_ERR(inode));
857 /* Read current file data version */
858 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
861 CDEBUG(D_HSM, "Could not read file data version of "
862 DFID" (rc = %d). Archive request ("
863 "%#llx) could not be done.\n",
864 PFID(©->hc_hai.hai_fid), rc,
865 copy->hc_hai.hai_cookie);
866 hpk.hpk_flags |= HP_FLAG_RETRY;
867 /* hpk_errval must be >= 0 */
868 hpk.hpk_errval = -rc;
872 /* Store in the hsm_copy for later copytool use.
873 * Always modified even if no lsm. */
874 copy->hc_data_version = data_version;
878 /* On error, the request should be considered as completed */
879 if (hpk.hpk_errval > 0)
880 hpk.hpk_flags |= HP_FLAG_COMPLETED;
882 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
885 /* Return first error */
886 RETURN(rc != 0 ? rc : rc2);
890 * Generic handler to do any post-copy work.
892 * It will send the last hsm_progress update to coordinator to inform it
893 * that copy is finished and whether it was successful or not.
896 * - for ARCHIVE request, it will sample the file data version and compare it
897 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
898 * will be considered as failed.
899 * - for RESTORE request, it will sample the file data version and send it to
900 * coordinator which is useful if the file was imported as 'released'.
902 * \return 0 on success.
904 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
906 struct ll_sb_info *sbi = ll_s2sbi(sb);
907 struct hsm_progress_kernel hpk;
912 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
913 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
914 * initialized if copy_end was called with copy == NULL.
917 /* Forge a hsm_progress based on data from copy. */
918 hpk.hpk_fid = copy->hc_hai.hai_fid;
919 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
920 hpk.hpk_extent = copy->hc_hai.hai_extent;
921 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
922 hpk.hpk_errval = copy->hc_errval;
923 hpk.hpk_data_version = 0;
925 /* For archive request, we need to check the file data was not changed.
927 * For restore request, we need to send the file data version, this is
928 * useful when the file was created using hsm_import.
930 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
931 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
932 (copy->hc_errval == 0)) {
934 __u64 data_version = 0;
936 /* Get lsm for this fid */
937 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
939 hpk.hpk_flags |= HP_FLAG_RETRY;
940 /* hpk_errval must be >= 0 */
941 hpk.hpk_errval = -PTR_ERR(inode);
942 GOTO(progress, rc = PTR_ERR(inode));
945 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
948 CDEBUG(D_HSM, "Could not read file data version. "
949 "Request could not be confirmed.\n");
950 if (hpk.hpk_errval == 0)
951 hpk.hpk_errval = -rc;
955 /* Store in the hsm_copy for later copytool use.
956 * Always modified even if no lsm. */
957 hpk.hpk_data_version = data_version;
959 /* File could have been stripped during archiving, so we need
960 * to check anyway. */
961 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
962 (copy->hc_data_version != data_version)) {
963 CDEBUG(D_HSM, "File data version mismatched. "
964 "File content was changed during archiving. "
965 DFID", start:%#llx current:%#llx\n",
966 PFID(©->hc_hai.hai_fid),
967 copy->hc_data_version, data_version);
968 /* File was changed, send error to cdt. Do not ask for
969 * retry because if a file is modified frequently,
970 * the cdt will loop on retried archive requests.
971 * The policy engine will ask for a new archive later
972 * when the file will not be modified for some tunable
974 hpk.hpk_flags &= ~HP_FLAG_RETRY;
976 /* hpk_errval must be >= 0 */
977 hpk.hpk_errval = -rc;
984 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
987 /* Return first error */
988 RETURN(rc != 0 ? rc : rc2);
992 static int copy_and_ct_start(int cmd, struct obd_export *exp,
993 const struct lustre_kernelcomm __user *data)
995 struct lustre_kernelcomm *lk;
996 struct lustre_kernelcomm *tmp;
997 size_t size = sizeof(*lk);
1002 /* copy data from userspace to get numbers of archive_id */
1003 OBD_ALLOC(lk, size);
1007 if (copy_from_user(lk, data, size))
1008 GOTO(out_lk, rc = -EFAULT);
1010 if (lk->lk_flags & LK_FLG_STOP)
1013 if (!(lk->lk_flags & LK_FLG_DATANR)) {
1014 __u32 archive_mask = lk->lk_data_count;
1017 /* old hsm agent to old MDS */
1018 if (!exp_connect_archive_id_array(exp))
1021 /* old hsm agent to new MDS */
1022 lk->lk_flags |= LK_FLG_DATANR;
1024 if (archive_mask == 0)
1027 count = hweight32(archive_mask);
1028 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1029 OBD_ALLOC(tmp, new_size);
1031 GOTO(out_lk, rc = -ENOMEM);
1033 memcpy(tmp, lk, size);
1034 tmp->lk_data_count = count;
1040 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1041 if (BIT(i) & archive_mask) {
1042 lk->lk_data[count] = i + 1;
1049 /* new hsm agent to new mds */
1050 if (lk->lk_data_count > 0) {
1051 new_size = offsetof(struct lustre_kernelcomm,
1052 lk_data[lk->lk_data_count]);
1053 OBD_ALLOC(tmp, new_size);
1055 GOTO(out_lk, rc = -ENOMEM);
1061 if (copy_from_user(lk, data, size))
1062 GOTO(out_lk, rc = -EFAULT);
1065 /* new hsm agent to old MDS */
1066 if (!exp_connect_archive_id_array(exp)) {
1069 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1070 GOTO(out_lk, rc = -EINVAL);
1072 for (i = 0; i < lk->lk_data_count; i++) {
1073 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1075 CERROR("%s: archive id %d requested but only "
1076 "[0 - %zu] supported: rc = %d\n",
1077 exp->exp_obd->obd_name, lk->lk_data[i],
1078 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1082 if (lk->lk_data[i] == 0) {
1087 archives |= (1 << (lk->lk_data[i] - 1));
1089 lk->lk_flags &= ~LK_FLG_DATANR;
1090 lk->lk_data_count = archives;
1093 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1099 static int check_owner(int type, int id)
1103 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1107 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1116 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1118 int cmd = qctl->qc_cmd;
1119 int type = qctl->qc_type;
1120 int id = qctl->qc_id;
1121 int valid = qctl->qc_valid;
1128 case LUSTRE_Q_SETDEFAULT:
1129 case LUSTRE_Q_SETQUOTAPOOL:
1130 case LUSTRE_Q_SETINFOPOOL:
1131 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1135 case LUSTRE_Q_GETDEFAULT:
1136 case LUSTRE_Q_GETQUOTAPOOL:
1137 if (check_owner(type, id) &&
1138 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1142 case LUSTRE_Q_GETINFOPOOL:
1145 CERROR("unsupported quotactl op: %#x\n", cmd);
1149 if (valid != QC_GENERAL) {
1150 if (cmd == Q_GETINFO)
1151 qctl->qc_cmd = Q_GETOINFO;
1152 else if (cmd == Q_GETQUOTA ||
1153 cmd == LUSTRE_Q_GETQUOTAPOOL)
1154 qctl->qc_cmd = Q_GETOQUOTA;
1160 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1161 sizeof(*qctl), qctl, NULL);
1164 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1165 sizeof(*qctl), qctl, NULL);
1168 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1169 sizeof(*qctl), qctl, NULL);
1171 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1173 sizeof(*qctl), qctl, NULL);
1185 struct obd_quotactl *oqctl;
1186 int oqctl_len = sizeof(*oqctl);
1188 if (LUSTRE_Q_CMD_IS_POOL(cmd))
1189 oqctl_len += LOV_MAXPOOLNAME + 1;
1191 OBD_ALLOC(oqctl, oqctl_len);
1195 QCTL_COPY(oqctl, qctl);
1196 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1198 OBD_FREE(oqctl, oqctl_len);
1201 /* If QIF_SPACE is not set, client should collect the
1202 * space usage from OSSs by itself */
1203 if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
1204 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1205 !oqctl->qc_dqblk.dqb_curspace) {
1206 struct obd_quotactl *oqctl_tmp;
1208 OBD_ALLOC_PTR(oqctl_tmp);
1209 if (oqctl_tmp == NULL)
1210 GOTO(out, rc = -ENOMEM);
1212 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1213 oqctl_tmp->qc_id = oqctl->qc_id;
1214 oqctl_tmp->qc_type = oqctl->qc_type;
1216 /* collect space usage from OSTs */
1217 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1218 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1219 if (!rc || rc == -EREMOTEIO) {
1220 oqctl->qc_dqblk.dqb_curspace =
1221 oqctl_tmp->qc_dqblk.dqb_curspace;
1222 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1225 /* collect space & inode usage from MDTs */
1226 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1227 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1228 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1229 if (!rc || rc == -EREMOTEIO) {
1230 oqctl->qc_dqblk.dqb_curspace +=
1231 oqctl_tmp->qc_dqblk.dqb_curspace;
1232 oqctl->qc_dqblk.dqb_curinodes =
1233 oqctl_tmp->qc_dqblk.dqb_curinodes;
1234 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1236 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1239 OBD_FREE_PTR(oqctl_tmp);
1242 QCTL_COPY(qctl, oqctl);
1243 OBD_FREE(oqctl, oqctl_len);
1249 int ll_rmfid(struct file *file, void __user *arg)
1251 const struct fid_array __user *ufa = arg;
1252 struct fid_array *lfa = NULL;
1255 int i, rc, *rcs = NULL;
1258 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1259 !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1261 /* Only need to get the buflen */
1262 if (get_user(nr, &ufa->fa_nr))
1264 /* DoS protection */
1265 if (nr > OBD_MAX_FIDS_IN_ARRAY)
1268 size = offsetof(struct fid_array, fa_fids[nr]);
1269 OBD_ALLOC(lfa, size);
1272 OBD_ALLOC_PTR_ARRAY(rcs, nr);
1274 GOTO(free_lfa, rc = -ENOMEM);
1276 if (copy_from_user(lfa, arg, size))
1277 GOTO(free_rcs, rc = -EFAULT);
1279 /* Call mdc_iocontrol */
1280 rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1282 for (i = 0; i < nr; i++)
1284 lfa->fa_fids[i].f_ver = rcs[i];
1285 if (copy_to_user(arg, lfa, size))
1290 OBD_FREE_PTR_ARRAY(rcs, nr);
1292 OBD_FREE(lfa, size);
1297 /* This function tries to get a single name component,
1298 * to send to the server. No actual path traversal involved,
1299 * so we limit to NAME_MAX */
1300 static char *ll_getname(const char __user *filename)
1305 OBD_ALLOC(tmp, NAME_MAX + 1);
1308 return ERR_PTR(-ENOMEM);
1310 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1313 else if (len > NAME_MAX)
1314 ret = -ENAMETOOLONG;
1317 OBD_FREE(tmp, NAME_MAX + 1);
1323 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1325 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1327 struct dentry *dentry = file_dentry(file);
1328 struct inode *inode = file_inode(file);
1329 struct ll_sb_info *sbi = ll_i2sbi(inode);
1330 struct obd_ioctl_data *data;
1334 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1335 PFID(ll_inode2fid(inode)), inode, cmd);
1337 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1338 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1341 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1343 case FS_IOC_GETFLAGS:
1344 case FS_IOC_SETFLAGS:
1345 RETURN(ll_iocontrol(inode, file, cmd, arg));
1346 case FSFILT_IOC_GETVERSION:
1347 case FS_IOC_GETVERSION:
1348 RETURN(put_user(inode->i_generation, (int __user *)arg));
1349 /* We need to special case any other ioctls we want to handle,
1350 * to send them to the MDS/OST as appropriate and to properly
1351 * network encode the arg field. */
1352 case FS_IOC_SETVERSION:
1355 case LL_IOC_GET_MDTIDX: {
1358 mdtidx = ll_get_mdt_idx(inode);
1362 if (put_user((int)mdtidx, (int __user *)arg))
1367 case IOC_MDC_LOOKUP: {
1368 int namelen, len = 0;
1372 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1377 filename = data->ioc_inlbuf1;
1378 namelen = strlen(filename);
1380 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1381 GOTO(out_free, rc = -EINVAL);
1384 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1386 CERROR("%s: lookup %.*s failed: rc = %d\n",
1387 sbi->ll_fsname, namelen, filename, rc);
1391 OBD_FREE_LARGE(buf, len);
1394 case LL_IOC_LMV_SETSTRIPE: {
1395 struct lmv_user_md *lum;
1404 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1409 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1410 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1411 GOTO(lmv_out_free, rc = -EINVAL);
1413 filename = data->ioc_inlbuf1;
1414 namelen = data->ioc_inllen1;
1417 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1418 GOTO(lmv_out_free, rc = -EINVAL);
1420 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1421 lumlen = data->ioc_inllen2;
1423 if (!lmv_user_magic_supported(lum->lum_magic)) {
1424 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1425 lum->lum_magic, -EINVAL);
1426 GOTO(lmv_out_free, rc = -EINVAL);
1429 if ((lum->lum_magic == LMV_USER_MAGIC ||
1430 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1431 lumlen < sizeof(*lum)) {
1432 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1433 filename, lumlen, lum->lum_magic, -EINVAL);
1434 GOTO(lmv_out_free, rc = -EINVAL);
1437 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1438 lumlen < sizeof(struct lmv_foreign_md)) {
1439 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1440 filename, lum->lum_magic, lumlen, -EFAULT);
1441 GOTO(lmv_out_free, rc = -EINVAL);
1444 mode = data->ioc_type;
1445 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1447 OBD_FREE_LARGE(buf, len);
1451 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1452 struct lmv_user_md lum;
1453 struct lmv_user_md __user *ulump =
1454 (struct lmv_user_md __user *)arg;
1457 if (copy_from_user(&lum, ulump, sizeof(lum)))
1460 if (lum.lum_magic != LMV_USER_MAGIC)
1463 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1467 case LL_IOC_LOV_SETSTRIPE_NEW:
1468 case LL_IOC_LOV_SETSTRIPE: {
1469 struct lov_user_md_v3 *lumv3 = NULL;
1470 struct lov_user_md_v1 lumv1;
1471 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1472 struct lov_user_md_v1 __user *lumv1p =
1473 (struct lov_user_md_v1 __user *)arg;
1474 struct lov_user_md_v3 __user *lumv3p =
1475 (struct lov_user_md_v3 __user *)arg;
1478 int set_default = 0;
1480 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1481 sizeof(struct lov_comp_md_v1));
1482 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1483 /* first try with v1 which is smaller than v3 */
1484 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1487 if (inode->i_sb->s_root == file_dentry(file))
1490 switch (lumv1.lmm_magic) {
1491 case LOV_USER_MAGIC_V3:
1492 case LOV_USER_MAGIC_SPECIFIC:
1493 lum_size = ll_lov_user_md_size(&lumv1);
1496 OBD_ALLOC(lumv3, lum_size);
1499 if (copy_from_user(lumv3, lumv3p, lum_size))
1500 GOTO(out, rc = -EFAULT);
1501 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1503 case LOV_USER_MAGIC_V1:
1506 GOTO(out, rc = -ENOTSUPP);
1509 /* in v1 and v3 cases lumv1 points to data */
1510 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1513 OBD_FREE(lumv3, lum_size);
1516 case LL_IOC_LMV_GETSTRIPE: {
1517 struct lmv_user_md __user *ulmv =
1518 (struct lmv_user_md __user *)arg;
1519 struct lmv_user_md lum;
1520 struct ptlrpc_request *request = NULL;
1521 struct ptlrpc_request *root_request = NULL;
1522 union lmv_mds_md *lmm = NULL;
1525 struct lmv_user_md *tmp = NULL;
1529 int max_stripe_count;
1533 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1536 max_stripe_count = lum.lum_stripe_count;
1537 /* lum_magic will indicate which stripe the ioctl will like
1538 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1539 * is for default LMV stripe */
1540 if (lum.lum_magic == LMV_MAGIC_V1)
1541 valid |= OBD_MD_MEA;
1542 else if (lum.lum_magic == LMV_USER_MAGIC)
1543 valid |= OBD_MD_DEFAULT_MEA;
1547 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1548 &request, &root_request, valid);
1550 GOTO(finish_req, rc);
1552 /* Get default LMV EA */
1553 if (lum.lum_magic == LMV_USER_MAGIC) {
1554 if (lmmsize > sizeof(*ulmv))
1555 GOTO(finish_req, rc = -EINVAL);
1557 if (copy_to_user(ulmv, lmm, lmmsize))
1558 GOTO(finish_req, rc = -EFAULT);
1560 GOTO(finish_req, rc);
1563 /* if foreign LMV case, fake stripes number */
1564 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1565 struct lmv_foreign_md *lfm;
1567 lfm = (struct lmv_foreign_md *)lmm;
1568 if (lfm->lfm_length < XATTR_SIZE_MAX -
1569 offsetof(typeof(*lfm), lfm_value)) {
1570 __u32 size = lfm->lfm_length +
1571 offsetof(typeof(*lfm), lfm_value);
1573 stripe_count = lmv_foreign_to_md_stripes(size);
1575 CERROR("invalid %d foreign size returned\n",
1580 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1582 if (max_stripe_count < stripe_count) {
1583 lum.lum_stripe_count = stripe_count;
1584 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1585 GOTO(finish_req, rc = -EFAULT);
1586 GOTO(finish_req, rc = -E2BIG);
1589 /* enough room on user side and foreign case */
1590 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1591 struct lmv_foreign_md *lfm;
1594 lfm = (struct lmv_foreign_md *)lmm;
1595 size = lfm->lfm_length +
1596 offsetof(struct lmv_foreign_md, lfm_value);
1597 if (copy_to_user(ulmv, lfm, size))
1598 GOTO(finish_req, rc = -EFAULT);
1599 GOTO(finish_req, rc);
1602 lum_size = lmv_user_md_size(stripe_count,
1603 LMV_USER_MAGIC_SPECIFIC);
1604 OBD_ALLOC(tmp, lum_size);
1606 GOTO(finish_req, rc = -ENOMEM);
1608 mdt_index = ll_get_mdt_idx(inode);
1610 GOTO(out_tmp, rc = -ENOMEM);
1612 tmp->lum_magic = LMV_MAGIC_V1;
1613 tmp->lum_stripe_count = 0;
1614 tmp->lum_stripe_offset = mdt_index;
1615 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1616 for (i = 0; i < stripe_count; i++) {
1619 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1620 if (fid_is_sane(&fid)) {
1621 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1623 GOTO(out_tmp, rc = mdt_index);
1625 tmp->lum_objects[i].lum_mds = mdt_index;
1626 tmp->lum_objects[i].lum_fid = fid;
1629 tmp->lum_stripe_count++;
1632 if (copy_to_user(ulmv, tmp, lum_size))
1633 GOTO(out_tmp, rc = -EFAULT);
1635 OBD_FREE(tmp, lum_size);
1637 ptlrpc_req_finished(request);
1638 ptlrpc_req_finished(root_request);
1642 case LL_IOC_REMOVE_ENTRY: {
1643 char *filename = NULL;
1647 /* Here is a little hack to avoid sending REINT_RMENTRY to
1648 * unsupported server, which might crash the server(LU-2730),
1649 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1650 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1651 * server will support REINT_RMENTRY XXX*/
1652 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1655 filename = ll_getname((const char __user *)arg);
1656 if (IS_ERR(filename))
1657 RETURN(PTR_ERR(filename));
1659 namelen = strlen(filename);
1661 GOTO(out_rmdir, rc = -EINVAL);
1663 rc = ll_rmdir_entry(inode, filename, namelen);
1666 ll_putname(filename);
1670 RETURN(ll_rmfid(file, (void __user *)arg));
1671 case LL_IOC_LOV_SWAP_LAYOUTS:
1673 case IOC_OBD_STATFS:
1674 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1675 case LL_IOC_LOV_GETSTRIPE:
1676 case LL_IOC_LOV_GETSTRIPE_NEW:
1677 case LL_IOC_MDC_GETINFO:
1678 case LL_IOC_MDC_GETINFO_OLD:
1679 case IOC_MDC_GETFILEINFO:
1680 case IOC_MDC_GETFILEINFO_OLD:
1681 case IOC_MDC_GETFILESTRIPE: {
1682 struct ptlrpc_request *request = NULL;
1683 struct ptlrpc_request *root_request = NULL;
1684 struct lov_user_md __user *lump;
1685 struct lov_mds_md *lmm = NULL;
1686 struct mdt_body *body;
1687 char *filename = NULL;
1688 lstat_t __user *statp = NULL;
1689 lstatx_t __user *stxp = NULL;
1690 __u64 __user *flagsp = NULL;
1691 __u32 __user *lmmsizep = NULL;
1692 struct lu_fid __user *fidp = NULL;
1695 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1696 cmd == IOC_MDC_GETFILEINFO ||
1697 cmd == IOC_MDC_GETFILESTRIPE) {
1698 filename = ll_getname((const char __user *)arg);
1699 if (IS_ERR(filename))
1700 RETURN(PTR_ERR(filename));
1702 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1703 &lmmsize, &request);
1705 rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1711 body = req_capsule_server_get(&request->rq_pill,
1713 LASSERT(body != NULL);
1718 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1719 cmd == LL_IOC_MDC_GETINFO ||
1720 cmd == IOC_MDC_GETFILEINFO_OLD ||
1721 cmd == LL_IOC_MDC_GETINFO_OLD)) {
1729 if (cmd == IOC_MDC_GETFILESTRIPE ||
1730 cmd == LL_IOC_LOV_GETSTRIPE ||
1731 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1732 lump = (struct lov_user_md __user *)arg;
1733 } else if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1734 cmd == LL_IOC_MDC_GETINFO_OLD){
1735 struct lov_user_mds_data_v1 __user *lmdp;
1737 lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1738 statp = &lmdp->lmd_st;
1739 lump = &lmdp->lmd_lmm;
1741 struct lov_user_mds_data __user *lmdp;
1743 lmdp = (struct lov_user_mds_data __user *)arg;
1744 fidp = &lmdp->lmd_fid;
1745 stxp = &lmdp->lmd_stx;
1746 flagsp = &lmdp->lmd_flags;
1747 lmmsizep = &lmdp->lmd_lmmsize;
1748 lump = &lmdp->lmd_lmm;
1752 /* If the file has no striping then zero out *lump so
1753 * that the caller isn't confused by garbage. */
1754 if (clear_user(lump, sizeof(*lump)))
1755 GOTO(out_req, rc = -EFAULT);
1756 } else if (copy_to_user(lump, lmm, lmmsize)) {
1757 if (copy_to_user(lump, lmm, sizeof(*lump)))
1758 GOTO(out_req, rc = -EFAULT);
1762 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1763 cmd == LL_IOC_MDC_GETINFO_OLD) {
1766 st.st_dev = inode->i_sb->s_dev;
1767 st.st_mode = body->mbo_mode;
1768 st.st_nlink = body->mbo_nlink;
1769 st.st_uid = body->mbo_uid;
1770 st.st_gid = body->mbo_gid;
1771 st.st_rdev = body->mbo_rdev;
1772 st.st_size = body->mbo_size;
1773 st.st_blksize = PAGE_SIZE;
1774 st.st_blocks = body->mbo_blocks;
1775 st.st_atime = body->mbo_atime;
1776 st.st_mtime = body->mbo_mtime;
1777 st.st_ctime = body->mbo_ctime;
1778 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1782 if (copy_to_user(statp, &st, sizeof(st)))
1783 GOTO(out_req, rc = -EFAULT);
1784 } else if (cmd == IOC_MDC_GETFILEINFO ||
1785 cmd == LL_IOC_MDC_GETINFO) {
1786 lstatx_t stx = { 0 };
1787 __u64 valid = body->mbo_valid;
1789 stx.stx_blksize = PAGE_SIZE;
1790 stx.stx_nlink = body->mbo_nlink;
1791 stx.stx_uid = body->mbo_uid;
1792 stx.stx_gid = body->mbo_gid;
1793 stx.stx_mode = body->mbo_mode;
1794 stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1797 stx.stx_size = body->mbo_size;
1798 stx.stx_blocks = body->mbo_blocks;
1799 stx.stx_atime.tv_sec = body->mbo_atime;
1800 stx.stx_ctime.tv_sec = body->mbo_ctime;
1801 stx.stx_mtime.tv_sec = body->mbo_mtime;
1802 stx.stx_btime.tv_sec = body->mbo_btime;
1803 stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1804 stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1805 stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1806 stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1807 stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
1810 * For a striped directory, the size and blocks returned
1811 * from MDT is not correct.
1812 * The size and blocks are aggregated by client across
1814 * Thus for a striped directory, do not return the valid
1815 * FLSIZE and FLBLOCKS flags to the caller.
1816 * However, this whould be better decided by the MDS
1817 * instead of the client.
1819 if (cmd == LL_IOC_MDC_GETINFO &&
1820 ll_i2info(inode)->lli_lsm_md != NULL)
1821 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1823 if (flagsp && copy_to_user(flagsp, &valid,
1825 GOTO(out_req, rc = -EFAULT);
1827 if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1829 GOTO(out_req, rc = -EFAULT);
1831 if (!(valid & OBD_MD_FLSIZE))
1832 stx.stx_mask &= ~STATX_SIZE;
1833 if (!(valid & OBD_MD_FLBLOCKS))
1834 stx.stx_mask &= ~STATX_BLOCKS;
1836 if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1837 GOTO(out_req, rc = -EFAULT);
1839 if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1841 GOTO(out_req, rc = -EFAULT);
1846 ptlrpc_req_finished(request);
1847 ptlrpc_req_finished(root_request);
1849 ll_putname(filename);
1852 case OBD_IOC_QUOTACTL: {
1853 struct if_quotactl *qctl;
1854 int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
1856 OBD_ALLOC(qctl, qctl_len);
1860 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1861 GOTO(out_quotactl, rc = -EFAULT);
1863 if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
1864 char __user *from = (char __user *)arg +
1865 offsetof(typeof(*qctl), qc_poolname);
1866 if (copy_from_user(qctl->qc_poolname, from,
1867 LOV_MAXPOOLNAME + 1))
1868 GOTO(out_quotactl, rc = -EFAULT);
1871 rc = quotactl_ioctl(sbi, qctl);
1873 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1877 OBD_FREE(qctl, qctl_len);
1880 case OBD_IOC_GETDTNAME:
1881 case OBD_IOC_GETMDNAME:
1882 RETURN(ll_get_obd_name(inode, cmd, arg));
1883 case LL_IOC_FLUSHCTX:
1884 RETURN(ll_flush_ctx(inode));
1885 case LL_IOC_GETOBDCOUNT: {
1887 struct obd_export *exp;
1889 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1892 /* get ost count when count is zero, get mdt count otherwise */
1893 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1894 vallen = sizeof(count);
1895 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1896 KEY_TGT_COUNT, &vallen, &count);
1898 CERROR("get target count failed: %d\n", rc);
1902 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1907 case LL_IOC_PATH2FID:
1908 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1909 sizeof(struct lu_fid)))
1912 case LL_IOC_GET_CONNECT_FLAGS: {
1913 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1914 (void __user *)arg));
1916 case OBD_IOC_FID2PATH:
1917 RETURN(ll_fid2path(inode, (void __user *)arg));
1918 case LL_IOC_GETPARENT:
1919 RETURN(ll_getparent(file, (void __user *)arg));
1920 case LL_IOC_FID2MDTIDX: {
1921 struct obd_export *exp = ll_i2mdexp(inode);
1925 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1929 /* Call mdc_iocontrol */
1930 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1931 (__u32 __user *)&index);
1937 case LL_IOC_HSM_REQUEST: {
1938 struct hsm_user_request *hur;
1945 /* We don't know the true size yet; copy the fixed-size part */
1946 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1951 /* Compute the whole struct size */
1952 totalsize = hur_len(hur);
1957 /* Final size will be more than double totalsize */
1958 if (totalsize >= MDS_MAXREQSIZE / 3)
1961 OBD_ALLOC_LARGE(hur, totalsize);
1965 /* Copy the whole struct */
1966 if (copy_from_user(hur, (void __user *)arg, totalsize))
1967 GOTO(out_hur, rc = -EFAULT);
1969 if (hur->hur_request.hr_action == HUA_RELEASE) {
1970 const struct lu_fid *fid;
1974 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1975 fid = &hur->hur_user_item[i].hui_fid;
1976 f = search_inode_for_lustre(inode->i_sb, fid);
1982 rc = ll_hsm_release(f);
1988 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1993 OBD_FREE_LARGE(hur, totalsize);
1997 case LL_IOC_HSM_PROGRESS: {
1998 struct hsm_progress_kernel hpk;
1999 struct hsm_progress hp;
2001 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
2004 hpk.hpk_fid = hp.hp_fid;
2005 hpk.hpk_cookie = hp.hp_cookie;
2006 hpk.hpk_extent = hp.hp_extent;
2007 hpk.hpk_flags = hp.hp_flags;
2008 hpk.hpk_errval = hp.hp_errval;
2009 hpk.hpk_data_version = 0;
2011 /* File may not exist in Lustre; all progress
2012 * reported to Lustre root */
2013 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2017 case LL_IOC_HSM_CT_START:
2018 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2021 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2022 (struct lustre_kernelcomm __user *)arg);
2025 case LL_IOC_HSM_COPY_START: {
2026 struct hsm_copy *copy;
2029 OBD_ALLOC_PTR(copy);
2032 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2037 rc = ll_ioc_copy_start(inode->i_sb, copy);
2038 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2044 case LL_IOC_HSM_COPY_END: {
2045 struct hsm_copy *copy;
2048 OBD_ALLOC_PTR(copy);
2051 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2056 rc = ll_ioc_copy_end(inode->i_sb, copy);
2057 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2063 case LL_IOC_MIGRATE: {
2064 struct lmv_user_md *lum;
2071 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
2075 data = (struct obd_ioctl_data *)buf;
2076 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2077 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2078 GOTO(migrate_free, rc = -EINVAL);
2080 filename = data->ioc_inlbuf1;
2081 namelen = data->ioc_inllen1;
2083 if (namelen < 1 || namelen != strlen(filename) + 1) {
2084 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2085 GOTO(migrate_free, rc = -EINVAL);
2088 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2089 if (lum->lum_magic != LMV_USER_MAGIC &&
2090 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2092 CERROR("%s: wrong lum magic %x: rc = %d\n",
2093 filename, lum->lum_magic, rc);
2094 GOTO(migrate_free, rc);
2097 rc = ll_migrate(inode, file, lum, filename);
2099 OBD_FREE_LARGE(buf, len);
2103 case FS_IOC_FSGETXATTR:
2104 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2105 case FS_IOC_FSSETXATTR:
2106 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2107 case LL_IOC_PCC_DETACH_BY_FID: {
2108 struct lu_pcc_detach_fid *detach;
2110 struct inode *inode2;
2113 OBD_ALLOC_PTR(detach);
2117 if (copy_from_user(detach,
2118 (const struct lu_pcc_detach_fid __user *)arg,
2120 GOTO(out_detach, rc = -EFAULT);
2122 fid = &detach->pccd_fid;
2123 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2124 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2126 /* Target inode is not in inode cache, and PCC file
2127 * has aleady released, return immdiately.
2129 GOTO(out_detach, rc = 0);
2131 if (!S_ISREG(inode2->i_mode))
2132 GOTO(out_iput, rc = -EINVAL);
2134 if (!inode_owner_or_capable(inode2))
2135 GOTO(out_iput, rc = -EPERM);
2137 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2141 OBD_FREE_PTR(detach);
2144 #ifdef HAVE_LUSTRE_CRYPTO
2145 case LL_IOC_SET_ENCRYPTION_POLICY:
2146 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2148 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
2149 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
2150 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2152 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
2153 case LL_IOC_ADD_ENCRYPTION_KEY:
2154 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2156 return llcrypt_ioctl_add_key(file, (void __user *)arg);
2157 case LL_IOC_REMOVE_ENCRYPTION_KEY:
2158 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2160 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
2161 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
2162 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2164 return llcrypt_ioctl_remove_key_all_users(file,
2165 (void __user *)arg);
2166 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
2167 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2169 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
2172 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2173 (void __user *)arg));
2177 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2179 struct inode *inode = file->f_mapping->host;
2180 struct ll_file_data *fd = file->private_data;
2181 struct ll_sb_info *sbi = ll_i2sbi(inode);
2182 int api32 = ll_need_32bit_api(sbi);
2183 loff_t ret = -EINVAL;
2191 offset += file->f_pos;
2197 offset += LL_DIR_END_OFF_32BIT;
2199 offset += LL_DIR_END_OFF;
2206 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2207 (!api32 && offset <= LL_DIR_END_OFF))) {
2208 if (offset != file->f_pos) {
2209 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2210 (!api32 && offset == LL_DIR_END_OFF))
2211 fd->lfd_pos = MDS_DIR_END_OFF;
2212 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2213 fd->lfd_pos = offset << 32;
2215 fd->lfd_pos = offset;
2216 file->f_pos = offset;
2217 file->f_version = 0;
2224 inode_unlock(inode);
2228 static int ll_dir_open(struct inode *inode, struct file *file)
2231 RETURN(ll_file_open(inode, file));
2234 static int ll_dir_release(struct inode *inode, struct file *file)
2237 RETURN(ll_file_release(inode, file));
2240 const struct file_operations ll_dir_operations = {
2241 .llseek = ll_dir_seek,
2242 .open = ll_dir_open,
2243 .release = ll_dir_release,
2244 .read = generic_read_dir,
2245 #ifdef HAVE_DIR_CONTEXT
2246 .iterate_shared = ll_iterate,
2248 .readdir = ll_readdir,
2250 .unlocked_ioctl = ll_dir_ioctl,