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) 2003, 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/
32 #ifndef LLITE_INTERNAL_H
33 #define LLITE_INTERNAL_H
35 #include <lustre_disk.h> /* for s2sbi */
36 #include <lustre_linkea.h>
38 /* for struct cl_lock_descr and struct cl_io */
39 #include <cl_object.h>
40 #include <lustre_lmv.h>
41 #include <lustre_mdc.h>
42 #include <lustre_intent.h>
43 #include <linux/compat.h>
44 #include <linux/aio.h>
45 #include <linux/parser.h>
46 #include <linux/seqlock.h>
47 #include <lustre_compat.h>
48 #include <lustre_crypto.h>
49 #include <range_lock.h>
51 #include "vvp_internal.h"
53 #include "foreign_symlink.h"
59 #ifndef HAVE_VM_FAULT_RETRY
60 #define VM_FAULT_RETRY 0
63 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
64 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
65 #ifndef LOOKUP_CONTINUE
66 #define LOOKUP_CONTINUE LOOKUP_PARENT
69 /** Only used on client-side for indicating the tail of dir hash/offset. */
70 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
71 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
73 /* 4UL * 1024 * 1024 */
74 #define LL_MAX_BLKSIZE_BITS 22
76 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
78 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
80 struct ll_dentry_data {
81 unsigned int lld_sa_generation;
82 unsigned int lld_invalid:1;
83 unsigned int lld_nfs_dentry:1;
84 struct rcu_head lld_rcu_head;
87 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
89 #define LLI_INODE_MAGIC 0x111d0de5
90 #define LLI_INODE_DEAD 0xdeadd00d
92 struct ll_getname_data {
93 #ifdef HAVE_DIR_CONTEXT
94 struct dir_context ctx;
96 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
97 struct lu_fid lgd_fid; /* target fid we are looking for */
98 int lgd_found; /* inode matched? */
101 struct ll_grouplock {
102 struct lu_env *lg_env;
104 struct cl_lock *lg_lock;
105 unsigned long lg_gid;
108 /* See comment on trunc_sem_down_read_nowait */
109 struct ll_trunc_sem {
110 /* when positive, this is a count of readers, when -1, it indicates
111 * the semaphore is held for write, and 0 is unlocked
113 atomic_t ll_trunc_readers;
114 /* this tracks a count of waiting writers */
115 atomic_t ll_trunc_waiters;
118 struct ll_inode_info {
119 __u32 lli_inode_magic;
122 volatile unsigned long lli_flags;
123 struct posix_acl *lli_posix_acl;
125 /* identifying fields for both metadata and data stacks. */
126 struct lu_fid lli_fid;
127 /* master inode fid for stripe directory */
128 struct lu_fid lli_pfid;
130 /* We need all three because every inode may be opened in different
132 struct obd_client_handle *lli_mds_read_och;
133 struct obd_client_handle *lli_mds_write_och;
134 struct obd_client_handle *lli_mds_exec_och;
135 __u64 lli_open_fd_read_count;
136 __u64 lli_open_fd_write_count;
137 __u64 lli_open_fd_exec_count;
139 /* Number of times this inode was opened */
140 u64 lli_open_fd_count;
141 /* When last close was performed on this inode */
142 ktime_t lli_close_fd_time;
144 /* Protects access to och pointers and their usage counters */
145 struct mutex lli_och_mutex;
147 struct inode lli_vfs_inode;
149 /* the most recent timestamps obtained from mds */
154 spinlock_t lli_agl_lock;
156 /* Try to make the d::member and f::member are aligned. Before using
157 * these members, make clear whether it is directory or not. */
161 /* metadata statahead */
162 /* since parent-child threads can share the same @file
163 * struct, "opendir_key" is the token when dir close for
164 * case of parent exit before child -- it is me should
165 * cleanup the dir readahead. */
166 void *lli_opendir_key;
167 struct ll_statahead_info *lli_sai;
168 /* protect statahead stuff. */
169 spinlock_t lli_sa_lock;
170 /* "opendir_pid" is the token when lookup/revalid
171 * -- I am the owner of dir statahead. */
172 pid_t lli_opendir_pid;
173 /* directory depth to ROOT */
174 unsigned short lli_dir_depth;
175 /* stat will try to access statahead entries or start
176 * statahead if this flag is set, and this flag will be
177 * set upon dir open, and cleared when dir is closed,
178 * statahead hit ratio is too low, or start statahead
180 unsigned short lli_sa_enabled:1;
181 /* generation for statahead */
182 unsigned int lli_sa_generation;
183 /* rw lock protects lli_lsm_md */
184 struct rw_semaphore lli_lsm_sem;
185 /* directory stripe information */
186 struct lmv_stripe_md *lli_lsm_md;
187 /* directory default LMV */
188 struct lmv_stripe_md *lli_default_lsm_md;
191 /* for non-directory */
193 struct mutex lli_size_mutex;
194 char *lli_symlink_name;
195 struct ll_trunc_sem lli_trunc_sem;
196 struct range_lock_tree lli_write_tree;
197 struct mutex lli_setattr_mutex;
199 struct rw_semaphore lli_glimpse_sem;
200 ktime_t lli_glimpse_time;
201 struct list_head lli_agl_list;
204 /* for writepage() only to communicate to fsync */
207 /* protect the file heat fields */
208 spinlock_t lli_heat_lock;
209 __u32 lli_heat_flags;
210 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
213 * Whenever a process try to read/write the file, the
214 * jobid of the process will be saved here, and it'll
215 * be packed into the write PRC when flush later.
217 * So the read/write statistics for jobid will not be
218 * accurate if the file is shared by different jobs.
220 char lli_jobid[LUSTRE_JOBID_SIZE];
222 struct mutex lli_pcc_lock;
223 enum lu_pcc_state_flags lli_pcc_state;
225 * @lli_pcc_generation saves the gobal PCC generation
226 * when the file was successfully attached into PCC.
227 * The flags of the PCC dataset are saved in
229 * The gobal PCC generation will be increased when add
230 * or delete a PCC backend, or change the configuration
231 * parameters for PCC.
232 * If @lli_pcc_generation is same as the gobal PCC
233 * generation, we can use the saved flags of the PCC
234 * dataset to determine whether need to try auto attach
237 __u64 lli_pcc_generation;
238 enum pcc_dataset_flags lli_pcc_dsflags;
239 struct pcc_inode *lli_pcc_inode;
241 struct mutex lli_group_mutex;
242 __u64 lli_group_users;
243 unsigned long lli_group_gid;
245 __u64 lli_attr_valid;
247 __u64 lli_lazyblocks;
251 /* XXX: For following frequent used members, although they maybe special
252 * used for non-directory object, it is some time-wasting to check
253 * whether the object is directory or not before using them. On the
254 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
255 * the "ll_inode_info" size even if moving those members into u.f.
256 * So keep them out side.
258 * In the future, if more members are added only for directory,
259 * some of the following members can be moved into u.f.
261 struct cl_object *lli_clob;
263 /* mutex to request for layout lock exclusively. */
264 struct mutex lli_layout_mutex;
265 /* Layout version, protected by lli_layout_lock */
266 __u32 lli_layout_gen;
267 spinlock_t lli_layout_lock;
269 __u32 lli_projid; /* project id */
271 struct rw_semaphore lli_xattrs_list_rwsem;
272 struct mutex lli_xattrs_enq_lock;
273 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
274 struct list_head lli_lccs; /* list of ll_cl_context */
275 seqlock_t lli_page_inv_lock;
278 #ifndef HAVE_USER_NAMESPACE_ARG
279 #define inode_permission(ns, inode, mask) inode_permission(inode, mask)
280 #define generic_permission(ns, inode, mask) generic_permission(inode, mask)
281 #define simple_setattr(ns, de, iattr) simple_setattr(de, iattr)
282 #define ll_inode_permission(ns, inode, mask) ll_inode_permission(inode, mask)
283 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
284 #define ll_getattr(ns, path, stat, mask, fl) ll_getattr(path, stat, mask, fl)
285 #endif /* HAVE_INODEOPS_ENHANCED_GETATTR */
286 #define ll_setattr(ns, de, attr) ll_setattr(de, attr)
289 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
291 atomic_set(&sem->ll_trunc_readers, 0);
292 atomic_set(&sem->ll_trunc_waiters, 0);
295 /* This version of down read ignores waiting writers, meaning if the semaphore
296 * is already held for read, this down_read will 'join' that reader and also
297 * take the semaphore.
299 * This lets us avoid an unusual deadlock.
301 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
302 * in Lustre, in order to exclude truncates. Some of these paths then need to
303 * take the mmap_lock, while still holding the trunc_sem. The problem is that
304 * page faults hold the mmap_lock when calling in to Lustre, and then must also
305 * take the trunc_sem to exclude truncate.
307 * This means the locking order for trunc_sem and mmap_lock is sometimes AB,
308 * sometimes BA. This is almost OK because in both cases, we take the trunc
309 * sem for read, so it doesn't block.
311 * However, if a write mode user (truncate, a setattr op) arrives in the
312 * middle of this, the second reader on the truncate_sem will wait behind that
315 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
316 * to the mode in which they take the semaphore):
317 * reader (holding mmap_lock, needs truncate_sem)
319 * reader (holding truncate sem, waiting for mmap_lock)
321 * And so the readers deadlock.
323 * The solution is this modified semaphore, where this down_read ignores
324 * waiting write operations, and all waiters are woken up at once, so readers
325 * using down_read_nowait cannot get stuck behind waiting writers, regardless
326 * of the order they arrived in.
328 * down_read_nowait is only used in the page fault case, where we already hold
329 * the mmap_lock. This is because otherwise repeated read and write operations
330 * (which take the truncate sem) could prevent a truncate from ever starting.
331 * This could still happen with page faults, but without an even more complex
332 * mechanism, this is unavoidable.
336 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
338 wait_var_event(&sem->ll_trunc_readers,
339 atomic_inc_unless_negative(&sem->ll_trunc_readers));
342 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
344 wait_var_event(&sem->ll_trunc_readers,
345 atomic_read(&sem->ll_trunc_waiters) == 0 &&
346 atomic_inc_unless_negative(&sem->ll_trunc_readers));
349 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
351 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
352 atomic_read(&sem->ll_trunc_waiters))
353 wake_up_var(&sem->ll_trunc_readers);
356 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
358 atomic_inc(&sem->ll_trunc_waiters);
359 wait_var_event(&sem->ll_trunc_readers,
360 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
361 atomic_dec(&sem->ll_trunc_waiters);
364 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
366 atomic_set(&sem->ll_trunc_readers, 0);
367 /* match the smp_mb() in wait_var_event()->prepare_to_wait() */
369 wake_up_var(&sem->ll_trunc_readers);
372 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
373 static inline void lli_clear_acl(struct ll_inode_info *lli)
375 if (lli->lli_posix_acl) {
376 posix_acl_release(lli->lli_posix_acl);
377 lli->lli_posix_acl = NULL;
381 static inline void lli_replace_acl(struct ll_inode_info *lli,
382 struct lustre_md *md)
384 write_lock(&lli->lli_lock);
385 if (lli->lli_posix_acl)
386 posix_acl_release(lli->lli_posix_acl);
387 lli->lli_posix_acl = md->posix_acl;
388 write_unlock(&lli->lli_lock);
391 static inline void lli_clear_acl(struct ll_inode_info *lli)
395 static inline void lli_replace_acl(struct ll_inode_info *lli,
396 struct lustre_md *md)
401 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
405 spin_lock(&lli->lli_layout_lock);
406 gen = lli->lli_layout_gen;
407 spin_unlock(&lli->lli_layout_lock);
412 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
414 spin_lock(&lli->lli_layout_lock);
415 lli->lli_layout_gen = gen;
416 spin_unlock(&lli->lli_layout_lock);
420 /* File data is modified. */
421 LLIF_DATA_MODIFIED = 0,
422 /* File is being restored */
423 LLIF_FILE_RESTORING = 1,
424 /* Xattr cache is attached to the file */
425 LLIF_XATTR_CACHE = 2,
426 /* Project inherit */
427 LLIF_PROJECT_INHERIT = 3,
428 /* update atime from MDS even if it's older than local inode atime. */
429 LLIF_UPDATE_ATIME = 4,
430 /* foreign file/dir can be unlinked unconditionnaly */
431 LLIF_FOREIGN_REMOVABLE = 5,
432 /* Xattr cache is filled */
433 LLIF_XATTR_CACHE_FILLED = 7,
437 int ll_xattr_cache_destroy(struct inode *inode);
438 int ll_xattr_cache_empty(struct inode *inode);
440 int ll_xattr_cache_get(struct inode *inode,
446 int ll_xattr_cache_insert(struct inode *inode,
451 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
453 #ifdef CONFIG_SECURITY
454 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
455 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
461 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
463 #ifdef CONFIG_SECURITY
464 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
468 /* Only smack and selinux is known to use security contexts */
469 static inline bool ll_xattr_is_seclabel(const char *name)
471 return !strcmp(name, XATTR_NAME_SELINUX) ||
472 !strcmp(name, XATTR_NAME_SMACK);
475 static inline bool ll_xattr_suffix_is_seclabel(const char *suffix)
477 return !strcmp(suffix, XATTR_SELINUX_SUFFIX) ||
478 !strcmp(suffix, XATTR_SMACK_SUFFIX);
481 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
482 const char **secctx_name, __u32 *secctx_name_size,
483 void **secctx, __u32 *secctx_size);
484 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
487 int ll_inode_notifysecctx(struct inode *inode,
488 void *secctx, __u32 secctxlen);
490 void ll_secctx_name_free(struct ll_sb_info *sbi);
492 int ll_secctx_name_store(struct inode *in);
494 __u32 ll_secctx_name_get(struct ll_sb_info *sbi, const char **secctx_name);
496 int ll_security_secctx_name_filter(struct ll_sb_info *sbi, int xattr_type,
499 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
501 #ifdef HAVE_LUSTRE_CRYPTO
502 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
503 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
509 static inline void obd_connect_set_enc(struct obd_connect_data *data)
511 #ifdef HAVE_LUSTRE_CRYPTO
512 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
516 static inline bool obd_connect_has_name_enc(struct obd_connect_data *data)
518 #ifdef HAVE_LUSTRE_CRYPTO
519 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
520 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT_NAME;
526 static inline void obd_connect_set_name_enc(struct obd_connect_data *data)
528 #ifdef HAVE_LUSTRE_CRYPTO
529 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT_NAME;
534 * Locking to guarantee consistency of non-atomic updates to long long i_size,
535 * consistency between file size and KMS.
537 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
540 void ll_inode_size_lock(struct inode *inode);
541 void ll_inode_size_unlock(struct inode *inode);
543 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
545 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
548 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
550 return ll_i2info(inode)->lli_pcc_inode;
553 /* default to use at least 16M for fast read if possible */
554 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
556 /* default read-ahead on a given client mountpoint. */
557 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(1024UL)
559 /* default read-ahead for a single file descriptor */
560 #define SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX MiB_TO_PAGES(256UL)
562 /* default read-ahead full files smaller than limit on the second read */
563 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
565 /* default range pages */
566 #define SBI_DEFAULT_RA_RANGE_PAGES MiB_TO_PAGES(1ULL)
568 /* Min range pages */
569 #define RA_MIN_MMAP_RANGE_PAGES 16UL
574 RA_STAT_DISTANT_READPAGE,
575 RA_STAT_MISS_IN_WINDOW,
576 RA_STAT_FAILED_GRAB_PAGE,
577 RA_STAT_FAILED_MATCH,
582 RA_STAT_MAX_IN_FLIGHT,
583 RA_STAT_WRONG_GRAB_PAGE,
584 RA_STAT_FAILED_REACH_END,
586 RA_STAT_FAILED_FAST_READ,
587 RA_STAT_MMAP_RANGE_READ,
592 atomic_t ra_cur_pages;
593 unsigned long ra_max_pages;
594 unsigned long ra_max_pages_per_file;
595 unsigned long ra_range_pages;
596 unsigned long ra_max_read_ahead_whole_pages;
597 struct workqueue_struct *ll_readahead_wq;
599 * Max number of active works could be triggered
600 * for async readahead.
602 unsigned int ra_async_max_active;
603 /* how many async readahead triggered in flight */
604 atomic_t ra_async_inflight;
605 /* Threshold to control when to trigger async readahead */
606 unsigned long ra_async_pages_per_file_threshold;
609 /* ra_io_arg will be filled in the beginning of ll_readahead with
610 * ras_lock, then the following ll_read_ahead_pages will read RA
611 * pages according to this arg, all the items in this structure are
612 * counted by page index.
615 pgoff_t ria_start_idx; /* start offset of read-ahead*/
616 pgoff_t ria_end_idx; /* end offset of read-ahead*/
617 unsigned long ria_reserved; /* reserved pages for read-ahead */
618 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
619 bool ria_eof; /* reach end of file */
620 /* If stride read pattern is detected, ria_stoff is the byte offset
621 * where stride read is started. Note: for normal read-ahead, the
622 * value here is meaningless, and also it will not be accessed*/
624 /* ria_length and ria_bytes are the length and pages length in the
625 * stride I/O mode. And they will also be used to check whether
626 * it is stride I/O read-ahead in the read-ahead pages*/
631 /* LL_HIST_MAX=32 causes an overflow */
632 #define LL_HIST_MAX 28
633 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
634 #define LL_PROCESS_HIST_MAX 10
635 struct per_process_info {
637 struct obd_histogram pp_r_hist;
638 struct obd_histogram pp_w_hist;
641 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
642 struct ll_rw_extents_info {
644 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
647 #define LL_OFFSET_HIST_MAX 100
648 struct ll_rw_process_info {
651 loff_t rw_range_start;
653 loff_t rw_last_file_pos;
655 size_t rw_smallest_extent;
656 size_t rw_largest_extent;
657 struct ll_file_data *rw_last_file;
660 enum stats_track_type {
661 STATS_TRACK_ALL = 0, /* track all processes */
662 STATS_TRACK_PID, /* track process with this pid */
663 STATS_TRACK_PPID, /* track processes with this ppid */
664 STATS_TRACK_GID, /* track processes with this gid */
668 /* flags for sbi->ll_flags */
670 LL_SBI_NOLCK, /* DLM locking disabled directio-only */
671 LL_SBI_CHECKSUM, /* checksum each page as it's written */
672 LL_SBI_LOCALFLOCK, /* local flocks instead of fs-wide */
673 LL_SBI_FLOCK, /* flock enabled */
674 LL_SBI_USER_XATTR, /* support user xattr */
675 LL_SBI_LRU_RESIZE, /* lru resize support */
676 LL_SBI_LAZYSTATFS, /* lazystatfs mount option */
677 LL_SBI_32BIT_API, /* generate 32 bit inodes. */
678 LL_SBI_USER_FID2PATH, /* fid2path by unprivileged users */
679 LL_SBI_VERBOSE, /* verbose mount/umount */
680 LL_SBI_ALWAYS_PING, /* ping even if server suppress_pings */
681 LL_SBI_TEST_DUMMY_ENCRYPTION, /* test dummy encryption */
682 LL_SBI_ENCRYPT, /* client side encryption */
683 LL_SBI_FOREIGN_SYMLINK, /* foreign fake-symlink support */
684 LL_SBI_FOREIGN_SYMLINK_UPCALL, /* foreign fake-symlink upcall set */
685 LL_SBI_NUM_MOUNT_OPT,
687 LL_SBI_ACL, /* support ACL */
688 LL_SBI_AGL_ENABLED, /* enable agl */
689 LL_SBI_64BIT_HASH, /* support 64-bits dir hash/offset */
690 LL_SBI_LAYOUT_LOCK, /* layout lock support */
691 LL_SBI_XATTR_CACHE, /* support for xattr cache */
692 LL_SBI_NOROOTSQUASH, /* do not apply root squash */
693 LL_SBI_FAST_READ, /* fast read support */
694 LL_SBI_FILE_SECCTX, /* file security context at create */
695 LL_SBI_TINY_WRITE, /* tiny write support */
696 LL_SBI_FILE_HEAT, /* file heat support */
697 LL_SBI_PARALLEL_DIO, /* parallel (async) O_DIRECT RPCs */
698 LL_SBI_ENCRYPT_NAME, /* name encryption */
702 int ll_sbi_flags_seq_show(struct seq_file *m, void *v);
704 /* This is embedded into llite super-blocks to keep track of connect
705 * flags (capabilities) supported by all imports given mount is
707 struct lustre_client_ocd {
708 /* This is conjunction of connect_flags across all imports
709 * (LOVs) this mount is connected to. This field is updated by
710 * cl_ocd_update() under ->lco_lock. */
712 struct mutex lco_lock;
713 struct obd_export *lco_md_exp;
714 struct obd_export *lco_dt_exp;
718 /* this protects pglist and ra_info. It isn't safe to
719 * grab from interrupt contexts */
721 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
722 spinlock_t ll_process_lock; /* ll_rw_process_info */
723 struct obd_uuid ll_sb_uuid;
724 struct obd_export *ll_md_exp;
725 struct obd_export *ll_dt_exp;
726 struct obd_device *ll_md_obd;
727 struct obd_device *ll_dt_obd;
728 struct dentry *ll_debugfs_entry;
729 struct lu_fid ll_root_fid; /* root object fid */
730 struct mnt_namespace *ll_mnt_ns;
732 DECLARE_BITMAP(ll_flags, LL_SBI_NUM_FLAGS); /* enum ll_sbi_flags */
733 unsigned int ll_xattr_cache_enabled:1,
734 ll_xattr_cache_set:1, /* already set to 0/1 */
735 ll_client_common_fill_super_succeeded:1,
738 struct lustre_client_ocd ll_lco;
740 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
742 /* Used to track "unstable" pages on a client, and maintain a
743 * LRU list of clean pages. An "unstable" page is defined as
744 * any page which is sent to a server as part of a bulk request,
745 * but is uncommitted to stable storage. */
746 struct cl_client_cache *ll_cache;
748 struct lprocfs_stats *ll_ra_stats;
750 struct ll_ra_info ll_ra_info;
751 unsigned int ll_namelen;
752 const struct file_operations *ll_fop;
754 struct lu_site *ll_site;
755 struct cl_device *ll_cl;
758 struct ll_rw_extents_info *ll_rw_extents_info;
759 int ll_extent_process_count;
760 unsigned int ll_offset_process_count;
761 struct ll_rw_process_info *ll_rw_process_info;
762 struct ll_rw_process_info *ll_rw_offset_info;
763 ktime_t ll_process_stats_init;
764 unsigned int ll_rw_offset_entry_count;
765 int ll_stats_track_id;
766 enum stats_track_type ll_stats_track_type;
769 /* metadata stat-ahead */
770 unsigned int ll_sa_running_max;/* max concurrent
771 * statahead instances */
772 unsigned int ll_sa_max; /* max statahead RPCs */
773 atomic_t ll_sa_total; /* statahead thread started
775 atomic_t ll_sa_wrong; /* statahead thread stopped for
777 atomic_t ll_sa_running; /* running statahead thread
779 atomic_t ll_agl_total; /* AGL thread started count */
781 dev_t ll_sdev_orig; /* save s_dev before assign for
784 struct root_squash_info ll_squash;
787 /* st_blksize returned by stat(2), when non-zero */
788 unsigned int ll_stat_blksize;
790 /* maximum relative age of cached statfs results */
791 unsigned int ll_statfs_max_age;
793 struct kset ll_kset; /* sysfs object */
794 struct completion ll_kobj_unregister;
797 unsigned int ll_heat_decay_weight;
798 unsigned int ll_heat_period_second;
800 /* Opens of the same inode before we start requesting open lock */
801 u32 ll_oc_thrsh_count;
803 /* Time in ms between last inode close and next open to be considered
804 * instant back to back and would trigger an open lock request
808 /* Time in ms after last file close that we no longer count prior opens*/
811 /* filesystem fsname */
812 char ll_fsname[LUSTRE_MAXFSNAME + 1];
814 /* Persistent Client Cache */
815 struct pcc_super ll_pcc_super;
817 /* to protect vs updates in all following foreign symlink fields */
818 struct rw_semaphore ll_foreign_symlink_sem;
819 /* foreign symlink path prefix */
820 char *ll_foreign_symlink_prefix;
821 /* full prefix size including leading '\0' */
822 size_t ll_foreign_symlink_prefix_size;
823 /* foreign symlink path upcall */
824 char *ll_foreign_symlink_upcall;
825 /* foreign symlink path upcall infos */
826 struct ll_foreign_symlink_upcall_item *ll_foreign_symlink_upcall_items;
827 /* foreign symlink path upcall nb infos */
828 unsigned int ll_foreign_symlink_upcall_nb_items;
830 /* cached file security context xattr name. e.g: security.selinux */
831 char *ll_secctx_name;
832 __u32 ll_secctx_name_size;
835 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
836 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
838 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT (5)
839 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MS (100) /* 0.1 second */
840 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS (60000) /* 1 minute */
843 * per file-descriptor read-ahead data.
845 struct ll_readahead_state {
847 /* End byte that read(2) try to read. */
848 loff_t ras_last_read_end_bytes;
850 * number of bytes read after last read-ahead window reset. As window
851 * is reset on each seek, this is effectively a number of consecutive
852 * accesses. Maybe ->ras_accessed_in_window is better name.
854 * XXX nikita: window is also reset (by ras_update()) when Lustre
855 * believes that memory pressure evicts read-ahead pages. In that
856 * case, it probably doesn't make sense to expand window to
857 * PTLRPC_MAX_BRW_PAGES on the third access.
859 loff_t ras_consecutive_bytes;
861 * number of read requests after the last read-ahead window reset
862 * As window is reset on each seek, this is effectively the number
863 * on consecutive read request and is used to trigger read-ahead.
865 unsigned long ras_consecutive_requests;
867 * Parameters of current read-ahead window. Handled by
868 * ras_update(). On the initial access to the file or after a seek,
869 * window is reset to 0. After 3 consecutive accesses, window is
870 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
871 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
873 pgoff_t ras_window_start_idx;
874 pgoff_t ras_window_pages;
876 /* Page index where min range read starts */
877 pgoff_t ras_range_min_start_idx;
878 /* Page index where mmap range read ends */
879 pgoff_t ras_range_max_end_idx;
880 /* number of mmap pages where last time detected */
881 pgoff_t ras_last_range_pages;
882 /* number of mmap range requests */
883 pgoff_t ras_range_requests;
886 * Optimal RPC size in pages.
887 * It decides how many pages will be sent for each read-ahead.
889 unsigned long ras_rpc_pages;
891 * Where next read-ahead should start at. This lies within read-ahead
892 * window. Read-ahead window is read in pieces rather than at once
893 * because: 1. lustre limits total number of pages under read-ahead by
894 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
895 * not covered by DLM lock.
897 pgoff_t ras_next_readahead_idx;
899 * Total number of ll_file_read requests issued, reads originating
900 * due to mmap are not counted in this total. This value is used to
901 * trigger full file read-ahead after multiple reads to a small file.
903 unsigned long ras_requests;
905 * The following 3 items are used for detecting the stride I/O
907 * In stride I/O mode,
908 * ...............|-----data-----|****gap*****|--------|******|....
909 * offset |-stride_bytes-|-stride_gap-|
910 * ras_stride_offset = offset;
911 * ras_stride_length = stride_bytes + stride_gap;
912 * ras_stride_bytes = stride_bytes;
913 * Note: all these three items are counted by bytes.
915 loff_t ras_stride_offset;
916 loff_t ras_stride_length;
917 loff_t ras_stride_bytes;
919 * number of consecutive stride request count, and it is similar as
920 * ras_consecutive_requests, but used for stride I/O mode.
921 * Note: only more than 2 consecutive stride request are detected,
922 * stride read-ahead will be enable
924 unsigned long ras_consecutive_stride_requests;
925 /* index of the last page that async readahead starts */
926 pgoff_t ras_async_last_readpage_idx;
927 /* whether we should increase readahead window */
928 bool ras_need_increase_window;
929 /* whether ra miss check should be skipped */
930 bool ras_no_miss_check;
933 struct ll_readahead_work {
934 /** File to readahead */
935 struct file *lrw_file;
936 pgoff_t lrw_start_idx;
940 /* async worker to handler read */
941 struct work_struct lrw_readahead_work;
942 char lrw_jobid[LUSTRE_JOBID_SIZE];
945 extern struct kmem_cache *ll_file_data_slab;
946 struct lustre_handle;
947 struct ll_file_data {
948 struct ll_readahead_state fd_ras;
949 struct ll_grouplock fd_grouplock;
953 /* openhandle if lease exists for this file.
954 * Borrow lli->lli_och_mutex to protect assignment */
955 struct obd_client_handle *fd_lease_och;
956 struct obd_client_handle *fd_och;
957 struct file *fd_file;
958 /* Indicate whether need to report failure when close.
959 * true: failure is known, not report again.
960 * false: unknown failure, should report. */
961 bool fd_write_failed;
962 bool ll_lock_no_expand;
963 /* Used by mirrored file to lead IOs to a specific mirror, usually
964 * for mirror resync. 0 means default. */
965 __u32 fd_designated_mirror;
966 /* The layout version when resync starts. Resync I/O should carry this
967 * layout version for verification to OST objects */
968 __u32 fd_layout_version;
969 struct pcc_file fd_pcc_file;
970 /* striped directory may read partially if some stripe inaccessible,
971 * -errno is saved here, and will return to user in close().
973 int fd_partial_readdir_rc;
976 void llite_tunables_unregister(void);
977 int llite_tunables_register(void);
979 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
981 return &lli->lli_vfs_inode;
984 __u32 ll_i2suppgid(struct inode *i);
985 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
987 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
989 #if BITS_PER_LONG == 32
991 #elif defined(CONFIG_COMPAT)
992 if (unlikely(test_bit(LL_SBI_32BIT_API, sbi->ll_flags)))
995 # ifdef CONFIG_X86_X32
996 /* in_compat_syscall() returns true when called from a kthread
997 * and CONFIG_X86_X32 is enabled, which is wrong. So check
998 * whether the caller comes from a syscall (ie. not a kthread)
999 * before calling in_compat_syscall(). */
1000 if (current->flags & PF_KTHREAD)
1004 return unlikely(in_compat_syscall());
1006 return unlikely(test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1010 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
1012 return test_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1015 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
1017 return test_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1020 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
1022 return test_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1025 static inline bool ll_sbi_has_foreign_symlink(struct ll_sb_info *sbi)
1027 return test_bit(LL_SBI_FOREIGN_SYMLINK, sbi->ll_flags);
1030 static inline bool ll_sbi_has_parallel_dio(struct ll_sb_info *sbi)
1032 return test_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1035 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
1037 /* llite/lcommon_misc.c */
1038 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
1039 enum obd_notify_event ev, void *owner);
1040 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
1041 struct ll_grouplock *lg);
1042 void cl_put_grouplock(struct ll_grouplock *lg);
1044 /* llite/lproc_llite.c */
1045 int ll_debugfs_register_super(struct super_block *sb, const char *name);
1046 void ll_debugfs_unregister_super(struct super_block *sb);
1047 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
1048 void ll_free_rw_stats_info(struct ll_sb_info *sbi);
1051 LPROC_LL_READ_BYTES,
1052 LPROC_LL_WRITE_BYTES,
1079 LPROC_LL_GETXATTR_HITS,
1081 LPROC_LL_REMOVEXATTR,
1082 LPROC_LL_INODE_PERM,
1084 LPROC_LL_INODE_OCOUNT,
1085 LPROC_LL_INODE_OPCLTM,
1086 LPROC_LL_FILE_OPCODES
1090 enum get_default_layout_type {
1091 GET_DEFAULT_LAYOUT_ROOT = 1,
1094 extern const struct file_operations ll_dir_operations;
1095 extern const struct inode_operations ll_dir_inode_operations;
1096 #ifdef HAVE_DIR_CONTEXT
1097 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1098 struct dir_context *ctx, int *partial_readdir_rc);
1100 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1101 void *cookie, filldir_t filldir, int *partial_readdir_rc);
1103 int ll_get_mdt_idx(struct inode *inode);
1104 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
1105 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
1106 __u64 offset, int *partial_readdir_rc);
1107 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1108 int quotactl_ioctl(struct super_block *sb, struct if_quotactl *qctl);
1111 extern const struct inode_operations ll_special_inode_operations;
1113 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1114 struct lustre_md *lic);
1115 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1116 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1117 void *data, int flag);
1118 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1119 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1120 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1123 int ll_writepage(struct page *page, struct writeback_control *wbc);
1124 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1125 int ll_readpage(struct file *file, struct page *page);
1126 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1127 struct cl_page *page, struct file *file);
1128 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1129 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1132 void ll_cl_add(struct inode *inode, const struct lu_env *env, struct cl_io *io,
1133 enum lcc_type type);
1134 void ll_cl_remove(struct inode *inode, const struct lu_env *env);
1135 struct ll_cl_context *ll_cl_find(struct inode *inode);
1137 extern const struct address_space_operations ll_aops;
1140 extern const struct inode_operations ll_file_inode_operations;
1141 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi);
1142 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1143 enum ldlm_mode l_req_mode);
1144 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1145 struct lustre_handle *lockh, __u64 flags,
1146 enum ldlm_mode mode);
1148 int ll_file_open(struct inode *inode, struct file *file);
1149 int ll_file_release(struct inode *inode, struct file *file);
1150 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1151 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1152 void ll_track_file_opens(struct inode *inode);
1153 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1154 struct ll_file_data *file, loff_t pos,
1155 size_t count, int rw);
1156 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
1157 int ll_getattr(struct user_namespace *mnt_userns, const struct path *path,
1158 struct kstat *stat, u32 request_mask, unsigned int flags);
1160 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1161 #endif /* HAVE_USER_NAMESPACE_ARG */
1162 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1163 unsigned int flags, bool foreign);
1164 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1165 struct posix_acl *ll_get_acl(struct inode *inode, int type
1166 #ifdef HAVE_GET_ACL_RCU_ARG
1168 #endif /* HAVE_GET_ACL_RCU_ARG */
1170 int ll_set_acl(struct user_namespace *mnt_userns, struct inode *inode,
1171 struct posix_acl *acl, int type);
1172 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1173 #define ll_get_acl NULL
1174 #define ll_set_acl NULL
1175 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1177 static inline int ll_xflags_to_inode_flags(int xflags)
1179 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1180 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1181 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1182 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1185 static inline int ll_inode_flags_to_xflags(int inode_flags)
1187 return ((inode_flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1188 ((inode_flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1189 ((inode_flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1190 ((inode_flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1193 int ll_migrate(struct inode *parent, struct file *file,
1194 struct lmv_user_md *lum, const char *name, __u32 flags);
1195 int ll_get_fid_by_name(struct inode *parent, const char *name,
1196 int namelen, struct lu_fid *fid, struct inode **inode);
1197 int ll_inode_permission(struct user_namespace *mnt_userns, struct inode *inode,
1199 int ll_ioctl_check_project(struct inode *inode, __u32 xflags, __u32 projid);
1200 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1202 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1204 int ll_ioctl_project(struct file *file, unsigned int cmd,
1207 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1208 __u64 flags, struct lov_user_md *lum,
1210 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1211 struct lov_mds_md **lmm, int *lmm_size,
1212 struct ptlrpc_request **request);
1213 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1215 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1216 int *lmm_size, struct ptlrpc_request **request,
1217 struct ptlrpc_request **root_request, u64 valid);
1218 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1219 struct ptlrpc_request **request, u64 valid);
1220 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1221 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1222 int ll_fid2path(struct inode *inode, void __user *arg);
1223 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1224 int ll_hsm_release(struct inode *inode);
1225 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1226 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1228 /* llite/dcache.c */
1230 extern const struct dentry_operations ll_d_ops;
1232 bool ll_d_setup(struct dentry *de, bool do_put);
1234 static inline bool lld_is_init(struct dentry *dentry)
1236 return ll_d2d(dentry);
1239 #define ll_d_setup(de, do_put) (true)
1240 #define lld_is_init(dentry) (true)
1243 void ll_intent_drop_lock(struct lookup_intent *);
1244 void ll_intent_release(struct lookup_intent *);
1245 void ll_prune_aliases(struct inode *inode);
1246 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1247 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1248 struct lookup_intent *it, struct dentry *de);
1250 /* llite/llite_lib.c */
1251 extern const struct super_operations lustre_super_operations;
1253 void ll_lli_init(struct ll_inode_info *lli);
1254 int ll_fill_super(struct super_block *sb);
1255 void ll_put_super(struct super_block *sb);
1256 void ll_kill_super(struct super_block *sb);
1257 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1258 void ll_dir_clear_lsm_md(struct inode *inode);
1259 void ll_clear_inode(struct inode *inode);
1260 int volatile_ref_file(const char *volatile_name, int volatile_len,
1261 struct file **ref_file);
1262 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1263 enum op_xvalid xvalid, bool hsm_import);
1264 int ll_setattr(struct user_namespace *mnt_userns, struct dentry *de,
1265 struct iattr *attr);
1266 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1267 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1269 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1270 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags);
1271 void ll_update_dir_depth(struct inode *dir, struct inode *inode);
1272 int ll_read_inode2(struct inode *inode, void *opaque);
1273 void ll_truncate_inode_pages_final(struct inode *inode);
1274 void ll_delete_inode(struct inode *inode);
1275 int ll_iocontrol(struct inode *inode, struct file *file,
1276 unsigned int cmd, unsigned long arg);
1277 int ll_flush_ctx(struct inode *inode);
1278 void ll_umount_begin(struct super_block *sb);
1279 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1280 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1281 void ll_dirty_page_discard_warn(struct inode *inode, int ioret);
1282 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
1283 struct super_block *sb, struct lookup_intent *it);
1284 int ll_obd_statfs(struct inode *inode, void __user *arg);
1285 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1286 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1287 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1289 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1290 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1291 struct inode *i1, struct inode *i2,
1292 const char *name, size_t namelen,
1293 __u32 mode, enum md_op_code opc,
1295 void ll_finish_md_op_data(struct md_op_data *op_data);
1296 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1297 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1298 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1299 struct lov_user_md **kbuf);
1300 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill);
1302 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req);
1304 /* Compute expected user md size when passing in a md from user space */
1305 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1307 switch (lum->lmm_magic) {
1308 case LOV_USER_MAGIC_V1:
1309 return sizeof(struct lov_user_md_v1);
1310 case LOV_USER_MAGIC_V3:
1311 return sizeof(struct lov_user_md_v3);
1312 case LOV_USER_MAGIC_SPECIFIC:
1313 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1316 return lov_user_md_size(lum->lmm_stripe_count,
1317 LOV_USER_MAGIC_SPECIFIC);
1318 case LOV_USER_MAGIC_COMP_V1:
1319 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1320 case LOV_USER_MAGIC_FOREIGN:
1321 return foreign_size(lum);
1327 /* llite/llite_nfs.c */
1328 extern const struct export_operations lustre_export_operations;
1329 __u32 get_uuid2int(const char *name, int len);
1330 struct inode *search_inode_for_lustre(struct super_block *sb,
1331 const struct lu_fid *fid);
1332 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1334 /* llite/symlink.c */
1335 extern const struct inode_operations ll_fast_symlink_inode_operations;
1338 * IO arguments for various VFS I/O interfaces.
1340 struct vvp_io_args {
1341 /** normal/sendfile/splice */
1344 struct kiocb *via_iocb;
1345 struct iov_iter *via_iter;
1355 struct ll_cl_context {
1356 struct list_head lcc_list;
1358 const struct lu_env *lcc_env;
1359 struct cl_io *lcc_io;
1360 struct cl_page *lcc_page;
1361 enum lcc_type lcc_type;
1362 pgoff_t lcc_end_index;
1365 struct ll_thread_info {
1366 struct vvp_io_args lti_args;
1367 struct ra_io_arg lti_ria;
1368 struct ll_cl_context lti_io_ctx;
1371 extern struct lu_context_key ll_thread_key;
1373 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1375 struct ll_thread_info *lti;
1377 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1378 LASSERT(lti != NULL);
1383 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env)
1385 return &ll_env_info(env)->lti_args;
1388 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1389 struct vvp_io_args *args);
1391 /* llite/llite_mmap.c */
1393 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1394 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1395 unsigned long addr, size_t count);
1396 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1399 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1401 /* don't need an addref as the sb_info should be holding one */
1402 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1404 return ll_s2sbi(sb)->ll_dt_exp;
1407 /* don't need an addref as the sb_info should be holding one */
1408 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1410 return ll_s2sbi(sb)->ll_md_exp;
1413 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1415 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1421 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1422 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1424 return ll_s2sbi(inode->i_sb);
1427 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1429 return ll_s2dtexp(inode->i_sb);
1432 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1434 return ll_s2mdexp(inode->i_sb);
1437 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1441 LASSERT(inode != NULL);
1442 fid = &ll_i2info(inode)->lli_fid;
1447 static inline bool ll_dir_striped(struct inode *inode)
1450 return S_ISDIR(inode->i_mode) &&
1451 lmv_dir_striped(ll_i2info(inode)->lli_lsm_md);
1454 static inline loff_t ll_file_maxbytes(struct inode *inode)
1456 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1459 return MAX_LFS_FILESIZE;
1461 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1465 extern const struct xattr_handler *ll_xattr_handlers[];
1467 #define XATTR_USER_T 1
1468 #define XATTR_TRUSTED_T 2
1469 #define XATTR_SECURITY_T 3
1470 #define XATTR_ACL_ACCESS_T 4
1471 #define XATTR_ACL_DEFAULT_T 5
1472 #define XATTR_LUSTRE_T 6
1473 #define XATTR_OTHER_T 7
1474 #define XATTR_ENCRYPTION_T 9
1476 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1477 int ll_xattr_list(struct inode *inode, const char *name, int type,
1478 void *buffer, size_t size, u64 valid);
1479 const struct xattr_handler *get_xattr_type(const char *name);
1482 * Common IO arguments for various VFS I/O interfaces.
1484 int cl_sb_init(struct super_block *sb);
1485 int cl_sb_fini(struct super_block *sb);
1487 enum ras_update_flags {
1491 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1492 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1496 #define LL_SA_RPC_MIN 2
1497 #define LL_SA_RPC_DEF 32
1498 #define LL_SA_RPC_MAX 512
1500 /* XXX: If want to support more concurrent statahead instances,
1501 * please consider to decentralize the RPC lists attached
1502 * on related import, such as imp_{sending,delayed}_list.
1504 #define LL_SA_RUNNING_MAX 256
1505 #define LL_SA_RUNNING_DEF 16
1507 #define LL_SA_CACHE_BIT 5
1508 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1509 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1511 /* per inode struct, for dir only */
1512 struct ll_statahead_info {
1513 struct dentry *sai_dentry;
1514 atomic_t sai_refcount; /* when access this struct, hold
1516 unsigned int sai_max; /* max ahead of lookup */
1517 __u64 sai_sent; /* stat requests sent count */
1518 __u64 sai_replied; /* stat requests which received
1520 __u64 sai_index; /* index of statahead entry */
1521 __u64 sai_index_wait; /* index of entry which is the
1522 * caller is waiting for */
1523 __u64 sai_hit; /* hit count */
1524 __u64 sai_miss; /* miss count:
1525 * for "ls -al" case, includes
1526 * hidden dentry miss;
1527 * for "ls -l" case, it does not
1528 * include hidden dentry miss.
1529 * "sai_miss_hidden" is used for
1532 unsigned int sai_consecutive_miss; /* consecutive miss */
1533 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1534 * is not a hidden one */
1535 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1537 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1539 sai_in_readpage:1;/* statahead is in readdir()*/
1540 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1541 struct task_struct *sai_task; /* stat-ahead thread */
1542 struct task_struct *sai_agl_task; /* AGL thread */
1543 struct list_head sai_interim_entries; /* entries which got async
1544 * stat reply, but not
1546 struct list_head sai_entries; /* completed entries */
1547 struct list_head sai_agls; /* AGLs to be sent */
1548 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1549 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1550 atomic_t sai_cache_count; /* entry count in cache */
1553 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1555 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1556 void ll_authorize_statahead(struct inode *dir, void *key);
1557 void ll_deauthorize_statahead(struct inode *dir, void *key);
1560 blkcnt_t dirty_cnt(struct inode *inode);
1562 int cl_glimpse_size0(struct inode *inode, int agl);
1563 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1564 struct inode *inode, struct cl_object *clob, int agl);
1566 static inline int cl_glimpse_size(struct inode *inode)
1568 return cl_glimpse_size0(inode, 0);
1571 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1572 * part of statahead */
1573 static inline int cl_agl(struct inode *inode)
1575 return cl_glimpse_size0(inode, 1);
1578 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1580 int cl_io_get(struct inode *inode, struct lu_env **envout,
1581 struct cl_io **ioout, __u16 *refcheck);
1583 static inline int ll_glimpse_size(struct inode *inode)
1585 struct ll_inode_info *lli = ll_i2info(inode);
1588 down_read(&lli->lli_glimpse_sem);
1589 rc = cl_glimpse_size(inode);
1590 lli->lli_glimpse_time = ktime_get();
1591 up_read(&lli->lli_glimpse_sem);
1595 /* dentry may statahead when statahead is enabled and current process has opened
1596 * parent directory, and this dentry hasn't accessed statahead cache before */
1598 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1600 struct ll_inode_info *lli;
1601 struct ll_dentry_data *ldd;
1603 if (ll_i2sbi(dir)->ll_sa_max == 0)
1606 lli = ll_i2info(dir);
1608 /* statahead is not allowed for this dir, there may be three causes:
1609 * 1. dir is not opened.
1610 * 2. statahead hit ratio is too low.
1611 * 3. previous stat started statahead thread failed. */
1612 if (!lli->lli_sa_enabled)
1615 /* not the same process, don't statahead */
1616 if (lli->lli_opendir_pid != current->pid)
1620 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1621 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1622 * For patchless client, lookup intent is not accurate, which may
1623 * misguide statahead. For example:
1624 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1625 * have the same intent -- IT_GETATTR, while one dentry should access
1626 * statahead cache once, otherwise statahead windows is messed up.
1627 * The solution is as following:
1628 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1629 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1630 * bypass interacting with statahead cache by checking
1631 * 'lld_sa_generation == lli->lli_sa_generation'.
1633 ldd = ll_d2d(dentry);
1634 if (ldd != NULL && lli->lli_sa_generation &&
1635 ldd->lld_sa_generation == lli->lli_sa_generation)
1641 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1642 enum cl_fsync_mode mode, int ignore_layout);
1644 static inline int ll_file_nolock(const struct file *file)
1646 struct ll_file_data *fd = file->private_data;
1647 struct inode *inode = file_inode((struct file *)file);
1649 LASSERT(fd != NULL);
1650 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1651 test_bit(LL_SBI_NOLCK, ll_i2sbi(inode)->ll_flags));
1654 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1655 struct lookup_intent *it, __u64 *bits)
1657 if (!it->it_lock_set) {
1658 struct lustre_handle handle;
1660 /* If this inode is a remote object, it will get two
1661 * separate locks in different namespaces, Master MDT,
1662 * where the name entry is, will grant LOOKUP lock,
1663 * remote MDT, where the object is, will grant
1664 * UPDATE|PERM lock. The inode will be attched to both
1665 * LOOKUP and PERM locks, so revoking either locks will
1666 * case the dcache being cleared */
1667 if (it->it_remote_lock_mode) {
1668 handle.cookie = it->it_remote_lock_handle;
1669 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1670 "(%p) for remote lock %#llx\n",
1671 PFID(ll_inode2fid(inode)), inode,
1673 md_set_lock_data(exp, &handle, inode, NULL);
1676 handle.cookie = it->it_lock_handle;
1678 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1679 " for lock %#llx\n",
1680 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1682 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1683 it->it_lock_set = 1;
1687 *bits = it->it_lock_bits;
1690 static inline int d_lustre_invalid(const struct dentry *dentry)
1692 return !ll_d2d(dentry) || ll_d2d(dentry)->lld_invalid;
1696 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1697 * ll_md_blocking_ast), it will be pruned by ll_prune_aliases() and
1698 * ll_prune_negative_children(); otherwise dput() of the last refcount will
1699 * unhash this dentry and kill it.
1701 static inline void d_lustre_invalidate(struct dentry *dentry)
1703 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1705 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1707 spin_lock(&dentry->d_lock);
1708 if (lld_is_init(dentry))
1709 ll_d2d(dentry)->lld_invalid = 1;
1710 spin_unlock(&dentry->d_lock);
1713 static inline void d_lustre_revalidate(struct dentry *dentry)
1715 spin_lock(&dentry->d_lock);
1716 LASSERT(ll_d2d(dentry));
1717 ll_d2d(dentry)->lld_invalid = 0;
1718 spin_unlock(&dentry->d_lock);
1721 static inline dev_t ll_compat_encode_dev(dev_t dev)
1723 /* The compat_sys_*stat*() syscalls will fail unless the
1724 * device majors and minors are both less than 256. Note that
1725 * the value returned here will be passed through
1726 * old_encode_dev() in cp_compat_stat(). And so we are not
1727 * trying to return a valid compat (u16) device number, just
1728 * one that will pass the old_valid_dev() check. */
1730 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1733 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1734 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1735 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1736 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1737 struct lu_extent *ext);
1739 int ll_xattr_init(void);
1740 void ll_xattr_fini(void);
1742 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1743 struct cl_page *page, enum cl_req_type crt);
1745 int ll_getparent(struct file *file, struct getparent __user *arg);
1748 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1749 enum op_xvalid xvalid, unsigned int attr_flags);
1751 extern struct lu_env *cl_inode_fini_env;
1752 extern __u16 cl_inode_fini_refcheck;
1754 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1755 void cl_inode_fini(struct inode *inode);
1757 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1758 u32 cl_fid_build_gen(const struct lu_fid *fid);
1760 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1762 return &ll_i2sbi(inode)->ll_pcc_super;
1765 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1767 return ll_i2pccs(ll_info2i(lli));
1771 /* The digested form is made of a FID (16 bytes) followed by the second-to-last
1772 * ciphertext block (16 bytes), so a total length of 32 bytes.
1773 * That way, llcrypt does not compute a digested form of this digest.
1775 struct ll_digest_filename {
1776 struct lu_fid ldf_fid;
1777 char ldf_excerpt[LL_CRYPTO_BLOCK_SIZE];
1780 int ll_setup_filename(struct inode *dir, const struct qstr *iname,
1781 int lookup, struct llcrypt_name *fname,
1782 struct lu_fid *fid);
1783 int ll_fname_disk_to_usr(struct inode *inode,
1784 u32 hash, u32 minor_hash,
1785 struct llcrypt_str *iname, struct llcrypt_str *oname,
1786 struct lu_fid *fid);
1787 int ll_revalidate_d_crypto(struct dentry *dentry, unsigned int flags);
1788 int ll_file_open_encrypt(struct inode *inode, struct file *filp);
1789 static inline char *xattr_for_enc(struct inode *inode)
1791 if (ll_sbi_has_name_encrypt(ll_i2sbi(inode)))
1792 return LL_XATTR_NAME_ENCRYPTION_CONTEXT;
1794 return LL_XATTR_NAME_ENCRYPTION_CONTEXT_OLD;
1796 #ifdef HAVE_LUSTRE_CRYPTO
1797 extern const struct llcrypt_operations lustre_cryptops;
1800 /* llite/llite_foreign.c */
1801 int ll_manage_foreign(struct inode *inode, struct lustre_md *lmd);
1802 bool ll_foreign_is_openable(struct dentry *dentry, unsigned int flags);
1803 bool ll_foreign_is_removable(struct dentry *dentry, bool unset);
1805 int ll_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf);
1807 #endif /* LLITE_INTERNAL_H */