2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
7 * Copyrights for code taken from ext2:
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
13 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * This file is part of exofs.
18 * exofs is free software; you can redistribute it and/or modify
19 * it under the terms of the GNU General Public License as published by
20 * the Free Software Foundation. Since it is based on ext2, and the only
21 * valid version of GPL for the Linux kernel is version 2, the only valid
22 * version of GPL for exofs is version 2.
24 * exofs is distributed in the hope that it will be useful,
25 * but WITHOUT ANY WARRANTY; without even the implied warranty of
26 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
27 * GNU General Public License for more details.
29 * You should have received a copy of the GNU General Public License
30 * along with exofs; if not, write to the Free Software
31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
37 #include <linux/time.h>
38 #include <linux/backing-dev.h>
39 #include <linux/pnfs_osd_xdr.h>
42 #define EXOFS_ERR(fmt, a...) printk(KERN_ERR "exofs: " fmt, ##a)
44 #ifdef CONFIG_EXOFS_DEBUG
45 #define EXOFS_DBGMSG(fmt, a...) \
46 printk(KERN_NOTICE "exofs @%s:%d: " fmt, __func__, __LINE__, ##a)
48 #define EXOFS_DBGMSG(fmt, a...) \
49 do { if (0) printk(fmt, ##a); } while (0)
52 /* u64 has problems with printk this will cast it to unsigned long long */
53 #define _LLU(x) (unsigned long long)(x)
56 struct osd_obj_id obj;
61 /* Our way of looking at the data_map */
70 struct exofs_components {
71 unsigned numdevs; /* Num of devices in array */
72 /* If @single_comp == EC_SINGLE_COMP, @comps points to a single
73 * component. else there are @numdevs components
76 EC_SINGLE_COMP = 0, EC_MULTPLE_COMPS = 0xffffffff
78 struct exofs_comp *comps;
79 struct osd_dev **ods; /* osd_dev array */
83 * our extension to the in-memory superblock
85 struct exofs_sb_info {
86 struct backing_dev_info bdi; /* register our bdi with VFS */
87 struct exofs_sb_stats s_ess; /* Written often, pre-allocate*/
88 int s_timeout; /* timeout for OSD operations */
89 uint64_t s_nextid; /* highest object ID used */
90 uint32_t s_numfiles; /* number of files on fs */
91 spinlock_t s_next_gen_lock; /* spinlock for gen # update */
92 u32 s_next_generation; /* next gen # to use */
93 atomic_t s_curr_pending; /* number of pending commands */
95 struct pnfs_osd_data_map data_map; /* Default raid to use
98 struct exofs_layout layout; /* Default files layout */
99 struct exofs_comp one_comp; /* id & cred of partition id=0*/
100 struct exofs_components comps; /* comps for the partition */
101 struct osd_dev *_min_one_dev[1]; /* Place holder for one dev */
105 * our extension to the in-memory inode
107 struct exofs_i_info {
108 struct inode vfs_inode; /* normal in-memory inode */
109 wait_queue_head_t i_wq; /* wait queue for inode */
110 unsigned long i_flags; /* various atomic flags */
111 uint32_t i_data[EXOFS_IDATA];/*short symlink names and device #s*/
112 uint32_t i_dir_start_lookup; /* which page to start lookup */
113 uint64_t i_commit_size; /* the object's written length */
114 struct exofs_comp one_comp; /* same component for all devices */
115 struct exofs_components comps; /* inode view of the device table */
118 static inline osd_id exofs_oi_objno(struct exofs_i_info *oi)
120 return oi->vfs_inode.i_ino + EXOFS_OBJ_OFF;
123 struct exofs_io_state;
124 typedef void (*exofs_io_done_fn)(struct exofs_io_state *ios, void *private);
126 struct exofs_io_state {
130 exofs_io_done_fn done;
132 struct exofs_layout *layout;
133 struct exofs_components *comps;
135 /* Global read/write IO*/
137 unsigned long length;
143 unsigned pages_consumed;
146 unsigned in_attr_len;
147 struct osd_attr *in_attr;
148 unsigned out_attr_len;
149 struct osd_attr *out_attr;
153 /* Variable array of size numdevs */
155 struct exofs_per_dev_state {
156 struct osd_request *or;
164 static inline unsigned exofs_io_state_size(unsigned numdevs)
166 return sizeof(struct exofs_io_state) +
167 sizeof(struct exofs_per_dev_state) * numdevs;
173 #define OBJ_2BCREATED 0 /* object will be created soon*/
174 #define OBJ_CREATED 1 /* object has been created on the osd*/
176 static inline int obj_2bcreated(struct exofs_i_info *oi)
178 return test_bit(OBJ_2BCREATED, &oi->i_flags);
181 static inline void set_obj_2bcreated(struct exofs_i_info *oi)
183 set_bit(OBJ_2BCREATED, &oi->i_flags);
186 static inline int obj_created(struct exofs_i_info *oi)
188 return test_bit(OBJ_CREATED, &oi->i_flags);
191 static inline void set_obj_created(struct exofs_i_info *oi)
193 set_bit(OBJ_CREATED, &oi->i_flags);
196 int __exofs_wait_obj_created(struct exofs_i_info *oi);
197 static inline int wait_obj_created(struct exofs_i_info *oi)
199 if (likely(obj_created(oi)))
202 return __exofs_wait_obj_created(oi);
206 * get to our inode from the vfs inode
208 static inline struct exofs_i_info *exofs_i(struct inode *inode)
210 return container_of(inode, struct exofs_i_info, vfs_inode);
214 * Maximum count of links to a file
216 #define EXOFS_LINK_MAX 32000
218 /*************************
219 * function declarations *
220 *************************/
223 int exofs_get_rw_state(struct exofs_layout *layout,
224 struct exofs_components *comps,
225 bool is_reading, u64 offset, u64 length,
226 struct exofs_io_state **ios);
227 int exofs_get_io_state(struct exofs_layout *layout,
228 struct exofs_components *comps,
229 struct exofs_io_state **ios);
230 void exofs_put_io_state(struct exofs_io_state *ios);
232 int exofs_check_io(struct exofs_io_state *ios, u64 *resid);
234 int exofs_sbi_create(struct exofs_io_state *ios);
235 int exofs_sbi_remove(struct exofs_io_state *ios);
236 int exofs_sbi_write(struct exofs_io_state *ios);
237 int exofs_sbi_read(struct exofs_io_state *ios);
238 int exofs_truncate(struct exofs_layout *layout, struct exofs_components *comps,
241 int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr);
242 extern const struct osd_attr g_attr_logical_length;
245 unsigned exofs_max_io_pages(struct exofs_layout *layout,
246 unsigned expected_pages);
247 int exofs_setattr(struct dentry *, struct iattr *);
248 int exofs_write_begin(struct file *file, struct address_space *mapping,
249 loff_t pos, unsigned len, unsigned flags,
250 struct page **pagep, void **fsdata);
251 extern struct inode *exofs_iget(struct super_block *, unsigned long);
252 struct inode *exofs_new_inode(struct inode *, int);
253 extern int exofs_write_inode(struct inode *, struct writeback_control *wbc);
254 extern void exofs_evict_inode(struct inode *);
257 int exofs_add_link(struct dentry *, struct inode *);
258 ino_t exofs_inode_by_name(struct inode *, struct dentry *);
259 int exofs_delete_entry(struct exofs_dir_entry *, struct page *);
260 int exofs_make_empty(struct inode *, struct inode *);
261 struct exofs_dir_entry *exofs_find_entry(struct inode *, struct dentry *,
263 int exofs_empty_dir(struct inode *);
264 struct exofs_dir_entry *exofs_dotdot(struct inode *, struct page **);
265 ino_t exofs_parent_ino(struct dentry *child);
266 int exofs_set_link(struct inode *, struct exofs_dir_entry *, struct page *,
270 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN],
271 const struct osd_obj_id *obj);
272 int exofs_sbi_write_stats(struct exofs_sb_info *sbi);
274 /*********************
275 * operation vectors *
276 *********************/
278 extern const struct file_operations exofs_dir_operations;
281 extern const struct inode_operations exofs_file_inode_operations;
282 extern const struct file_operations exofs_file_operations;
285 extern const struct address_space_operations exofs_aops;
288 extern const struct inode_operations exofs_dir_inode_operations;
289 extern const struct inode_operations exofs_special_inode_operations;
292 extern const struct inode_operations exofs_symlink_inode_operations;
293 extern const struct inode_operations exofs_fast_symlink_inode_operations;
295 /* exofs_init_comps will initialize an exofs_components device array
296 * pointing to a single exofs_comp struct, and a round-robin view
297 * of the device table.
298 * The first device of each inode is the [inode->ino % num_devices]
299 * and the rest of the devices sequentially following where the
300 * first device is after the last device.
301 * It is assumed that the global device array at @sbi is twice
302 * bigger and that the device table repeats twice.
303 * See: exofs_read_lookup_dev_table()
305 static inline void exofs_init_comps(struct exofs_components *comps,
306 struct exofs_comp *one_comp,
307 struct exofs_sb_info *sbi, osd_id oid)
309 unsigned dev_mod = (unsigned)oid, first_dev;
311 one_comp->obj.partition = sbi->one_comp.obj.partition;
312 one_comp->obj.id = oid;
313 exofs_make_credential(one_comp->cred, &one_comp->obj);
315 comps->numdevs = sbi->comps.numdevs;
316 comps->single_comp = EC_SINGLE_COMP;
317 comps->comps = one_comp;
319 /* Round robin device view of the table */
320 first_dev = (dev_mod * sbi->layout.mirrors_p1) % sbi->comps.numdevs;
321 comps->ods = sbi->comps.ods + first_dev;