[linux-block.git] / fs / orangefs / dir.c

/*
 * Copyright 2017 Omnibond Systems, L.L.C.
 */

#include "protocol.h"
#include "orangefs-kernel.h"
#include "orangefs-bufmap.h"

struct orangefs_dir_part {
	struct orangefs_dir_part *next;
	size_t len;
};

struct orangefs_dir {
	__u64 token;
	struct orangefs_dir_part *part;
	loff_t end;
	int error;
};

#define PART_SHIFT (24)
#define PART_SIZE (1<<24)
#define PART_MASK (~(PART_SIZE - 1))

/*
 * There can be up to 512 directory entries.  Each entry is encoded as
 * follows:
 * 4 bytes: string size (n)
 * n bytes: string
 * 1 byte: trailing zero
 * padding to 8 bytes
 * 16 bytes: khandle
 * padding to 8 bytes
 *
 * The trailer_buf starts with a struct orangefs_readdir_response_s
 * which must be skipped to get to the directory data.
 *
 * The data which is received from the userspace daemon is termed a
 * part and is stored in a linked list in case more than one part is
 * needed for a large directory.
 *
 * The position pointer (ctx->pos) encodes the part and offset on which
 * to begin reading at.  Bits above PART_SHIFT encode the part and bits
 * below PART_SHIFT encode the offset.  Parts are stored in a linked
 * list which grows as data is received from the server.  The overhead
 * associated with managing the list is presumed to be small compared to
 * the overhead of communicating with the server.
 *
 * As data is received from the server, it is placed at the end of the
 * part list.  Data is parsed from the current position as it is needed.
 * When data is determined to be corrupt, it is either because the
 * userspace component has sent back corrupt data or because the file
 * pointer has been moved to an invalid location.  Since the two cannot
 * be differentiated, return EIO.
 *
 * Part zero is synthesized to contains `.' and `..'.  Part one is the
 * first part of the part list.
 */

static int do_readdir(struct orangefs_inode_s *oi,
    struct orangefs_dir *od, struct dentry *dentry,
    struct orangefs_kernel_op_s *op)
{
	struct orangefs_readdir_response_s *resp;
	int bufi, r;

	/*
	 * Despite the badly named field, readdir does not use shared
	 * memory.  However, there are a limited number of readdir
	 * slots, which must be allocated here.  This flag simply tells
	 * the op scheduler to return the op here for retry.
	 */
	op->uses_shared_memory = 1;
	op->upcall.req.readdir.refn = oi->refn;
	op->upcall.req.readdir.token = od->token;
	op->upcall.req.readdir.max_dirent_count =
	    ORANGEFS_MAX_DIRENT_COUNT_READDIR;

again:
	bufi = orangefs_readdir_index_get();
	if (bufi < 0) {
		od->error = bufi;
		return bufi;
	}

	op->upcall.req.readdir.buf_index = bufi;

	r = service_operation(op, "orangefs_readdir",
	    get_interruptible_flag(dentry->d_inode));

	orangefs_readdir_index_put(bufi);

	if (op_state_purged(op)) {
		if (r == -EAGAIN) {
			vfree(op->downcall.trailer_buf);
			goto again;
		} else if (r == -EIO) {
			vfree(op->downcall.trailer_buf);
			od->error = r;
			return r;
		}
	}

	if (r < 0) {
		vfree(op->downcall.trailer_buf);
		od->error = r;
		return r;
	} else if (op->downcall.status) {
		vfree(op->downcall.trailer_buf);
		od->error = op->downcall.status;
		return op->downcall.status;
	}

	/*
	 * The maximum size is size per entry times the 512 entries plus
	 * the header.  This is well under the limit.
	 */
	if (op->downcall.trailer_size > PART_SIZE) {
		vfree(op->downcall.trailer_buf);
		od->error = -EIO;
		return -EIO;
	}

	resp = (struct orangefs_readdir_response_s *)
	    op->downcall.trailer_buf;
	od->token = resp->token;
	return 0;
}

static int parse_readdir(struct orangefs_dir *od,
    struct orangefs_kernel_op_s *op)
{
	struct orangefs_dir_part *part, *new;
	size_t count;

	count = 1;
	part = od->part;
	while (part && part->next) {
		part = part->next;
		count++;
	}

	new = (void *)op->downcall.trailer_buf;
	new->next = NULL;
	new->len = op->downcall.trailer_size -
	    sizeof(struct orangefs_readdir_response_s);
	if (!od->part)
		od->part = new;
	else
		part->next = new;
	count++;
	od->end = count << PART_SHIFT;

	return 0;
}

static int orangefs_dir_more(struct orangefs_inode_s *oi,
    struct orangefs_dir *od, struct dentry *dentry)
{
	struct orangefs_kernel_op_s *op;
	int r;

	op = op_alloc(ORANGEFS_VFS_OP_READDIR);
	if (!op) {
		od->error = -ENOMEM;
		return -ENOMEM;
	}
	r = do_readdir(oi, od, dentry, op);
	if (r) {
		od->error = r;
		goto out;
	}
	r = parse_readdir(od, op);
	if (r) {
		od->error = r;
		goto out;
	}

	od->error = 0;
out:
	op_release(op);
	return od->error;
}

static int fill_from_part(struct orangefs_dir_part *part,
    struct dir_context *ctx)
{
	const int offset = sizeof(struct orangefs_readdir_response_s);
	struct orangefs_khandle *khandle;
	__u32 *len, padlen;
	loff_t i;
	char *s;
	i = ctx->pos & ~PART_MASK;

	/* The file offset from userspace is too large. */
	if (i > part->len)
		return -EIO;

	while (i < part->len) {
		if (part->len < i + sizeof *len)
			return -EIO;
		len = (void *)part + offset + i;
		/*
		 * len is the size of the string itself.  padlen is the
		 * total size of the encoded string.
		 */
		padlen = (sizeof *len + *len + 1) +
		    (8 - (sizeof *len + *len + 1)%8)%8;
		if (part->len < i + padlen + sizeof *khandle)
			return -EIO;
		s = (void *)part + offset + i + sizeof *len;
		if (s[*len] != 0)
			return -EIO;
		khandle = (void *)part + offset + i + padlen;
		if (!dir_emit(ctx, s, *len,
		    orangefs_khandle_to_ino(khandle),
		    DT_UNKNOWN))
			return 0;
		i += padlen + sizeof *khandle;
		i = i + (8 - i%8)%8;
		BUG_ON(i > part->len);
		ctx->pos = (ctx->pos & PART_MASK) | i;
	}
	return 1;
}

static int orangefs_dir_fill(struct orangefs_inode_s *oi,
    struct orangefs_dir *od, struct dentry *dentry,
    struct dir_context *ctx)
{
	struct orangefs_dir_part *part;
	size_t count;

	count = ((ctx->pos & PART_MASK) >> PART_SHIFT) - 1;

	part = od->part;
	while (part->next && count) {
		count--;
		part = part->next;
	}
	/* This means the userspace file offset is invalid. */
	if (count) {
		od->error = -EIO;
		return -EIO;
	}

	while (part && part->len) {
		int r;
		r = fill_from_part(part, ctx);
		if (r < 0) {
			od->error = r;
			return r;
		} else if (r == 0) {
			/* Userspace buffer is full. */
			break;
		} else {
			/*
			 * The part ran out of data.  Move to the next
			 * part. */
			ctx->pos = (ctx->pos & PART_MASK) +
			    (1 << PART_SHIFT);
			part = part->next;
		}
	}
	return 0;
}

static int orangefs_dir_iterate(struct file *file,
    struct dir_context *ctx)
{
	struct orangefs_inode_s *oi;
	struct orangefs_dir *od;
	struct dentry *dentry;
	int r;

	dentry = file->f_path.dentry;
	oi = ORANGEFS_I(dentry->d_inode);
	od = file->private_data;

	if (od->error)
		return od->error;

	if (ctx->pos == 0) {
		if (!dir_emit_dot(file, ctx))
			return 0;
		ctx->pos++;
	}
	if (ctx->pos == 1) {
		if (!dir_emit_dotdot(file, ctx))
			return 0;
		ctx->pos = 1 << PART_SHIFT;
	}

	/*
	 * The seek position is in the first synthesized part but is not
	 * valid.
	 */
	if ((ctx->pos & PART_MASK) == 0)
		return -EIO;

	r = 0;

	/*
	 * Must read more if the user has sought past what has been read
	 * so far.  Stop a user who has sought past the end.
	 */
	while (od->token != ORANGEFS_ITERATE_END &&
	    ctx->pos > od->end) {
		r = orangefs_dir_more(oi, od, dentry);
		if (r)
			return r;
	}
	if (od->token == ORANGEFS_ITERATE_END && ctx->pos > od->end)
		return -EIO;

	/* Then try to fill if there's any left in the buffer. */
	if (ctx->pos < od->end) {
		r = orangefs_dir_fill(oi, od, dentry, ctx);
		if (r)
			return r;
	}

	/* Finally get some more and try to fill. */
	if (od->token != ORANGEFS_ITERATE_END) {
		r = orangefs_dir_more(oi, od, dentry);
		if (r)
			return r;
		r = orangefs_dir_fill(oi, od, dentry, ctx);
	}

	return r;
}

static int orangefs_dir_open(struct inode *inode, struct file *file)
{
	struct orangefs_dir *od;
	file->private_data = kmalloc(sizeof(struct orangefs_dir),
	    GFP_KERNEL);
	if (!file->private_data)
		return -ENOMEM;
	od = file->private_data;
	od->token = ORANGEFS_ITERATE_START;
	od->part = NULL;
	od->end = 1 << PART_SHIFT;
	od->error = 0;
	return 0;
}

static int orangefs_dir_release(struct inode *inode, struct file *file)
{
	struct orangefs_dir *od = file->private_data;
	struct orangefs_dir_part *part = od->part;
	orangefs_flush_inode(inode);
	while (part) {
		struct orangefs_dir_part *next = part->next;
		vfree(part);
		part = next;
	}
	kfree(od);
	return 0;
}

const struct file_operations orangefs_dir_operations = {
	.llseek = default_llseek,
	.read = generic_read_dir,
	.iterate = orangefs_dir_iterate,
	.open = orangefs_dir_open,
	.release = orangefs_dir_release
};
Commit	Line	Data
5db11c21	1	/*
382f4581	2	* Copyright 2017 Omnibond Systems, L.L.C.
5db11c21 MM	3	*/
	4
	5	#include "protocol.h"
575e9461 MM	6	#include "orangefs-kernel.h"
575e9461 MM	7	#include "orangefs-bufmap.h"
5db11c21	8
480e3e53 MB	9	struct orangefs_dir_part {
	10	struct orangefs_dir_part *next;
	11	size_t len;
	12	};
	13
	14	struct orangefs_dir {
	15	__u64 token;
	16	struct orangefs_dir_part *part;
	17	loff_t end;
	18	int error;
	19	};
	20
	21	#define PART_SHIFT (24)
	22	#define PART_SIZE (1<<24)
	23	#define PART_MASK (~(PART_SIZE - 1))
	24
5db11c21	25	/*
382f4581 MB	26	* There can be up to 512 directory entries. Each entry is encoded as
	27	* follows:
	28	* 4 bytes: string size (n)
	29	* n bytes: string
	30	* 1 byte: trailing zero
	31	* padding to 8 bytes
	32	* 16 bytes: khandle
	33	* padding to 8 bytes
382f4581 MB	34	*
	35	* The trailer_buf starts with a struct orangefs_readdir_response_s
	36	* which must be skipped to get to the directory data.
480e3e53 MB	37	*
	38	* The data which is received from the userspace daemon is termed a
	39	* part and is stored in a linked list in case more than one part is
	40	* needed for a large directory.
	41	*
	42	* The position pointer (ctx->pos) encodes the part and offset on which
	43	* to begin reading at. Bits above PART_SHIFT encode the part and bits
	44	* below PART_SHIFT encode the offset. Parts are stored in a linked
	45	* list which grows as data is received from the server. The overhead
	46	* associated with managing the list is presumed to be small compared to
	47	* the overhead of communicating with the server.
	48	*
	49	* As data is received from the server, it is placed at the end of the
	50	* part list. Data is parsed from the current position as it is needed.
	51	* When data is determined to be corrupt, it is either because the
	52	* userspace component has sent back corrupt data or because the file
	53	* pointer has been moved to an invalid location. Since the two cannot
	54	* be differentiated, return EIO.
	55	*
	56	* Part zero is synthesized to contains `.' and `..'. Part one is the
	57	* first part of the part list.
5db11c21	58	*/
5db11c21	59
480e3e53 MB	60	static int do_readdir(struct orangefs_inode_s *oi,
	61	struct orangefs_dir od, struct dentry dentry,
	62	struct orangefs_kernel_op_s *op)
382f4581	63	{
382f4581	64	struct orangefs_readdir_response_s *resp;
382f4581 MB	65	int bufi, r;
382f4581 MB	66
ee3b8d37	67	/*
382f4581 MB	68	* Despite the badly named field, readdir does not use shared
	69	* memory. However, there are a limited number of readdir
	70	* slots, which must be allocated here. This flag simply tells
	71	* the op scheduler to return the op here for retry.
ee3b8d37	72	*/
382f4581 MB	73	op->uses_shared_memory = 1;
	74	op->upcall.req.readdir.refn = oi->refn;
	75	op->upcall.req.readdir.token = od->token;
	76	op->upcall.req.readdir.max_dirent_count =
7d221485	77	ORANGEFS_MAX_DIRENT_COUNT_READDIR;
5db11c21	78
382f4581 MB	79	again:
	80	bufi = orangefs_readdir_index_get();
	81	if (bufi < 0) {
382f4581 MB	82	od->error = bufi;
382f4581 MB	83	return bufi;
5db11c21	84	}
5db11c21	85
382f4581	86	op->upcall.req.readdir.buf_index = bufi;
5db11c21	87
382f4581 MB	88	r = service_operation(op, "orangefs_readdir",
382f4581 MB	89	get_interruptible_flag(dentry->d_inode));
5db11c21	90
382f4581	91	orangefs_readdir_index_put(bufi);
ee3b8d37	92
382f4581 MB	93	if (op_state_purged(op)) {
	94	if (r == -EAGAIN) {
	95	vfree(op->downcall.trailer_buf);
	96	goto again;
	97	} else if (r == -EIO) {
	98	vfree(op->downcall.trailer_buf);
382f4581 MB	99	od->error = r;
	100	return r;
	101	}
5db11c21 MM	102	}
5db11c21 MM	103
382f4581 MB	104	if (r < 0) {
382f4581 MB	105	vfree(op->downcall.trailer_buf);
382f4581 MB	106	od->error = r;
	107	return r;
	108	} else if (op->downcall.status) {
	109	vfree(op->downcall.trailer_buf);
382f4581 MB	110	od->error = op->downcall.status;
	111	return op->downcall.status;
	112	}
	113
480e3e53 MB	114	/*
	115	* The maximum size is size per entry times the 512 entries plus
	116	* the header. This is well under the limit.
	117	*/
	118	if (op->downcall.trailer_size > PART_SIZE) {
	119	vfree(op->downcall.trailer_buf);
	120	od->error = -EIO;
	121	return -EIO;
	122	}
	123
382f4581 MB	124	resp = (struct orangefs_readdir_response_s *)
	125	op->downcall.trailer_buf;
	126	od->token = resp->token;
480e3e53 MB	127	return 0;
480e3e53 MB	128	}
382f4581	129
480e3e53 MB	130	static int parse_readdir(struct orangefs_dir *od,
	131	struct orangefs_kernel_op_s *op)
	132	{
	133	struct orangefs_dir_part part, new;
	134	size_t count;
	135
	136	count = 1;
	137	part = od->part;
	138	while (part && part->next) {
	139	part = part->next;
	140	count++;
382f4581 MB	141	}
382f4581 MB	142
480e3e53 MB	143	new = (void *)op->downcall.trailer_buf;
	144	new->next = NULL;
	145	new->len = op->downcall.trailer_size -
	146	sizeof(struct orangefs_readdir_response_s);
	147	if (!od->part)
	148	od->part = new;
	149	else
	150	part->next = new;
	151	count++;
	152	od->end = count << PART_SHIFT;
	153
382f4581 MB	154	return 0;
382f4581 MB	155	}
9f5e2f7f	156
480e3e53 MB	157	static int orangefs_dir_more(struct orangefs_inode_s *oi,
	158	struct orangefs_dir od, struct dentry dentry)
	159	{
	160	struct orangefs_kernel_op_s *op;
	161	int r;
	162
	163	op = op_alloc(ORANGEFS_VFS_OP_READDIR);
	164	if (!op) {
	165	od->error = -ENOMEM;
	166	return -ENOMEM;
	167	}
	168	r = do_readdir(oi, od, dentry, op);
	169	if (r) {
	170	od->error = r;
	171	goto out;
	172	}
	173	r = parse_readdir(od, op);
	174	if (r) {
	175	od->error = r;
	176	goto out;
	177	}
	178
	179	od->error = 0;
	180	out:
	181	op_release(op);
	182	return od->error;
	183	}
	184
	185	static int fill_from_part(struct orangefs_dir_part *part,
382f4581 MB	186	struct dir_context *ctx)
382f4581 MB	187	{
480e3e53	188	const int offset = sizeof(struct orangefs_readdir_response_s);
382f4581 MB	189	struct orangefs_khandle *khandle;
382f4581 MB	190	__u32 *len, padlen;
72f66b83	191	loff_t i;
382f4581	192	char *s;
480e3e53 MB	193	i = ctx->pos & ~PART_MASK;
	194
	195	/* The file offset from userspace is too large. */
	196	if (i > part->len)
	197	return -EIO;
	198
	199	while (i < part->len) {
	200	if (part->len < i + sizeof *len)
	201	return -EIO;
	202	len = (void *)part + offset + i;
382f4581 MB	203	/*
	204	* len is the size of the string itself. padlen is the
	205	* total size of the encoded string.
	206	*/
	207	padlen = (sizeof len + len + 1) +
480e3e53 MB	208	(8 - (sizeof len + len + 1)%8)%8;
	209	if (part->len < i + padlen + sizeof *khandle)
	210	return -EIO;
	211	s = (void )part + offset + i + sizeof len;
382f4581	212	if (s[*len] != 0)
480e3e53 MB	213	return -EIO;
480e3e53 MB	214	khandle = (void *)part + offset + i + padlen;
382f4581	215	if (!dir_emit(ctx, s, *len,
480e3e53 MB	216	orangefs_khandle_to_ino(khandle),
480e3e53 MB	217	DT_UNKNOWN))
382f4581	218	return 0;
72f66b83 MB	219	i += padlen + sizeof *khandle;
72f66b83 MB	220	i = i + (8 - i%8)%8;
480e3e53 MB	221	BUG_ON(i > part->len);
	222	ctx->pos = (ctx->pos & PART_MASK) \| i;
	223	}
	224	return 1;
	225	}
	226
	227	static int orangefs_dir_fill(struct orangefs_inode_s *oi,
	228	struct orangefs_dir od, struct dentry dentry,
	229	struct dir_context *ctx)
	230	{
	231	struct orangefs_dir_part *part;
	232	size_t count;
	233
	234	count = ((ctx->pos & PART_MASK) >> PART_SHIFT) - 1;
	235
	236	part = od->part;
	237	while (part->next && count) {
	238	count--;
	239	part = part->next;
	240	}
	241	/* This means the userspace file offset is invalid. */
	242	if (count) {
	243	od->error = -EIO;
	244	return -EIO;
	245	}
	246
	247	while (part && part->len) {
	248	int r;
	249	r = fill_from_part(part, ctx);
	250	if (r < 0) {
	251	od->error = r;
	252	return r;
	253	} else if (r == 0) {
	254	/* Userspace buffer is full. */
	255	break;
	256	} else {
	257	/*
	258	* The part ran out of data. Move to the next
	259	* part. */
	260	ctx->pos = (ctx->pos & PART_MASK) +
	261	(1 << PART_SHIFT);
	262	part = part->next;
	263	}
382f4581	264	}
382f4581	265	return 0;
382f4581	266	}
5db11c21	267
382f4581 MB	268	static int orangefs_dir_iterate(struct file *file,
	269	struct dir_context *ctx)
	270	{
	271	struct orangefs_inode_s *oi;
	272	struct orangefs_dir *od;
	273	struct dentry *dentry;
	274	int r;
5db11c21	275
382f4581 MB	276	dentry = file->f_path.dentry;
	277	oi = ORANGEFS_I(dentry->d_inode);
	278	od = file->private_data;
5db11c21	279
382f4581 MB	280	if (od->error)
382f4581 MB	281	return od->error;
5db11c21	282
382f4581 MB	283	if (ctx->pos == 0) {
	284	if (!dir_emit_dot(file, ctx))
	285	return 0;
	286	ctx->pos++;
5db11c21	287	}
382f4581 MB	288	if (ctx->pos == 1) {
	289	if (!dir_emit_dotdot(file, ctx))
	290	return 0;
480e3e53	291	ctx->pos = 1 << PART_SHIFT;
5db11c21 MM	292	}
5db11c21 MM	293
480e3e53 MB	294	/*
	295	* The seek position is in the first synthesized part but is not
	296	* valid.
	297	*/
	298	if ((ctx->pos & PART_MASK) == 0)
	299	return -EIO;
	300
382f4581 MB	301	r = 0;
382f4581 MB	302
72f66b83 MB	303	/*
	304	* Must read more if the user has sought past what has been read
	305	* so far. Stop a user who has sought past the end.
	306	*/
7b796ae3	307	while (od->token != ORANGEFS_ITERATE_END &&
480e3e53	308	ctx->pos > od->end) {
72f66b83 MB	309	r = orangefs_dir_more(oi, od, dentry);
	310	if (r)
	311	return r;
	312	}
7b796ae3	313	if (od->token == ORANGEFS_ITERATE_END && ctx->pos > od->end)
72f66b83	314	return -EIO;
72f66b83 MB	315
72f66b83 MB	316	/* Then try to fill if there's any left in the buffer. */
480e3e53	317	if (ctx->pos < od->end) {
382f4581 MB	318	r = orangefs_dir_fill(oi, od, dentry, ctx);
	319	if (r)
	320	return r;
5db11c21 MM	321	}
5db11c21 MM	322
72f66b83	323	/* Finally get some more and try to fill. */
7b796ae3	324	if (od->token != ORANGEFS_ITERATE_END) {
382f4581 MB	325	r = orangefs_dir_more(oi, od, dentry);
	326	if (r)
	327	return r;
	328	r = orangefs_dir_fill(oi, od, dentry, ctx);
5db11c21 MM	329	}
5db11c21 MM	330
382f4581	331	return r;
5db11c21 MM	332	}
5db11c21 MM	333
8bb8aefd	334	static int orangefs_dir_open(struct inode inode, struct file file)
5db11c21	335	{
382f4581 MB	336	struct orangefs_dir *od;
	337	file->private_data = kmalloc(sizeof(struct orangefs_dir),
	338	GFP_KERNEL);
5db11c21 MM	339	if (!file->private_data)
5db11c21 MM	340	return -ENOMEM;
382f4581	341	od = file->private_data;
7b796ae3	342	od->token = ORANGEFS_ITERATE_START;
480e3e53 MB	343	od->part = NULL;
480e3e53 MB	344	od->end = 1 << PART_SHIFT;
382f4581	345	od->error = 0;
5db11c21 MM	346	return 0;
	347	}
	348
8bb8aefd	349	static int orangefs_dir_release(struct inode inode, struct file file)
5db11c21	350	{
382f4581	351	struct orangefs_dir *od = file->private_data;
480e3e53	352	struct orangefs_dir_part *part = od->part;
8bb8aefd	353	orangefs_flush_inode(inode);
480e3e53 MB	354	while (part) {
	355	struct orangefs_dir_part *next = part->next;
	356	vfree(part);
	357	part = next;
	358	}
382f4581	359	kfree(od);
5db11c21 MM	360	return 0;
	361	}
	362
8bb8aefd	363	const struct file_operations orangefs_dir_operations = {
72f66b83	364	.llseek = default_llseek,
5db11c21	365	.read = generic_read_dir,
382f4581	366	.iterate = orangefs_dir_iterate,
8bb8aefd	367	.open = orangefs_dir_open,
382f4581	368	.release = orangefs_dir_release
5db11c21	369	};